Embed Size (px)
Citation preview
Book of Abstracts
Innovation and Society
Statistical methods for service evaluation
Firenze, 30 May ‐ 1 June 2011
IES 2011
Innovazione e Società Metodi statistici per la valutazione dei servizi
Innovation and society Statistical methods for service evaluation
30 May – 1 June 2011 Firenze (Italy)
Book of Abstracts
ii
IES 2011 COMMITTEES
Scientific Program Committee
Pietro Amenta, Università di Benevento Matilde Bini, Università Europea di Roma Mario Bolzan, Università di Padova Giuseppe Boari, Università Cattolica di Milano Eugenio Brentari, Università di Brescia Gabriele Cantaluppi, Università Cattolica di Milano Maurizio Carpita, Università di Brescia Luigi D'Ambra, Università di Napoli "Federico II" Bruno Chiandotto, Università di Firenze Enrico Ciavolino, Università del Salento Giorgio Cusatelli, Università di Bari Francesco Domenico d'Ovidio, Università di Bari Amedeo De Luca, Università Cattolica di Milano Tonio Di Battista, Università di Chieti-Pescara Michele Gallo, Università di Napoli "L'Orientale" Leonardo Grilli, Università di Firenze Donata Marasini, Università di Milano-Bicocca Stefania Mignani, Università di Bologna Giovanna Nicolini, Università di Milano Carla Rampichini, Università di Firenze Sergio Scippacercola, Università di Napoli "Federico II" Biagio Simonetti, Università del Sannio
Local Organizing Committee
Bruno Bertaccini, Università di Firenze Matilde Bini, Università Europea di Roma Anna Gottard, Università di Firenze Leonardo Grilli, Università di Firenze Carla Rampichini, Università di Firenze Roberta Varriale, Università di Firenze
iii
IES 2011 Conference Agenda
Monday, May 30, 2011
09:00 – 10:00 Registration
10:30 – 11:00 Opening ceremony
11:00 – 12:30 Specialised A Business and Bank
Contributed 1 Education I
Contributed 2 Economics and labour market
12:30 – 13:30 Lunch
13:30 – 15:00 Round table 1 Research evaluation
15:00 – 16:00 Plenary session 1 Performance measurements for healthcare services
16:00 – 16:30 Coffee break
16:30 – 18:00 Specialised B Ordinal data models
Contributed 3 Latent variable models
Contributed 4 Education II
19:30 – 21:30 Welcome aperitif – Biblioteca Le Oblate
Tuesday, May 31, 2011
09:00 – 10:30 Specialised C Public transports evaluation
Specialised D Rasch Analysis
Contributed 5 Health and social services
10:30 – 11:00 Coffee break
11:00 – 12:30 Specialised E Latent variable models
Specialised F Evaluation of local authorities performance
Contributed 6 Quality and risk
12:30 – 14:00 Lunch
14:00 – 15:30
Specialised G Teaching evaluation in the Italian University systems
Contributed 7 Transports
Contributed 8 Latent variable models and customer satisfaction
15:30 – 16:00 Coffee break
16:00 – 17:30 Round table 2 Public administration
17:30 – 18:00 JAQM 2010 Award
18:00 – 19:00 Meeting SVQS
20:00 – 23:00 Social Dinner – Ristorante Pennello
iv
Wednesday, June 1, 2011
09:00 – 10:30 Specialised H Law system
Contributed 9 Education III
Contributed 10 Socio-economic topics
10:30 – 11:00 Coffee break
11:00 – 12:00 Plenary session 2 The evaluation of the educational system
12:00 – 13:30 Specialised I Evaluation of cultural services
Specialised L Evaluation of knowledge transmission via web
Contributed 11 Ordinal data and latent variables
Structure of the sessions Plenary sessions 1h (40’ speaker + 20’ discussant) Round tables 1h e 30’ Specialised sessions with 3 speakers: 20’ each presentation + 15’ discussant
+ 15’ questions Specialised sessions with 4 speakers: 20’ each presentation + 10’ questions Contributed sessions with 4 speakers: 20’ each presentation + 10’ questions Contributed sessions with 5 speakers: 16’ each presentation + 10’ questions
Welcome aperitif (Monday, 30 May) will be held at: Biblioteca Le Oblate Via dell'Oriuolo, 26, Firenze Social dinner (Tuesday, 31 May) will be held at: Ristorante Pennello, Via Dante Alighieri, 4, Firenze
v
IES 2011 Conference venue
Fig. 1 IES 2011 Conference venue
vi
Fig. 2 Map of the “Polo delle Scienze Sociali”, Via delle Pandette, 9 – FIRENZE
Abstracts
5
Plenary Sessions
Performance measurements for healthcare services ............................................................ 12 Giorgio Vittadini, Paolo Berta
The evaluation of the educational system ............................................................................. 13 Giacomo Maria Elias
Specialised Sessions Specialised session A ‐ Business and Bank
Banking services evaluation: a dynamic analysis ................................................................ 17 Michela Lacangellera, Caterina Liberati, Paolo Mariani
Ordinal logistic regression response functions with main and interaction effects in the conjoint analysis................................................................................................................... 18 Amedeo De Luca, Sara Ciapparelli
Measuring media reputation for private and public institutions .......................................... 19 Paola Cerchiello
The confidence ellipses in decomposition Non-Symmetrical Correspondence Analysis for the evaluation of the innovative performance of the Manufacturing Enterprises in Campania ............................................................................................................................. 20 Antonello D’Ambra, Anna Crisci
Specialised session B ‐ Ordinal data models
A review of multilevel models for ordinal data .................................................................... 22 Leonardo Grilli, Carla Rampichini
A categorical data model to assess critical points ............................................................... 23 Luisa Stracqualursi, Stefania Mignani, Paola Monari
Modelling multivariate ordinal data: some experience with CUB models .......................... 24 Marcella Corduas
Specialised session C ‐ Public transports evaluation
Preliminary studies to define critical threshold for customer satisfaction and loyalty ........ 26 Giovanni Monaco, Valeria Scaramuzzi, Luce Allorio
Mixed logit models using Gaussian mixtures with covariate-dependent weights ................ 27 Luisa Scaccia
An evaluation of Customer Satisfaction in public train transport by complex principal component analysis .............................................................................................................. 28 Pasquale Sarnacchiaro, Roberta Di Gennaro
Optimal Sample strategies in public transports assessment ............................................... 29 Tonio Di Battista, Riccardo Di Nisio
6
Specialised session D ‐ Rasch Analysis
Motivation to achieve academically: Rasch analysis and first results regarding correlation with students outcomes at the university of Udine ............................................................... 31 Enrico Gori, Laura Pagani, Michela Battauz
Effects of DIF items on Rasch measures .............................................................................. 32 Silvia Golia
Impact of educational test features on item difficulties by the Linear Logistic Test Model . 33 Daniela Marella, Carlo Di Chiacchio, Giuseppe Bove
Specialised session E ‐ Latent variable models
A Bootstrap procedure to test the cograduation between two ordinal scales: a proposal of two new indices. ................................................................................................................... 35 Andrea Bonanomi
Combining PLS and GME to estimate Structural Equation Models .................................... 36 Giuseppe Boari, Gabriele Cantaluppi, Enrico Ciavolino
SImultaneous Factor Analysis Across Several Populations (SIFASP):identification criteria for latent structures .............................................................................................................. 37 Hans Schadee
Specialised session F ‐ Evaluation of local authorities performance
Methodology and Tools for Local Government Services Monitoring: the IQuEL project experience ............................................................................................................................ 39 Gianluca Vannuccini, Ciro Annicchiarico
The customer auditing system in Milan, a certificated managing procedure ISO 9001 ...... 40 Maria Morena Montagna
ELISA PROGRAM - LOCAL INNOVATION SYSTEM measuring the quality of the services of local Public Administrations ............................................................................................ 41 Angelina Tritto
Innovation and Quality in Local Authorities: A combination to be pursued ........................ 42 Flora Raffa
Specialised session G ‐ Teaching evaluation in the Italian university system
Descriptive analysis of student ratings ................................................................................ 44 Piero Quatto
Students’ Evaluation of Teaching. The need for adjusted measures in comparative evaluations ........................................................................................................................... 45 Mariano Porcu, Isabella Sulis
The evaluation of educational university processes and teaching activities in Italy: aims, surveys and open problems .................................................................................................. 46 Luigi Biggeri
7
Specialised session H ‐ Law system
The Italian judicial offices productivity in 130 years of cognition civil procedures ............ 48 Carlo Cusatelli
Evaluating the Administrative Efficiency of Justice Courts in Italy ..................................... 49 L. Antonucci, C. Crocetta, F. d’Ovidio, M. Lorizio
The participatory evaluation in the analysis of the judicial system: the georeferencing of the opinions ................................................................................................................................ 50 Simone Di Zio, Antonio Pacinelli
Specialised session I ‐ Evaluation of cultural services
A Structural Equation Model Proposal for evaluating Visitor Satisfaction at an Exhibition.............................................................................................................................................. 52 Gabriele Cantaluppi, Barbara Bianchi, Domenico Piraina, Francesca La Placa
Between accessibility and marketing: online ticketing for entertainment events ................. 53 Fabrizio Chirico, Luca Bottini
An Exchange market model to improve the impact of Lyric Opera theatres in the social and working life .......................................................................................................................... 54 Paola Fandella, Enrico Girardi
Specialised session L - Evaluation of knowledge transmission via web
Java visual tools for economic, social and epidemiologic statistical model simulation ..... 56 Stefano Rosignoli
Evaluating peculiar lexicon for medical record sections identification ............................... 57 Flora Amato, Antonino Mazzeo, Sara Romano, Sergio Scippacercola
The assessment of knowledge transfer via WEB ................................................................. 58 Antonino Mazzeo, Flora Amato, Sergio Scippacercola
Contributed Sessions Contributed session 1 ‐ Education I
A new proposal to assess evaluation models ........................................................................ 61 Paolo Giudici, Emanuela Raffinetti
Impact evaluation of University grants ................................................................................ 62 Maria Luisa Maitino, Nicola Sciclone
Differential variability of test scores among schools: a multilevel analysis of the 5th grade Invalsi test using heteroschedastic random effects .............................................................. 63 Claudia Sani, Leonardo Grilli
University careers evolution. A multistate modeling for a perspective study of the Italian situation ................................................................................................................................ 64 Matilde Bini, Bruno Monastero, Margherita Velucchi
8
Contributed session 2 ‐ Economic and Labour market
A Statistical Analysis of Accidents at Work and Sectoral Performance in the European Economy ............................................................................................................................... 66 Luigi D'Ambra, Antonio Frenda
A composite index used for measuring intellectual capital of SMEs from Romania ............ 67 Bogdan Ileanu, Alexandru Isaic-Maniu
The analysis of informal economy and its implication on the services sector. The case of Romania as a country in transition ...................................................................................... 68 Tudorel Andrei, Marius Profiroiu, Andreea Iluzia Iacob
Combining Statistical and Algorithmic Models for Latent Variables Analysis: A Look at the Fairness of Work .................................................................................................................. 69 Maurizio Carpita, Marika Vezzoli
Contributed session 3 - Latent variable models
A Latent Class Approach for Estimating Labour Market Mobility in the Presence of Multiple Indicators and Retrospective Interrogation ........................................................... 71 Francesca Bassi, Marcel Croon, Arianna Pittarello
Latent growth models with multiple indicators: a longitudinal analysis of student ratings. 72 Leonardo Grilli, Roberta Varriale
SEM and IRM procedures to assess the relationship between latent traits .......................... 73 Anna Simonetto
Latent Markov models from a potential outcome prospective for causal inference in dynamic settings ................................................................................................................. 734 Francesco Bartolucci, Fulvia Pennoni, Giorgio Vittadini
Contributed session 4 - Education II
The Course organizational structure as a determinant of academic success. Some evidences from Padova University ....................................................................................................... 76 Renata Clerici, Anna Giraldo, Elisa Visentin
Measures for Ph.D. Evaluation: the Recruitment Phase ...................................................... 77 A. D’Agostino, G. Ghellini, L. Neri
Student Satisfaction Indicators: a Delphi approach ............................................................ 78 Laura Antonucci, Corrado Crocetta, Patrizia Soleti, Ernesto Toma
A two level structural equation model for evaluating the external effectiveness of PhD ..... 79 Lucio Masserini Contributed session 5 ‐ Health and social services
Text-mining: an application for classifying pathology reports ............................................ 81 Claudio Sacchettini, Bruno Bertaccini
NANOVA as a new tool for the evaluation of health services .............................................. 82 Gabriella Milone, Luigi D’Ambra
9
Domiciliary assistance satisfaction among aged and disabled beneficiaries: a Rasch analysis................................................................................................................................. 83 Annalina Sarra
Performance assessment in healthcare as a management tool ............................................ 84 Sabina Nuti
Contributed session 6 ‐ Quality and risk
Statistical Analysis of the Perceived Quality and Customer Satisfaction of a ski school: the Sesto Survey 2010 ................................................................................................................ 86 Stefano Bonnini, Luigi Salmaso, Francesca Solmi
Risk Profile using Rasch Analysis ........................................................................................ 87 Valeria Caviezel, Sergio Ortobelli, Lucio Bertoli Barsotti
Safety at work and industrial accidents in Europe: an efficiency analysis .......................... 88 Eugenia Nissi, Agnese Rapposelli
The front-office services provided by some municipalities of the area of Florence: An evaluation of the quality from the supply side ...................................................................... 89 Daniele Vignoli, Bruno Bertaccini, Ciro Annicchiarico
Contributed session 7 - Transports
Consumers’ satisfaction with railway transport: a Bayesian Network approach. ............... 91 Giovanni Perucca, Silvia Salini
An overall passenger satisfaction measure through a Structural equation model with high-order latent variables ........................................................................................................... 92 Enrico Ciavolino, Mariangela Nitti
Passenger satisfaction: a multi-group analysis ................................................................... 93 Laura Antonucci, Corrado Crocetta, Francesco D. d’Ovidio, Ernesto Toma
Using the Disco index for the determination of the Passenger Satisfaction ........................ 94 Biagio Simonetti, Antonio Lucadamo
Contributed session 8 - Latent variable models and customer satisfaction
PLS models: importance of the stability during the time, how was insured in the case history of Hera. .................................................................................................................... 96 Giovanni Monaco, Fabio Marotta, Roberto Riccardi , Stefano Baldassini
The customer satisfaction in INAIL ...................................................................................... 97 Rosa Maria Lacquaniti, Maria Cristina Paoletti
Measuring the administrative compliance burden on enterprises. The Italian experience .. 98 Antonio Pavone, Paola Pianura
Common Components and Specific Weights Analysis of cross-product tables for the Full Multi Modules Customer Satisfaction Evaluation ................................................................ 99 Pietro Amenta
10
Contributed session 9 ‐ Education III
Evaluation of performance at university with Rasch Analysis ........................................... 101 Piergiorgio Mossi, Sara Calogiuri, Paola Tondo
Ability effect influence on the Italian graduates’ labour income: ATT estimation and sensitivity analysis .............................................................................................................. 102 Antonino Di Pino, Patrizia Pulejo
The impact of the "3+2" reform on degrees ....................................................................... 103 Maria Luisa Maitino, Giulia Peruzzi
The Estimation of Dimension and Factors of School Abandon in the Romanian Development Regions ......................................................................................................... 104 Tudorel Andrei, Alina Profiroiu, Andreea Iluzia Iacob, Bogdan Ileanu
Contributed session 10 ‐ Socio-economic topics
The system of “performance evaluation” of the Ministry of Labour and Social Policy: an analysis of the production function of output inspection .................................................... 106 Anna Maria Frasca, Piergiorgio Mossi, Enrico Ciavolino
Assessment of relevance and efficiency of scientific research in Universities departments............................................................................................................................................ 107 Laura Antonucci, Francesco Domenico d’Ovidio
A Structural Equation Model to analyze the Household Budget: a case study .................. 108 Giovanni Di Trapani, Pasquale Marrone, Pasquale Sarnacchiaro, Ilaria Sasso
Research Projects Evaluation. A Small Case Study on “Ideas” Competition from National Research Program of Romania .......................................................................................... 109 Claudiu Herteliu, Tudorel Andrei
Particularities and typologies of the development level of Romanian’s villages from the ethnic affiliation view point ................................................................................................ 110 Claudiu Herteliu, Bogdan Ileanu
Contributed session 11 ‐ Ordinal data and latent variables
Performance Assessment of Social Agents. A goal-planned approach .............................. 112 Giulio D’Epifanio
The EN index for the use of the normal, exponential, beta or gamma distribution in the indirect quantification ........................................................................................................ 113 Antonio Lucadamo, Giovanni Portoso
A PLUM model for the evaluation of university teaching .................................................. 114 Angela Alibrandi, Massimiliano Giacalone
Complementary use of different methods to evaluate Customer Satisfaction of Services with subgroups data ................................................................................................................... 115 M. Chiara Zanarotti, Laura Pagani
AUTHOR INDEX ............................................................................................................... 116
11
Plenary Sessions
12
Performance measurements for healthcare services
Giorgio Vittadini1, Paolo Berta 2 1 Università degli Studi di Milano- Bicocca, [email protected] 2 CRISP, Università degli Studi di Milano- Bicocca, [email protected]
Performance measurements for healthcare services, the i mprovement of healt hcare quality are becoming more and more important.
The m ost traditional m ethods for quality evaluation are the e x-ante methods, as accreditation of excellence methods in the fields of health. However, in order to obtain a nd to eval uate t he heal th st ructures we al so need be nchmarking results d efined as e x post evaluation based on quantitative indicators of outputs and outcomes.
Efficiency consists of t he comparison between production input and business output in terms of both quality (days, cases etc.); Customer satisfaction, used to measure consumer’s satisfaction regarding as pects rel ated t o a product or a serv ice; relative effectiv eness is defined as th e ability to p rovide treatment to patients, improving healthcare ou tcome and improving the ability to modify the patient’s state (Donabedian, 1988).
In particular relative effectiveness needs to be adjusted for patient-specific and hospital-specific variables. For th is goal g iven also th e m ultilevel n ature of data, in th e n ineties, numerous au thors pro posed u tilizing th e Mu ltilevel Mo del (Go ldstein & Sp iegelhalter, 1996; Marshall & Spiegelhalter, 2001, Vittadini, Minotti, 2005).
However, t he pre vious proposals co ncern sm all sam ples of patients affect ed by particular di seases and t herefore are n o s ufficient fo r e valuating be nchmark hos pitals. Moreover also the use of the death rate as health outcome has been discussed (Goldstein, Spiegelhalter, 1996). Finally, the use of risk adjusted comparisons for benchmarking health structures has been strongly criticized (Lilford, Spiegelhater et al., 2004).
In order to overcome these problems: 1. In stead of u tilizing sm all sa mples, we base ou r effectiv eness an alysis o n larg e
administrative data sets (Health discharge cards). 2. We build the intrahospital mortality ratio and we also use context outcomes obtained
from administrative data giving more complete information about effectiveness. 3. We perform our analysis for the single DRG getting less generic and confused case-
mix corrections and more robust risk adjusted comparisons. 4. Instead of ranking health structures we classify them in few e ffectiveness subgroups
(inferior, superior and equal to regional average). 5. We present all the ex ante and ex post indicators in a different, simple and synthetic
way (radar scheme). All the indicators are tested on Lombardy hospitals.
References
Donabedian A. (1988). The Quality of Care. How can it be assessed?. JAMA, 260(12), 1743-1748. Goldstein H., Spiegelhalter D.J. (1996). League Table and their Limitations: Statistical I ssues in co mparisons of
Institutional Performances (with discussion). J. Roy. Statistical Society, 159(5), 385-443. Lilford R. , Mohammed M.A., Spiegelhalter D.J., Thomson R. (2004). Use and M isuse of Process and Outcome
Data. I n: Managing Per formance of Acute M edical Ca re: Avoiding I nstitutional Stig ma, The L ancet, 364, 1147-1154.
Marshall E.C., Spiegelhalter D.J. (2001). Institutional Performance, Goldstein H., Leyland A.H. (eds.), Multilevel Modelling of Health Statistics, Wiley, Chichester, 127-142.
Vittadini, G., Min otti, S. (2005). A methodology for measuring the relative effectivene ss of healthcare services. IMA, http://imaman.oxfordjournals.org/papbyrecent.dtl.
13
The evaluation of the educational system
Giacomo Maria Elias1 1 Dr. Engineer, PHD, Former Full Professor of Applied Physics at the Milan University, Former President of INVALSI and member of CNVSU
The exposition of this issue as far as I’m concerned requires two premises that derive from more than a decade of studies and experiences on the field.
First the educational system should be understood as a continuum throughout the life: it helps, tog ether with th e experien ces, to th e fo rmation of Hu man Cap ital (HC ), i.e. t he "culture" of each.
The second is that the concept of evaluation is not absolute, but "measure" on a case-by-case basis the achievement of one or more objectives (measured with suitable indicators). These must be established by means of a "policy" (for example education policy) that must be declared and adopted by those who has the responsibility of a given service and, for this reason, it promotes the evaluation of the results achieved.
I am convinced that, as all social systems, educational one possesses the peculiarities of a C AS (C omplex Adaptive Sy stem). C onsequently, i t cann ot be co nsidered a sy stem "closed" (linear approach), bu t reciprocally interacts with th e env ironment, understood in the broadest sense (non-linear approach, network).
This leads us to say that the policy adopted for the educational system, on the one hand, is indissolubly linked to the one chosen for the development of t he Country, on the other hand it must take account of a high num ber of stres ses that com e from outside. For the latter reason, it must be sufficiently flexible in order to respond to the continuing challenges from outside and at the same time be capable to reconcile the right of everyone to have an adequate education with the educational and cultural autonomy of educational institutions that operate in the territory and from this are influenced.
In order that the effects on individuals are to be ef fective (it is necessary to distinguish between efficiency and effic acy), the educational system must not present discontinuities, although in creases in th e UC th at it d etermines, always p ositive, can b e different i n different p eriods. For ex ample, stu dies have b een developed at INVALSI, primary and secondary education (by L. D'Ambra e A. Paletta), and at CNVSU, university (for example, by G. Vittadini), and they are very difficult to be connected between them.
It is ev ident that th e p rocess of assessing the effectiveness of the system cannot (a nd must not) be fragmented (for example, referred to the different stages of development from childhood to maturity), nor it may present solutions of continuity or, worse, be entrusted to structures between them are not coordinated, as it is happening today.
We must not misunderstand: every phase of the development of the person (education is a service to the person) has different needs and involves different targets and indicators for measuring the achievement, but the asse ssment of the preceding phase must represent the point "zero" of the next.
Taking account o f the above, it is v ery difficult to imagine and design a system for the evaluation of the university system without having input data consistent with the objectives assigned to i t (for the t ime, very general). This may be partially tru e only if it in tends to assess the organizational efficiency of it or the quality of teaching. It is absolutely wrong if we wish t o as sess efficacy, i.e. the a dequacy betwee n resources (human a nd financi al) invested and the "product" obtained.
A fi nal co nsideration co ncerns t he c omplexity and t he cost of t he creation of a n organizational structure to collect the data necessary to calculate one or more indicators (for
14
example the CU, a f unction of many variables) which distinguish each person throughout life. So me att empt, ev en of go od quality, is alread y underway (see fo r example Al ma Laurea a nd Alma Diplo ma), but each of them still suffers from the incom pleteness a t national level and from the need of a greater accent on assessment.
In conclusion, i think that it must be cont inued and increased research on indicators of efficacy (for e xample the CU) t o find viable sol utions, which c ould be im plemented on national plan and t hat, at t he sam e t ime, ha ve t o be more ob jectively defi ned by t he Government po licies, th e resulting go als and in struments for assessi ng th e edu cation service.
15
Specialised Sessions
16
Specialised session A
Business and Bank
17
Banking services evaluation: a dynamic analysis
Michela Lacangellera1, Caterina Liberati2, Paolo Mariani3 1 University of Milano-Bicocca, [email protected] 1 University of Milano-Bicocca, [email protected] 1 University of Milano-Bicocca, [email protected]
Customer centric vision has been successfully applied in the last years in ba nking sector. Such a conce pt has customer satisfaction as the most im portant asse t of a com pany. According with t his i dea banks h ave i ncreased t heir products di fferentiation i n order t o match with potential clien ts requests or expectations. Th is resu lted in offer ho mologation making necessary a focus on service attributes differentiation.
Such scenario leads banks to employ intelligent systems to monitor their own clients in order to build and preserve a robust relationships with them. That generated an operative improvement in term s of effi ciency and ec onomic retu rns. However, as it is well k nown, the bank-consumer is an “ever-c hanging” relationship due to both environment and ac tors evolution.
Therefore the core of our contribution focuses on analysing the evolution of customer satisfaction and track patterns of customer evaluations related to bank service features. The tested hypothesis is the loss of bank retail services competitiveness, probably due to a drop of customer satisfaction.
A m ultiway f actorial m odel run on data co llected b y an Italian bank h as em ployed information coming from a sat isfaction survey repeated in 3 different temporal waves on a sample panel composed by 27.000 instances per wave. The questionnaire has been framed according to SERVQUAL model. The multiway technique rationale is based on a particular decomposition o f th e t otal variability: wit hin and between groups. This secon d p art i s modelled through a linear regression where the different times represent the observations of the covariate. In our case a tim e series with only three observations is in adequate for th e regression m odel. Th us we propose a t echnique based on principal c omponents w hich enables the definition of a “ compromise plan” in order to perform a j oint analysis of th e three waves data.
This dynamic study has been carried out on a th ree-way data matrix Xijk (professional segments (i), ser vice attributes (j), w aves ( k)) of 9x24x3 d imensions. The interstructure analysis shows a high similarity among the waves and it justifies the search for a common “compromise” matrix through the intrastructure analysis. In this phase on a co mmon plan are re presented the indivi duals trajectories in the three waves accordi ng t o two fac tors (84%) summarizing service dimensions in terms of expectations and evaluations.
Results sho w an in creasing satisfaction between the 1 st an d t he 2nd wav e which highlights a n effect ive p ositive im pact of m anagement act ions. However t he ori entation discontinuity in individual tack patterns suggests a future expectation fall (third wave) and, in a medium term a consistent customer satisfaction drop.
References
Coppi R., Bolasco S. (1989). Multiway Data Analysis, Amsterdam: North-Holland. Munari L. (1999). Custo mer Satisfaction e redditività nelle banche. In: Ne wfin W orking Paper n.1, Ma rch,
Università Commerciale L. Bocconi. Berry L. L., Parasuraman, A., Zeit haml, V., A. (1988). SERVQ UAL: a multiple-item scale for measuring
consumer perceptions of service quality, Journal of Retailing, 64 (1), 12-37.
18
Ordinal logistic regression response functions with main and interaction effects in the conjoint analysis
Amedeo De Luca1, Sara Ciapparelli2 1 Università Cattolica del Sacro Cuore di Milano, [email protected] 2 Agos Ducato, [email protected]
In the Conjoint Analysis (COA) in order to link the categories of overall evaluation to th e factor lev els, we ado pt a cum ulative lo git m odel (De Luca, 2010) at the aggre gate level (pooled model, Moo re, 1980). Th e novelty v alue in our a pproach i s that one set of aggregated part- worths is estimated in connection with each category Yk
The n umber of profiles S (c ombinations o f levels of the M attributes or factors of a product), co nstitutes a full-factorial expe rimental desi gn. It i s assum ed t hat t he overall evaluation (Y) o f a produ ct consists in the choice of one of the ordere d categories k = 1, 2,…K. Th e effects of th e facto rs exp ress the v ariations of th e probabilities Pks associated with the vector sz corresponding to the combination s (s = 1, 2, …, S). The kth cumulative response probability is: )|( sks kYP z = Fk( sz ) = 1( sz ) + 2( sz ) + … + k( sz ).
The cumulative logits )~( skL z of the first (K-1) probabilities are:
)~( skL z = )]~([itlog zkF ln ])~(1
)~([
sk
sk
F
F
z
z
=
0
1
1 ~)~(...)~(
)~(...)~(k
sKsk
sks
zz
zzz~~ ' ,
with k = 1, 2, …, K-1; sz~ is the vector of the dummy explanatory variables relative to
the profile s in the reduced matrix Z~ ; 0
~k is the constant term, ' is the unknown vector
of regression coefficients of the factor levels. The al gebraic f orm of the response functions with m ain a nd first-order interactio n
effects is: )~( skL z = 0~
k + )(
1 2
)( ~~ mls
M
m
l
l
ml z
m
+ ),(
1 2 2
),( ~~ pmshl
M
m
l
l
h
h
pmlh z
m p
se ; k = 1,
…, q, h = 1, 2 , …, mh , p = m + 1, m + 2, …, M; where: 0~
k is the constant ter m
associated wit h the re ference category; )(~ ml and ),(~ pm
lh are the unknown re gression coefficients, respectively, for the lth level of the m factor and for the first-order interaction of the factors m and p; )(~ m
lsz and ),(~ pmlhsz are the dummy variables, respectively, for t he lth
level of the m factor and the h level of the p factor in the stim ulus s; kse is the error term pertinent to the sth sti mulus. We provide an application to real data wit h PLUM-Ordinal regression procedure of SPSS.
References
De Luca A., (2010). Ordinal logistic regression for the estimate of the r esponse functions in the conjoint analysis, in Joint-Meeting GfK1 – Cladag 2010, Book of Short Papers, 8-10 September, Firenze, 224-226.
Green, P.E ., Rao, V. R. ( 1971), Conjoint measurement fo r quantifying judg mental da ta, Jour nal o f M arketing Research, 8, 355-363.
Moore W.L. (1980). Levels of Aggregation in Conjoint Analysis: an Empirical Comparison, Journal of Marketing Research, 17, 516-523.
19
Measuring media reputation for private and public institutions
Paola Cerchiello1 1 University of Pavia, [email protected]
Reputation can be define d as ho w an entity (private or public) is perceived by each of its stakeholder groups and reputation risk as t he risk that an event will negatively influence stakeholder p erceptions. Si nce re putation i nvolves intangi ble assets (pu blic opi nion, perception, reliability, merit), it is not si mple to define and consequently to measure and to monitor the c orrelated risk. Because of the novelty of the problem, the scientific literatur e on the topic i s li mited and/ or not com pletely sh ared. In th is co ntribution we pr opose statistical models based o n or dinal data ai med at measuring ef fectively reputation and reputational risk. Such m odels are applied to real data on Italian public com panies taken from financial media corpora. We propose two parallel approaches rooted in the context of, respectively, no n parametric and parametric statistics. The form er allows us t o employ a scorecard ap proach base d on flexible i ndexes with the final aim of creating a ranking. Besides non parametric models, we need a parametric model whose estimation allows not only to describe and rank reputation, but also to predict and, therefore, prevent, reputational risks. In particular we need a parametric model suited for ordinal variables, as most reputational data is typical ly available in su ch format. W e propose a m ixture model that extends the CUB model proposed by D'Elia and Picc olo in 2 005, particularly useful when covariates are not available.
In order t o enable the application of th e scoreca rd approach and of th e CUB- CUBB models in this context we have collaborat ed with the It alian m arket leader com pany in financial and economic communication, ''IlSole24ORE'' and with DFKI a German Research Centre for Artificial Intelligence in t he framework of the European research project MUSING. The obj ective is to evaluate the corporate reputation of 40 Italian c ompanies listed as Blue Chips in the Italian Stock market, on the basis of newspaper articles delivered by ''IlSole24Ore'' and analyzed by means of an opinion mining tool. The OM result pursues data structured according to the following ordinal scale: 1 (very bad news), 2 (bad news), 3 (neutral news), 4 (good news) and 5 (very good news). Such output represents our variable of interest , which we use to assess the scor ecard a pproach a nd the mixture m odel. By means of the two parallel approaches we can rank Italian public companies and evaluate the latent component named reputation awareness contained in the analyzed newspaper.
References
Cerchiello P., D equarti E., Giudici P., Magni C.(2010) . Scorecard models to evaluate perceived quality of academic teaching, Statistica e Applicazioni.
D'Elia A. , Piccolo D. (2005). A mixture model for preference data analysis. Co mputational Statistics and D ata Analysis, 49, 917--934.
Siegel and Castell an, (1988). Nonpara metric sta tistics for the behavioral sciences, 2nd edition. McGraw-Hi ll, London.
20
The confidence ellipses in decomposition Non-Symmetrical Correspondence Analysis for the evaluation of the innovative performance of the Manufacturing Enterprises in Campania
Antonello D’Ambra 1, Anna Crisci2 1 Second University of Naples, antonello.d’[email protected] 2 Second University of Naples, [email protected]
Non-Symmetric Correspondence Analysis-NSCA (D’Ambra L. & Lauro, 1989) is a useful technique for anal yzing a two-way contingency t able. Th ere ar e m any re al-life app lications where it is n ot appropriate to p erform classical corresponden ce an alysis becaus e of the obvious as ymmetry of the association between the variables.
The key d ifference between the sy mmetrical an d non-sy mmetrical versions of correspondence analysis res ts o n the m easure of the as sociation used to quantif y the r elationship between the variables. For a two-wa y, or multi-way, contingency table, t he Pearson ch i-squared stat istic is commonly us ed when it c an be as sumed that t he ca tegorical v ariables ar e s ymmetrically r elated. However, for a two-way table, it ma y be that one variable can be treated as a pr edictor variable and the second variable can be considered as a response variable.
Yet, for such a variable structure, the Pearson chi-squared statistic is not an appropriate measure of the association. Instead, one may consider the Goodman-Kruskal tau index. In the case that there are more than two cross-classified variables, multivariate versions of the Goodman-Kruskal tau index can be consid ered. These include Marcotorchino’s index (Marco torchino, 1985) and Gray -Williams’ index (Gray & Williams, 1975).
In the present paper, the Multiple Non- Symmetric Correspondence Analysis- MNSCA (Gray, L. N., Williams, J. S,1975), along with the decomposition of the TAU by Gray-Williams in main effects and interaction (D’Ambra, L. et al., 2010), is used for the evaluation of the innovative performance of the manufactur ing enterprises in Campania. In novation represents a ver y important element fo r the competition of the enterprises and economic growth. Only the enterprises which are able to innovate regularly can have at their disposal a range of mo re and more appealing products for the customers. Moreover, only a constant inn ovation provides the constant effici ency of the proc esses and the optimization of the production costs. Final ly, the use of the ellipse confid ence (Beh , 2010) has allowed to identify a category which is statistically significant.
References
Beh, E.J. (2010), Elliptical confidence regions for simple correspondence analysis, Journal of Statistical Pl anning and Inference.
Beh, E .J., D’Ambra L .(2010), No n Sy mmetrical Co rrespondence Analy sis with concatenation and linear constraints, Australian & New Zeland Journal of Statistics, 52(1),27-44.
D’ambra, L., Lauro N. (1989), Non Symmetrical Analysis Of Three Way Contingency Tables. In Multiway Data Analysis, Eds. R. Coppi And S. Bolasco, North Holland, 301–315
D’Ambra L., D’Ambra A., Sarnacchiaro P. (2011) Analysis of main effects and interaction term in multiple non symmetrical correspondence analysis, submitted.
D’Ambra L., Beh E. J., Amenta P. (2005), Catanova For Two-Way Contingency Tables With Ordinal Variables Using Orthogonal Polynomials, Communication In Statistics (Theory And Methods), 34, 1755-1969.
Gray, L. N., Williams, J. S. (1975), Goodman And Kruskals Tau B: Multiple And Partial Analogs, American Statistical Association (Proceedings Of The Social Statistics Section), 444-448.
Light, R. J. & Margolin, B. H. (1971), An Analysis Of Variance For Categorical Data, Journal Of The American Statistical Association, 66, 534-544.
21
Specialised session B
Ordinal data models
22
A review of multilevel models for ordinal data
Leonardo Grilli1, Carla Rampichini2 1 Dipartimento di Statistica ‘G. Parenti’ Università di Firenze, [email protected] 2 Dipartimento di Statistica ‘G. Parenti’ Università di Firenze, [email protected]
An ordinal va riable is just a categorical v ariable supplemented with inform ation on the ordering of t he categories. The statistical methods for ordinal variables are desi gned to exploit such information. Here we focus on regression models for an ordinal response, with special emphasis on cumulative models, namely models based on cumulative probabilities.
We review random effects cumulative models for multilevel data and we discuss several issues peculiar to the random effects extension such as the distinction between marginal and conditional e ffects, t he measures of unob served cluster-le vel heter ogeneity, the consequences of adding covariates, and the main types of predicted probabilities. W e also briefly consider the topics of estim ation, inference an d prediction, with a brief look on available software.
The issues are illustrated through an analysis of student ratings on university courses.
References
Agresti A. (2010). Analysis of Ordinal Categorical Data, 2nd edition. New York: Wiley. Agresti A., Natarajan R. (2001). Modeling clustered ordered categorical data: A sur vey. International Statistical
Review 69, 345-371. Hedeker D. (2008 ). Multilevel Models for Ordinal and Nominal Variab les. In Handbo ok of Multilevel A nalysis
(ed. De Leeuw J and Meijer E), 237–274. New York: Springer.
23
A categorical data model to assess critical points
Luisa Stracqualursi1, Stefania Mignani2, Paola Monari3 1 Department of Statistics P. Fortunati, University of Bologna, [email protected], 2 Department of Statistics P. Fortunati, University of Bologna, [email protected] 3 Department of Statistics P. Fortunati, University of Bologna, [email protected]
In this paper the scholastic performance of young students during their compulsory studies has been studied.
The analysis is based on the data collected by the Scholastic Observatory of Province of Bologna in the recent years. They include individual perf ormances for e very student monitored over time.
The relations between sc holastic results a nd predictive varia bles as gender, place of origin, previous school mobility, family status, and type of school have been studied.
Interesting conclusions about the effects of some variables of interest have been drawn by using logistic models and classification trees.
In particular, the im portance of m onitoring the stude nt histories of life to identify critical points in the scholastic careers emerged. Moreover, the inclusion of one student in a given class of a classification tree can permit to forecast the further steps of his educational path.
References
Agresti A. (2002). Categorical Data Analysis. New York: Wiley-Interscience. Balakrishnan N. (1991). Handbook of the Logistic Distribution. Marcel Dekker, Inc Barabási A.L . et al. ( 2003), Scale-Free and hier archical structures in co mplex networks, Modeling of Com plex
Systems: Seventh Granada Lectures, AIP, Melville New York. Bartolomew DJ, Knott M. (1999). Latent Variable Models and Factor Analysis, Hodder Arnold , London. Breiman L. , Fr iedman J. , Olshen R. , Stone C. ( 1983). Classification and r egression t rees, Wadsworth and
Brooks/Cole, Monterey. Field J. 82006). Lifelong Learning and the New Educational Order, Trentham Books EU Commission (2005) Towards a European Qualifications Framework for Lifelong Learning, Commission Staff
Working Document, http://ec.europa.eu/education/policies/2010/doc/consultation_eqf.pdf Hilbe J.M. (2009). Logistic Regression Models. Chapman & Hall/CRC Press Jones H. ( 2005). L ifelong L earning in E uropean Union: W hiter the L isbon Str ategy? European Jour nal of
Education, 40, 3. Kruskal J.B., Wish. (1978). Multidimensional Scaling, Newbury Park: Sage Publications. Shelley M.C. (2007). Multivari ate Techniques for Dichoto mous Dependent Vari ables: An Application to Public
Policy, in Gerald J. Miller and Kaifeng Yang (Eds.), pp. 489-513 of Handbook of Research Methods in Public Administration (2nd ed.) (London: Taylor & Francis).
24
Modelling multivariate ordinal data: some experience with CUB models
Marcella Corduas1 1 Department TEOMESUS, University of Naples Federico II, [email protected]
A mixture model for ordinal data modelling (denoted CUB) have be en recently introduced in the literature (D’Elia and Piccolo, 2005; Piccolo, 2006). Specifically, the ratings are represented by means o f a d iscrete m ixture of a Uniform and s hifted B inomial random variables. T he parameters characterizing the mixture have an interesting interpretation in terms of ‘uncertainty’ and ‘degree of liking/dis liking’ expressed by the raters with respect to a certain item . Moreover, those parameters can be related to c ovariates so that further flexibility to the final model is added (see Corduas et al. 2010 for a discussion).
However, in real applicati ons the modelling is performed on single items separately. In complex surveys, instead, when subjects a re asked to evaluate a certain object with respect to several features or to expre ss their satisfaction with respect to various objects, it is often the case that responses are associated.
In t his pa per, we discuss the p roblem of i ncorporating C UB m odels withi n a multivariate distribution useful for the anal ysis of correlated ordinal data. In pa rticular, we consider the Plackett distribu tion (Plackett, 1965; Dale, 1 986; M olensberg a nd Lesa ffre, 1994) in order to con struct a class of d istribution with CUB models as m arginal distributions. The approach is interesting si nce it prov ides a fram e to move fr om the bivariate case to the multivariate one. In ad dition, it allows to retain som e useful interpretation and devices of marginal models and to use signif icant cova riates for improving final distribution fitting.
The paper presents s ome methodological and com putational as pects related to s uch technique and the results obtained from an empirical study. Acknowledgments: The p aper h as been p artly supported b y a MIUR grant within th e PRIN2008 project of Research Unit of Un iversity of Naples Federico II ( code 2008WKHJPK -PUC number E61J10000020001).
References
Dale J. R. (1986). Global cross-ratio models for bivariate, discrete, ordered responses. Biometrics, 42, 909-917. Corduas, M ., I annario M ., Piccolo D. ( 2010). A class of statistical models for evaluating ser vices and
performances. In P. M onari, M. Bini, D. Piccolo, St atistical Methods for the E valuation of E ducational Services and Quality of Products, Springer, Heidelberg-Berlin, 99-115.
D’Elia A., Piccolo D. (2005). A mixture model for preference data analysis, Co mputational Statistics and Data Analysis, 49, 917-934.
Molensberg G., Lesaffre E. (1994). Ma rginal modeling of correlated ordinal data usin g multivariate Plac kett distribution. Journal of the American Statistical Association, 89, 633-644.
Plackett R.L. (1965). A class of bivar iate distributions. Journal of the Am erican Statistical Association, 60, 516-522.
Piccolo D. (2006). Observed information matrix for MUB models. Quaderni di Statistica, 8, 33-78.
25
Specialised session C
Public transports evaluation
26
Preliminary studies to define critical threshold for customer satisfaction and loyalty
Giovanni Monaco1, Valeria Scaramuzzi2, Luce Allorio3
1CFI GROUP ITALIA, [email protected] 2CFI GROUP ITALIA, [email protected] 3CFI GROUP ITALIA, [email protected]
Company ofte n de bate on which is th e sat isfaction le vel to be reached to m ake sure of customer loyalty. The determin ation o f such a thres hold could be d one through different methodologies, empirical or mathematical. The article shows two ways t o do that: first one is used an d proved along many years, and is based on the measurement of the decrease in the custom er satisfaction in function of the percei ved quality. Second one is m ore theoretical but has m ore developing perspectives, and is born observing the gap between objective evaluations, global satisfaction and loyalty declared from the customer.
References
Fornell C. and C ha J. ( 1994). Par tial L east Square, Richard Bagozzi (E d.), Advanced M ethods of M arketing, pp.52-78.
Cronbach, L. J. (1951). Coefficient alpha and the internal structure of tests. Psychometrika, 16, 297-334. Fornell C. et alii, 1996. The Am erican Custo mer Satis faction Index: Nature, Purpos e and Findings. J ournal of
Marketing October.
27
Mixed logit models using Gaussian mixtures with covariate-dependent weights
Luisa Scaccia1 1 Dipartimento di Istituzioni economiche e finanziarie, Università di Macerata,, [email protected]
The m ultinomial logit (MNL) m odel has been widely used in the analysis of discrete choices and has found large application in tr ansport studies (McFadden (1974)). However, its restrictive assumptions, such as independence from irrelevant alternatives and preference homogeneity across respondents, have motivated the development of more flexible model structures that allow for an increasingly realistic representation of travel behaviour. Among these, a primary r ole is play ed by m ixed l ogit (M MNL) m odels (M cFadden a nd T rain (2000), Train (1998)), in which the utility of each individual is a function of the alternat ive attributes, with attribute coefficients that are random and reflect individual preferences.
In MMNL models a crucial issue is that of specifying an appropriate mixing distribution of the ra ndom coefficient s that m ay be interpreted as representing random t aste heterogeneity. Popular specifications have been the normal, triangular, uniform, lognormal distributions. However, any of them has shown its deficiencies (Hess et al. (2005)).
To deal with t his issue , F osgerau a nd Hess (2 009) proposed to m ake use of a sem i-parametric mixing distri bution c onsisting of a discrete m ixture of normal distributions (MOD). Scaccia and Marcucci (2010) considered the MOD approach and illustrated how to estimate this model in a Bayesian framework. Moreover, they extended the approach to the case in which multiple random coefficients, potentially correlated, are present in the model.
Here, we de velop the M OD approach to allow f or the de pendence of the observations on subject-specific covariates. In practice, we let the wei ghts of the mixture depend on the covariates through a logit-type function, so that the weights can vary between observations.
The model is applied to a data set referring to a study carried out in Urbino (Italy) to analyse the attributes of the local public transport and investigate possible interventions to improve the se rvice (Marcucci and Scaccia (2005), Scaccia (2009)). Five attributes of the service we re considere d: cost of m onthly ticke t, headway, first and last run, real ti me information di splays, bus s helters. C ovariates suc h as gender, m onthly bu dget, public transport usage frequency, availability of other transport means were also accounted for.
References
Fosgerau M., Hess S. (2009). A co mparison of methods for representing random taste h eterogeneity in discr ete choice models. European Transport/Trasporti Europei, 42, 1-25.
Hess S. , Bier laire M., Polak J.W . (2005). Estimation o f va lue of tr avel time savings using m ixed logit models. Transportation Research Part A: Policy and Practice, 39 (2-3), 221-236.
Marcucci E., Scac cia L. (2005). Alcune applicazioni d ei modelli a scelt a discreta al settore dei trasporti. In: I modelli a scelta discreta nel settore dei trasporti. Teoria, metodologia e applicazioni, Eds: E. Marcucci, Carocci editore, Roma.
McFadden D. (1974). Conditio nal logit analy sis of q ualitative choice behaviour. In: Frontiers in Econo metrics, Eds: P.C. Zarembka, Academic Press, New York, 105–142.
McFadden D., Train K. (2000). Mixed MNL models for discrete response. Journal of Applied Econometrics, 15 (5), 447-470.
Scaccia L. (2009). Rando m para meters logit models a pplied to p ublic transport de mand. Global & Loca l Economic Review, 13, 147-166.
Scaccia L., Marcucci E. (2010). Flexible Modelling of Mixed Logit Models. In: Proceedings of COMPSTAT'2010 - 19th International Conference on Co mputational S tatistics, Eds: Y. Lechevallier, G. Saporta, Physica -Verlag, Heidelberg, 1613-1620.
Train K. (1998). Recreation demand models with taste differences over people. Land Economics, 74 (2), 230-239.
28
An evaluation of Customer Satisfaction in public train transport by complex principal component analysis
Pasquale Sarnacchiaro1, Roberta Di Gennaro2 1 Università Unitelma Sapienza, [email protected] 2 Università degli studi di Napoli Parthenope, [email protected]
For a company, the knowledge of Customer Satisfaction (CS) regarding a given product or service, represents an im portant starting po int f or e very busi ness strat egy. I n o rder to measure CS many methods have been proposed: Servqual, Servperf and etc (Franceschini & Rossetto, 1996). In many of t hese methods, the survey allows to have two sets of data collected in two fully matched matrices (Amenta P. & Sarnacc hiaro P., 2001). In order to analyze this schem a of dat a, a fi rst sol ution c onsists i n a nalyzing t he tw o m atrices separately u sing Pri ncipal Component A nalysis (PC A). If we wa nt t o a nalyze the two matrices at the sam e time, we can perform a PCA on a new matrix obtained by horizontal juxtaposition of t he two data matrices. Other m ore appr opriate symme trical and non symmetrical s olutions, t hat allow to a nalyze the two set s of data joint ly, are: Co-Inertia Analysis, Co-Structure Analysis and Partial Least Square (Viv ien, 2001). The information obtained by means of the m entioned analysis is of different types, in order to have almost all the information embedded in only one analysis, it is possi ble to use Complex Principal Component Analysis (CPCA) (Horel, 1984). An application e nhancing the interpretativ e gain of the results will concern the evaluation of public train transport CS.
References
Amenta P., Sar nacchiaro P. ( 2001), T ensorial Co- Structure Anay sis for the f ull multi modules Custo mer Satisfaction Evaluation, Atti del Convegno intermedio della SIS
Escoufier Y. Gr ourud A. ( 1980), Analyse Factor ielle de s matrices car res non sym etriques, Diday ( Ed.) Data Analysis and Informatics, NY North Holland
Franceschini F., Rossetto S.(1996), Qualità nei servizi: un metodo per l a valutazione e il controllo in li nea del differenziale tra Qualità attesa e percepita. “De Qualitate”, anno V, n. 3, pp. 53-64
Horel J.D.(1984), Complex Principal Component Analysis: Theory and Examples. Journal of Climate and Applied Meteorology, Vol.23
Vivien M .(1999), Nouvelles appr oches en analy se m ulti-tableaux, R apport de Stage DEA de Biostatistique, Universitè de Montpellier
29
Optimal Sample strategies in public transports assessment
Tonio Di Battista1, Riccardo Di Nisio2
1 Department of Quantitative Methods and Economic Theory, [email protected] 2 Department of Quantitative Methods and Economic Theory, [email protected]
In Local Transportation Planing (LTP) with the aim to co ntribute to en courage the use of public transport and to redu ce the pollution i n a city, it’s important to u nderstand t he expectation of people living in delineated study area so to i ncrease the num ber of passengers.
Because the busi ness is strictly correlate d to the level of quality percei ved, it’s important to define an approach that could be allow us to capture t he diversity insight the people expectation’s concerni ng public transport in term s of quality. This means that the aspect of interest may not be just the “cust omer satisfaction” itself but the interact ions between expectations and the environmental stress gradient in a specific area.
As a consequence, traditional measures of custom er satisfactions m ay be not solve us problem because they may do not take into account the different ordering of people in terms of their expectations. From this way the results of comparative studies may depend on the selected index of diversity that it incorporat es a particular degree of sensitivity to rare and common expectations. A s olution is gi ven by the use of parametric families of indic es of diversity bo rrowed f rom the ecolo gical di versity (Gov e et al, 199 4) which are usually referred to the diversity pr ofiles which co nsist of a seque nce o f m easurements allowing different aspects of c ommunity structure s o that they take back all the differences in t o a single diversity spectrum.
Therefore, the diversity measure can be seen as a c urve unlike the i ndex of “c ustomer satisfaction” which is a scalar. This approa ch c ould em phasizes the im portance o f using such an approach that does not collapse the inform ation of a multidimensional item into a singular number.
In this framework t he Functional data anal ysis (FDA) approach will be considered (Gattone and Di Battista, 2009). As known (Ramsay at al l, 1996), the aims of FDA is that the observed data could be thought as single en tities rather than sequences of observations. Starting from this point of view in the LTP field the goal could be investigate the customer satisfaction diversity by m eans of the diversity profile (Patil and Taillie, 1982). The approach can be inferential or censorious.
In this work we deal with circum stances whe re the diversity insight into the expectations is a kn own function s uch as diversity profi les. In t his case the focus is to estimate the diversity function by means of a suitable sampling design. A f ixed population sampling point of view will be assumed and we will provide statistical inference based on the Horvits-Thompson estimator of the functional diversity profile.
References
J.O. Ramsey , B. W. Silverman (2005). Functional Data Analysis. Springer Series in Statistics Dennis B, Patil G.P., Rossi O., Stheman S and Taille C. (1979) A bibliography of literature on ecological diversity
and related methodology. In Ecologi cal Diversity in Theory and Practice , J.F. Grassl e, G.P. Patil, W. Smith and C. Taille (Eds.), International Co-operative Publishing House, Fairland (MD), 319-354.
Gove J.H., Patil G.P., Swi ndel B.F. and Taille C. (1994) Ecological diversity a nd forest management, In Handbook o f Sta tistics, Vol 12 ( Environmental Statistic s), G. P. Patil and C. R. Rao ( Eds.), E lsevier, Amsterdam, pp. 409-462.
Gattone, Di Battista (2009). A functional approach to diversity profiles. Journal of the Royal Statistical Society
30
Specialised session D
Rasch Analysis
31
Motivation to achieve academically: Rasch analysis and first results regarding correlation with students outcomes at the university of Udine
Enrico Gori1, Laura Pagani2, Michela Battauz3 1 Dipartimento di Scienze Economiche e Statistiche, Università di Udine, [email protected] 2 Dipartimento di Scienze Economiche e Statistiche, Università di Udine, [email protected] 3 Dipartimento di Scienze Economiche e Statistiche, Università di Udine, [email protected]
Academic achievem ent is rel ated to vari ous f actors a nd motivation is expecte d to be an important determinant. This work aims to measure motivation of university students and to study the correlation between motivation and academic outcomes.
The sample is composed of 99 students enrolled in the first academic year of the faculty of Ec onomics of t he U niversity of Udine. Data were c ollected du ring the lessons of statistics.
The questionnaire used to measure motivation is the questionnaire developed by Waugh (2002), th at ap plied Rasch analysis. Th is questionnaire is com posed by 24 item s on an ordinal scale with four categories. In the original version, each item requires two responses: What I aim for and What I actually do. A r ating scale m odel (Andrich, 1978) was used to analyse the data. Consistently w ith Waugh, the item s rel ative to What I aim for resulted easier than What I actually do. I n o rder to av oid dependencies between item s, only the responses relat ive to What I actually do, that presented di fficulties more adequate to the motivation levels, were considered. Ten items were discarded due to poor fit statistics. Item difficulties estimated with our data are sim ilar to those obtained by Waugh (correlation = 0.44).
In order to validate the scale obtained, the relation bet ween students outcome and the measure for motivation was studied using a multiple regression model. Students outcome is defined as the sum of the grades obtained by the students in the e xams passed i n the first academic year multiplied by the credits. The other variables that resulted significant are the intake level (measured using the score obtained in t he enrollment test) and the sec ondary degree grade. All the explanatory variables are standardized. Measurement error present in the measure for m otivation and in the i ntake level was adjusted using the SIMEX m ethod (Cook and Stefanski, 1994). The method requires the knowledge of the measurement error variance. For motivation, t he m easurement er ror va riance can be considered k nown and equal to the variance of the person parameter estimates. Since the res ponses given to the enrollment questionnaire were not available but only the total score was, a Rasch analysis was not possible for this vari able. Consecut ively, the measurement error variance for t he intake level is not known and a sensitivity a nalysis with reliabilities equal to 0.7, 0.8 and 0.9 was performed. The m easure for motivation obtained resulted positively related to the academic outcom e ( p-value=0.01), irrespectiv e of the reliability a ssumed for t he intake level.
References
Andrich, D. (1978). A rating formulation for ordered response categories. Psychometrika, 43, 357-374. Cook, J., & Stefanski, L.A. (1994). A simulation extrapolation method for parametric measurement error models.
Journal of the American Statistical Association, 89, 1314–1328. Waugh, R.F. (2002). Creating a scale to measure motivation to achiev e acade mically: Linking attitudes a nd
behaviours using Rasch measurement. British Journal of Educational Psychology, 72, 65-86.
32
Effects of DIF items on Rasch measures
Silvia Golia1 1 Quantitative Department, University of Brescia, [email protected]
Differential item functioning (DIF) is understood to be present when something about the characteristics of a test taker interferes w ith the relationshi p between ability and ite m responses. For a given level of trait, the probability of endorsing a specified item response should be independent of subgroup membership; if it does not happen, then that item is said to exhibit DIF. Two types of DIF can be identified: uniform and nonuniform (Mellenbergh, 1982). Uniform DIF (UDIF) occurs when an item is endorsed at a consis tently higher level by o ne g roup over the other gr oup at all levels o f the underlying trait . Nonuniform DIF (NUDIF) means that at certain levels of th e underlying trait, one g roup has hi gher scores, while at o ther levels the opposite is the case. In a typical DIF study, subgroups are studied in pairs, labeled the reference and focal group.
When DIF is present, an impact on the estimated ability measure could be expected. The present simulation study addresses the issue in a ssessment of the impact of both kinds of DIF on the measures obtained applying the Rasch model when the questionnaire is formed by polytomous items. There are several methods to asses DIF in polytomous items; good reviews are given by Potenza and Dorans (1995) and Penfield and Lam (2000).
The Rating Scale Model (Andrich, 1978) is the one used to simulate responses to items without DIF and to estim ate the RSM parameters. The data generating m echanism that allows to sim ulate data affected by UDIF or NUDIF are based on adding to the lo g-odds ratio of two adjacent categories an extra term which depends on the interaction between the item and the dummy variable group, coded as 1 if the s ubject belongs to the focal group and 0 otherwise, in the case of UDIF, and between the i tem, the subj ect ability and the variable group for NUDIF.
In th e pr esent stu dy 1 000 subjects and 15 item s with 6 response categories are considered. Two different cases of UDIF are analysed; in the first one the DIF sign and size compensate each ot her whereas in the second one t he focal group shows a relative advantage (disadvantage) over the reference group for al l th e item s exhibiting DIF. For each case, 500 data sets were simulated and analyzed and 500 sets of estimated abilities and item difficulties were com puted. The sim ulation shows that when the item s exhibit UDIF with DIF sign and size whi ch com pensate each othe r, the esti mates o f ability are not significantly influenced by the presence of DIF; different results show up when DIF sign and size do not co mpensate each other. When the items exhibit NUDIF, there is an im pact on the measures.
The results highlight the importance to identify the nature of DIF in order to decide the best strategies to face it.
References
Andrich, D. (1978). A rating formulation for ordered response categories. Psychometrika, 43, 357-374. Mellenbergh, G.J. (1982). Contingency table models for assessing item bias. Journal of E ducational Statistics, 7,
105-107. Penfield, R.D., Lam, T.C.M. (2000). Assessing differ ential item functioning in per formance assessment: Review
and recommendations. Educational Measurement: Issues and Practice, 19, 5-15. Potenza, M.T., Dorans, N.J. (1995). Evaluation DIF assessment for polytomously scored items: a fr amework for
classification and evaluation. Applied Psychological Measurement, 19, 23-37.
33
Impact of educational test features on item difficulties by the Linear Logistic Test Model
Daniela Marella1, Carlo Di Chiacchio2, Giuseppe Bove3 1 Dipartimento di Scienze dell’Educazione, Università degli studi Roma Tre, [email protected] 2 Invalsi, [email protected] 3 Dipartimento di Scienze dell’Educazione, Università degli studi Roma Tre, [email protected]
Educational testing studies focus on latent variables, usually named abilities. Remarkable examples are reading ability or mathematical ability. A primary goal of these studies is how much of such abilities persons possess and to this aim a test consisting of a number of items (questions) is developed. Item response theory (IRT) is essentially a theory of the relat ion between item responses and the underlying abilities (or traits). Mathematically, the relation is described by a function (named item characteristic curve, ICC) linking the probability of correct response to an item a nd t he ability scale. Since test ite ms a re not necessa rily equivalent in difficulty or validity in measuring the underlying trait, ite m parameters are included in IC C. Whe n only one difficulty parameter for each item is consi dered and a logistic model is adopted for the ICC we obt ain the famous Rasch model, whose attractive theoretical properties have been extensively studied (e.g. Fischer 1995).
Linear logistic test model (LLTM) proposed by Fischer (1973) is a Rasch-family model that includes parameters for the impact of cognitive design variables and other test variables on item difficulty. LLTM breaks down the ite m difficulty parameter of the Rasch m odel into a linear com bination of cer tain hypothesized elementary pa rameters. Apart fr om the primary application o f generating items, there ar e many other potential applications of the LLTM (see e.g. Kubinger 2008, Xie & Wilson 2008, Daniel & Em bretson 2010). In this communication some recent developments and potentialities of the model are reviewed and an application to Italian PISA 2006 sample is provided.
References
Daniel R. C., E mbretson S. E. (2010). Designing cogni tive complexity in mathematical p roblem-solving item s. Applied Psychological Measurement. 34 (5), 348-364.
Fischer, G. H. (1973). The linear logistic test model as an instrument in educational r esearch. Acta Psychologica, 37, 359-374.
Fischer, G. H. (1995). Derivations of the Rasch model. In Fischer, G. H., & Molenaar, I. W. (eds), Rasch models. Foundations, Recent Developments, and Applications. New York: Springer, 15-38.
Kubinger H. D. ( 2008). On the r evival of the Rasc h model.based L LTM: Fr om constructing tests usin g it em generating rules to measuring items administration effects. Psychology Science Quarterly. 50 (3), 311-327.
Xie Y ., Wilson M. (2008). Investiga ting DIF and exte nsions usi ng an LLTM approach an d also an individual differences approach: an international testing context. Psychology Science Quarterly. 50 (3), 403-416.
34
Specialised session E
Latent variable models
35
A Bootstrap procedure to test the cograduation between two ordinal scales: a proposal of two new indices.
Andrea Bonanomi1 1 Dipartimento di Scienze Statistiche, Università Cattolica del Sacro Cuore di Milano, [email protected]
In psychometric sciences the choice of a good response scale is a typical problem . Several studies (see, in particular, Bonanomi, 2004) proved that different measurement scales lead to hig hly dissi milar evaluations of g oods/services, in particular in t he measurement of observable variables in late nt variables models. So, this requires that proper cograduation indices should be available, in order to com pare the attit ude of different m easurement scales (typically ordinal scale s) in assigni ng the vario us and different number of allo wed response modalities. For this purpose, an inno vative proposal is given, consisting not only on the measurement of the natural concordance between two scales in the evaluation of the same good/service, but in the possi ble propens ity and attitude of a scale to provi de more positive or negative evaluations: two new different indices are proposed, Ip1 and Ip2. The evaluation of an ite m with two response scales, S1 (with modalities s11,…,s1I) and S2 (with modalities s21,…,s2J), with I J, is proposed to the same sample of n subjects, randomizing the presentation sequence of the scales, in order to avoid memory effects. Consi dering a generic item, let nij be the number of s ubjects which assigned the i-th modality evaluating with S1 an d the j-th with S2, in a I J c ontingency ta ble. Ip1 is based o n a com parison between frequencies in th e inferior and in the superi or t riangle in t he case of a squa re contingency table. A proper procedure to determine the inferior and superior triangle in a I J contingency table (I ≠ J) is also implemented.
1sup.triangle inf.triangle
ij ijIp n n n
Ip2 is determined by a com parison of the empirical cu mulative distribution functions, having previously normalized the scales i n the [0;1] interval. With reference of the two Lorenz curves, Ip2 is obtained as the region between the cumulative functions. Let F be the relative empirical cumulative function of S1 and G of S2; Ip2 is
112
1 12 2
I Jj ji i
i j
G GF FIp
I J.
Ip1 and Ip2 are then rescaled in the domain [-1;1]; zero indicates the perfect concordance of the two ordinal scales. In order to verify the attitude of a scale to assign a mainly positive or negative rating c ompared to a diffe rent one , a n on parametric test is set up. T he n ull hypothesis is referred to t he perfect co ncordance bet ween the two c onsidered scales ( Ip1 and Ip2 e qual to zer o). A bootstrap procedure is useful in this co ntext: n on-parametric bootstrap confidence interval s ar e used (Diciccio, Efr on, 19 96), in the ir particular B ca version (bias-corrected acce lerated pe rcentile method), since they ha ve acc uracy a nd correctness higher than other similar bootstrap intervals.
References
Bonanomi A. (2004). Variabili ordinali e trasformazioni di scala, con partic olare riferi mento alla sti ma dei parametri dei modelli interpretativi con variabili latenti, Tesi di dottorato di Ricerca in Statistica Metodologica ed Applicata, sede amministrativa Università degli Studi di Milano – Bicocca.
Diciccio T.J., Efron B. (1996). Bootstrap Confidence Intervals, Statistical Science, Vol. 11, No. 3, pp 189-212. Landenna G. (1994). Fondamenti di statistica descrittiva, Bologna, Il Mulino.
36
Combining PLS and GME to estimate Structural Equation Models
Giuseppe Boari1, Gabriele Cantaluppi2, Enrico Ciavolino3 1 Dipartimento di Scienze statistiche, Università Cattolica del Sacro Cuore, Milano, [email protected] 2 Dipartimento di Scienze statistiche, Università Cattolica del Sacro Cuore, Milano, [email protected] 3 Dipartimento di Filosofia e Scienze Sociali, Università del Salento, Lecce, [email protected]
Structural E quation M odels with Latent Variables (SE M-LV) are c ommonly used in frameworks, e.g. Customer Satisfaction (C S) analyses, where, once de fined a construct by means of a pat h model, the interest is m ainly focused on the estimation of the parameters indicating the strength of the relationships among the considered unobservable entities.
Partial Least Squares (PLS) and LISREL are the estim ation proce dures usually considered (t he first method per forms the parameter estim ation afte r a latent score reconstruction; while parameter estimates are directly obtained with the latter method from the covariance structure of the observed variables). Possible drawbacks may ensue from the presence of multi-collinearity in the following cases: in the outer m odel, only for the PLS algorithm, when constructs of the formative type are present; in th e inne r m odel, w hen some explicative latent variables are highly correlated, possibly due also to the correlations among the observable variables pertaining to different blocks.
The Generalized Maximum Entropy (GME) method (Golan et al., 199 6) r epresents a semi-parametric estimation method for the SEM (Ciavolino, Al Nasser, 2009) which works well in case of ill-behaved or nonlinear da ta. The GME approach considers t he re-parameterization of th e unknown p arameters an d t he distu rbance term s as a co nvex combination of the expected value of discrete random variables. The estimation is achieved by the maximization of the Shannon’s entropy function.
Our proposal is to define a combined algorithm that use first PLS to estimate the latent variables and, then, GME t o estimate the para meters in the inner and outer models, t hus overcoming the possible presence of multi-collinearity.
Moreover the algorithm will deal with the presence of missing values, by considering in the PLS step the procedure proposed by Boari et al. (2007).
The behaviour o f the p roposed procedure i s evaluated by using the ‘mobile’ data set, proposed by Tenenhaus et al. ( 2005) a nd available in Sanchez, T rinchera (2010), also simulating the presence of missing values.
References
Boari G. , Cantalu ppi G., De Lauri A. (2007). Handling missing valu es in PLS path modelling: co mparison of alternative pr ocedures. I n: Classification and Data A nalysis 2007, B ook o f Shor t Paper s, M eeting of the Classification and Data Analysis Group of the Italian Statistical Society, 641-644.
Ciavolino, E ., Al-Nasser , A.D., (2009). Co mparing Generalized Maximum Entropy and Par tial L east Squar es methods for Structural Equation Models, Journal of Nonparametric Statistics, 21 (8), 1017-1036.
Golan, A., Judge, G.G., Mille r, D. (1996). Maxi mum Entropy Econo metrics: Robust Estim ation with Li mited Data. New York: John Wiley & Sons.
Sanchez G., Trinchera L. (2010). plspm: Partial Least Squares Data Analysis Methods. R package version 0.1-11. http://CRAN.R-project.org/package=plspm.
Tenenhaus M., Esposito Vinzi V. , Chatelin Y-M., Lauro C. (2005). PLS path modeling. Computational Statistics & Data Analysis, 48, 159-205.
37
SImultaneous Factor Analysis Across Several Populations (SIFASP): identification criteria for latent structures
Hans Schadee1 1 Dipartimento di Psicologia , Ateneo Milano-Bicocca, [email protected]
A growing n umber of i nternational surveys put t he same questions (in translation) to national random samples. The number of countries (groups) involved often are too few for a multilevel approach, even when researchers know the relevant higher level variables, but too many to handle each sample (group) on its own. In such contexts multisample analyses can b e useful. ( Jöreskog 197 1) The m any ways in which factor m odels can be speci fied frequently lead in m ultisample solutions t o un clear procedures and interpretations. This paper as sumes that variables are similar b ut vary, across groups, th rough lin ear transformations, and explores the consequences of this for factor analytic models.
Consider a multisa mple covariance m odel w ith latent variables where a num ber of variables ( j) a re similar in all gr oups (g). ygj = a hj+bhjyhj : the responses, ygj and y hj for a similar variable (j) in different groups g and h, have the same meaning but measurements are affine tra nsforms of one another. It sp ecifies the question of Ahmavaara (1954) a bout factorial inva riance in such situations. T he (p artial) answer offered here is put in the framework of (strongly) constrained solutions. The covariance model used differs from the common factor analytic model (CFA) in that it includes a speci fic linear scaling coefficient for the variables, different in the various groups, (using G diagonal matrices Bg with entries changing ov er th e G groups 1…G) so : Varg(Y) = Va r(Bggg) + Var(Dg), a m odel suggested by W iley et.al.(1973). The m odel requires sc ale invariant e stimation m ethods (ML, GLS, WLS, but not principal components, principal factors or ULS). For single factor models with t he same number of observed va riables this model can be scaled s o that all loadings a re the same within eac h g roup and across groups. The variance of the (single) factor can however vary. For congeneric factor models the rescaling of loading is as for the single factor model, but the different constructs underlying the observ ed variables can still have varying variances and covariances (correlations). In models with at least three latent factors the covariances among factors in different groups can be put equal, thus also fixing the variances of the latent variables. For the common factor model a s ubset of the model satisfying the criteria of the congeneric model is define d, and the rescaling c onstraints are applied for that submodel; parameters not fixed or constrained this way provide remaining degrees of f reedom for testing further sim ilarity constraints. More general models with constraints on uni que variances or c orrelated er rors can be dealt with in a si milar way. Means can be handled separately afterwards. Fit and partial fit indices for these procedures will be discussed. An example are gi ven on estimating and interpreti ng such models with standard SEM software (LISREL, Jöreskog, Sörbom 1988), usi ng ten ‘trust in institution’ variables from the vario us waves of the Eu ropean Social Science su rvey in ten count ries. (van der Veld, Saris 2009). Ahmavaara Y., (1954) The mathematical theory of factorial invariance under selection Psychometrika 19:27-23 Jöreskog K G, (1971) Simultaneous factor analysis in several populaztons Psychometrika 36:400-42 Jöreskog K G, SöRBOM D. (1988) Lisrel a guide to the program (SPSS Science Inc) Van der Veld W, Saris W (2009) Methodlogical aspects of the cross national evaluation of a theory on the causes
of generalized social trust (paper for QMSS2 seminar Bolzano-Bozen, Italy , june 11-12) Wiley D E , Schmidt W H, Bramble W J, (9173) Studies of a class of covariance models ‘ Journal of the American
Statistical Association 68:317-232.
38
Specialised session F
Evaluation of local authorities performance
39
Methodology and Tools for Local Government Services Monitoring: the IQuEL project experience
Gianluca Vannuccini1, Ciro Annicchiarico2 1 Comune di Firenze, Firenze, Italia, [email protected] 2 Comune di Firenze, Firenze, Italia, [email protected]
The eGovernment proj ect IQuEL (Innovazione e Qualità per Enti Locali) funded by the Italian Dipartimento Affari Regionali (DAR) of the Presidenza del Consiglio dei Ministri, and ended in year 2009 , produced a qualitative- and quan titative-analysis methodology for the monitoring of services provided by Local Governments. Such a methodology starts by a co-design with stakeh olders, in o rder t o de fine variables and m etrics, then ends with the delivery of reports a nd dashboards e nabling decisions on the p rovided se rvice u nder examination.
Such a m ethodology is enabl ed and powere d by means of a softwa re platform - which was de veloped within the project itself - c omposed of a tool f or data collection (citizen satisfaction surveys, data collection through online forms, datasets or external surveys import from other sources) and an open-source tool for decision support reporting.
The p roject f ocused on five dif ferent municipal services, nam ely, kin dergarden subscription, change of address within the same municipality, road maintenance, request to start building and commercial activities, and public and institutional communication.
By adopting the proposed approach, for each service it will be possi ble to exploit the results of the analysis (bot h in terms of perceived quality and internal quality), in order to highlight the strengths and the main critical factors, thus supporting with processed data the managers conduct, for a continuous improvement of the services under their scope.
The IQuEL project was ai med at devel oping B usiness Intelligence for the Public Administration, thus f ollowing guidelines coming from the Italian M inistry fo r Pu blic Administration (l.6 9/09, Dl gs 15 0/09), tha t mainly regard performance, transpare ncy, accountability, and continuous process improvement.
40
The customer auditing system in Milan, a certificated managing procedure ISO 9001
Maria Morena Montagna1 1Responsabile Servizio Customer Care Settore Qualità e Semplificazione Comune di Milano
For some years commune of Milan has been using quality controlling means on its services and other self-evaluating means such as CAF and EFQM; since 2006 Comune of Milan has started an articulated iter for certification ISO 9001-2008 and ISO 14001-2004 which has resulted in the standardization of every Board service.
More precisely: through the a doption of UNI ISO standards and indicators Comune of Milan has m ade its commit ments about quality and custom er’s auditing clear to its customers and other city users.
Goal of the process was that to integrate the managing means with the contract of service set up by the system of volunteer certification in order to m ake the 2 a reas (law – required and volunteer) c onsistently orient ed to t he deve lopment of t he quality syste m inside the local public services.
The system has its strength in the process of t he c ustomer aud iting; it’s not only a managing proced ure of SGQ, it also re presents a strong sign al of the need for no-self reference, identification and meeting our customers and city users needs and e xpectations keeping up and improving the global pe rformances. The s ystem is transv erse to the wh ole administration and is formed by the complaint and data collecting desk. Complaint desk is a multichannel net system (fax, on line, m ail, direct delivery) for c omplaint reception which was born in 2007 and has dealt with m ore th an 25 .000 complaints givi ng the custom ers concrete and guaranteed answers in average time limit of 6 days; it has also perm itted to collect and monitorate more than 3000 events.
The customer satisfaction col lecting activity is addressed to the users of the municipal services and undertakings; it foresees a yearly planning of the investigations arranged with the different directions and services.
These in vestigations have been drawn int o legal form thro ugh data col lecting f orms with passages, form alities a nd responsibilities of the necessary ac tivities to the surveys which ha ve been cleare d in the “m anaging procedure” and i ntegrated th rough t heir guidelines.
Each survey is planned both for the m ost appropriate surv ey mean (e.g. que stionnaire, focus group, phone calls, emoticon) and for the example plan which must consider every time the goals of: our research, our users’ target, the time management and of the resources of the involved services. Keywords for both the adopted methods are: Economicity: all the activities are realized and managed with internal resources of the
board Attention to the technological support: complaint managing e monitoring/reading of
the custom er satisfaction work on tech nological ap plications w hich have been planned a nd im plemented than ks t o the s ynergies betw een the B oard sectors a nd services Optimization of the listeni ng skills: will-power to change criticis ms into im proving
actions and strategic goals
41
ELISA PROGRAM - LOCAL INNOVATION SYSTEM measuring the quality of the services of local Public Administrations
Angelina Tritto1 1Manager P.O.R.E., c/o Premier of Council of Ministers, Dipartiment of Regional Affairs, Rome -Italy
Elisa Program, managed by the P residency of the Council of Ministers - Department fo r Regional Affai rs, and in particular the stru cture of the PORE (Project Opport unities for Regions in Europe) with the technical assi stance of Invitalia SpA, has funded projects on innovation and development of l ocal aut horities also on the m easurement of quality of services provided by local government.
IQUEL – The city of Parma has successfully completed the IQUEL project, which was attended by a large num ber of m unicipalities and provinces, aggregated toget her (23 administrations of cen tral and nor thern I taly). T he project aim s to provi de local governments tools to an optimal management of the services on different channels. IQUEL achieved the following results: id entifying and sharing a set of statistical indicators fo r measuring access and performance on different channels (web, desk, telephone, etc.); a set of indicators of customer satisfaction, a set of features of Citizen Relationship Management to help optimize the delivery of services to users.
ELISTAT – ELISTAT is a wor k in p rogress p roject, managed by the Pr ovince o f Brescia as Responsi ble aggreg ation, with a local identity as syste mic, involvi ng 42 provinces from 1 2 d ifferent r egions and a catchment area of m ore than 22 m illion inhabitants. The initiative pa rticipating the local authorities of various kinds: metropolitan provinces and reduced size, a newly created provinces, provin ces with high intensity of small municipalities. ELISTAT proj ect intends to design, develo p and publish on the Internet an integrated system of statistical indicators covering all the functions and services provided by the provinces, with particular attention to services for small municipalities, for a constant measurement of performance, cost and induced benefits throughout the country.
42
Innovation and Quality in Local Authorities: A combination to be pursued
Flora Raffa1 1 Comune di Parma, [email protected]
The transformation of Public Administration, which began in Italy since the ea rly 90's in times of financial crisis, was the solution to meet the dual objectives of reducing the costs of administrative cars as wel l as the public services and by offering quality services with the continuity to the real needs of the citizens. A necessity-oriented culture is established in policy plannings for improved performance.
Parma has created a Citizen ’s Service Centre that is innov atively ef ficient in Italy, by using new technologies and interaction-sensitive services.
For further advancem ent of the quality services offered to the citizens, the City of Parma has prom oted the "IQuEL" project, funded by the first noti ce o f ELISA pr ogram, acting as the coordinator of more than twenty local administrations represented mostly from the national territory.
To implement a continuous improvement of the service, IQuEL has c reated an efficient detection system of the local administrations, starting from the quality level of the services offered. B runetta’s Dec ree highlighted t he nee d to com bine t he asse ssment of the its performance with the capability of the administration to be "accountable" to the citizens.
Quality assessment becomes an obligation to address the need to give tangibility to the achieved results so as to prep are a social r eport whic h hi ghlights its contrib ution to the welfare of the citizens.
The IQuEL Project has developed significant experiences on three lines of action: - to know and to describe the services objectively (quality supplied) - to know the perceived quality (Customer Satisfaction) - to set up the service on the needs of the customer (CRM) A quality-oriented Public Ad ministration updates t he ci tizen-customer the tim e of
testing and the re-organization of the service. The City of Parma has given its specific contribution in analyzing and implementing the
CRM softwar e (C itizen R elationship M anagement) whi ch re presents its contact center, designed to be the main a ccess of the citizen to a multi-channel contact with the Administration.
The CRM was founded in the tradi ng scheme as a high-tech system created to collect and process client-related i nformations and aimed at im proving the ser vices o ffered reformulated based on the characteristics and the value of the clients.
The PA ce rtainly does not inten d to discri minate am ong its ”clients” but ex ploit the CRM to manage the complexity:
- of the numerous contact channels that are available to the citizens - of a num ber of internal a nd external inform ation creators that certification of sources
are mandatorily provided - of the proce sses ne cessary to fulfill a si ngle re quirement that m akes it necessary to
trace the process of handling the request The CRM is a tool to help improve the quali ty by ensuring the uniformity and certainty
of the response. It allows y ou to extract performance indices from the supplied responses and suppo rts th e outbound as a too l to co llect th e cu stomer’s satisf action or to carr y ou t proactive steps.
43
Specialised session G
Teaching evaluation in the Italian university system
44
Descriptive analysis of student ratings
Piero Quatto1 1 Dipartimento di Statistica, Università degli Studi di Milano - Bicocca, [email protected]
Let X be a statistical variable re presenting the student rat ings of university teaching. It is natural to assume for X an ordinal scale consisting of k categories (in ascending order of student satisfaction).
At first glance, student ratings can be summarized by a location index (such as the mode or the m edian of X) associa ted with a convenie nt m easure of d ispersion ( Leti, 198 3) o r asymmetry (Leti, 1983; Arnold and Groeneveld , 1995). For instance, the median of X may be associated with th e asymmetry index of Leti, resulting in a synt hesis that takes i nto account the ordinal nature of ratings and also c ommunicates information in an effec tive way.
More generally, there are infinitely many indexes (such as the ordinal entropy) that can be properly employed to measure the ordinal dispersion.
In addition, on the basis of any measure of ordinal dispersion, it is possible to define the corresponding index of ordinal asymmetry following Leti (1983).
On the other hand, stu dent ratings a re oft en conve rted i nto scores and treated as a quantitative variable. M ore generally, it is po ssible to measure stud ent satisfaction by means of a s uitable real-valued function (Agresti, 1984; Bross, 1958; Capursi et al., 2 001; Cerchiello et al., 2010; Chiandotto et al., 2000; Civardi et al., 2006), which we denote by S. It turns out that S is naturally defined on the standard simplex
k
iiik pippp
11 0 ,1:,...,
where pi is the relative frequ ency of the category i (i = 1,2,…,k). B esides, it seem s necessary that such a function satisfies so me appropriate conditi ons. For e xample, the function S is expected to reach its minimum at (1,0,…,0) and its maximum at (0,…,0,1).
Finally, each measure of student satisfaction can be a ssociated with a suitable measure of variability.
References
Agresti A. (1984). Analysis of Ordinal Categorical Data, Wiley, New York. Arnold B.C., Groeneveld R.A. (1995). Measuring Skewness with Respect to the Mode. The American Statistician,
49 (1), 34-38. Bross I.D.J. (1958), How to use ridit analysis. Biometrics, 14 (1), 18-38. Capursi V., Porcu M. (2001). La didatti ca universitaria valutata dagli studenti: un indicator e basato su misure di
distanza fra distribuzioni di giudizi. In: Atti Convegno Intermedio della Società Italiana di Statistica “Processi e Metodi Statistici di Valutazione”, Roma 4-6 giugno 2001.
Cerchiello P., De quarti E., Giudici P., Magni C. (2 010). Scorecard models to evaluat e perceived qualit y of academic teaching. Statistica & Applicazioni, 2, 145-155.
Chiandotto B., Gola M. (2000). Questionario di base da utilizza re pe r l'attuazione di u n progra mma per la valutazione della didattica da par te degli studenti. Rapporto di Ricerca RdR 1-00. Gruppo di Ricerca MIUR-CNVSU.
Civardi M., Crocetta C., Zavarrone E. (2006), Summary indicators of opinion expressed by the user s of a given service, Statistica. LXVI (4), 373-388.
Leti G. (1983). Statistica descrittiva, il Mulino, Bologna.
45
Students’ Evaluation of Teaching. The need for adjusted measures in comparative evaluations
Mariano Porcu1, Isabella Sulis2 1 Dipartimento di Ricerche Economiche e Sociali, Università di Cagliari, [email protected] 2 Dipartimento di Ricerche Economiche e Sociali, Università di Cagliari, [email protected]
Private stakeholders (families and enterprises) and public financing system constantly ask for a system of indicators suitable for evaluating, in comparative terms, the performance of the university system. These indicators are fre quently us ed to s upport and e nhance only those act ors (uni versities, faculties, depart ments) who satisfy specific quality standards (defined in terms of efficiency and efficacy).
Some of t he indicat ors c onsidered for the c omparative eval uation of teachi ng institutions in terms of efficiency pertain to the dom ain of the St udents’ Evaluati on of Teaching (SET). It has already been argued by different authors that students’ evaluations are infl uenced by sources of heterogeneity whic h are e xternal to the teaching processes: students’ cultural background, their previous school/university career, if they are part-ti me students etc. From these considerations a rises the demand to m ake available comparative indicators which take i nto account the e ffects o f th e so called Perf ormance Con founding Factors – PCF (Draper and Gittoes, 2004) that operate on bot h macro (e.g. degree programs) and micro (students) levels. From the second half of the Eighties there has been a growi ng i nterest on studies on the effectiv eness of educational institutions (Goldest ein and Spiegelhalter, 1996; Bratti et al., 2004; Bird et al., 2005) and on the use of statistical methodologies to assess the effect of PCF on the outcome variable at different levels of the data structure.
The approach underlying the building up of ad justed comparative indicators relies o n regression analysis models that allow us t o m easure t he infl uence played by eval uators’ characteristics on the outcome variables used to monitor the performances. In this work we present some applications on data referred to SET i n order t o explore the potentiality of latent variable modelling approaches belonging to the families of the Item Response Theory – IR T (R ijmen an d B riggs, 20 04) a nd Latent C lass R egression Analysis – LC RA (Bartholomew, 1999; Hagenaars JA and McCutcheon AL, 2002).
References
Bartholomew, D. J. and Knott, M . ( 1999). L atent Var iable M odels and Factor Analy sis. Kendall’s L ibrary of Statistics. Hodder Arnold Publication, London.
Bird, M., Cox, D., Goldstein, H., Holt, T. and P. Smith, P. C. (2005). Performance indicators: good, bad, and ugly, Journal of the Royal Statistical Society. A, 168, 1, 1–27.
Bratti, M., McKnight, A., Naylor, R. and S mith, J. (2004). Higher education outcomes, graduate employment and university performance indicators. Journal of the Royal Statistical Society. A , 167, 3, 475–496.
Draper, D. and Gittoes, M. (2004). Statistical analy sis of performance indicators in UK higher education. Journal of the Royal Statistical Society. A, 167, 3, 449–474.
Hagenaars, J. A. and M cCutcheon A. L . ( 2002). Applied Latent Class Analy sis. Cam bridge Univer sity Pr ess, Cambridge.
Goldstein, H., and Spiegelhalter, D. (1996). League tables and their limitations: Statistical issues in comparisons of institutional performance. With discussion. Journal of the Royal Statistical Society, A, 159, 385-443.
Rijmen, F . and Briggs, D. (2004). Explanatory Item Response Models, chapter Multiple pe rson dimensions and latent item predictors, 111–166. Springer.
46
The evaluation of educational university processes and teaching activities in Italy: aims, surveys and open problems
Luigi Biggeri
In Italian Universities, the evaluation process started in the early 90s, with D. Lgs. n. 29/93 followed by L. n. 537/93 introducing Internal Evaluation Units (IEUs) in each University and the Observatory for University Evaluation System (UESO) to coordinate them , which was transformed into the National University Evaluation Committee (NUEC) in 1999.
Since the beginning, it was clear that the evaluation of t he teaching ac tivity and its quality had to be conducted inside of a general fram ework of the university educational process, taking into account all the factors that affect t he last o ne’s. In particular in the analysis of the university process of human capital formation emphasis has to be placed on the fact that the stude nt is also a user of the service and, therefore, he plays a more or less active role in carrying out the process and affects the results.
Along these lines, the Law 370/99 foreseen that each university must collect the opinion of the students about the teaching activity implemented in every course programs and the NUEC underlined the importance of using a “minimum set of general questions”, in order to make the methods of collection of the data homogeneous among the Italian universities.
This paper has three main aims. First, an explanation of the place and importance of the students’ opinion surveys and of the evaluation of the quality of teaching on the framework of the evaluation of the educational process and its “accreditation” is pre sented. Second, a description of the differe nt methods fo r the collection of the stu dents’ o pinions implemented by the di fferent faculties and universities is illustrated, in order to verify the homogeneity of the surveys. Proposals co ming from a recent sem inar on this topic organized by the NUEC are also presented. Third, attention is devoted to the analysis of the results of the students’ opinion surveys and to their use f or the improvement of the quality of the teaching activity.
As regard to each objective, the paper point out the issues still open and some possible solutions.
References
Biggeri L. (1999), Autono mia e val utazione dell’insegna mento nel sist ema universitario italiano, in: Atti della Giornata di Studio “L’insegnamento universitario in Italia”, Accademia dei Lincei, Roma.
Biggeri L. (2006), La valutazione del sistem a universitario, in M. Ca rpita, L. D’Ambra. M. Vichi, G . Vittadini, Valutare la qualità, Cap. 6 – pp. 143-167; Guerini e Associati, settembre 2006.
Cnvsu (2010), La valutazione della didattica da parte de gli st udenti: l o svol gimento della rilevazione via w eb, Incontro-Seminario svoltosi a Roma il 23 giugno 2010 (vedi: www.cnvsu.it).
47
Specialised session H
Law system
48
The Italian judicial offices productivity in 130 years of cognition civil procedures
Carlo Cusatelli1 1 Dipartimento di Scienze statistiche “Carlo Cecchi”, Università degli Studi di Bari Aldo Moro, [email protected]
In I taly, on e of th e h inge poin ts o n wh ich the conce pt of State leans - the justice - has slipped in a deep crisis m ore and m ore since remarkable difficulties in its i nternal reorganization are accom panied to the natural pr ocess of re view in the civil society in evolution. The more evident external aspects of such crisis are translated in the slowness of the judicial mechanism, in the high cost of its antiquated procedures and in the difference of the sentences for degrees of judgment and for districts of Appeals Court.
To comfort or to contradict this or that th esis, also, sometimes statistic data are brought in contrast from each other, because of what they define is not well specified. Wanting to give clarity, it is first of all necessary to del imit this analysis to th e procedure of cognition, essential unit of the civil trial activity, for an objective knowledge of the phenomenon from the quantitative po int of view, to be able t o supply stable term s of reference for a better interpretation of the facts and a more serious search of the causes and the effects, reaching a suitable territorial distribution of the enquirer personnel, judging or not.
The analysis of som e statistic indicators (e.g., th e pr ocedures du ration, th e in dex of disposal, the p ercentage vari ation of pen ding) de rived by the data related to supervened, exhaustions and pending al lows to estimate the productivity of t he judicial office s in comparison to the j ustice demand. In the centen nial oscillation of the civil procedures of cognition (and particularly of th e relative qu otients for 100.000 inhabitants), both in first degree a nd appeals, a growth is established, especially in the last twenty-thirty years, of supervened and exhausted procedures, and still more than t hose leani ng t hat am ong the 1991/2000 decade and the average value of the last seven years go over the doubling. The average life of the ci vil procedures in e very degree o f judgment that on the contrary has gone growing since 1881 to t oday, even though with occasional lowering events. The civil procedures of cognition have reached by now the average duration of 3.000 days, and this means around eight years of waiting for the definitive sentence.
Inefficiency of the judicial system in its complex certainly depends on the backwardness of farraginous rules and from the shortage of human and material resources, but also - and perhaps above all - from how much it reveals to us, at last, the different productivity of the 29 Appeals Court districts.
References
Cecchi C. (1975). Analisi statistica dei procedimenti civili di cognizione in Italia, Roma-Bari: Laterza. Corrado S. (1993). Statistica giudiziaria, Rimini: Maggioli. Giacalone M. (2009). Manuale di statistica giudiziaria, Roma: Bel-Ami edizioni. ISTAT (1959/2011). Annuari di statistiche giudiziarie, Roma.
49
Evaluating the Administrative Efficiency of Justice Courts in Italy
L. Antonucci 1, C. Crocetta 2, F. d’Ovidio3, M. Lorizio4 1 Dipartimento di Scienze mediche e del lavoro, Università di Foggia, [email protected] 2 Dipartimento di Scienze economiche, matematiche e statistiche, Università di Foggia, [email protected] 3 Dipartimento di Scienze statistiche “C.Cecchi”, Università degli studi di Bari Aldo Moro, [email protected] 4 Dipartimento delle Scienze giuridiche privatistiche, Università di Foggia, [email protected]
Efficiency and effectivene ss ar e central goals for the adm inistration of j ustice in Italy. Efficiency means economically applying available resources to accomplish statutory goals, while effectiveness refers to carrying out justice system activities with proper regard for equity, proportionality and public safety.
A commonly held view is t hat the j ustice system in Ital y is chaotic and rathe r poorly administered. The justice system is busy, with a very large number of transactions taking place annually.
Unanimous agreement exists that the justice system ought to be efficient, effective, and fair (Cook, 1982; Snipes, 1980, Vito, 1972). Less accord, however, exists about how best to secure these essential qualities or how to measure whether they have been achieved.
One of the main goals of the Italian government is to plan expenditure and measurement activities of the justice system. There is an attempt of determining measurement criteria and performance indicators, according to some best practices implemented by some courts.
In this work, using data collected by Istat and Ministero di giustizia (Lupo, 2011), we propose an aggregate index (Molteni, 1999, Sc iacca, 2007), obtained through a weighted synthesis of standardized ratios which should preserve the relevance of the context.
Those synthetic indicators (Wilodhorn, 1977) can be decomposed in sub index, in order to o btain a dashboard o f i ndexes useful to eval uate the ef ficiency of t he different subsystems.
The ra nking of the Italian justice courts so obtained ca n be use ful to com pare the different departments and Corti d’Appello to reward to the courts which best perform.
A c ritical analysis, from the statistical point of view, finally evaluate s the different alternatives proposed.
References
Cook T.J. et al. (1982) Meas uring Court Performance. Research Triangle Institute, Research Triangle Park, No rth Carolina. USA
Lupo E. (2011) R elazione su ll’amministrazione della Giustizia nell’ anno 2010, http://www.cortedicassazion e.it/ Documenti/Relazione%20anno%20giudiziario%202010.pdf
Molteni M. (1999) Attività e perform ance nelle az iende di servizi alla persona, in E. Go ri e G. Vittadini e ds., Qualità e valutazione nei servizi di pubblica utilità, Etas, Milano, 49-120.
Procura della Repubblica di Bolzano ( 2008) Be st pr actices negli uf fici gi udiziari, http://www.giustizia.it/giustizia/it/mg_2_9_4.wp
Sciacca M. (2007) Gli strumenti di efficienza del sistema giudiziario e l’incidenza della capacità organizzativa de l giudice, in Rivista di Dirittto Processuale, 643-661.
Snipes L.L. et al. (1980) Managing to Reduce Court Delay. National Center for State Courts. Vito D.P.A. (1972) An experiment in the use of court statistics. Judicature 56, (2) August-September. Wilodhorn L., Wilodhorn P. (1977) Indicators of Justice. Rand Corporation Research Study.
50
The participatory evaluation in the analysis of the judicial system: the georeferencing of the opinions
Simone Di Zio1, Antonio Pacinelli2 1 University “G. d’Annunzio”, Chieti-Pescara, and Telematic University “L. da Vinci”, [email protected] 2 University “G. d’Annunzio”, Chieti-Pescara, and Telematic University “L. da Vinci”, [email protected]
The evaluation of the judicial system affects both the ability to make impartial sentences in compliance with the requ irements of substantive law, and the containment of the costs in terms of using public and pr ivate resources. The im partiality of the sentences and the compliance with the substantiv e law are related to the final decision and t he "soc ial sharing", while the other component regards the efficiency of the judicial services.
In this context, the subj ective data is of great importance, both because is the basis of "social sharing", and because it is decisive in cases where objective data are not available.
The e fficiency, in the broadest se nse, is the rationalization and opti mization of available resources in terms of employees, facilities and technology, time and resources. I n this context it is appropriate to distinguish between administrative/organizational efficiency and procedural efficiency (R omei Pasetti, 2 001). The e fficacy concerns the evaluation of final results, in terms of how much the final goal has been reached (Cicala, 2000).
In this c ontext, we co nsider desirable f or t he im mediate future, by m eans of special surveys, the collection of the citizens opinions about the gravity of the offenses, following an a pproach to the policy that p asses thr ough th e surv eying of t he need for justice (Pacinelli, 1995).
As it happens in other areas of the e valuation, for some particular as pects (such as the perceived risk of the crimes) it is also important to take into account the spatial component. Very often a problem is perc eived differently depending on the locati on or spatial cont ext to which it is related. Thus, for example, it is known that the perception of risk of certain crimes, such as robbe ry or t heft, in a qua rter with high crime rate is h igher in respect t o another quarter of the sam e city that is considered more secure. Or, in a large r scale, some cities or regions are considered m ore dangerous than others. In these cases, in addition to the classic tools for opinion survey, it is possible to use other systems that take into account the spatial com ponent, to de tect the views of each person (whether citizen or e xpert) with geographical refere nces: he re we propose the “spatial questionnaire”. T herefore, we c an collect differe nt opi nions o n the sam e subject depending on the locati on to which they relate.
In this way it is possi ble to improve the ex-ante evaluation to identify different degrees of plausibility/feasibility of inte rventions on di fferent places of t he sa me territory or to reconstruct the relevance of the needs and desiderata in a greater spatial detail. This would lead to a diversification of the interventions on t he territory th at, other things being equal, would produc e (ex-post) both an im provement of the effi ciency and efficacy of the interventions and a gathering of end-user satisfaction geographically diversified.
References
Cicala M. (2000) Effettività dei diri tti ed efficacia del le decisioni nell’o rdinamento ita liano di fronte alla sfi da europea, Corriere Giuridico, 6.
Pacinelli A. (1995) Sulla rilevazion e del fabbiso gno di beni e servizi pubblic i, Atti dell’Incontro di Statistica Economica, Napoli: 127-133.
Romei Pasetti M. (2001) L’art. 111 Cost. e il princip io di e fficienza de ll’organizzazione giudiziaria, Giustizi a Civile.
51
Specialised session I
Evaluation of cultural services
52
A Structural Equation Model Proposal for evaluating Visitor Satisfaction at an Exhibition
Gabriele Cantaluppi1, Barbara Bianchi, Domenico Piraina2, Francesca La Placa2 1 Dipartimento di Scienze statistiche, Università Cattolica del Sacro Cuore, Milano, [email protected] 2 Comune di Milano, Servizio Mostre e Coordinamento Attività Espositive - Palazzo Reale, Milano
The main problems in measuring visitor satisfaction in a museum context are related t o the characteristics of intang ible services, in the sense that visitors gi ve only an intri nsic psychological evaluation, having made no experience of use. The concept of loyalty, in its common sense, may seldom be encountered; it is sometimes associated to the concept of intensification, usually measured through the evaluation of the satisfaction in consum ption of extra services – like purchases of guides, publications, souvenirs or presents –, that are assumed to be related to the word of mouth and thus to the confirmation and transfer of the repurchase and revisit intentio n also to other people. Sampling problems may arise: unless tickets are purchased on the web, there is no database of t he visitors, and when it exists it can be of little use, e.g. when visitors are tourists. The sole impressions of people filling the questionnaires are available and not all visitors m ay be interested before they attend the event. A kind of systematic sampling is usually adopted, which does not ensure optimality criteria, though giving a representative sample for the visitors’ population.
Structural Equation Models with Latent Va riables (SEM-LV) have been commonly and successfully applied to Customer Satisfaction (CS) a nalyses. To build a SEM-LV we have first to define a construct consisting of a net of concepts related to CS – which are classified in determinants and consequences of CS – and the causal relationships existing among the involved c oncepts. Here we pr opose a c onstruct to eval uate the satisfaction level for an exhibition: we consider the ‘Information availability’, the ‘Impression at the visitor arrival’, the ‘Logistic aspects’, the ‘Exp ositive Route’ and the ‘Hall Personnel’ as exogenous lat ent variables, the ‘Visitor satisfaction’ and ‘Extra Services’ as endogenous ones.
The parameter estim ates confirm what is reported in the international literature; ‘Expositive Route’ has t he strongest impact on ‘Visitor satisfaction’. The observed socio-demographic variables may be useful to study the different behaviour of clusters of visitors: (e.g., it is possible to create a segmentation based on the visitor art knowledge background).
Future developm ents will regard t he purification of th e scales and t he possibility to include in the construct also the modern vision of the service user: he is considered a co-creator of satisfaction, an aspect assuming great importance for art connoisseurs.
References
de Rojas C., Camarero C. (2008). Visitor’s experience, mood and satisfaction in a heritage context: Evidence from an interpretation center. Tourism Management, 29, 525-537.
Boari G., Cantaluppi G. (2010). Co nstruction of Balanced Scorecards b y using Struct ural Equation Models with latent var iables. Electronic Journal of Applied Statis tical Analysis: De cision Support Syste ms and Services Evaluation EJASA:DSS, 1 (1), 66-78.
Harrison P. , Shaw R. ( 2004). Con sumer Satisfaction a nd Post- purchase I ntentions: An Exploratory Study of Museum Visitors. International Journal of Arts Management, 6 (2), 23-32.
Johnson M.D., Gustafsso n A. (200 0). Improving Cu stomer Satisfaction, Lo yalty and Profit: An Integrat ed Measurement and Management System, San Francisco: Jossey Bass.
Magagnoli U., Cantaluppi G. (2007). Problemi e ruolo dei metodi statistici per la valutazione dei servizi collegati ai sistemi di produzione dei beni. Rivista di Economia e Statistica del Territorio, 3, 183-199.
53
Between accessibility and marketing: online ticketing for entertainment events
Fabrizio Chirico1, Luca Bottini2
1 Dipartimento di Scienze dell’Economia e della Gestione Aziendale, Università Cattolica del Sacro Cuore di Milano, [email protected] 2 Dipartimento di Scienze dell’Economia e della Gestione Aziendale, Università Cattolica del Sacro Cuore di Milano, [email protected]
As part of the larger phenomenon of Internet diffusion and e-commerce development, sale of tickets for entertainments events has gradually grown in the network. The paper aims to assess, through a sur vey, the lev el o f perceived effectiveness of on line sales systems and consumer/audience satisfaction with respec t to these services. In particular, we intend to analyze whether a nd how such se rvices might increase the accessibility of cultural e vents (live concerts, theatrical events, operas, ballets, film shows) for consumer / audience and as such, however, are to ols of a n inte grated m arketing im plementation by e nterprises operating in this sector. With the ter m accessibility we int end to refer to the possi bility of overcoming the ty pical economic, social or cu ltural ba rriers which tend to make a elitist consumption the attendance at cultural eve nts. This issue is of particular interest because the scientific l iterature has i nvestigated many aspects of the e-c ommerce but has not ye t considered the specific case of cultural products.
References
Foglio A. ( 2010), E -commerce e web marketing: str ategie di web marketing e tecniche di vendita i n inter net, Milano Franco Angeli.
Lubbe S., Van Heerden J. M. (2003), The economic and social impacts of e-commerce, Hershey Idea Group. Moraru M., (2008), E-commerce, Romanian Economic and Business Review, 3 (4), 44-50. Friberg R. , Ganslandt M ., Sandstrom M (2001), Pr icing Strategies in E -Commerce: Br icks vs. Clicks, Research
Institute of Industrial Economics Working Papers, n. 559.
54
An Exchange market model to improve the impact of Lyric Opera theatres in the social and working life
Paola Fandella1, Enrico Girardi2 1 Dipartimento di Scienze dell’Economia e della Gestione Aziendale, Università Cattolica del Sacro Cuore, Milano, [email protected] 2 Dipartimento di Scienze dell’Economia e della Gestione Aziendale, Università Cattolica del Sacro Cuore, Milano, [email protected]
The main object of this work is to introduce a possible model describing how an Exchange secondary market for the lyric opera productions may work, in order also to originate a new channel of fu nd raising public or p rivate. In pa rticular, th e model should identify all the auto-financing capabilities and, at the same ti me, define how to m easure the value of the specific “product” offere d by lyric opera theatres: namely the product is not to be considered only an object for a single specific production but also an entity with a value for and in the market, both in a cultural sense and also from a financial point of view. It seems to be clear, fi rst at all, that t his model of market can be constructed only for the tangible asset of the ly ric pr oduction. In othe r w ords, the fundamental topics for this model are basically two: the first one is certainly related to the financial problem that seems to appear as the main problem for the cultural world, not only in our country: the financial constraint require to c ontinuously find new financial sources allowing lyric opera theatres to create new pr oductions; th e second f actor is th at in m any cases th e “o ld” productions – also historical and with artistic value and recognition – are subject to a progre ssive disinvestment program due to th e nu merous ti mes th at on e sin gular theatre has used that specific production and also to the cost of conservation and storage of the inventory related to the production.
There is no model implementing the above issues at the moment, only an hybrid m odel based on the l ogic of c o-production has b een proposed that can be signe d a s a m odel intended to reduce basic and real timing costs of a specific production but do not consider the future benefit to realize the production from an economic point of market.
The model that we pr opose makes it possi ble to define a real market and c ompute the value for thes e pr oductions. In this way it is possible to evaluate: 1) t he m arket artistic value. This is the fi rst time that we can cons ider the tangible asset of a lyr ic production as an artistic m arket product 2) the value of the re-proposi ng the production at other thea tres 3) the value of the possibility of re-proposing to other public institutions.
References
Sicca L.M., Zan L., (2005). Much Ado About Management: Managerial Rhetoric in the Transformation of Italian Opera Houses, International Journal of Arts Management, vol.7, n.3, 46-64.
Financing Opera Europe, Rapporto 2007.
55
Specialised session L
Evaluation of knowledge transmission via web
56
Java visual tools for economic, social and epidemiologic statistical model simulation
Stefano Rosignoli1 1 Tuscany’s Regional Institute for Economic Planning (IRPET), [email protected]
During the last twenty years we have assisted to a constant growth of software for statistical analysis and econometric forecast and simulation. These programs also have been boosted by many additional tools and technical procedures. Moreover, the introduction of WYSIWYG (what you see is what you get) form into these applications has made their access easier also for non-specialists. In spite of this greater s implicity, thes e s oftware need a bi t of learn ing tim e and a m inimum leve l of t echnical knowledge. Th is contribute puts forward so me simple java applications (jav a stand-alon e desktop applications) useful to make simulation in th e economic, social and epidemiologic fields. These applications, which do not require an y previous technical learning, consist of single jar-f iles that do not need to b e ins talled and may be dis tributed as e-m ail attachments or al so b y a cop y-paste operation. They only requir e th e JRE (Java Ru ntime Environment), which is g enerally installed in every computer with internet co nnection. Opening these jar-files (by double-clicking) an extr emely intuitive main form will appear, in which it is possible to change exogenous variables and see directly the corresponding change of end ogenous ones. The underlying simulation model may be complex or not, bu t i ts use for th e sim ulation appl ication will r emain ex tremely sim ple and intui tive. These products could be useful for po licy makers , managers or for o ther categor ies o f users that do not possess a high technical and statis tical knowledge but are no netheless inter ested in using the simulation model. The presentation will show three of these java-tools developed in IRPET. Economic simulation and forecast by Input-Output Model of Tuscany. Simulation tool of a input-
output model (Miller, Blair 2009) to see the evolution of four endogenous variables (growth rate of gdp, regional import, foreign import, and em ployees) by change of the exogenous variables (growth rate of final demand composed b y household exp enditure, governmen t exp enditure, investments, regional and foreign exports).
Simulation of health services related to demographic population structure in Tuscany. Simulation too l to see th e evo lution of government health expenditure and of th e number and type of health services at ever y chang e of the population level and ag e str ucture plus th e evolution of the capital needed to face the request of health services by population.
Simulation of outbreak and spread of infectious diseases in populations. Simulation tool based on a SIR Model (developed first time in 1927 by W. O. Kermack and A. G. McKendrick) to see the evolution of an infectious diseases dep ending to some par ameters (number of susceptible people, number of daily mean contacts per person, infection probability at each contact, starting infected people, share of vaccinated people, incubation period). The tool will show, by graph an table, at each change of parameter, the number of ill people and the incidence of infection.
References
Cay S. Horstmann (San Jose State Univ.) Progettazione del software e design pattern in Java, Apogeo 2004 Kermack W. O., McKendrick A. G. "A Contribution to the Mathematical Theory of Epidemics." Proc. Roy. Soc.
Lond. A 115, 700-721, 1927. Lorenzini S., Patacchini V. “La proiezione della domanda e della spesa sanitaria pubblica in Toscana” in “Toscana
2020. Una regione verso il futuro” , IRPET 2005 Miller R. E., Blair P. D. Input-Output Analysis: Foundations and Exte nsions, 2nd editio n. Cambridge University
Press, 2009 Page B. , Kreutzer W. The Java Sim ulation Handbook: Simulating Discrete Event Syste ms with UML and Ja va,
Shaker Verlag GmbH, Germany 2005
57
Evaluating peculiar lexicon for medical record sections identification
Flora Amato1, Antonino Mazzeo2, Sara Romano3, Sergio Scippacercola4 1 Dipartimento di Informatica e Sistemistica, Università di Napoli “Federico II”, [email protected] 2 Dipartimento di Informatica e Sistemistica, Università di Napoli “Federico II”, [email protected] 3 Dipartimento di Informatica e Sistemistica, Università di Napoli “Federico II”, [email protected] 4 Dipartimento di Informatica e Sistemistica, Università di Napoli “Federico II”, [email protected]
Information and communication technologies applied to health care domain had lead to the so-called Electronic Health (E-Health) with the challenge to provide value-added services, enhancing efficiency and reducing the costs of complex informative systems. Actually, the e-health is going to chang e the interactions between pati ents and healthcare pr oviders. Since many efforts are currently devoted to extract knowled ge from tex ts in order to enh ance some features provided b y sev eral s ystems, knowledge management applications, dealing with acqu iring, maintaining, and accessing knowled ge within da ta, hav e a central ro le in the e-health context. In o rder to im prove the eff ectiveness of medical record management procedures, techn iques for automatic comprehensio n of textual con tent are r equired. The automatic h andling of tex tual contents involves the adop tion of sev eral text-processing disciplines that wo rk considering complex and s trongly in ter-dependent s yntactic, semantic and pr agmatic aspects. In order to ex tract knowledg e from textual me dical records, it is necessary to identify domain relevant terms, their meanings, and the relationships among them.
In general the first activity for knowledge de rivation fro m text has as requirement the identification of the p eculiar le xicon, which is a terminological vocabular y r epresentative of the domain (in our case the medical one). For peculiar le xicon id entification, diff erent kinds of text analysis based o n NLP techniqu es are requir ed. In this f eld, methods adopted are related to cross-disciplinary perspectives including Statistical Linguistics (Balbi et al.), (Bolasco and Pavone), (Lebart et al .) and Co mputational Linguistics (Giovan netti et a l.), whose objec tive is the stud y and the analysis of natural language and its functioning through computational tools and models.
In this work, we propose a methodolog y fo r semi-automatic deriva tion of knowledge fro m medical records by means of both statistical and lexical approaches. Moreover, we propose a statistic-based methodology for the peculiar lexicon extracted quality evaluation. The evaluation is performed by means of a semantic distance, based on 2 statistical measure, between the l exicon extracted and the corpus composed b y th e set of medical record an alyzed. The methodo logy can be used for automatic medical record section s identification in order to improve the in teractions among different actors belonging to the health care domain.
References
Balbi S. , Bolasco S. , Verde R. , (2002) Text Mining on elementary forms in co mplex lexical str uctures", JADT 2002. Actes des 6es Jo urnées int ernationales d’ Analyse statistique des Données T extuelles, Saint- Malo, IRISA, pp. 89-100.
Bolasco S., Pavone P. (2007), Automatic dictionary and rule-based systems for ex-tracting information from text, Classification and Data Analy sis, B ook o f Shor t Pap er, Meeting of th e Classification and Data Analy sis Group of the Italian Statistical Society, 255-258.
Giovannetti E., Marchi S., Monte magni S., Bartolini R . (2008) Ontology Learning a nd Semantic Annotation : a Necessary Symbiosis. LREC 2008: Proceedings of the Sixth International Conference on Language Resources and Evaluation, European Language Resources Association (ELRA), Ma rrakech, Morocco, May 26-1 June 2008.
Lebart L., Salem A., Berry L. (1998), Exploring Textual Data, Kluwer Academic Publishers, Dordrecht, NL.
58
The assessment of knowledge transfer via WEB
Antonino Mazzeo1, Flora Amato2, Sergio Scippacercola3
1 Dipartimento di Informatica e Sistemistica,Università di Napoli “Federico II”, [email protected] 2 Dipartimento di Informatica e Sistemistica, Università di Napoli “Federico II”, [email protected] 3 Dipartimento di Informatica e Sistemistica, Università di Napoli “Federico II”, [email protected]
Nowadays e-learni ng is bec oming increasi ngly im portant in the learni ng e nvironments, thanks t o its undeniable advantages respect to traditional traini ng. Unfortunately the transfer of the knowledge through learning-object in the various environments is missing of reference standards for the assessment.
The propose d indicators (Sc alise, et al., 2006; Selim , 2007) that are use d today are inadequate for a com prehensive assessm ent of the service. The traditional or e-learning training can be evaluated at fo ur progressive levels: 1. Reaction, 2. Learning, 3. Behavior and 4. R esults (Kir kpatrick 1979). I n pa rticular, this work focuses on t he sec ond an d fourth level.
For the second level, it is suggested to ident ify in the learning-obj ect the critical points (Critical Success Factors - CSF) (Champy, Hammer, 1993; Bracchi, et al., 2001; Selim, 2007) that allow a com plete learning. Al so, these CSF indicators sh ould be useful to confirm the validity of learning-object or to direct the eventual reengineering.
For the fourth level, it is nec essary to define an indicator System that is activated each time you start the transfer of knowledge to the user who connects via the web.
Aim of this wor k is the design of a System of simple and complex indicators by means of suitable Algorithms.
The indicators proposed in this paper are of two types: indirect (the critical tasks, the number of accesses to the module, the usage time of
a session, the mode of use, etc..) and direct (the ave rage response tim e to ques tions, the num ber of atte mpts before you
answer correctly ,etc. ..). The system of indicators and the Critical Success Factors identified for learning-object
have been implemented for a first-year University Course. Thus it was possible t o evaluate the succe ss of knowledge tra nsfer by means of t he
learning-objects.
References
Bracchi G., Francalanci C., Motta G. (2001). Sistemi informativi e aziende in rete, McGraw-Hill. Champy J. , Ham mer M. ( 1993). R eengineering the Corporation. A manifesto for busi ness r evolution, Har per
Collins Business, New York. Kirkpatrick, D. (1979). Techniques for evaluating training programs. Training and Development Journal 33(6), 78
– 92. Scalise K., Giffor d B. (2006). Com puter-Based Assessment in E-Learning: A Fr amework for Constr ucting
"Intermediate Co nstraint" Questions and T asks for Technology Platform s, The Jour nal of T echnology, Learning and Assessment, vol. 4, no. 6.
Selim H. M. (2007). Critical succes s factors for e-lear ning acceptance: confirmatory factor models, Conputers & Education, vol.49, 2, 396-413.
59
Contributed Sessions
60
Contributed session 1
Education I
61
A new proposal to assess evaluation models
Paolo Giudici1, Emanuela Raffinetti2 1 University of Pavia, [email protected] 2 University of Pavia, [email protected]
In this paper we introduce a novel multivariate concordance index t hat can be usefully employed to study th e d ependence between a response va riable a nd a number of explanatory ones. In order to achieve this g oal one can re sort to som e s pecific statistical tools such as t he concordance curve and the Lorenz curves. Let us su ppose to ha ve a k -variate ra ndom vector ),...,,( 11 kXXY and let us desc ribe the relationship am ong the response variable Y and the explanatory va riables 11,..., kXX through the multiple linear regression model. More precisely, let us suppose that the response variable Y assumes non-negative values. Furt hermore, this approach will be applied when the m ost relevant explanatory variables have categorical nature and are always characterized by non-negative values representing the corresponding assigned label valu es. Once built the response variable Lorenz curve and its dual (obtained by ordering all the response variable values in a decreasing sense), one proceeds to the concordance curve construction defined as the set
of ordered pairs ))/1(,/(1
*
i
jjY ynMni where *
iy is the Y variable values ordered according
to the ranks assigned t o th eir respective esti mates. Th is curve m oves betwee n t he Y Lorenz curve and its dual. A multivariate concordance index can be provided:
,))/(1(/
))/(1(/
1
1 1
*)(
1
1 1
*
,...,, 11
n
i
i
jjY
n
i
i
jjY
XXY
ynMni
ynMni
Ck
where )( jy represents the first j , with ij ,...,1 , Y values ordered i n a n increasing
sense. T his in dex ca n be very usef ul as a measure o f fi t whe n the rel evant e xplanatory variables have categorical nature because it is based on the response variable values ordered according to the ra nks assigned to t heir corresponding estimated values rather tha n on the euclidean distance. An application in the evaluation context will be developed.
References
Giudici P., Raffinetti E . (2011). Multiv ariate Rank s-based conc ordance indexes. Published in the volu me of selected papers “Statistical Methods for the analysis of large data-sets”. Springer
62
Impact evaluation of University grants
Maria Luisa Maitino1, Nicola Sciclone2 1 IRPET, [email protected] 2 IRPET, [email protected]
Is the scholarships an effective tool to pr event stud ents f rom d ropping out of higher education? The paper deals with the im pact evaluation of university scholarships financed by the Tuscany Region during the last years.
The data used in the empirical analysis are drawn from the three Tuscany universities – Florence, Pisa and Siena-, and collect all the individual data of the freshmens from 2000 to 2008.
We used seve ral statistical methodologies (probit models, pr opensity score m atching, regression-discontinuity design) to compare the performances of students who obtained the scholarship with the other ones.
The analysis shows how the scholarships reduce drop-outs for two cohorts of freshmen and encourage to their graduation. Scholarships’ ef fects may otherwise vary according to the uni versity and to t he am ounts, m oreover som e scholarshi p‘s am ounts are to o lo w to prevent low-income students from dropping out.
References
Catalano G., Biggeri L. (2006), L’e fficacia delle politiche di sostegno agli studenti univ ersitari: l’esperi enza italiana nel panorama internazionale, Il Mulino, Bologna
Mealli F., Mele S., Rampichini C., Sciclone N. (2006), “ La valutazione di efficacia del sistema universitario”, in Biggeri L., Catalano G. (a cura di), L’efficacia delle politiche di sostegno agli studenti,, Il Mulino, Bologna
Rosenbaum P. R., Rubin D. B. (1983), “The central role of the propensity score in observational studies for causal effect”, Biometrika, n. 70, pp. 41-55
63
Differential variability of test scores among schools: a multilevel analysis of the 5th grade Invalsi test using heteroschedastic random effects
Claudia Sani1, Leonardo Grilli1 1 Dipartimento di Statistica “G. Parenti”, Università di Firenze, [email protected]
The performance of a sc hool system is usually evaluated through the learning levels of the pupils, focusing on school av erage scores. The vari ability of the scores across schools i s rarely studied in detail, though it is a cruci al aspect: especially in prim ary schools, a low variability acr oss schools helps to guarantee equal rights. To investigate the patterns variability in Italy, we anal yse data from In valsi, t he Italian national institute for the evaluation of the school system, which regularly carries out standardized tests to assess the learning levels of pupils at v arious grades. We consider the mathematics test ad ministered to 5th grade pupils at th e end of the 20 08/2009 year, along with a pup il's questionnaire for measuring s ocio-economic f actors. The sa mple includes abo ut 1 000 s chools a nd 40000 pupils (Invalsi 2009).
The analysis is performed using a two-level linear model (Hox 2010) on the Rasch score of the mathematics test, with pupil-level errors depending on gender and school-level errors depending on the ge ographical area. T he model include s seve ral s ocio-demographic an d economic covariates and som e context ual co variates obtained as sc hool m eans of pupil variables. The results are essentially in line with the literature and the expectations based on the knowledge of the It alian situation. Nevert heless, th e increase in the variance am ong schools w hen goi ng f rom Nort h to So uth is as tonishing, pointin g o ut a serious iss ue of segregation in Southern Italy. Specifically, the South/Isles area (Basilicata, Calabria, Sicilia and Sardegna) has a low m ean score and a hig h between-school standard deviation (0.486 versus 0.197 of the North-West area), whereas the South area (Abruzzo, Molise, Campania and Puglia) has a mean score similar to Northern Italy but a hu ge between-school standard deviation (0.652). As a consequence, the Southern regions have both the best and the worst schools in Italy.
References
Hox J. (2010). Multilevel Analysis: Techniques and Applications (Second Edition), New York: Routledge. Invalsi ( 2009). S ervizio Nazionale di Valutazione a.s. 2008- 2009. R ilevazione degli a pprendimenti. Scu ola
primaria. Sintesi rapporto. Downloadable from http://www.invalsi.it.
64
University careers evolution. A multistate modeling for a perspective study of the Italian situation
Matilde Bini1, Bruno Monastero2, Margherita Velucchi3 1 European University of Rome, [email protected] 2 Politecnico di Torino, [email protected] 3 New York University, [email protected]
University teachers a re a fundam ental component for t he research and teaching acti vities development. For this reason it is important for the Ministry of research and policy makers of the Universities to adopt strategic policies of recruitment and retirement of the academic personnel in order to improve the turn-over, to satisfy the needs of planning of research and teaching activities, as well as to improve the efficiency of the related economic activities. In this work we propose a depiction of the evolution of the Italian academic staff from 1988 to 2008 years, and provide a pe rspective of this evolution for the ne xt five years. To d o that we implement a multistate modeling using information such as sex, age, length of service, job position, academic branch and salaries, provided from Ministry of Research and Italian Consortium of Universities (CINECA).
References
Caswell H. (2001), Matrix Population Models. Sinauer Associates Inc., Sunderland, MA 01375 USA. CHEPS, Center f or Higher Education Policy Studie s ( 2001), Academ ic car eer: a co mparative per spective,
http://www.utwente.nl/cheps/publications/. Comitato Nazionale per la Valutazio ne del Sistema Universitario, (2007), Ottavo rapporto sullo stato del Siste ma
Universitario, Ministero dell’istruzione, dell’università e della ricerca, Roma. Consales B. , De Rosa C. , Ger li S. , M inieri S. , ( 2006), Manuale di L egislazione uni versitaria, Si mone E dizioni
Giuridiche, Napoli. Luenberger D. G. (1979), Introduction to dynamic systems theory. Models & Applications, Wiley.
65
Contributed session 2
Economics and labour market
66
A Statistical Analysis of Accidents at Work and Sectoral Performance in the European Economy
Luigi D’Ambra1, Antonio Frenda2 1 Università degli Studi di Napoli “Federico II”, [email protected] 2 Italian National Statistical Institute (ISTAT) - University of Bergamo, [email protected]; [email protected]
Our research describes safety at wo rk as an i ssue to be addressed in terms of legislation , with due statistical kn owledge o f the phenomenon. T hrough t he a pplication of m ultiple indicators while investigating the rate of homogeneity and no n-homogeneity of available data, especially at in ternational level, the co ncept at study can be analyzed using specific scientific methods.
The Total Fre quency Rate of accidents could be influe nced by a country’s industrial structure, so as the adde d value growth rate (gross domestic product). The standardization of industrial structures into NACE divisions or sub-divisions (and not only on an aggregate activity level), has proved very useful as an integral part of the statistical infrastructure used within the European statisti cal syste m for producing co mparable statistics. Mo reover, thanks to its relation to t he International Standard Industrial Classification of all Economic Activities this fram ework i s also an i mportant tool for com paring statist ical dat a on economic activities at international level.
Looking at the distribution of the Total Freq uency Rates of accidents in differe nt world regions, t he picture is q uite different, as th e p henomenon is by n o m eans e venly sp read across the globe. Fatalities are proportionately much higher in some regions than i n others. Carrying o ut a cou ntry-by-country analy sis wo uld in no doubt re veal g reater variations. Occupational accidents and work-related diseases in some European countries are twice as high as in some others, while in the Middle East and Asia, these phenomena are the biggest component as fatality rates rise four tim es high er tha n thos e in the sa fest industrialized countries.
In Italy, h owever, tha nks to progressive im provements in the last thirty years, i njury levels have fal len considerably below the European a verage. However, the im pact of an excessive economy and the influence of organized crime in the South, which could alter the basis for re porting work-related accidents and occupational safety performance indicators, has led to the consideration of the actual rate of accident s in Italy slightly higher tha n the European average.
Finally, we highlight that, in the European economy, total accident rate is influenced by sectoral breakdown of gross value added. It is relevant to analyze the relationship between black econom y a mplitude and the rate of inju ries: in Italy it is well above t he national average for all regions of the South. Within the entire Central and Northern regions (with the exception of Lazio and Piedmont), it is below national average. Southern regions having a higher level of ir regularity based on estimates by ISTAT, should, logically, have hig her accident rates, or at least e qual to that of the northe rn regions, which ha ve with l ower prevalence of underground economy. Sinc e the data show the opposite, the presence of a large num ber of accide nts not reporte d could be easily identifie d in these abnorm alities, especially in Calabria, Sicily and Campania.
In Europe t he existence of signi ficant underground ec onomies and new areas t o investigate, pa rticularly in Greece, Hungary, Lithuania, Rom ania, Sout hern Italy, Spain, explains the need to make Eurostat indicators more comparable.
67
A composite index used for measuring intellectual capital of SMEs from Romania
Bogdan Ileanu1, Alexandru Isaic-Maniu2 1Academy of Economic Studies of Bucharest, [email protected] 2 Academy of Economic Studies of Bucharest, [email protected]
The aim of the pape r is to determ ine a com posite index used to m easure the intellectual capital of Romanian enterprises. The composite index is determined testing more statistical methods such as: DEA, C onjoint Analysis, Partial Least Squa res or Principal Component Analysis combined with Classical multiple regression model(PCA&CMRM) or Structural Equation Modelling (SEM).
The study continues the previous researches made in this domain. Statistical techniques are applied on a larg e set of indicators proposed in pre vious approaches. All the analyses were made on a representative sample of SMEs from Romania.
Our database contains 10-15 indicators used to measure intellectual capital. From all the methods mentioned we decided to a pply and compare the results o nly for PCA&CMRM and SEM because due to the context of data availability and nature of indicators it seem s that these methods are the most appropriate.
Applying these methods we are trying to find out if there is a method which can indicate how to obtain the final composite indicator for measuring Intellectual capital.
Our st udies re veal that none o f this m ethod is best. T he su ggestions a re to u se both methods because: PC A m ethod is m ore a dvantageous when the number of i ndicators is small and it shows the natural association of components. In most of the cases the Principal components are ove rlapping the principal components of IC-inde x found in the literature . The p roblem of this m ethod is that th e ne w com ponents achie ved are losi ng the practical/economical significance and in most of the cases the standardization is required.
SEM modeling is more com plex an d the re ar e m any m athematical restrictions. Since we, are not able to choose m ultiple variable or large sam ples because of multiple reasons related mostly of the economic context of SMES this method is hard to be applied. Even if it is hard to be applied it has a big advantage compared with PCA method. SEM modeling shows in str uctural f orm or in re duced form the natural expressi on and the latent correspondence betwee n the vari ables as t hey are in the eco nomy or as they cannot be measured in the reality (but could be seen in the reduced form).
In conclusion we recommend in the first steps a PCA analysis in order to see the natural associations of the factors and then SEM m odeling starting from the ideas achieved after PCA results.
68
The analysis of informal economy and its implication on the services sector. The case of Romania as a country in transition
Tudorel Andrei1, Marius Profiroiu2, Andreea Iluzia Iacob3 1 University of Economics, Bucharest, Romania, [email protected] 2 University of Economics, Bucharest, Romania, [email protected] 3 University of Economics, Bucharest, Romania, [email protected]
In t his study we had t he goal to estim ate the inform al econom y of Romania using a monetary model. The o bjectives f ollowed by this method were t o k nown better t he dimension of the informal economy during 2000 and 2010 and to see th e impact of it on other m acroeconomic variables. A special a ttention was given fo r the evaluation of the informal economy of the services sector. This importance was given for the services sector because it is the m ost im portant sector on R omanian GDP value formation and an estimation of i nformal economy, without a cohe rent analysis of t his sector, is inadequate. The large weight of its co ntribution, the heterogeneity and complexity of this sector m akes the analysis more difficult. For these purposes we u sed specific econometr ical techniques such as: Granger causality, VAR etc.
69
Combining Statistical and Algorithmic Models for Latent Variables Analysis: A Look at the Fairness of Work
Maurizio Carpita1, Marika Vezzoli2 1 Department of Quantitative Methods, University of Brescia, Italy, [email protected] 2 Department of Quantitative Methods, University of Brescia, Italy, [email protected]
The fairness is a psychological construct for asse ssing the quality of work that has a key role in the workers well-being (Tortia, 2008; Jones and Martens, 2009). Statistically, it can be represented with som e latent variables to be in ferred by s pecific questio ns to the workers which are next projected onto homogeneous groups of items (Likert-type scales).
In this st udy, using a com prehensive Italian social cooperatives workers data collected in 2007, t he so called ICSI 2007 (Carpita, 2009), we combine statistical models, i.e. Rating Scale Model (Andrich, 1978; Brentari and Golia, 2008) and CatPC A (Michailidis and de Leeuw, 1998; Meulman et. al, 2004) with a data mining algorithm, such as Random Forest (Breiman, 2001), to inspect the w ork fairness. T he reas on o f our proposal, which m ixes together two very diffe rent statistical ap proaches, is because when combining t he two classes o f models we offer more insights than using them sepa rately. Indeed, on the one hand, using R ating Scale M odel a nd C atPCA w e first con struct s ome work intensity measures for grouping the subjects within three homogeneous clusters ( low, medium and high work intensity) and then identify which aspects of fair ness are m ore or less easy to endorse. On t he other ha nd, R andom Fore sts an d, i n particular, its va riable im portance indicator (C arpita and Z uccolotto, 2007), identifies which drivers of the fairness have a strong impact on the overall distributive and procedural fairness items.
For the overall distributive fairness, o ur re sults sho w tha t for lo w an d medium wo rk intensity, the drivers having the m ajor importance are Responsibility, Effort and Training (with t rivial shift on t heir order), w hile f or people wit h high work intensity, the most important driver is the Economic resources owned by the cooperative. For th e overall procedural fairness, th e Respect is the m ost i mportant driver for all the work i ntensity categories, however for the m edium and hig h cl usters the difference relative to the remaining drivers, having approximately the same importance, appears to be significant.
References
Andrich D. (1978). A rating scale formulation for ordered response categories, Psychometrika, 43, 561-573. Brentari E., Golia S. (2008). Measuring job satisfaction in the Social Service Sector with the Rasch Model, Journal
of Applied Measurement, 9, 1, 45-56. Breiman, L. (2001). Random forests, Machine Learning, 45, 5-32. Carpita M. (Ed.) (2009). La qualità de l lavoro nelle cooperative sociali. Mi sure e modelli statistici, Mil ano,
FrancoAngeli. Carpita M ., Z uccolotto P. ( 2007). Mining the dr ivers of job satis faction using algor ithmic var iable im portance
measures, in Me todi, Modelli, e Tecnologie dell ’Informazione a Supporto delle Decisioni, Vol. I: Metodologie, eds. L. D’Ambra, P. Rostirolla, M. Squillante, Milano, FrancoAngeli, 63-70.
Jones D.A., Martens M.L. (2009). The mediating role of overall fairness and the moderating role of trust certainty in justice-crite ria relationships: the f ormation and us e of f airness heuristics in the work place, Journal of Organizational Behavior, 30, 8, 1025-1051.
Meulman J. J., Van der Kooij A. J., Heiser , W.J. ( 2004). Principal Co mponent Analy sis with nonli near opt imal scaling transformations for ordinal and nominal data. In D. Kaplan (ed.), The Sage Handbook of Quantitative Methodology for the Social Sciences. Sage, London.
Michailidis G., de Leeuw J. (1998). The Gifi sy stem of descriptive multivariate analysis, Statistical Science, 13, 307-336.
Tortia E. ( 2008). Worker Satisfaction and Per ceived Fa irness in Public and Non- Profit Or ganizations: Sur vey-Based Findings from Italy, Journal of Socio-Economics, vol. 37, 2080-2094.
70
Contributed session 3
Latent variable models
71
A Latent Class Approach for Estimating Labour Market Mobility in the Presence of Multiple Indicators and Retrospective Interrogation
Francesca Bassi1, Marcel Croon2, Arianna Pittarello3 1 University of Padova, Italy, [email protected] 2 University of Tilburg, The Netherlands M.A:[email protected], 3 University of Padova, Italy, [email protected]
With panel data analy sts can estim ate labour force gross flo ws – i.e., transitions in time between different states.
Measurement (classification) errors in the observed state can induce substantial bias in the estimation of gross flows, thus leading to er roneous conclusions about labour market dynamics (Bound, Brown, & Mathiowetz, 2001).
A large body of literature on gross flows estimation is based on the assum ption that errors a re unc orrelated ove r time. According to that assu mption, classifi cation errors produce spurious transitions and consequently induce overestimation of changes.
However, the independent classification errors (ICE) assumption is not realistic in many contexts, because of survey design and data collection strategies. This is especially relevant when panel data are collected by retrospective interrogation (Tourangeau, Rips, & Rasinski, 2000).
We use a m odel-based approach to a djusting o bserved gr oss flo ws for classification errors, e ventually correlated . A conv enient fr amework fo r formulating our m odel is provided by latent class analysis, specifi cally latent cl ass Markov models (Bassi & Trivellato, 2009).
We apply our approach to data collected on the Italian labour market from January 2004 to Octo ber 2007 with th e C ontinuous Qu arterly Labou r Force Su rvey, wh ich is cross-sectional with a 2-2-2 rotating design yielding two-wave panels one quarter, three quarters and one year apart. The survey collects information about lab our market participation on a sample of respondents from the resident non-institutional population.
The questionnaire allows to dispose of multiple indicators of labour force condition for each quarter: (i) each respondent is classified as employed, unemployed or out of the labour market according to the de finition of the In ternational Labour Office on the ba ses of answers given to a group of questions (ii) each respondent is aske d to classify himself as employed, unemployed or out of the labour market, the so-called self-perceived c ondition; and (iii) a retrospective questi on asks about condition in t he labour market one year before the interview.
Our approach provides a means to esti mate labour market mobility taking into account correlated measurement errors and the rotating design of the survey. Special attention will be given to the validity of the retrospective questions in relation to the actual measurements made at each time point.
References
Bassi F. , T rivellato U. ( 2009). A latent class appr oach for estimating gr oss fl ows in t he p resence of cor related classification errors, in P. Lynn ED Methodology of Longitudinal Surveys, Chichester, Wiley, 367- 380.
Duncan G. J., Kalton, G. ( 1987). I ssues of desig n and an alysis of sur veys acr oss tim e. I nternational Statistical Review, 55, 97-117.
Tourangeau R. , Rips L .J., Rasinski K. A. (2000). T he psy chology of s urvey response. Cam bridge: Cam bridge University Press.
72
Latent growth models with multiple indicators: a longitudinal analysis of student ratings
Leonardo Grilli1, Roberta Varriale2 1 Department of Statistics “G.Parenti”, University of Florence, [email protected] 2 Department of Statistics “G.Parenti”, University of Florence, [email protected]
Latent Growth Curve (LGC) models aim at modelling change across time. While traditional LGC models are based on a single observed indicator, we focus on a multivariate extension, namely a LGC model with multiple indicators for modelling change across ti me of a laten t factor which is measured at different occasions by multiple items. This model is also known as ‘second-order LGC’ or ‘curve-of-factors’, see McAr dle ( 1988), Mer edith an d Tisak (1990), Muthén (2004), Bollen and Curran (2006), Steele (2008).
When fitting LGC m odels with m ultiple indicators we need t o account for both t he interrelationships of the observe d va riables (indicat ors) within eac h occasion a nd t he interrelationships of the same indicator ac ross occasions in order to measure change in the latent variable (facto r) ac ross ti me. In this work, we consider a widely used form : the structural model specifying t hat the latent variable grows accordi ng to a ra ndom slope linear m odel, com bined wit h a measurement model speci fying t hat the latent variable is measured at each occasion by a conventional factor model with time-invariant loadings.
The specificat ion of a m ultiple-indicator LGC m odel i nvolves several interrelated choices. In particular, the features of the structural model, such as the functional form of the growth, a re li nked t o the f eatures of the m easurement m odel, suc h as the c orrelation structure across time of the measurement errors. In t his work, we investigate the empirical implications o f different s pecification st rategies through an application to the ch ange of student satisfaction about univ ersity courses. Speci fically, we analy se stude nt ratings collected in four academic years over the period 2005-2008, concerning 380 courses of the faculty of Economics of the University of Florence.
References
Bollen K.A., Curran P.J. (2006) Latent Curve Models: A Structural Equation Perspective, Hoboken, N.J. : Wiley. McArdle, J. J. (1988). Dynamic but structural equation modeling of repeated measures data. In J. R. Nesselroade
& R.B. Cattell (Eds.), Handbook o f multivariate experimental psychology (2nd ed., pp. 561–614). New York: Plenum.
Meredith, W., Tisak, J. (1990). Latent curve analysis. Psychometrika, 55, 107–122. Muthén B. (2004). Latent variable analysis: Growth mixture modeling and related techniques for longitudinal data.
In: Handbook of quantitative methodology for the so cial sciences. Ne wbury Park, CA: Sage Publications, 345-368.
Steele F. (2008). Multilevel models for longitudinal data. Journal of the Royal Statistical Society A, 171 (1), 5-19.
73
SEM and IRM procedures to assess the relationship between latent traits
Anna Simonetto1 1 Department of Quantitative Methods, University of Brescia, [email protected]
The prese nt work is part of the measurement of subjecti ve perceptions, using statisti cal models to understand their dependencies.
The aim of this study is to analyze the relationship between the l atent varia bles (estimated through m easures af fected by m easurement errors) th at und erlie m any of th e observed socio-economic phenomena.
Being able to measure these relationships is very important in practical applications: for example to evaluate custom er or job satisfa ction we ca n directly measure them through a battery of ite ms, but we h ave to kee p i n m ind that the assessm ent sho uld be str ongly influenced by other latent variables, such as motivations or expectations.
We will focus on bias and st andard error of som e parameters est imators of the regression models with variables affected by m easurement errors and we m ake a comparative analysis of the two different approaches: the Item Response Model (IRM) and the Structural Equation Model (SEM).
The com parison between the tw o proposed procedures will state that the two-step procedure (based on IRM), a t the cost of a s mall loss in efficiency and accuracy, allows more flexibility and provides som e useful i ndications about the reliability of obtai ned measures.
References
Andrich D. (1978). A rating formulation for ordered response categories. Psychometrika, 43, 561-73. Carpita M. (2009). La Qualità del Lavor o nelle Coo perative Sociali - Misure e Modelli Statistici (a cura di),
Milano: FrancoAngeli. Carroll, R.J., Ruppert, D., Stefanski, L.A. and Crainiceanu, C.M. (2006). Measurement Error in Nonlinear Models:
A Modern Perspective. 2nd Edition. Chapman and Hall CRC Press. Fuller, W. A. (1987). Measurement Error Models. New York: Wiley. Gibbons R. D. , B ock R. D., Hedeker D. , Weiss D. J ., Se gawa E., Bhau mik D. K. , Kupfer D. J. , Fr ank E. ,
Grochocinski V. J. , Stover A. ( 2007). Full- Information Item Bifactor Analysis of Gr aded Response Data. Applied Psychological Measurement; 31(1): 4 - 19.
Linacre J.M. (1997). KR-20 / Cronbach Alpha or Rasch Reliability: Which Tells the "Truth"? Rasch Measurement Transactions; 11:3 p. 580-1 1997 .
Muthén, B. (1984). A gener al structural equation model with dichoto mous, ordered categorical, and continuous latent variable indicators; Psychometrika 49, 115-132.
Samejima, F. (1969). Esti mation of latent ability us ing a response pattern of graded s cores. Psy chometrika Monograph, 17.
74
Latent Markov models from a potential outcome prospective for causal inference in dynamic settings
Francesco Bartolucci1, Fulvia Pennoni2, Giorgio Vittadini3 1 Department of Economics, Finance and Statistics, University of Perugia, [email protected] 2 Department of Statistics, University of Milano-Bicocca, [email protected] 3 Department of Quantitative Methods for Business and Economic Sciences, University of Milano-Bicocca, [email protected]
The latent M arkov m odel was pr oposed by Wi ggins ( 1973) as a to ol f or the a nalysis of longitudinal c ategorical data . The basic a ssumption of this m odel is that the response variables are co nditionally in dependent given an u nobservable Markov ch ain, wh ich represents the evolution of the individual characteristic of interest.
The initial version of the latent Markov m odel has been extended in several ways; for a review see Bartolucci, Fa rcomeni & Pe nnoni (2010). One of the most recent extensions is to multilevel data; this extension was dealt with by Bartolucci, Pennoni & Vittadini (2010), who exploited it to set up an educational evaluation system.
In this paper, we ref ormulate the la tent M arkov m odel from a potential outcom e perspective, so as to better justify the u se of t his m odel in ev aluation c ontexts. Our approach follows the sam e lines as that of Hec kman (2010), who, i n a more general context, s hows the e quivalence between eco nometric structural m odels and potentia l outcome models (R ubin, 1974). Note t hat, o nly rece ntly, both ap proaches have been extended to longitudinal contexts (Gill and Robins, 2001; Abbring & Van Den Berg, 2003), when the effect of repeated treatments over time is of interest.
From the m ethodological point o f view, the proposed causal f ormulation of the laten t Markov m odel is base d on s ome assum ptions o f c onditional i ndependence between the potential outcomes and the treatment variables, given the occasion-specific latent variables. Similar assu mptions were adop ted in the approaches for cau sal infere nce of B artolucci (2010).
We illustrate the proposed framework through an application to a dataset concerning the labor m arket history of y oung in dividuals residents in the Lom bardy R egion. I n this application, the interest is in assessing how university education impacts on job careers.
References
Abbring, J. and Van den Berg, G. (2003). The nonparametric identification of treatment effects in duration models. Econometrica, 71, 1491-1517.
Bartolucci, F. (2010). On the conditional logistic estimator in two-arm experimental studies with non-co mpliance and before-after binary outcomes. Statistics in Medicine, 29, 1411-1429.
Bartolucci, F. , Farcomeni, A. and Pennoni, F. ( 2010). An over view of latent M arkov models for longitudi nal categorical data. Techinical Report Arxiv 1003.2804.
Bartolucci, F., Pennoni, F. and Vittadini, G. (2010). Assessment of school performance through a multilevel latent Markov Rasch model. Journal of Educational and Behavioral Statistics, in press.
Heckman, J. J. ( 2010). Building br idges between str uctural and pr ogram evaluation appr oaches to evaluating policy. Journal of Economic Literature, 48, 356-398.
Gill, R.D. and Robins, J.M. (2001). Causal inference for complex longitudinal data: the continuous case. Annals of Statistics, 29, 1785–1811.
Rubin, D.B. (1974). Estimating causal effects of tr eatments in randomized and nonrandomized studies. Journal of Educational Psychology, 66, 688-701.
Wiggins, L.M. (1973). Panel analysis: latent probability models for attitude and behavious processes. Amsterdam: Elsevier.
75
Contributed session 4
Education II
76
The Course organizational structure as a determinant of academic success. Some evidences from Padova University
Renata Clerici1, Anna Giraldo2, Elisa Visentin3 1 Dipartimento di Scienze Statistiche, Università di Padova, [email protected] 2 Dipartimento di Scienze Statistiche, Università di Padova, [email protected] 3 Dipartimento di Scienze Statistiche, Università di Padova, [email protected]
University drop-out is one of the major problems in Italian University. According to data of MIUR/CNVSU (2011), in 2008/0 9 only 6 0% of st udents ha d a regula r career in tertiary first degree level of education and 18% of students withdraw from university after the first year. The perc entage o f dr op-out was quite high be fore the Uni versity ref orm of 2 001. Drop-out had a te mporary sm all decrease afte r 200 1 but, at p resent, th e p roportion of students who enter tertiary education without o btaining a f irst degree is st ill below OECD and EU19 average (OECD, 2010).
In the literature some authors present studies that aim to detect the subjective factors of academic failure / success (se e for example MIUR/CNVSU, 2010; Cingano and Cipollone, 2007). Furthermore, some studies analyse the data at most at the Faculty level or separately for 1st and 2nd cycle degree.
In this contribution we present an a nalysis by cohort of students and by courses using data from the administrative archives of Pa dova University in a l ongitudinal perspective. The goal is t o e xamine the influe nce of different o rganizational characteristics of the didactic paths on educational failure / success.
We examine the trends of th ree specific phenom ena: withdraw, change of course and delay, referring to five cohorts of students (2001/02-2005/06) enrolled in 84 undergraduates courses (ex D.M. 509/1999). We carry out the analysis in two steps.
First of all we use Multiple-decrement life tables (Garcia, 1994) to describe, by means of survival rates and cumulative decrement rates, levels of withdraw, change of course and delay, to obtain a segmentation of the courses in homogenous groups.
Then we analyse, by m eans of hie rarchical r egression m odels, th e effects of context variables that, with personal characteristics, infl uence t he diffe rent phenom ena we are studying. In this way we can in tegrate in an interpretative frame the dimensions that affect withdraw, change and delay. According to the emerging typology we can propose specific intervention strategies to contrast academic failure.
References
Cingano F., Cipollone P. (2007). University drop-out: The case of I taly, Temi di discussione del Ser vizio Studi, 626, Roma: Banca d’Italia.
Garcia P. (1994). Predicting College Enrollement. Results from a variant of the L ife Table, in Kintner H.J. et al. , Demographics. A Casebook for Business and Government, Boulder, San Francisco, Oxford: Westview Press 307-326.
MIUR/CNVSU (2010).Vincoli dei si stemi formativi e socia li che pr ovocano abbandoni e r itardi all’Università e possibili interventi risolutivi, www.cnvsu.it.
MIUR/CNVSU (2011). Undicesimo rapporto sullo stato del sistema universitario, www.cnvsu.it. OECD (2010), Education at a glance 2010. OECD Indicators, www.oecd.org.
77
Measures for Ph.D. Evaluation: the Recruitment Phase
A. D’Agostino1, G. Ghellini 2, L. Neri3 1 Dipartimento Statistica e Matematica per la Ricerca Economica, Università Napoli “Parthenope” [email protected] 2 Dipartimento Economia Politica, Università degli Studi di Siena, [email protected] 3 Dipartimento Economia Politica, Università degli Studi di Siena, [email protected]
In the last years the quality of Higher Education (HE) system and its ev aluation have been key issues of the political and scientific debate on education policies all over Europe. Apart from the individual political orientation, some of the several themes discussed are crit ical for the renewal of t he enti re ed ucation s ystem purs uing the c ontinuous im provement perspective, both on efficiency and efficacy. Following this perspective the measurement of the statu s quo of th e HE Italian system ta kes an ev en greater im portance. Th erefore no t only the concepts and the overall set of the assessment measures should be reviewed but it should be als o desi gned a cohe rent in formation syste m able to provide the needed information for implementing Quality Assurance (QA) for HE and its components.
In the wide landscape that involves the entire HE system we draw attention on t he third level of the organization of it, i. e. the Ph.D., that re presents one of the crucial aspect of the overall education process. Indeed the Ph.D. can be viewed as a production process whose task is to respond to the increasing n eed of k nowledge in all econom ic, social and institutional activities. This statement lead us to recognize the importance of monitoring the whole Ph.D. syste m from the recruitment to the placement phase t hrough the traini ng process. In this paper we focus our attention just on the former phase whose main aspects are Ph.D. attractiveness and selection policies.
Referring to the Ph.D. attractiveness, it is crucial to define a basic core of measures able to classify each Ph.D. course or school. We propose some indicators on attractiveness for quantifying an d qu alifying app licants. Instead, consi dering the selection policy the measures considere d are principally related to transpa rency, fair an d consistency of the defined institutional guidelines and objectives.
The empirical analysis refers to t he XXV cohort of Ph.D. schools of the Univer sity of Siena. It is bas ed just on administrative data that allow us to compute the over-m entioned measures for e ach Ph.D. schools. The n comparing these indicators we are able to provide useful tools for detecting weakness and strength of each Ph.D. and to aggregate the results at scientific area level. Th e preliminary results show ex cellences am ong Ph.D. sc hools which seem to be reasonable and in particular useful for the QA process.
References
ENQA, European Standard and Guidelines, 2006, www.enqa.eu/pubs_esg.lasso Ghellini G., Neri L., D’Agostino A., (2009). Towards a longitudinal survey design for Ph.D. evaluation, Quaderni
di Statistica, vol.11, 2009, 127-143. Phillips J.J., Stone R.D. (2002). How to Measure Training Results, New York: MacGraw-Hill.
78
Student Satisfaction Indicators: a Delphi approach
Laura Antonucci1, Corrado Crocetta2, Patrizia Soleti3, Ernesto Toma4 1 Dip. di Scienze mediche e del lavoro, Univ. degli studi di Foggia, [email protected] 2 Dipartimento di Scienze economiche, matematiche e statistiche, Univ. degli studi di Foggia, [email protected] 3 Ufficio di supporto al Nucleo di Valutazione, Univ. degli studi di Bari Aldo Moro, [email protected] 4 Dipartimento di Scienze statistiche “C.Cecchi”, Univ. degli studi di Bari Aldo Moro, [email protected]
In the last decad es the demand for information about student satisfaction has con siderably increased. A large par t of the information available about th e efficiency and the eff ectiveness of universities is referred to data collected on objective bases, but there is an increasing need for subjective data and in particular of student satisfaction judgments.
There is a stron g interest in m easuring the re putation of a univ ersity (Giuditta and Costabile, 2006; Iezzi, 2005; Milioli and Zani, 2003). This kind of information is usually difficult to collect, but our proposal is devoted to build a monitoring system very cheap, extensive and reliable.
Our proposal can be implemented by a university interested in monitoring the student satisfaction about same aspect as: informa tion and communication techn ology available, librar y resources, accommodation facilities offered b y the cam pus or the cit y, l aboratories, cl assrooms, parking and public transport ation ava ilable, feeling with a cademic inst itutions, pla cement o f graduate students, quality of life in the area, efficiency and effectiveness of services offered and international relations.
In order to build a system of monitoring reliable and inexpensive it is possible to follows a Delphi like approach. As known the Delphi m ethod is an interac tive forecasting method which relies on a panel o f exp erts (Dalkey and Helmer, 1963 ). In th e classical version , th e exp erts answ er questionnaires in two or m ore rounds. At the end of each round, a fac ilitator provides an anonymous summary of the experts’ forecasts from the prev ious round as well as the reasons they provided for their judgments. Thus, experts are encouraged to revise their earlier answers in lig ht of the r eplies of other members of their pan el. I t is expect ed that during this pr ocess the range of the answers will decrease and the group will converge towards the "correct" answer (Pacinelli, 2008).
We will use a version of the Delphi method adapted for web interviews. The student’s opinion can be collected by using a CAWI method (Biffignandi and Pratesi, 2003) which is able to split the different aspects considered in many questionnaires that can be randomly administered to the students at the time of b ooking exams. The monitoring system can be administered d irectly by the university that wi ll adapt the exam reservation system to t he needs of the monitoring system. Data co llection will take place during the year. The data will be published in an annual report and in the un iversity website. It will be possible to do some online queries to obtain the information requested. The annual report will contain dashboard ind icators useful to compare the student satisfaction with the d ifferent aspects considered. The procedure proposed is inexpensive, easy to deploy and manage. It provides a continuous and reliable monitoring system.
References
Biffignandi S., Pratesi M. (2003). Te mpestività e qualità: as petti concettuali e co mportamenti di risposta nelle indagini via Internet. Un’applicazione ad un’indagine regionale sulle imprese. In: S. Biffignandi S. e G. Gozzi (eds.) Qualità e Informazione Statistico Economica Territoriale: Aspetti del Processo di Formazione dei Dati e delle Metodologie di Analisi. F. Angeli. Milano.
Dalkey N. C ., Helm er O. ( 1963). An E xperimental Application o f the Delphi M ethod t o the Use o f E xperts. Management Science, 9: 458-467.
Giuditta A. and Costabile M. (2006). L’orienta mento ai clienti dell e università rifles sioni s ulla custom er satisfaction delle imprese che utilizzano ALMALAUREA. In: Consorzio Interuniversitario ALMALAUREA, VIII Rapporto sulla condizione occupazionale dei laureati – I laureati di primo livello alla prova del lavor o. Il Mulino, Bologna.
Iezzi D.F. (2005). A new method t o measure the quality on teaching evaluation of the university sy stem: the Italian case. Social Indicators Reasearch, Vol. 73, 3: 459-477
Milioli M. A., Zani S. (2000). Analisi della “student satisfaction" nella Facoltà di Economia di Parma. In Atti della XL Riunione Italiana di Statistica, Processi e metodi statistici di valutazione. Firenze: 701-704.
Pacinelli A. (2008). Metodi per la ricerca sociale partecipata, Franco Angeli, Milano.
79
A two level structural equation model for evaluating the external effectiveness of PhD
Lucio Masserini1 1 University of Pisa, [email protected]
PhD is t he t hird level of university educati on and is the highest degree of specialization offered by the universities for academic and research careers. In recent years the number of PhDs in Italy has grown significantly and purpose of PhD program have expanded from the traditional ones. The analysis of t he contribution of PhD for em ployment is an im portant tool f or e valuating the quality and ef fectiveness o f d octoral programs. For t his reas on, knowledge of the em ployment status and c areer of PhD graduates becomes essential and can help to reduce the gap between academia and labor market. The aim of this paper is to build a tw o level str uctural equation m odel w ith latent variables to assess the e xternal effectiveness of P hD. T he analysis is pe rformed usin g data f rom the resea rch "C urrent situation and employment prospects o f PhDs", commissioned by National Committee for the Eval uation of the University Syste m ( CNVSU) to th e De partment of Statistics "G. Parenti" of the University of Florence. The proposed measure of "external effectiveness" is a latent variable obtained by evaluating the level of satisfaction with the employment status of doctoral stu dents who graduated in 200 8. Th e opinion was expr essed one year after graduation on a ten orde red p oint scale. Ex ternal effectiveness indicators use d are "consistency with studies", "use of acquired skills" and "meet the cultural interests".
References
Asparouhov, T. e Muthén, B. (2007) . Computationally efficient estimation of multilevel high-dimensional latent variable models. In Proceedings of the J oint Sta tistical Meeting in Salt Lake City. ASA Section on Biometrics.
Chiandotto, B. (2004). Sulla misura della qualità della formazione universitaria. Studi e Note di Economia. vol. 3. Kaplan, D. (2009). Structural equation modelling (2nd ed.). Thousand Oakes, CA: Sage. Muthén, B. ( 1984). A gener al str uctural equation model with dichot omous, or dered catego rical and contin uous
latent variable indicators. Psychometrika, 49, 115-132.
80
Contributed session 5
Health and social services
81
Text-mining: an application for classifying pathology reports
Claudio Sacchettini1, Bruno Bertaccini2 1 ISPO, Istituto per lo Studio e la Prevenzione Oncologica, [email protected] 2 Department of Statistics “G. Parenti”, University of Florence, [email protected]
This work presents a text-minig application for the automatic classification of the pathology reports related to oncological diseases. The reports belong to the laboratories of analyses of the T uscan r egion a nd are very heterogeneous, depending both o n t he tum our ty pe and location and on the subj ective ability in the description of t he analysis results. The technique used to perform the classification is n amed “text categorization” and belongs to the “machine learning” algorithms. Starting from a sample of reports already classified, this technique of t ext mining allows t o train a classifier able to assign ne w reports to cla sses defined on the tumour location.
References
De Bruijn L. M.; Verheijen E.; van Nes F. L.; Arends J. W. (1996). Assigning snomed codes to natur al language pathology reports. Proc. of Medical Informatics Europe, Copenaghen, pp. 198-202
Feldman R.; Sanger J. (2007). The text mining handbook. Cambridge University Press, Cambridge. Joachims T. (2002). Learing to classify text using support vetor machines. Kluwer Academic Publisher, Norwell.
82
NANOVA as a new tool for the evaluation of health services
Gabriella Milone1, Luigi D’Ambra2 1 Università degli Studi di Napoli “Federico II”, [email protected] 2 Università degli Studi di Napoli “Federico II”, [email protected]
In a context of eval uation of ser vices da ta are often expressed as nominal. Nominal responses are rated the natural way of expressing opinions or actions.
A definition of variance for nominal data was first proposed by Gini (1939) In the past, since the nominal responses do not generate numerical data, they have been
underutilized in behavi oural research. This is the case in which nom inal responses are elicited, the responses are customarily aggregated over people or trials so that large-sample statistics can be employed (Keppel, 1991).
This p aper present an innov ative an alysis th at d irectly asso ciates d ifferences am ong responses with particular sources in factorial designs.
In particular, in this pa per we consider data from a surve y on the e valuation of health services. The collected data are nominal expression levels of judgment on the evaluation of the hospitals. The importance of data processing and are rated the performance of a newly developed methodology the analysis of variance for nominal data, NANOVA (Weiss, 2009) that will be applied to the opinions expressed in the evaluation of hospitals in Naples.
In these paper the design presented is characterized by: 9 hospitals, each with a score in 5 domains. T he n ull hy potheses w ould be that neither hospital (9 le vels) nor do main ( 5 levels) affects the numbers obtained. The analysis set it up as a 2-fact or, repeated-measures design with hospitals taking on the role of subjects.
The projected actions, reported nominally, were analyzed with the NANOVA computer program (Weiss, 2009).
References
Gini, C. (1939). Variabilità e concentrazione: Vol. 1 di. Memorie di metodologia statistica. Milano: Giuffrè. Keppel, G. (1991). Design and analysis: A researcher’s handbook. Upper Saddle River, NJ: Prentice Hall. Weiss D.J. (2009). Nominal analysis of “variance”, Cambridge: Behavior Research Methods, 901-908.
83
Domiciliary assistance satisfaction among aged and disabled beneficiaries: a Rasch analysis
Annalina Sarra1 1 Department of Quantitative Methods and Economic Theory, University “G. d’Annunzio”, Pescara, [email protected]
In the E uropean Union countries we assist to a gradual but stable inc rease of the aged segment in the society compos ition. Local governm ents have undertaken program s to improve the quality of life of elderly and disabled. Accordingly, policy makers are focusing more on quality of assistance for social services.
An im portant com ponent of th e quality of care is satisf action with services. This satisfaction is related to how bene ficiaries experience the received assistance com pared to their expectation. Ge nerally, measuring sat isfaction with home based a ssistance is not an easy task. First of all reliabl e measures and data on quality for domiciliary assistance ar e not readily available. In add ition, devel oping m easures of qu ality for home assistance i s difficult partly because of the special characteristics of the service.
In this study we want to measure home assistance satisfaction among aged and disabled people by developing an appropriate questionnaire. We also aim at identifying aspects of the social service achievi ng a low quality score. These characteristics might require immediate action if they w ere give n a r ank of prim ary im portance. In this respect, a questionnaire i s administrated to a sam ple of 117 elderly and disabled people with home assistance services, provi ded by social services of eight municipalities of Pescara (Italy) district. The questionnaire is composed of 34 items related to various aspects of domiciliary assistance, grouped according to six differe nt dimensions. The importance and satisfac tion of these quality aspects is measured by four-point Likert type items.
To comply wi th the aim s of this study, the Rasch m odel is e mployed as a statistical, appropriate tool for calibrating the questionnaire itself. As known, the Rasch model has two notable advantages over the tr aditional methods: the resulting measure is on interval scale and the extent to which data fit the model is assessed. He nce through the Rasch analysis, we are able, i n this st udy, to evaluate the items unidimensionality for each dimension and the model fit, to estimate the items and subjects parameters as well as the thresholds values. Finally, in order to obtain a “strengths and weakness” analysis, the Rasch item evaluation in each dimension is crossed with the importance level, also required in the questionnaire.
References
De Battisti F., Nicolini G., Salini (2005). The Rasch model to measure the service quality. The Journal of Services Marketing, 3 (3), 58-80.
Fisher G.H., Molenaar I.W(1995) Rasch Models: Fou ndations, Recent Developments and Applications, Springe r Verlag, Berlin
RUMM2010 (2001) Rasch Unidimensional Measuremnt Models Manual, Rumm Laboratory Pty Ltd.
84
Performance assessment in healthcare as a management tool
Sabina Nuti1 1 Direttore Laboratorio Management e Sanità, Scuola Superiore Sant’Anna, [email protected]
Performance evaluatio n allo ws p olicy to be linke d t o management and t o o rient output results in order to achieve outcomes. In the public healt hcare system s multidimensional performance evaluation systems and be nchmarking analysis are dem onstrated to be useful tools to link performance measurement to performance management.
The paper reports the effects of the application of performance evaluation system first in Tuscany Region and then in other eight Italian regions, as a tool to support decision makers in public health care systems.
Methodology: a multidimensional performance evaluation system has been extensively and systematically used since 2005 within Tuscan health care sector both at regional and local level. This system is based on 130 indicators classified in six dimensions: population health status; capacity to pursu e regional strategies; quality ; patient satisfaction; staff satisfaction; efficiency and financial pe rformance. T hese i ndicators, routinely collected, allow benchmarking analysis across all the health care organizations.
With years im provements were ac hieved i n m ost of the m onitored indicators of the performance evaluation syste m and the high variability of performance across HAs has been reduced allowing a more e quitable system. M oreover, a f urther evide nce o f the importance o f usin g performance evaluat ion was t he presence o f a negative c orrelation between overa ll perform ance and costs tha t i ndicates that decision m akers should focus their actions on improving quality, effectiveness, and efficacy in order to reduce costs.
Practice implications: the performance evaluation system, implemented first in Tuscany, then in other eight Italian regions and now at the national level on behalf of the health care ministry, has become a public policy tool that helps, on one side, the regional government to eval uate its strategic actions and, on th e other, to promote a “managed” com petition among the HAs. It has been demonstrated to be helpful to enhance innovation and improve results, to increase efficiency, effectiveness and cost containment.
85
Contributed session 6
Quality and risk
86
Statistical Analysis of the Perceived Quality and Customer Satisfaction of a ski school: the Sesto Survey 2010
Stefano Bonnini1, Luigi Salmaso2, Francesca Solmi3 1 Dipartimento di Economia, Università di Ferrara, [email protected] 2 Dipartimento di Tecnica e Gestione dei Sistemi Industriali, Università di Padova, [email protected] 3 Dipartimento di Scienze Statistiche, Università di Padova, [email protected]
For m anufacturers or se rvice pr oviders who wis h to prove their ca pacity of pr oviding a product that meets the customer needs and also want to increase t he customer satisfaction, the quality monitoring and the statistical analys is of data are needful. In particular the quality system of a serv ice provider needs to measure the quality of process and of service through customer satisfaction surveys.
This is the case of the ski schools of A lto Adige w hich pr omoted a pilot survey on customer satisfaction, performed in 20 10 at the Ski School of Sest o-Bozen (north of Italy near the border with Austria). Specifically the parents of young children under the a ge of 13, w ho p articipated in s ki co urses o rganized in Sesto, were as ked t o a nswer a questionnaire to express their level of satisf action about som e aspects of the service. The data processing is mainly aimed to two goals:
1. To calculate a global index of quality, as synthesis of the custom er satisfaction for the various evaluated aspects;
2. To estimate the degree of “feeling” toward the service and the degree of uncertainty of the res pondents an d to detect if a nd how the pe rsonal characteristics of the customers can affect t hese t wo psy chological com ponents, acc ording t o the idea that customer satisfaction can measure the “perceived” quality of the service.
To perform p oint (1) t he NPC m ethodology wa s a pplied s o t hat pa rtial and global complex indicators can be calculated for each customer segment (Arboretti et al., 2007). To study the psychological mechanism (feeling and uncertainty) on which the choices of the respondents are based (point (2)), the t heory of the CUB model and a suitable test o n the covariates of the model can to be applied (Bonnini et al., 2011).
Acknowledgements
Authors wish to thank the University of Padova (CPDA092350/09) and the Italian Ministry for U niversity and R esearch (2 008WKHJPK/002) for providing the financial sup port f or this research.
References
Arboretti G R, Bonnini S, Salmaso L (2007). A performance indicator for multivariate data, Quaderni di Statistica, 9, 1-29
Bonnini S, Piccolo D, Salm aso L , Solm i F ( 2011). Perm utation infe rence for a class of mixture models, Communication in Statistics: Theory and Mehods, accepted for publication
87
Risk Profile using Rasch Analysis
Valeria Caviezel1, Sergio Ortobelli2, Lucio Bertoli Barsotti3
1 University of Bergamo, [email protected] 2 University of Bergamo, [email protected] 3 University of Bergamo, [email protected]
In this paper we propose an evaluation of the investors’ risk profile in order to account the minimal requi rements that It alian financial ins titutions must satisfy by law (d. lgs. 164, 2007). Therefore we investigate all the aspe cts characterizing the so called risk profile: the investor’s knowledg e and h is f inancial exp erience ( about f inancial in struments an d th eir use); the fi nancial obj ectives, the tem poral horizon and th e personal predisposition to risk/earn.
The methodology to deal the ri sk profile in financial literature is essentiall y based on basic statistics that do not consider an y psychological asp ect. In order to acc ount the investor preferences and his psychological attitude we propose a Rasch analysis (Fischer & Molenaar, 1994) characterizing the main features of the individual risk profile. In particular we identify a questionnaire whos e item s desc ribe different c haracteristics of t he l atent variable (predisposition to risk/earn). From a first Rasch analysis – conducted using RUMM 2020 (Andrich et al., 2003) - we could observe that there exist at least three latent variables that characteri ze the risk profile. T hus, we su ggest to a nalyze: the investors’ financial knowledge (financial products, in stitutions, etc.), the investor s’ te mporal horizon and the personal aversion to risk us ing a Generalized Multidim ensional Rasch Model with wit hin-items multidimensionality - that m ay be tr eated as an instance of a Multidi mensional Random C oefficient M ultinomial Logit M odel ( Adams et al., 1997). These analyses are conducted using ConQuest (Wu et al., 2007).
The Rasch analysis applied to results of the test permits t o better define the invest or’s risk profile. In particular, given the m ultivariate position of each investor with respects the three latent traits (personal knowledge, risk predisposition and tem poral horizon) we can represent his position with respect the possible investments proposed from the bank. Therefore opportunely rescaling the t hree dimensions we can prospect different situations that respect the investors risk profile characterizing better the typical investor choice.
References
Adams R.J., Wilson M., Wang W. (1997). The Multidimensional Random Coefficients Multinomial Logit Model. Applied Psychological Measurement, 21, 1-23.
Andrich D., Lyne A., Sheridan B., Lou G. (2003). RUMM 2020 [Computer Software], Perth, Australia: RUMM Laboratiry.
Fischer G.H., Molenaar I.W. (Eds.) (1994). Rasch Models: Foundations, Recent Developments and Applications, New York: Springer-Verlag.
Wu M .L., Adam s R. J., Wilson M .R., Haldane S.A. (2007). ConQuest: Gener alized I tem Response M odeling Software, ACER Press.
88
Safety at work and industrial accidents in Europe: an efficiency analysis
Eugenia Nissi1, Agnese Rapposelli2 1 Università G. D’Annunzio” di Chieti-Pescara, [email protected] 2 Università G. D’Annunzio” di Chieti-Pescara, [email protected]
Safety and health at wor k is no w o ne of t he most important and m ost hig hly devel oped aspects o f Eu ropean Union’s policy on em ployment and social affairs. For t he European Union, in fact, the im provement of working conditions a nd t he pre vention of workplace accidents are amongst the primary objectives to pursue, as stipulated in the Treaty of Rome (article 1 36) a nd confirmed by the F ramework Dir ective 8 9/391. A gain at the Lis bon European C ouncil in Mar ch 2 000, th e objective that the European Union set itself was “creating more and better jobs” (Commission of the European Communities, 2002).
In this c ontext, the present pa per exam ines t he performance of fifteen E uropean countries with respect to t he number of industrial accident. To this purpose, we apply the non-parametric approac h to efficiency m easurement, rep resented by D ata Envelo pment Analysis (DEA), to European data for the year 2005.
However, while in traditional DEA m odels we have two categories o f factors (inputs and outputs), now we consider a third kind of factor, an undesirable output, represented by the num ber of accident at work . The ordinary effi ciency measures are not suitable in contexts where at least one of the varia bles that have to be radially contracted or expanded is not a “good”. In the standard DEA models, decreases in outputs are not allowed and only inputs are allowed to decrea se (sim ilarly, increases in inputs ar e no t allo wed an d only outputs are allowed to increase). Hence, our objective is to adapt the DEA technique to the problem at hand, where outputs do not refer only to goods, but we also consider undesirable outputs. To ef fect ran kings, we pr opose theref ore a ne w m odel ty pe of D EA, where undesirable and desirable outputs will be treated differently.
References
Banker R.D., Charnes A. , Cooper W.W. (1984). Some Models for Estimating Technical and Scale I nefficiencies in Data Envelopment Analysis. Management Science, 30, 1078-1092.
Charnes A. , Cooper W.W., Rhodes E . ( 1978). Measuring the E fficiency of Decision M aking Units. E uropean Journal of Operational Research, 2, 429-444.
Coli M., Nissi E., Rapposelli A. (2 008). Performance measurement by means of Data En velopment Analysis: a new fr ontier for undesir able outputs . I n: Mantr i J.K. (Eds.), Research methodology on Data E nvelopment Analysis, Boca Raton, FL: Universal-Publishers.
Cooper W .W., Seifor d L .M., T one K. ( 2000). Data Envelo pment Analysis: a co mprehensive text with models, applications, references and DEA-Solver software, Boston: Kluwer Academic Publishers.
Scheel H. (2001). Undesirable Outputs in E fficiency Valuations. European Journal of Operational Research, 132, 400-410.
89
The front-office services provided by some municipalities of the area of Florence: An evaluation of the quality from the supply side
Daniele Vignoli1, Bruno Bertaccini2, Ciro Annicchiarico3
1Department of Statistics “G. Parenti”, University of Florence, [email protected] 2Department of Statistics “G. Parenti”, University of Florence 3Statistical Office, Florence
The objective of this work is to highlight the efficiency of the front-office services provided by the “o pen office for citizenship” (Uffici Relazioni con il Pubblico, URP) and from the “population registry system ” (A nagrafe) i n som e municipalities of the Joint Statistica l Office of the Area Florence (Ufficio di Statistica Associato) for which the Statistical Office of Florence serves as a coor dination office. In pa rticular, the selected municipalities are Calenzano, Impruneta, Scandicci, and Sesto Fiorentino. The study is conducted under the project Q UA.SER, financed by t he T uscany re gion, th at has the scope to establi sh a common set of qualitative and quantitative indicators of cust omer satisfaction am ong the public administrations that joined the project.
In this vain, we conducted two focus groups with the aim of finding words and contents that can be helpful to describe the peculiarities of t he front office services. We studied the text pr oduced alon g the f ocus g roups th rough text ual analysis statistical techniques, following a mixed method approach(i.e., quantitative-qualitative).
The findings are in line with standard litera ture on local welfare-state systems. In the analyzed text we clearly found the classica l dichotomization between demand and supply, complemented by a subjective dim ension. Interestingly, the persons tha t joined the focus groups (th at b y defi nition represent the sid e of the supply) tend to see them selves as a citizen that make potential usage of such services (and so from the side of the demand).
Our results gave us useful inputs in order to p repare a s pecific ad hoc questionnaire t o evaluate the efficiency of suc h front-office services among the resident population of each municipality.
References
Bolasco S. (1999 ): Analisi multidimensionale dei dati. Metodi, strat egie e criteri d’int erpretazione, Ro ma, Carocci.
Giuliano L. (2004), L’analisi automatica dei dati testuali. Software e istruzioni per l’uso, Milano, LED.
90
Contributed session 7
Transports
91
Consumers’ satisfaction with railway transport: a Bayesian Network approach.
Giovanni Perucca1, Silvia Salini2 1 Department of Economics, Business and Statistics, [email protected] 2 Department of Economics, Business and Statistics, [email protected]
Background. In the last years the i nterest for c onsumers’ expe rience with Services of General Interest (SGI) largely increased. Several surveys have been conducted in order to observe how consumers’ satisfaction differs across EU countries. Judgements are based on personal perceptions and evaluations which, in tur n, depend on several unobservable and subjective fact ors, such as indi vidual cha racteristics, group-specific fe atures a nd social norms, as pointed out by several contributions (Bertrand and Mullainathan, 2001).
Research motivation. So me stu dies (Fiorio and Flo rio 2010) alr eady fo cused on th e connections between consum ers’ satisfaction with SGI, resp ondents’ characte ristics and socio-economic indicators. This literature us ually makes use of ec onometric m odels for categorical dependent variable. Our work tries to analyse the same issue through Bayesian Networks (Kenett and Salini 2009). Compared with other procedures, this approach allows to remark any interdependency between the variables included in the model. The objective of this work consists in checking the consistency of our results with previous findings.
The data. We use Eurobarometer data, and in pa rticular the results from three surveys conducted in 2000, 200 02 an d 2004 abo ut SGI . Th ese surveys wer e sp onsored by th e European Commission for monitoring and planning purposes and record individual level of satisfaction wi th SGI, com bined with s ome inform ation ab out res pondents’ in dividual characteristics. Among all SGI we focus on railway transport. Moreover, we include in our network some macroeconomic variables, in order to und erline any link between the socio-economic environment on satisfaction.
Method. In t his application we use the pa ckage R and the library bnlearn (Sc utari, 2010). Compared with other softwa re, the a dvantage of using this package relies in the possibility to perform both constrained-based and score-based m ethods on our sam ple, testing several and alternative algorithms.
Results. O ur results sh ow how networks’ estim ations slightly differ based o n the learning algorithm applied. C oncerning individual characteristics, most of the findings are consistent wit h the results pointed out by pr evious lit erature. T he i nclusion of socio -economic indicators is appa rently m ore com plex and highlights the l imitations of t his approach when dealing with mixed datasets.
References
Marianne Ber trand M ., M ullainathan S. ( 2001). Do P eople Mean What They Say? I mplications for Subjec tive Survey Data, The American Economic Review, 91-2, 67-72.
Fiorio C. , Flor io M. ( 2010). Fior io, C. V. and Flor io, M . (2010), «Would y ou say that the pr ice y ou pay for electricity is fair?» Consumers' satisfaction and utility reforms in the EU15, Energy Economics, forthcoming.
Salini S., Kenett R. (2009). Bayesian networks of customer satisfaction survey data, Journal of Applied Statistics, 36-11, 1177-1189.
Scutari M. (2010), Learning Bayesian Networks with the bnlearn R Package, Journal of Statistical Software, 35-3.
92
An overall passenger satisfaction measure through a Structural equation model with high-order latent variables
Enrico Ciavolino1, Mariangela Nitti2
1 Dipartimento di Filosofia e Scienze Sociali, Università del Salento, Lecce, [email protected] 2 Dip. di Scienze Pedagogiche, Psicologiche e Didattiche, Univ. del Salento, Lecce, [email protected]
The aim of th is pape r is to propose a ne w con cept of global m easure for the Passenger Satisfaction (PS), by considering the approach of the high order latent variables (Henseler et al., 2010).
The paper will introduce the idea of the high order Latent Variable (LV) in the context of the Structural Equation Models (SEMs), by considering the Partial Least Squares (PLS) estimation method. We will present the t wo main approaches consi dered in literature: the repeated indicator and the two-step approach, showing the advantage and the disadvantage of choosing one method rather than the other.
After the presentation of the statistical method, we p ropose a new para digm to define the global measure of PS by using an high order LVs with Path PLS. The global measure of PS, rather than being conceived as one LV directly measured by manifest variables i n the questionnaire, is here modelled as the meaning underlying all the dimensions of satisfaction measured by the questionnaire (ie. Tangibility, Reliabil ity, Responsiveness, Assurance and Empathy, for the SERVQUAL approach. Parasuraman et al., 1985).
We will com pare som e PS surveys kno wn in literature (Camminatiello et al., 2 010; Gallo et al., 2009; Ciavolino et al., 2010), where the PS model is defined via SEM and the global measure is identified by a specific or derived LV.
The com parative analysis, between the high order model and t he PS m odels, will be made by evaluating the accuracy of the LVs estimated and the efficiency of the parameters, drawing some conclusions on the empirical results obtained.
Moreover, the results of a simulation study, aimed at evaluating the effect of a reflective and formative measurement and the relationships between the first and high order LVs, will be shown and commented.
References
Camminatiello I., D’Ambra L. (2010). Visualization of the signi ficant explicative categories using CATANOVA method and No n-simmetrical Cor respondence analy sis fo r evaluation o f passe nger satisf action. Jour nal of Applied Quantitative Method, 5 (1), 64–72.
Ciavolino E., Nitti M. (2010). High-order constructs for the structural equation model. In: V meeting on dynamics of social and eco nomic s ystems, DYSES 2010. Book of the abstr act, Fifth I nternational W orkshop on Dynamics of Social and Economical Systems.
Gallo M., Ciavolino E. (2009). Multivariate statistical approaches for the customer satisfaction into tr ansportation sector. Global & Local Economic Review. Vol. 13, Issue 2.
Henseler J., Chin W.W. ( 2010). A Comparison of Appr oaches for the Analy sis of I nteraction E ffects Between Latent Variables Using Partial Least Squares Path Modeling. Structural Equation Modeling, 17, 82–109.
Parasuraman A., Zeithaml V. and Berry L. (1985). A conceptual model of service quality and its implications for future research. Journal of Marketing, 49, 41–50.
93
Passenger satisfaction: a multi-group analysis
Laura Antonucci1, Corrado Crocetta2, Francesco D. d’Ovidio3, Ernesto Toma4 1Università degli studi di Foggia, [email protected] 2Università degli studi di Foggia, [email protected] 3Università degli studi di Bari Aldo Moro, [email protected] 4Università degli studi di Bari Aldo Moro, [email protected]
To analyse the level of passenger satisfaction of a public local transport service a PLS Path Modelling (Lohmöller, 1989; Tenenhaus et al., 2005) was adopted.
The results obtained were used in a classific ation tree model (CHAID algorithm, Kass, 1980) t o detect whic h as pects most influen ce the passenger satisfactio n, or better , which latent variables can be used to segment the travellers population in subgroups.
The main goal of this paper is to verify the opportunity of matching the PLS results with a CHIAD algorithm and to veri fy, through a multi-group anal ysis, the invariance of t he classification rules with respect to some variables.
In the m odel that we hypoth esized, the passenger satisfaction is the result of m any unobservable factors. The influence of each factor was estimated by using a PLS algorithm.
The passengers are found t o be very sensitive to the level of the service organization and to t he way the service is delivered (punct uality and regularity of the buses and short waiting time).
The safety and reliability of buses, th e level of com fort and cleanness and the professionalism and co urtesy of staff had, also, a big weight to determinate the passenger satisfaction.
Applying a se gmentation an alysis, such la tent variables were used to detect the best classification of passe nger satisfaction, i.e. which latent factors are better able to segm ent the travellers i nto different subgroups in term s of satisfaction. It has been marked that the results of the two procedures do no t o verlap, r ather th ey are com plementary in the characterization of the passenger satisfaction.
The organization of the service was found the best splitte r variable, divi ding the population into three subgroups: very unsatisfied, medium satisfied an d very satisfied. I n this way it wa s possible to s tudy the behaviour of the subgroups in order to discover the reason of the satisfaction or of the dissatisfaction and to improve the level of the service.
References
Kass G. V. (1980). An Exploratory Technique for Inv estigating Large Quantitie s of Categ orical Data, Journal of Applied Statistics, Vol. 29, 2: 119-127.
Lohmoller J.-B. (1989). Latent Variables Path Modeling with Partial Least Squares, Physica-Verlag, Heildelberg. Tenehaus M., Esposito Vinzi V. , Chatelin Y. , Lauro C. (2005). PLS path modeling, Computational Statistics &
Data Analysis, 48: 159-205.
94
Using the Disco index for the determination of the Passenger Satisfaction
Biagio Simonetti1, Antonio Lucadamo2 1 Dipartimento di Analisi dei Sistemi Economici e Sociali, Università del Sannio, [email protected] 2 Dipartimento di Analisi dei Sistemi Economici e Sociali, Università del Sannio, [email protected]
In the local public transport t he monitoring of the opinions expressed from the passe ngers on the quality of the service offered is of st rategic relevance. The passenger satisfaction is , in fact, very important to keep the m arket, to increase the use of the public transportation system, to attract new travell ers, to improve the image of the company in the public and to ensure financial results in the future years.
Obviously the m easurement of the satisfa ction is c omplex because of its subjective nature and the presence of other factors that are not always easily separable.
The global satisfaction of the users of a se rvice is dete rmined by different factors that influence the judgement (acco rding to Ze ithmal, Parasuram an and Berry, 1991, fi ve categories – assurance, em pathy, tangi bles, reliability and responsiveness – affect the perception of t he quality by the use rs). Of course, each researcher modifies the classical questions and the categories according to the aim of his survey, but the fact that the overall quality is influenced by many features of the services is incontrovertible.
The bu ilding of a g ood questionnaire is then of fundam ental importance to detect the satisfaction that a user, in our case a passenger, perceive by the service. At the same time it is i mportant t o use the ri ght statistical t echniques to measure the s atisfaction a nd to underline the varia bles that have a bi gger we ight in the determ ination of the pe rceived quality. Many techniques have bee n used duri ng the y ears to reach this purpose, with benefits and pitfalls.
In this paper we desc ribe as the disco index introduced by Raveh in 1983, developed subsequently ( Guttman, 1 988; Rav eh, 1989 ), and app lied pr evalently in th e con text o f discriminant analysis (Choul akian et al., 2001; Sim onetti et al. 2003) can be an im portant instrument that can be use d by the research er to discover if som e variables have more importance in the determ ination of the overa ll satisfaction for the pa ssenger of a local public transport firm. The disco coefficient will be used in a di scriminant analysis and in a logistic regression model in which the dependent variable will be the passenger satisfaction (satisfied or not satisfied) a nd the indepe ndent variab les will be some features of the service.
References
Arbolino R., Si monetti B., (2007). La misura della P assenger Satisfaction con l’ausili o d i Tecniche Statisti che Multivariate.
Choulakian V., J. Almhana, (2001). An Algorithm for Nonmetric Discriminant Analysis. Computational Statistics and Data Analysis Journal, Jan. 2001, 253-264.
Guttman L., (1988). Eta, disco, odisco, and F. Psychometrika, 53, 393-405. Raveh A., (1983). Preference structure analysis: A non metric approach. Pattern Recognition 16 (2), 253-259. Raveh A., (1989). A non metric approach to linear discriminant analysis. J. Amer. Statist. Assoc. 84, 176-183. Simonetti B., Choulakian V. (2003). Disc riminat anal ysis for spectrosc opic data. Relazio ne invitata, Att i del
Convegno intermedio SiS 2 003: Analisi Statistica Multivar iata per le Scienze Econo mico-Sociali, le Scienz e Naturali e la Tecnologia, Napoli.
Zeihtaml A., Parasuraman A., Berry L.L. (1991). Servire Qualità, Milano., Mc Graw – Hill.
95
Contributed session 8
Latent variable models and customer satisfaction
96
PLS models: importance of the stability during the time, how was insured in the case history of Hera
Giovanni Monaco1, Fabio Marotta2, Roberto Riccardi3 , Stefano Baldassini4 1 [email protected] 2 [email protected] 3 [email protected] 4 [email protected]
Hera C omm, multiutility le ader i n water, en ergy and environmental services based in Bologna, monitors its clients satisfaction using a P LS models developed with CFI Group Italia since 2005. The multivariate model built for Hera is very complex and peculiar due to latent variables with different number of respondents. To compare scores and impacts along the years is ve ry important they are stable a nd consistent. The causal models are processed using PLS (Partial Least S quare) algorit hm that ensures stable and consistent re sults throughout the years. The article shows how the m odel has been built and which statistical parameters have to be considered to obtain stable and consistent models.
References
Fornell C. and Cha J. (1994). Partial Least Square, Richard Bagozzi (Ed.), Advanced Methods of Marketing, pp.52-78.
Cronbach, L. J. (1951). Coefficient alpha and the internal structure of tests. Psychometrika, 16, 297-334. Fornell C. et alii, 1996. The American Customer Satisfaction Index: Nature, Purpose and Findings. Journal of
Marketing October.
97
The customer satisfaction in INAIL
Rosa Maria Lacquaniti1, Maria Cristina Paoletti2 1Dipartimento di Statistica, SAPIENZA Università di Roma ,[email protected] 2INAIL- Consulenza Statistico Attuariale, [email protected]
The Institute has always been in the forefront of adopting innovative services to its users and, therefore, careful to maintain a high level of quality.
This continuing resear ch aimed to raise the quality pe rceived by users is a challenge increasingly difficult with increasing complexity of the system (channels detection, increase in services, integration and cooperation with other agencies, etc.).
Every year the Institute is responsible for the periodic survey of customer satisfaction. The survey is one of the tools to measure the de gree of satisfaction compared to the
service provided for the improvement of quality services. Starting in 2010, according to new regulations, the surve y has also bec ome a tool for
evaluating staff Inail and this was extended to all the offices. The Inail’s universe is made up of 222 offices of different sizes located throughout the
country, about 20 million policyholders and 3.9 million companies of which approximately 900,000 injuries annually.
The survey regards all INAIL’s offices and the questionnaire is delivered through three channels of detection: the interview at the headquarters, the interview by phone and e-mail. The aim is to obtain a n opinion as objective as p ossible about the se rvices pr ovided by Headquarters.
For the new project has developed a unique questionnaire, for each place was set, the minimum sample size according to the office portfolio and channels of detection.
The main criteria analyzed are: com munication, trustworthiness, well-timed, clearness. The customers are profiled depending on the: kind (workers and companies), age and sex.
The total data are proces sed rega rdless o f the means o f detection used, w ould be included at trial on the work of reference.
To m easure th e de gree of sa tisfaction with aspects m onitored, p rompting t he user to express their opinion through a Lickert scale with 5 values.
It’s been set a s an objective opi nion "fai rly satisfied" against which t o evaluate the performance of each single Office.
In order t o a nalyze the data are used different statistica l indicators such as: wei ght mean, standard deviation, standard error, confidence interval 95% .
This article shows a s the INAIL ‘s experience in recent years has ac hieved significant results regarding the quality of services provided.
References
Belson A.W. (1981). The design and understanding of survey questions, London. Gower Ed. Cicchitelli C.,Herzel A., Montanari G.E.(1992). Il campionamento statistico, Bologna: Il Mulino. ISTAT Misurazione d iretta: l e Cu stomer Satisfaction Su rvey –Le inda gini di CS e la s tatistica uf ficiale -La
misurazione della CS negli Enti Locali”, http://www.istat.it/strumenti/metodi/lineeguida/index.html. Presidenza Del C onsiglio Dei M inistri-Dipartimento Della Funzione Pubblica- Uffi cio P er l’I nnovazione d ella
P.A. (2003). La C ustomer Satisfacti on nelle Am ministrazioni Pubbliche – valutare la qualità percepita dai cittadini-. Roma: I Manuali Rubettino.
Parasuraman, Zeit haml, Malhotra (2004) “E-S-Qual: “A Multiple-Ite m S cale For Assessing Electronic Se rvice Quality”, Msi Reports, 04-003.
Steyaert, J. C. ( 2004). M easuring the Per formance of E lectronic Gover nment Ser vices, I nformation & Management, 41, 369-375.
98
Measuring the administrative compliance burden on enterprises. The Italian experience
Antonio Pavone1, Paola Pianura2 1 ISTAT, [email protected] 2 ISTAT, [email protected]
Regulation is an im portant tool to sup port public policies (taxation, environm ental protection, he alth an d safe ty and em ployment rights) and creates ben efits f or th e participants in an economy because it ca n correct market failures. Howe ver, badly design administrative regulation or overly complex imposes excessive costs in t erms of tim e and money. The challenge for government is to maximize the potential economic benef ts from regulations while keeping cost as low as possible.
In 2007 t he It alian Govern ment launched a new national Multiannual Plan (2007 – 2010) for the Measurem ent of Adm inistrative Costs a nd quantitative reduction of administrative burdens fo r enter prises. A ccording to t he Law no. 133/08 article 25 (so called “Taglia-oneri” – “Cutting red-tape”) the Italian government has approved a number of m easures a imed to cost cutting, st ructure sim plification a nd a dministrative pr ocesses streamlining t hrough t he specifically project labelled “ Measurement of A dministrative Costs” (Misurazione Oneri Amministrativi). The effort is to id entify the regulat ory requirements that could be measured and simplified, without altering the expected public interest objectives.
Quantifying these costs is difficult. Enterprises tend to a pply the term of administrative burden in a much wider sense, incorporating many of the perceived direct and indirect costs of regulation. Fr om a methodological vi ewpoint, t here are two wa ys for m easuring regulatory burdens: “top-down” and “bottom-up”. Top-down approaches typically involve surveys of organizations. Bottom-up approaches typically involve in-depth discussions with a sm aller num ber of in dividuals an d groups a nd case st udies t o e xamine the im pacts of specific regulations.
The Italian e xperience combines advantages of both approaches. A t wo-phase national survey has been carried out only on sm all and m edium enterprises. First of all, a screener questionnaire is used to id entify enterpri ses that have complied with at least one administrative burden for the estimation year. This first survey also provi des an estimation of enterprises affected by the regulation and categorizes enterprises in relevant ec onomic segments. In the sec ond phase, a subset of e ligible enterprises is randomly selected. The aim is to esti mate the costs incurred by e nterprises i n o rder to com ply with information obligations. Th e methodology u sed is b ased on th e EU “stan dard cost m odel” ( SCM). I t breaks down information obligation into a range of m easurable components and activities and costs are t hen appo rtioned in the identif ied activities. Questionnai res and i nterviews guides in volve a co ntinuous pr ocess o f s pecific co nsultation of sta keholders a nd public administrations. More over, t o re duce m easurement erro rs, ex pert assess ments pr ovide a warning level alert when ente rprise i nterviewed declares that a dministrative costs are exceeding the expected threshold.
Public authorities but also private enterpri ses demand reliable statistic information for decision-making: Istat effort s are m eant to satisfy th e requests. Between 2007 an d 201 0, Istat has e stimated costs c aused by 54 administrative proce dures a ssociated t o e ight different sectors. Some Central Administrations, coordinated by the Department of Public administration, have adopted Reduction Plans, in order to implement appropriate actions to reduce administrative burdens and costs for enterprise.
99
Common Components and Specific Weights Analysis of cross-product tables for the Full Multi Modules Customer Satisfaction Evaluation
Pietro Amenta1 1 Department for the Studies of Economic, Law and Social Systems, University of Sannio, [email protected]
Servqual model (Parasuraman et al., 1994) involves a set of five dimensions that have been ranked by custom ers to be most i mportant for service quality: Tangi bility, Rel iability, Responsiveness, Assurance and Empathy. Many multidimensional statistical methods have been proposed in literature for the description and the exploratory study o f these kinds of data. Basic aims are to estimate the multidimensional aspects of the investigated system and the introduction of criteria of judgment. A lot of them are based on the maximization of the covariance c oefficient bet ween linear c ombinations of the variables of t wo m atrices: t he Co-inertia Analysis (Chessel, Mercier, 1993) and the C o-Inertia of Fully Matched Tabl es (Torre, Chessel, 1995). Amenta (2009) suggests to consider the Common Components and Specific Weights A nalysis ( CCSWA) (Qannari et al. , 20 00, 200 1; Hanafi and Qannari, 2008) in order to estimate the multidimensional aspects of the investigated system by taking in account the num ber of di mensions directly in the criteria. The rati onale behind t his method is the existence of a common structure to t he data tables, where data tables reflect the service quality dim ensions. T herefore, the method determines a com mon space of representation for all the data sets. Each t able is allowe d having a s pecific weig ht (or salience) associated with eac h dimension of this common space. Eac h salience represe nts the percentage of total varian ce of each tabl e (service quality dimension) explained by t he jth com mon com ponent. A deepe r anal ysis can be con sidered (e .g. in the stu dent satisfaction) if we have questionnaires taken in different K modules or times (as example, during a post -degree traini ng c ourse with differe nt teachers): we have K pr e-service matrices (one for each module) and K post-service matrices (at the end of each module).
Aim of this p aper is to p ropose a new a pproach in order to analyse this com plex expectations-perceptions syst em (Full Mult i Modules). This two-step st rategy consists in performing a CCSW A analysis of the m atrices of cross-covariances between the variables of eac h ex pectations/perceptions system. In the first step, we use C o-Inertia Analysis K times to com pute the se quence of K cross-covariance tables, and the n CCSWA to analyse this sequence to highlight the common structure of the data tables.
References
Amenta P. (2009), Evaluating custo mer satisfaction by Common Component and Specific Weight Analysis, Atti Convegno “IES2009 - Inn ovazione e Società 20 09. Met odi e p olitiche per la valutazi one dei servizi”. Università di Brescia.
Hanafi M ., Qannar i E .M. ( 2008), N ouvelles pr opriétés de l’ analyse en co mposantes co mmunes et poids spécifiques, Journal de la Société Française de Statistique, 49, 2, 75-97.
Qannari E .M., Wakeling I ., Cour coux P. , M acFie H.J. H. ( 2000), Defining the u nderlying sensor y dim ensions, Food Quality and Preference, 11, pp.151-154.
Qannari E.M., Courcoux P., Vigneau E. (2001), Common components and specif c weights analysis performed on preference data, Food Quality and Preference, 12, 365-368.
Chessel D., Mecie r P. (1993). Coupl age de triplets statistiques et liaisons especes-environement, in: Bio metrie et environment, Lebreton J.D., Asselain B. (Eds) Masson Paris.
Parasuraman A., Z eithmal V. & Berry L . ( 1994) Reassessement of ex pectations as a co mparison standar d in measuring service quality: implications for research, Journal of Marketing, 58
Torre, F., Chessel, D. (1995). Co-structure de deux tab leaux totalement appariés, Revue de Statistique Appliquée, XLIII (1)
100
Contributed session 9
Education III
101
Evaluation of performance at university with Rasch Analysis
Piergiorgio Mossi1, Sara Calogiuri1, Paola Tondo1 1 Dipartimento di Scienze Pedagogiche, Psicologiche e Didattiche Università del Salento, [email protected]
In 2007/2008 the University of Salento has started a degree course in Psychology. R ight from the start, the pe ople in charge of the course have been focusing on preventing early dropouts by stimulating a better professional awareness and by ensuring the best pos sible course experience (Venuleo et alii, 2009). To do so, t he students were asked increasing performance standa rds to p revent the choice of university studies as m ere alternative t o work.
Various managing activities have been undertaken, some (Salvatore, Mossi et alii, 2008) regarding the training experience (seminar approach to the studies, debt recovery activities, tutoring, w orkshops based o n the u niversity experie nce) and others wit h the p urpose of controlling the sam e training process (m onitoring career). This study refers specifically to the latter aspect or the arrangem ents for monitoring t he academ ic c areer t hrough the analysis of perform ance studies. This has a double utili ty: on one hand it calls for the recruitment of competent student function in the degree course, on the other hand it makes clear to the students themselves that the intervention is not reduced only to the operational dimension, but it also im plies the dimension of organizational management of the trai ning process (Salvatore & Scotto di Carlo, 2005).
Normally a review of developments in the curriculum in rela tion to explanat ory variables is performed using statistical infere ntial model whe re t he dependent variable is the grade obtained in the different disciplines or even the credits acquired by the student (Kulatunga-Moruzi & Norman, 2001).
The particular structure of the activity since its inception allows t he acquisition of the results of a career not only in terms of quantitative result s, or through t he allocation of a vote or the s ubsequent de rivation of t he cr edits, b ut als o by exam ining the re porting of failures recorded by the students during the academic career. That's how it was possible to introduce the evaluation of the c urriculum using Rasch Analysis. The model involves the insertion of the exam s’ outco mes analysed in ter ms of su ccess / failure examination for each of the constituent disci plines of studies. So by treati ng the exam s as ite ms of the analysis, the Rasch model allows to assign a value to each student and to estimate logit that defines the degree of performance of the career.
This m odel establishes a m ore reliable cr iteria for t he ve rification of dif ferential variables useful for monitoring the training process (for example: final grade to high school diploma, sex, age, university registration renewal , etc.) and the subsequent management of the training process.
References
Kulatunga-Moruzi, C. e Norman, G. R. (2001). Validity of Admissions Measures in Predicting Performance Outcomes: The Contribution of Cognitive and Non-Cognitive Dimensions. Teaching and Learning in Medicine, 14(1), 34-42.
Salvatore, S., Mossi, P., Venuleo, C., Guidi, M. (2008)"Pregnanza e sensatezza della formazione", Verso un approccio psicosociale alla qualità in Università, a cura di G. Venza, Franco Angeli, Milano.
Salvatore, S. and M. Scotto di Carlo (2005). L’intervento psicologico nella scuola. Modelli, metodi, strumenti. Roma, ICA.
Venuleo, C. Guidi, M. Mossi, P. & Salvatore, S. (2009) L’istituzione generativa. L’impatto di un setting riflessivo sulle immagini anticipatorie del corso di laurea in psicologia. Psicologia Scolastica, 8 (2), 197-222.
102
Ability effect influence on the Italian graduates’ labour income: ATT estimation and sensitivity analysis
Antonino Di Pino1, Patrizia Pulejo2 1 Dipartimento di Economia, Statistica, Matematica e Sociologia, Università di Messina, [email protected] 2 Dipartimento di Economia, Statistica, Matematica e Sociologia, Università di Messina, [email protected]
Several proble ms occur in estimating ca usal effect of educatio n o n indivi dual labour income by uti lizing traditional es timation procedures such as OLS and IV. Th e str ong dependence between individual ability and schooling m akes it difficult t o estim ate their separate effects on income, mainly as a consequence of the omission, in the specification of the income equation, of the “latent” individual cognitive ability.
The difficulty to fi nd an i nstrument suita ble as an unbiased m easurement of ability produce a sp urious relatio nship betw een labou r inc ome, as depen dent va riable, and education le vel and working expe rience as expl anatory varia bles, in t he se nse that the estimated relationship partly reflects the influence of the latent unobservable ability on both income and explanatory variables. This circumstance implies potential inconsistency of the education effect estimates obtained by applying an IV procedure (Hec kman and Vlytacil, 2001, inter alia). The aim of this contri bution is to i mprove the IV approach in estim ating the influence of education on labour income. To this purpos e, we estimate the difference between the labour incom e earned by an It alian graduate and a n Italian holder of a high-school di ploma wh o are si milar in ter ms of a ge, fam ily size, status in the fam ily and geographical area of resi dence. We consider graduates as a group of treated and the high-school diploma holders as a control group. In this context, education level can be assumed as a dichotomous selection criterion for the assignment of the subjects to the treatm ent. We suggest t o util ize the results of a preliminary reduced form esti mation of t he conditional probability to receive the treatm ent or not as a matching proce dure to esti mate the difference between a g raduate and a diplo ma holder in t heir perceived labour income. In this way , the im pact of several fact ors on lab our i ncome gap c an be estim ated distinguishing the influence of individual ability, measured by the difference in high school test score, between a t reated (graduate) a nd an untreated (di ploma holders ) s ubject both characterized by a similar socio-economic profile.
Treated and untreated subjects are dra wn respectively from two different dataset. We select a surve y sa mple of “ treated” (graduat es) subj ects from the 2007 Istat Survey on Italian Graduates in the y ear 2004. The control group, holders of high school diploma, is drawn from the Bank of Italy SHIW Surveys in the years 2004 and 2006.
We suggest the use of a sensitivity test (cfr. Rosembaum, 2002) to verify the robustness of m atching results if the Ignorability condition (m atching is not infl uenced by om itted variables) is v iolated. T o this pu rpose, we verify the robustness of th e ATT esti mation results simulating the i nfluence of “hidden” and potentially “confounding” factors on the assignment to the treatment.
References
Heckman J. J., Vytlacil E. (2001). Identifying the Role of Cognitive Ability in Explaining the Level of and Change in the Return of Schooling, The Review of Economics and Statistics, 83 (1), 1-12.
Rosenbaum P.R. (2002) Observational Studies, 2d ed., Springer, New York
103
The impact of the "3+2" reform on degrees
Maria Luisa Maitino1, Giulia Peruzzi2
1 IRPET, [email protected] 2 REGIONE TOSCANA, [email protected]
The aim of t his research is t o evaluate the i mpact of the It alian University Reform (the so called 3+2) on the rate of degrees achieved in the scheduled time. We obtained our data from the register office of the three Tuscan Un iversities (Firenze, Pisa, Siena), which contains all the personal information of the students.
Propensity Score Matching has become a popular approach to estimate causal treatment effects. It applies for all situations where one has a treatment, a group of treated individuals and a g roup o f unt reated in dividuals. Its basic idea is to find in a large group of non-participants those individual s who are si milar to the participants in all relevant pre-treatment characteristics. That being done, differences in outcomes of this well selected and thus adequate cont rol group and of partici pants can be attributed t o the treatm ent. Our analysis refers to the group of freshmen enrolled during the academic year 2000/01: this student’s c ohort is the only one that co uld ch oose, afte r the ref orm, if to shift the new system or not.
In order to perform propensity score matching, we consider as “treated” the group of the students that shifted i nto the new system, while the untreated group consists of simila r students in all relevant p re-treatment char acteristics (age, ge nder, m ark a nd kind of diploma, etc.).
The first results show a 24% increase in the rate of degrees obtained from the shift into the new system, even if the average effect of treatment varies across different universities. However, there is not a satisf actory effect on the probability of obtaini ng the l ong degree (laurea magistrale) within 7 years.
References
Rosenbaum P. R., Rubin D. B. (1983), “The central role of the propensity score in observational studies for causal effect”, Biometrika, n. 70, pp. 41-55
104
The Estimation of Dimension and Factors of School Abandon in the Romanian Development Regions
Tudorel Andrei1, Alina Profiroiu2, Andreea Iluzia Iacob3, Bogdan Ileanu4 1 University of Economics, Bucharest, Romania, [email protected] 2 University of Economics, Bucharest, Romania, [email protected] 3 University of Economics, Bucharest, Romania, [email protected] 4 University of Economics, Bucharest, Romania, [email protected]
During the transition period in Romania the dim ension of school abandon had risen. The main goals of the study are: to estim ate the school abandon rate by each educational level on th e de velopment region s from R omania, to identify the factors which affect sc hool abandon; to de termine the co rrelation between the sch ool abandon characteristics and the gipsy population percentage, to analy ze the effect of gove rnmental strategies. In the sa me time the analysis had followed also the temporal component by including in the database of the last decade statistical information. The school abandon was measured as the differe nce between the numbers of pupils/students found at t he end of the sc hool year and the same category enrolled in the beginning of the same year.
105
Contributed session 10
Socio-economic topics
106
The system of “performance evaluation” of the Ministry of Labour and Social Policy: an analysis of the production function of output inspection
Anna Maria Frasca1, Piergiorgio Mossi2, Enrico Ciavolino3 1 Direzione Regionale del Lavoro Puglia, Ministero del Lavoro e Politiche sociali, [email protected] 2 Dipartimento di Scienze Pedagogiche, Psicologiche e Didattiche Università del Salento, [email protected] 3 Dipartimento di Filosofia e Scienze Sociali Università del Salento, [email protected]
For several years it has become the practice of Public Administration quality assessment as part of the inst itutional activities carried out by offices (Bank of Italy, 2002). Each Pub lic Administration has there fore p repare a ppropriate inst ruments to enable an effec tive measure of its products (Morvillo, 2008).
The Ministry of Labour and Social Affairs has adopted a system of “quality assessment” based on indicators selected as representative of the inspection carried out in branch offices (eg. nu mber of in spections carried out, un documented and “ black” workers, num ber of unprotected minors found in the company, requested administrative injuries received in the offices, non-payment contributions found during inspection access, etc.).
This scoring system - called Project Quality - is defined by three “synthetic indicators” determined periodically to the 92 Provincial Labour Departments operating in the co untry. It does indeed have a rating system that defines a ranking between the provincial offices. A pilot study carried out on the basis of this national ra nking has s uggested the po ssible influence e xerted by the “local variables”, ie those relat ing to t he ge o-socio-economic differentials, in explaining the efficiency of inspection.
The purpose o f this paper is to present the results of re search work in collaboration between the Regional Directorate of Labour of Puglia (peripheral organs of the Ministry of Labour and Social Policy) with the University of Salento, in or der to analy tically describe the per formance level of th e Pro vincial L abour De partments of Pu glia on the basi s of variables related inspection and local variables (Gazzei et al., 1997; Amenta et al., 2008).
The data anal yzed com e from the Provi ncial Directorates 5 Puglia (Foggia, Bari, Taranto, Brindisi and Lecce). The measurement of perform ance according to the va riable inspection and local was formalized through a structural equation model.
The results of the modeling and analysis wi ll also be c ompared with those achieved in the project as defined at the national level by the Ministry.
References
Morvillo A. (200 8). L’analisi co mparativa delle perf ormance nella P ubblica Am ministrazione Atti del la Conferenza del C onsiglio Nazional e delle Ricerche – Istitu to di ricerc he sulle attività terziarie , Napoli 15 ottobre 2008.
Banca d’Italia, (2002). L’efficienza nei servizi pubblici, Ufficio Studi. Amenta P, Ciavoli no E., Si monetti B. (2008). La Quantifi cazione degli indici di Co ngruità e degli Scosta menti
Ammissibili. In: Cristina Sunna. La Lotta al Lavoro Nero nell'esperien za legislativa e amministrativa dell a Regione Puglia. (pp. 35-46). Cacucci Editore, Bari.
Gazzei D.S., Lemmi A., Viviani A. (1997). Misure statistiche di performance produttiva, Ceup, Padova.
107
Assessment of relevance and efficiency of scientific research in Universities departments
Laura Antonucci1, Francesco Domenico d’Ovidio2 1Università degli studi di Foggia, [email protected] 2Università degli studi di Bari Aldo Moro, [email protected]
The last legislative actions of the Italian government have started a revolution and a race to the i mprovement by the Universities, in order to avoid penalties. In this context, some doubts arise on scientific research evaluation and his costs: the 2001-2003 VTR exercise, conducted by CIVR in 2005-2006, was very expensive, providing a very partial pictur e of the Italian scientific reality. The next exerci se of evaluation (considerably more detailed, but surely far from complete) should cost much more.
The key to an y assessment must tend the im provement, but scientific research has fast dynamics; it is thus necessary evaluate it with continuity: ideally, ev ery year or e very two years. But the continuous assessment of the research quality seems almost impossible: the cost of obtaining data in this field, with the criteria previo usly set by CIVR, appears huge. But other aspects could (and maybe should) be evaluated in the scientific research, although they were confused often with the quality (or the effectiveness): importance and efficiency of research, which, in lack of information on inputs, could be estimated throught the output of the research, i.e. the number of scientific products.
In this paper we propose to estimate of th e actual scientific produ ctivity of institutions thru a simple indicator born as an evolution of VPS, proposed in 1999 by the Observatory for the Eval uation of University System. This indicator starts from the concept of equivalent product of a single researc her,
k
kwk a
wEP , where ak is the n umber of holders
of the kth scientific product (i .e. the number of au thors of a paper), defining his "degree of property", while wk is the weight of scientific potential of such product or its relevance.
The equivalent productivity of a research Department is simply obtained by adding all
the EPkw values of the researchers who belong to it:
jxnp
k jk
xjkc
jj
wx a
awEP
11, where c are few
categories of scientific relevan ce (with regard to scientific pu blications, we identified such categories using data a vailable for each researcher in t he database provided by CINECA). The num erator axjk is the n umber of researchers belonging to s uch department that are "owners" of the same product (e.g., co-authors of a paper).
Two use ful in dexes m ay be easily obtained fr om such equivalent productivity value: the Product Relevance Index, PRIx=EPx
w/npx (which maximum is given by the m aximum value of w) and the Structure Productivity Index, SPIx=EPx
w/ncx, where the term ncx is the total number of researchers belonging to that structure, including those without products.
Such indicators, whic h are easy to calculate (with low cost or zero cost), are not explicitly oriented t o the evalua tion of the quality of scientific research, but they hi ghlight the productivity and the relevance of su ch r esearch. Th ey c an be used to ev aluate in a simple way the research potential and productivity in the uni versity departm ents and to allocate internal funds for research, as already done somewhere with several criteria.
An e xperiment sho wed t hat PR I an d SP I va lues lea d to alm ost balanced rankings between resea rch departments of dif ferent orientation ("medical & health science s", "science & technologies", "ot her orientation"), althou gh it wa s li mited t o the database of publications. Better results could derive by using data on all scientific products.
108
A Structural Equation Model to analyze the Household Budget: a case study
Giovanni Di Trapani1, Pasquale Marrone2, Pasquale Sarnacchiaro3, Ilaria Sasso4 1 Istituto di Ricerche sulle Attività Terziarie, [email protected] 2 ISTAT, [email protected] 3 Università Unitelma Sapienza, [email protected] 4 ISTAT, [email protected]
The Household Budget Survey (HBS) constitutes the basic information to describe, analyse and interpret consumption expenditure of households.
The survey, based on a sam ple of households, collects data on household expenditures for co nsumption, with a particular foc us on social an d e conomic aspects of h ouseholds’ living conditions. The survey a llows both the qua litative and quantitative analysis of living standards and consumption behaviours of the households, refer red to different typologies and territorial and social contexts.
The main purpose of the survey is, in fact, to collect information on the structures and levels of c onsumption e xpenditure by th e main socio, ec onomic and territ orial characteristics of t he households; all expe nditures afforded by the households to purchase goods and services are registered. The definition of consumption expenditure includes also goods com ing from the ho useholds ga rden or fa rm directly consum ed by the ho usehold itself (self-consumption), the goods and services provided by the employers as salary , the imputed rent for houses occupied by the owner or used without a charge.
In order to draw a c omplete picture, the s urvey collects data on e xpenditures for food and beverages, housing, furniture, clothes and shoes, health, transport and communication, recreation, culture and education and other goods and services not already mentioned, in addition to inform ation on the household members (relationshi p, age, marital sta tus, education level, professional condition and position) and on housing characteristics.
The results are presented by groups or categories of expenditure (those considered more relevant), analysed by territorial breakdown, household size and typology, occupational and professional con dition of th e ho usehold h ead, as well as by those c haracteristics wh ich, more than other, influence and characterise the consumption levels and behaviour.
The data obtained have been elaborated by a Factorial Analysis in order to identify the main aspects that infl uence the habits of Ita lian Household. Starting from these results, we hypothesised the relationship among latent variables and we used structural equation model to analyze the data gathered.
Reference
Bollen, K. A. (1989). Structural equations with latent variables. New York: John Wiley & Sons. Joreskog K.G. (1981), Basic issue in the application o f LISREL, Data, Comunications computer analysis, pp. 73-
89,. Wold H., Soft Modelling: the basic design and so me extensions, Joreskog K.G. & Wold H. (ed.), Systems under
indirect observation: Causality, structure, prediction, North Holland, Amsterdam, 2 pp. 1-54, 1982.
109
Research Projects Evaluation. A Small Case Study on “Ideas” Competition from National Research Program of Romania
Claudiu Herteliu1, Tudorel Andrei2 1 University of Economics, Bucharest, Romania, [email protected] 2 University of Economics, Bucharest, Romania, [email protected]
In this paper we will briefly present an analysis of the eval uation process fo r the research projects from “Ideas” 2008 Co mpetition held in Rom ania w ithin the National Research Program. W e will focus on th e socio-economic do main. The competition started in 2008 and t here were 675 subm itted grants. The projects were constructed on a m ultiannual scheme (36 m onths). After the initial evalua tion there we re accepted for financial support 208 grants. After the first year of implementation (2009) there was a yearly evaluation and, subsequently, 68 projects we re selected for receiving financial support for the second year of implementation (2010), while from 75 proj ects that ha d been receiving a “ promise” for financial suppo rt and had to b e d elivered in 2011, 45 projects wer e “su spended” an d th e rest of 19 projects were stopped. At th e end of 2010 there was a nother yearly evaluation and, according to the outc omes of this process, for fi nal implementation year – 2011 – a number of 186 projects will be receiving financial support.
110
Particularities and typologies of the development level of Romanian’s villages from the ethnic affiliation view point
Claudiu Herteliu1, Bogdan Ileanu2 1 University of Economics, Bucharest, Romania, [email protected] 2 University of Economics, Bucharest, Romania, [email protected]
This paper is using a quantita tive methodology and aim s to analyse the way in which the ethnic affiliati on of Rom anian villages has an im pact over t heir development level. The development level use d ( measured by C ommunes Develo pment In dex – C DI) was designed by Dumitru Sandu (2009). The CDI was build, for all 3000 Romanian villages, on a similar structure with Hum an Development Index (HDI) having three components: life expectation, education and economic level.
111
Contributed session 11
Ordinal data and latent variables
112
Performance Assessment of Social Agents. A goal-planned approach
Giulio D’Epifanio1 1 Dipart. di Economia, Finanza e Statistica, Università degli Studi di Perugia, [email protected]
In evaluating social agents, an approach is outlined to construct a performance index, which is based on g oals pla nning. Goal- planning s hould be gr ounded on specificatio ns of requirements, from some policy-commitment (PC) (e.g. a cert ain institutional stakeholder), at which the policy-maker ( PM) w ould re spond. In a sim plified v ersion, go al-planning sheets might be de picted by Pert-like (direct acyclic graphs) DAG diagrams (D’Epifanio, 2010), where nodes are testable binary goals and the ultim ate-node is identifiable with the complete obtainm ent of the overall-goal. Using lo gical ope rations, taking int o account specifications of priorities and preferences from the PC, a sequence of binary testable goals may be distilled, which is Guttman’s ordered and value-increasingly. Data are provided by sensors, which chec k t he goals acr oss th e plan ned Guttman-ordered sequence, possibly conditional on context and subject condit ions. Th en, over su ch a sequen ce of go als, a graduated scale of performance may be operationalized, by using the “intrinsic worthiness” criterion (D’Epifanio, 2008), which is calibrated on statistical behaviour of reference gold-standards, to graduate the worthiness of subjects on the decision-maker's goal-oriented trait. Formally provided by a C hoquet's integral, an index is finally proposed which may be practically useful in referenced m onitoring o f ad vance of subje cts, against the programmed path of o perative goals, towards the decision-maker's overall-goal. Examples of distilling processes are outlined, over various types of Pert-like diagrams, to construct types of o rdinal performance scale, e.g. in educational evaluations. Finally, based o n the programmed goal-planning, a social i ndex may be derived which uses the Quiggin-Yaari functional from the “rank dependent expected utility” theory.
References
D'Epifanio G. (2009), Implicit Social Scaling. From an institutional perspective. Social Indicator Research, Social Indicator Research 94: 203-212
D'Epifanio G. (2008), Worthiness Based I nterpretation of Equi-distanced Performance Scales, in: M TISD 2008; Methods, Models and Information Technologies for Decision Support System
siba2.unile.it/ese/issues/324/722/MTISD\_2008\_p152.pdf; e-ISBN: 978-88-8305-060 D'Epifanio G. (2010), Sviluppo di un indice multi-attributo per la valutazione del merito, technical report, unipg
113
The EN index for the use of the normal, exponential, beta or gamma distribution in the indirect quantification
Antonio Lucadamo1, Giovanni Portoso2 1 Università del Sannio, [email protected] 2 Università del Piemonte Orientale, [email protected]
The evaluation of capabilities, attitudes, customer satisfaction, in different fields of the real life, is one of t he problems that, during the last years, have been studied in social sciences. In fact, these qualitie s are not directly observable, but they are expr essed with the modalities of ordinal scales. In these circumstances, if we want to perform the evaluation, it is necessary to assign some scores to the categorical modalities of the analysis.
A sim ple technique is the direct quantifica tion: this tec hnique hy pothesizes that th e modalities of a qualitative character are at the sa me distance, but this hypothesis is not respected in many situations. For this reason it is preferabl e to use an alternative techni que that consists in assigni ng real numbers to the categories of the qualita tive variable. In this type of quantification the numbers are not equidistant but they depend on a latent variable. Different m easurement techniques have been developed du ring t he y ears, based on the hypothesis that the model is normally distributed.
This assum ption is not real istic when the j udgments are all extremely positive or extremely negative. In this case the assum ption that t he latent varia ble is exponent ially distributed could be a solution.
Obviously the choice bet ween the tw o di stributions is not easy , b ut the EN inde x (introduced by Portoso in 2003), can be a good instrument to detect the right assum ption. The EN index assumes values between -1 and 1 and, in previous works, it has been showed that if the absolute value of t he index is be tween 0 and 0.30, the normal distribution is the better o ne; instead if the inde x ha s val ues bet ween 0.60 a nd 0.90, the exp onential distribution seems to be the more appropriate.
It is clear that there are intermediate values of t he index (0.30-0.60) for which neither the normal distribution nor the exponential distri bution can be considered adequate. In this paper we will show how, in this case, the beta or the gamma distribution could be the most appropriate latent variables in the indirect quantification.
References
Cerri M., Zanella A. (2000), La misura di Custo mer Satisfaction: Qualche r iflessione su lla scelta delle scale di punteggio, Atti de lla giornata di studio “Valutazione della Qualità e C ustomer Satisfaction: il ru olo della Statistica, Vita e Pensiero, Milano, pp. 217-231.
Lovaglio P.G. (2001) The esti mate of latent outcomes, Atti del Conve gno Inte rmedio della Società Italian a di Statistica”, Processi e Metodi di Valutazione, CISU, Roma, pp.393-396.
Lucadamo A. , Po rtoso G. ( 2011). Valori soglia dell’ indice E N per la scelta della distr ibuzione norm ale o esponenziale nella quantificazione i ndiretta. Accepted for the publicatio n on Rivista Italiana di Econo mia, Demografia e Statistica.
Marbach G. (1974). Sulla presunta equidistanza degli intervalli nelle scale di valutazione. Metron, XXXII, 1-4. Portoso G. (2003 a). La quantificazione determinata indiretta nella custo mer satisfaction: un ap proccio bas ato
sull’uso alternativo della normale e dell’esponenziale, Quaderni di dipartimento SEMeQ, 53. Portoso G. (2003b). Un indicatore di addensamento codale di frequenze per variabili categoriche ordinali basate su
giudizi, Quaderni di dipartimento SEMeQ, 66.
114
A PLUM model for the evaluation of university teaching
Angela Alibrandi1, Massimiliano Giacalone2 1 Department of Economic, Financial, Social, Environmental and Territorial Sciences, University of Messina 2 University of Catanzaro “Magna Graecia”
The teaching evaluation by students attending the course, conducted using questionnaire in the Italian uni versities m ight seem like j ust a bureaucratic process, aimed at sim ple compliance with m inisterial du ties. T his f orm of assessm ent is, howe ver, an im portant monitoring tool, available to the institutions responsible for governance and management of formative pr ocesses: Faculty C ouncils, d egree c ourse Councils and individual tea chers (Chiandotto et al, 2005).
The need for teaching e valuation, in addition to being perceived as very important by the university syste m itself, is also required by law; in particular we refer to La w 370/99, art. 1 (Internal Evaluation of Universities).
The aim of this paper is to show the phases of the statisti cal detection which led to the satisfaction of knowledge needs on the students, shared for their Faculties and for degree courses. B y u sing a ppropriate statistical methods, it is pos sible to i dentify fact ors that facilitate / hinder learning by students. This monitoring process is aimed at the adoption of correction measures, where such evidence is necessary. This study takes into acc ount the evaluation of the “internal eff ectiveness” of teaching, based on the opinions of stude nts, using the evaluation questionnaires, completed by students attending the courses.
Data we re col lected from th e academ ic ye ar 2009/ 2010 by the E valuation Group of “Magna Graecia University” of Catanzaro. We consider twenty-four courses belonging to the Degree in Business Administration; in order to protect their p rivacy, only their codes are sh own (EC09001, EC 09002, EC09024 .. .). Th e questionnaire is structured in th e following five sections: Section 1: Course organization; Section 2: Teaching organization; Section 3: Educational Activi ties and studies; Section 4: Infrastructure skills; Section 5: Interest and satisfaction.
The response is measured on an ordinal scale with four categories: Definitely not; More no than yes, More yes than no, Definitely yes.
From a methodological point of view, given the nature of the data, we decided to use the Polytomous Universal Model (or “PLUM ” pr ocedure), that is the procedure for the estimation o f ordinal re gression m odels (i mplemented in SPSS); it allows to m odel th e dependence of an or dinal response on a set of categorical and scale independe nt variables (Marija Norusis, 2010). I n particular we used the logit as link f unction a nd, s o, we estimated an ordinal logit model.
The preliminary descriptive an alysis shows that m ost of t he stude nts conside red insufficient the background knowledge; they perceived as to o high the overall study load and, there fore, do n’t consider it appr opriate to the claim s assigned to t he matter subject. They retain useful interim tests and supplementary educational activities; they declare to attend more than 75% of the teachings lessons. Finally, they often don’t feel involve d by the teacher during lessons bu t they j udge largely positive the teache rs availability for clarifications and explanations.
References
Chiandotto B., Grilli L., Ra mpichini C. (2005), Valut azione dei proces si form ativi di te rzo livello: contributi metodologici, n. 12, Collana ValMon, Dipartimento di Statistica “G. Parenti”, Firenze.
Marija J. Norusis (2010), PASW Statistics 18 Statistical Procedures Companion, Pearson Higher Education.
115
Complementary use of different methods to evaluate Customer Satisfaction of Services with subgroups data
M. Chiara Zanarotti1, Laura Pagani2 1 Department of Statistical Science, Catholic University of Milan, Italy, [email protected] 2 Department of Economic and Statistical Science, University of Udine, [email protected]
A lot of di fferent m odels and m ethods have been p urposed in last y ears to measure Customer Satisfaction (CS) in various contexts. Satisfaction is strictly linked to quality and it is necessary to refer to an abstract construct to m easure service quality. Often different approach to the measure problem are suggested as alternative: in this contribute we suggest the complementary use of different methods to evaluate services quality with the ai m to create a synergic tool of anal ysis in situatio n in which data are c ollected in s ubgroups and the objecti ve is not only to evaluate a sing le subgroup t o im prove t he service for less satisfactory aspects but also to com pare perform ances and to m onitor the whole service supplier during the time.
To analyze da ta separately by subgroup, we use a m ethod o f items ranking based on median an d he terogeneity in dex (median-HI criterion) si milar to the so called scorecard method ( Giudici 200 7, R ampichini et al., 20 00). After t his prelim inary analy sis we use Rasch Model for the whole service to obtain an overal l measure of i tem quality le vel (Rasch items parameters) to compare previous subgroups items order with the overall items order and to calibrate the questionnaire (Pagani and Zanarotti, 2003).
The last method of a nalysis we consi der is based o n a Dissimilarity Index bet ween ordinal distributions. The observed distribution of responses is compared with a t heoretical one, selected as com parative m odel. C apursi a nd P orcu ( 2001) su ggested t o use as theoretical distribution the optimal one, so called because is the distribution in which all subjects c hoose the hi gher response category. A dissimilarity index is obtaine d for each item and a comparative performance indicator (CPI, in Capursi e Porcu) for each subgroup is then calculated as simple mean of the complement to one of each dissimilarity index (one for each ite m). Through t hese CPI indexes it is possible to compare global satisfaction for different subgroups a nd to range subgroups according to global satisfa ction. Finally, a global satisfaction index (taking values between zero and one) for the whole supplier of the service is obtained as arithmetic mean of subgroups-CPIs.
An application of the above procedure to public services data is then proposed.
References
Capursi V, Porcu, M. (2001). La didattica universitaria valutata dagli st udenti: un indicator e basato su misure di distanza fra distribuzioni dei giudizi. In: “Processi e metodi statistici di valutazione”, SIS 2001, Roma.
Giudici, P., (2007) Governo dei Rischi: Il ruolo dei modelli statistici. Istituto Lo mbardo (Rend. Lett.), 141, 361-376.
Pagani L., Zanarotti M.C. (2003).Analisi della qualità di un servizio: un confro nto tra scale mediante il modello di Rasch, Statistica & Applicazioni, 2, 35-54.
Rampichini, C., Grilli. L. Petrucci, A.(2000) Analisi della qualità della didattica attraverso modelli multilivello, in Civardi M. e Fabbris L. (a cura di), Valutazione della didattica con sistemi computer-assisted, Cleup, Padova.
116
AUTHOR INDEX
Alibrandi; 114 Allorio; 26 Amato; 57; 58 Amenta; 99 Andrei; 68; 104; 109 Annicchiarico; 39; 89 Antonucci; 49; 78; 93; 107 Baldassini; 96 Bartolucci; 74 Bassi; 71 Battauz; 31 Berta; 12 Bertaccini; 81; 89 Bertoli Barsotti; 87 Bianchi; 52 Biggeri; 46 Bini; 64 Boari; 36 Bonanomi; 35 Bonnini; 86 Bottini; 53 Bove; 33 Calogiuri; 101 Cantaluppi; 36; 52 Carpita; 69 Caviezel; 87 Cerchiello; 19 Chirico; 53 Ciapparelli; 18 Ciavolino; 36; 92; 106 Clerici; 76 Corduas; 24 Crisci; 20 Crocetta; 49; 78; 93 Croon; 71 Cusatelli; 48 D’Agostino; 77 D’Ambra A.; 20 D’Ambra L.; 66; 82 D’Epifanio; 112 d’Ovidio; 49; 93; 107 De Luca; 18 Di Battista; 29 Di Chiacchio; 33 Di Gennaro; 28 Di Nisio; 29 Di Pino; 102
Di Trapani; 108 Di Zio; 50 Elias; 13 Fandella; 54 Frasca; 106 Frenda; 66 Ghellini; 77 Giacalone; 114 Giraldo; 76 Girardi; 54 Giudici; 61 Golia; 32 Gori; 31 Grilli; 22; 63; 72 Herteliu; 109; 110 Iacob; 68; 104 Ileanu; 67; 104; 110 Isaic-Maniu; 67 La Placa; 52 Lacangellera; 17 Lacquaniti; 97 Liberati; 17 Lorizio; 49 Lucadamo; 94; 113 Maitino; 62; 103 Marella; 33 Mariani; 17 Marotta; 96 Marrone; 108 Masserini; 79 Mazzeo; 57; 58 Mignani; 23 Milone; 82 Monaco; 26; 96 Monari; 23 Monastero; 64 Montagna; 40 Mossi; 101; 106 Neri; 77 Nissi; 88 Nitti; 92 Nuti; 84 Ortobelli; 87 Pacinelli; 50 Pagani; 31; 115 Paoletti; 97 Pavone; 98
Pennoni; 74 Perucca; 91 Peruzzi; 103 Pianura; 98 Piraina; 52 Pittarello; 71 Porcu; 45 Portoso; 113 Profiroiu; 68; 104 Pulejo; 102 Quatto; 44 Raffa; 42 Raffinetti; 61 Rampichini; 22 Rapposelli; 88 Riccardi; 96 Romano; 57 Rosignoli; 56 Sacchettini; 81 Salini; 91 Salmaso; 86 Sani; 63 Sarnacchiaro; 28; 108 Sarra; 83 Sasso; 108 Scaccia; 27 Scaramuzzi; 26 Schadee; 37 Sciclone; 62 Scippacercola; 57; 58 Simonetti; 94 Simonetto; 73 Soleti; 78 Solmi; 86 Stracqualursi; 23 Sulis; 45 Toma; 78; 93 Tondo; 101 Tritto; 41 Vannuccini; 39 Varriale; 72 Velucchi; 64 Vezzoli; 69 Vignoli; 89 Visentin; 76 Vittadini; 12; 74 Zanarotti; 115
