The determinants of apprenticeship training with …doku.iab.de/zaf/2007/2007_2-3_zaf_dietrich_gerner.pdfThe determinants of apprenticeship training with particular reference to business

The determinants of apprenticeship training withparticular reference to business expectations*

Hans Dietrich and Hans-Dieter Gerner**

Whilst in applied empirical research, training in general human capital is mainly ex-plained by structural characteristics of firms, this paper introduces business expecta-tions as an additional explanatory factor. Business expectations are strictly time-variateand firm-specific and reflect both a firm’s development in competitive markets and inthe business cycle. We assume that a firm’s business expectations strongly modify thecost-utility concept for firms’ decisions as regards providing apprenticeship places.

When controlling for firms’ structural characteristics, static econometric models sup-port our assumption that a change in business expectations leads to an asymmetricadjustment process of firms’ qualitative decisions regarding apprenticeship training.Concerning the quantitative decision as to how many apprenticeship places a firmprovides we found a significant but not asymmetric response to a change in businessexpectations.

A dynamic approach confirms the results obtained in the static models of a symmetricquantitative adjustment process in a short-term perspective. In a longer perspectivethe dynamic model supports the assumption of an asymmetric quantitative adjustmentprocess. Further on an application shows that an increasing uncertainty regarding busi-ness expectations tends to reduce the apprenticeship training at firm level.

* This paper was released for publication in September 2007.** We thank Lutz Bellmann, Anette Haas, Thorsten Schank, and two anonymus referees for valuable com-ments and suggestions. The usual disclaimer applies.

Contents

1 Introduction

2 Theoretical considerations

3 Econometric analysis Ð Part 1: static approach

3.1 Econometric strategy, data, variables and hypothesis

3.2 Results of the static analysis

4 Econometric analysis Ð Part 2: dynamic approach

4.1 Econometric model, data and variables

4.2 Results of the dynamic analysis

5 Application of the estimates

6 Conclusions and further options

References

Appendix

ZAF 2 und 3/2007, S. 221Ð233 221

The determinants of apprenticeship training Hans Dietrich and Hans-Dieter Gerner

1 Introduction

Since the mid-eighties firms’ provision of appren-ticeship places has decreased dramatically in the oldGerman Länder. Whilst in 1986 firms provided some700,000 new apprenticeship contracts, there wereonly about 450,000 in 2006. This process has foundits continuation in unified Germany since the earlynineties. Both sectoral change and the restructuringof enterprises have been identified as the drivingmechanisms behind this development. In addition tothis structural change, microeconomic studies pro-vide some evidence that firms’ human capital pro-file, their internal on-the-job training and furthertraining or recruitment strategies and firm-size char-acteristics explain firms’ provision of apprenticeshipplaces more or less stably over time (Neubäumer/Bellmann 1999; Dietrich 2000; Beckmann 2002;Euwals/Winkelmann 2004; Niederalt 2004; Dietrich/Gerner 2005; Muehlemann et al. 2007).

Besides these firm-specific characteristics of busi-ness organisation, recent literature indicates somearguments that uncertainty concerning the furtherdevelopment of business success influences firms’future provision of new apprenticeship places(DIHK 2006; Bellmann/Hartung 2005). The aim ofthis paper is to introduce business expectations anduncertainty into microeconomic models explainingfirms’ qualitative and quantitative decisions aboutproviding apprenticeship places. According to ourconsiderations, firms’ business expectations reflectboth firms’ expected performance in competitivemarkets and macroeconomic factors such as the ex-pected progress of the business cycle. In this respectbusiness expectations as a core element of firms’ ra-tional decisions are associated with uncertainty ina more general sense. A more specific concept ofuncertainty, discussed in more detail in section 5, isadapting a risk concept formulated by Rothschildand Stiglitz (1970).

In the remainder of the paper, on the basis of hu-man capital theory, explanations of firms’ provisionof apprenticeship places (Becker 1964; Harhoff/Kane 1997; Acemoglu/Pischke 1998, 1999a and b;Clark/Fahr 2002; Niederalt 2004) are extended byintroducing business expectations as an exogenousvariable. Compared with more static variables de-scribing firms’ characteristics of organisation andworkforce, business expectations seem to be morevariable over time. Furthermore we adopt theoreti-cal and empirical considerations about asymmetricprocesses of adjustment of firms’ training decisionsto changes in business expectations.

In the empirical section of this paper we use datafrom the IAB Establishment Panel. In a first analyti-

222 ZAF 2 und 3/2007

cal step, a static approach is applied to answer twoquestions: a) what factors motivate firms to train ap-prentices (qualitative decision) and b) what happensto the number of apprentices recruited by a firmwhen these causal factors vary over time (quantita-tive decision). Subsequently a dynamic approach isused to model the adjustment process for the quan-titative decision. In an application we use our resultsto formulate assumptions about the effect of uncer-tainty regarding business expectations over time ona firm’s quantitative and qualitative decision as towhether to provide new apprenticeship places in agiven year or not.

2 Theoretical considerations

Why do firms pay for apprenticeship training (regu-lated by the German vocational training act, BBiG),even if apprenticeship training is an investment ingeneral human capital? Becker already raised thisquestion in the early sixties (see Becker 1962; 1964)and Acemoglu/Pischke (1999a, b) and Niederalt(2004) reviewed in detail the ongoing research espe-cially related to the German apprenticeship system.

In particular two types of training costs can be dis-tinguished: direct and indirect costs. Direct costs arefor example expenditure on firm-based training fa-cilities or wages for apprentices. Indirect costs occurwhile workers spend part of their working time ontraining apprentices instead of on production. A netcost perspective takes into account the fact thateven apprentices may be productive while beingtrained. Calculations by Beicht and Walden (2004)present a remarkable variation of net training costs,explained partly by firm size, industry and the typeof occupation that apprentices are being trained in.Concepts motivated by net cost are closely relatedto assumptions derived from production-theory con-siderations (see Lindley 1975; Fougere/Schwerdt2002; Zwick 2007). Also, a specific type of firm isable to gain net returns from investments in generaltraining already during the training period.

In general, production-motivated explanations arenot sufficient to explain firms’ training activities inapprenticeship training. So it is also worth notingthat considerations oriented towards net cost onlytake into account costs and returns occurring duringthe institutionally defined period of contract-basedapprenticeship training. Opportunity costs or trans-action costs (e.g. recruitment costs for skilled em-ployees) may occur, however, when this type oftraining is completed. These types of costs are notconsidered here. Investment-motivated considera-

Hans Dietrich and Hans-Dieter Gerner The determinants of apprenticeship training

tions (Stevens 1994; Franz/Soskice 1995; Timmer-mann 1998; Fougere/Schwerdt 2002) reflect the re-turns on training after the completion of an appren-ticeship. From this viewpoint, apprenticeship train-ing improves not only productivity during the train-ing period but also future productivity. Highmobility costs for workers, institutional barriers andlow labour turnover in Germany (Harhoff/Kane1997), asymmetric information on the ability of theapprentices and other market imperfections gener-ate compressed wage structures and allow firms pro-viding apprenticeship training to take a rent fromemployees trained within the firm compared withfirms who recruit employees on the labour market(Acemoglu/Pischke 1998, 1999a, 1999b; Beckmann2002; Clark/Fahr 2002). Following Franz/Soskice(1995) and Werwatz (1996) apprenticeship trainingprovides additional firm-specific skills. According toassumptions on high factor specificity (Holtbrügge2004; Williamson 1990) training firms may have ad-ditional advantages from this firm-specific humancapital which is generated en-passant.1

Increasing or decreasing business expectations, how-ever, affect both the cost and the reward sides ofapprenticeship training. Decreasing business expec-tations indicate to training firms an expected reduc-tion in the productive contribution of apprentices tothe firm. As a consequence the net costs of appren-ticeship training may increase and training firms willreduce their provision of training capacities. A pre-condition for long-term returns on firm-based train-ing is that firms will be able to retain apprenticesafter the training period. As labour demand dependon the demand for goods (Freemann 1972), the fu-ture market position must be sufficient to retain theapprentices who are currently undergoing training.When offering new apprenticeship places, firmshave to make their decisions under uncertainty. Sobusiness expectations could be a relevant indicatorin firms’ decisions. Expanding on Acemoglu/Pischke(1999a, 1999b), a reduction in business expectationsshould decrease not only the probability of firms re-taining their own apprentices afterwards but also theprobability of using them productively during theapprenticeship training. As a consequence this willincrease the expected net costs of training.

Supported by empirical evidence, assumptions frombehavioural finance suggest that a deficit of a givenamount is perceived by actors more sorely, thanearnings of the same amount increase individual

1 Alternative training motives such as a reputation motive (Sa-dowski 1980), a stock-keeping motive especially of larger firms(Backes-Gellner 1992) or industry-specific arguments (Büchel/Neubäumer 2001) are not taken into account here.

ZAF 2 und 3/2007 223

utility Ð this paradox motivates so-called loss aver-sion (Bank 2003; Gul 1991). Given a loss-averse be-haviour, actors will renounce investments more eas-ily with a given decrease in business expectationsthan they will make investments in the oppositecase. In this respect, firms’ training decisions shouldcorrespond with asymmetric investment or adjust-ment decisions. We assume that firms will reducetheir training facility to a greater extent than theywould intensify their training facilities in the casethat business expectations improved to the same ex-tent.

From a human-capital theory perspective, firms’training behaviour depends on enterprise-specificcharacteristics such as the human capital distribu-tion of the employees, recruiting strategy, industryor firm size (Neubäumer/Bellmann 1999; Dietrich2000; Dietrich/Gerner 2005; Niederalt 2004, 2005;Zwick 2007).

3 Econometric analysis Ð Part 1:static approach

3.1 Econometric strategy, data, variablesand hypothesis

Analytically there are two basic management deci-sions Ð first a qualitative decision: either the firmtrains apprentices or it does not, and second thequantitative decision: how many apprentices are tobe trained by an individual firm (Niederalt 2004,2005).

To model the qualitative decision, we apply logittechniques. To use the structure of the data, we con-trol for unobserved heterogeneity by estimatingboth a logit model with random effects (see Conway1990; Liu/Pierce 1994) and a logit model with fixedeffects (Chamberlain 1980). A Hausman test is em-ployed to identify the most appropriate approachfor interpretation, given the assumptions made.2

The quantitative decision is modelled by using linearregressions. Again fixed effects and random effectsmodels are calculated to control for unobserved het-erogeneity (Arellano 2003). A Hausman test sup-ports the decision regarding the model adequacy.

The data used for the econometric analysis are de-rived from the IAB Establishment Panel. These data

2 A Hausman test is applied to decide whether the coefficients ofa random effects model Ð which is the default Ð are biased. Inthis cased fixed effects models are preferred.


are collected annually by means of personal inter-views with managers of a randomised sample offirms. In western Germany the first wave of the IABEstablishment Panel was carried out in 1993; since1996 the sample has also covered eastern Germany.The IAB Establishment Panel is a multiply stratifiedsample of all German firms which have at least oneemployee covered by social security (Kölling 2001).The fact that the panel is unbalanced (Bellmannet al. 2002) is ignored in our analysis.3 The subsam-ple used for this paper does not include firms withfewer than ten employees. Such smaller firms wouldhardly be able to recruit new apprentices every year.

The endogenous variable in our logit analyses is adummy which is zero for enterprises which do notrecruit an apprentice in a given year and is one ifat least one new training contract is concluded. Theendogenous variable in the linear regressions isgiven by the rate of change of the number of newapprenticeship contracts against the previous year.4

In both models we introduce a given set of exoge-nous variables. The variable “short-term businessexpectations” uses information which is measureddirectly. In each panel wave since 1993 firms havebeen asked to answer the question: “How do youexpect the business volume to develop in the currentyear (t), as compared with the previous year (t-1)?Is it expected to remain constant, to increase or todecrease?” In the case of change, the change in per-centage points is asked for. For our analysis wemake use of the given raw scores. We assume a posi-tive correlation of the exogenous variable “changein business expectations” with both endogenoustraining variables. In addition a dummy variable isconstructed which has the value of one if the sign ofthe change in business expectations is negative andthis variable interacts with the business expecta-tions.5 The influence of this new variable is expectedto be positive.6 This strategy enables us to identify

3 Following Nijman/Verbeek (1992) unit non response in unbal-anced panels should be at random; which is the case with the IABestablishment panel (Hartmann/Kohaut 2000).4 Ratet = NtÐNtÐ1 /NtÐ1; with Nt = number of new apprentices intime t.5 Alternative models were estimated, including the dummy itself;the results show no significant effect either of the dummy itselfor of the interaction term for the OLS model; we therefore usethe reduced form here. Analogously we estimated both variantsfor the logit model. Again, from a qualitative viewpoint, the twomodels show identical results.6 As e. g. Ai/Norton (2003) mentioned, the use of interactionterms in nonlinear models such as logit models causes some fun-damental problems. In the given case, however, the coefficientfor business expectations and the interaction term identifying anegative change should just be interpreted simultaneously. Addi-tional effects result from the control variables. In detail a mar-ginal reduction in business expectations will be compared to thereference situation in the representative firm; the reference situa-tion is defined by business expectations of 0 % and the represent-

224 ZAF 2 und 3/2007

different coefficients depending on the sign of thebusiness expectations (descriptive statistics of theexogenous variables can be found in Tables A1 andA2 in the appendix).

A further set of exogenous variables serves mainlyto control for observed heterogeneity of firms. Theshare of skilled employees among all employees ran-ges between zero and one and is an indicator offirms’ demand for skilled workers. As we assume,firms recruit at least some of their skilled staff fromtheir apprentices, the correlation between this varia-ble and the training behaviour of the firms shouldbe positive.

The technical state of the firm should influence thetraining decision in a positive way (Franz et al.2000). As there is an expected complementarity be-tween human capital and physical capital (Filer etal. 1996), especially firms using the latest technologyare limited when recruiting skilled staff on the la-bour market. Firms with state-of-the-art technologyare coded with a value of one whereas firms usingobsolete technology are coded with five.

The staff turnover rate as the share of personneloutflow as a percentage of the total number of em-ployees has a serious impact on the investment mo-tive of apprenticeship training. In case the invest-ment motive is decisive for the training decision weassume a negative effect, as firms’ returns fromtraining seem to be limited. Analogously we expectthe effect of staff turnover for the quantitative deci-sion.

The share of atypical workers, such as freelancersor agency workers, in the workforce should have anegative influence because enterprises which makeconsiderable use of these groups of workers are ex-pected to recruit know-how from outside. A rate of100 % is given a value of one.

Collective bargaining often includes negotiationsabout firm-based apprenticeship training. Firmswhich are involved in such negotiations are morelikely to opt for apprenticeship training and will alsorecruit more apprentices.

The probability of recruiting at least one apprenticeis simply correlated with firm size, measured in

ative firm by the mean vector of the remaining exogenous varia-bles. Statistical significance of the gross effect of a negativechange is to be identified by a simple Wald test; the identificationof the standard error using the so-called delta method (see Ai/Norton 2003; Norton et al. 2004; Xu/Long 2005; Oehlert 1992) isnot required.


terms of the number of employees; however thereare no precise arguments concerning the intensity ofapprenticeship training.

Industry dummy variables control for the primary,manufacturing and services sectors. Compared withother sectors the manufacturing industry has a longtradition of apprenticeship training and up to nowapprenticeship training has taken place there in veryspecific training occupations; furthermore training isless school-based and more practical in this field. Fi-nally time dummies are included.

3.2 Results of the static analysis

The results of the logit estimates can be found intable 1. The signs for significant coefficients areequal in both the random effects and the fixed ef-

ZAF 2 und 3/2007 225

fects model. A Hausman test favours the randomeffects model; so the interpretations are based onthis model.

Whereas improvements in business expectations donot have an influence on the probability of recruit-ing at least one new apprentice, an influence canbe seen if business expectations are decreasing. Ina representative firm the probability decreases byaround 0.3 percentage points if business expecta-tions fall by one percentage point; this effect ishighly significant.

The technical state of the firm and the existence ofcollective bargaining do have the expected positiveinfluence on the training decision. Following the as-sumption of a complementary relationship betweenphysical and human capital, the training probabilityincreases with the technical state. What is remark-


able is the fact that this probability increases if anenterprise changes from the status “there is no col-lective bargaining” to the status “there is collectivebargaining” and vice versa Ð the marginal effect inthe representative firm being four percentagepoints.

The results confirm a positive correlation betweenthe number of employees and a firm’s training activ-ity. It is more or less plausible that the training prob-ability increases with firm size. The relative entrycost into apprenticeship training decreases as firmsize increases.

The share of skilled employees improves a firm’sprobability of recruiting new apprentices, and firms in

226 ZAF 2 und 3/2007

the manufacturing sector are more likely to train ap-prentices than other firms. Firms with a high staffturnover tend to abstain from concluding new ap-prenticeship contracts. Finally no significant influencecan be found from the share of atypical workers.

Table 2 displays the results of the linear regressions;a Hausman test prefers the fixed effects model.However, the signs of the significant coefficients areagain the same in the two estimated models.

An improvement in the business expectations by onepercentage point induces an improvement in the vo-cational training engagement of 0.4 percentagepoints Ð an asymmetric behaviour like in the logitmodels cannot be found. Staff turnover has a decreas-


ing effect. The share of atypical workers has a signifi-cant positive influence, which is not in line with ourexpectations. Plausible, however, is the positive corre-lation between a change in the number of employeesand a change in the vocational training engagement.Moreover the estimations indicate a positive influ-ence of the share of skilled employees and the exis-tence of collective bargaining, whereas the coeffi-cients are supposed to be inconsistent. The weak ex-planatory power, however, is unsatisfying; both mod-els only explain about one per cent of the variance.

Altogether it can be concluded that structural deter-minants obviously do have a greater influence onthe decision as to whether a firm should provide atleast one new apprenticeship than on the extent ofengagement in vocational training. Thereforechanges in the German economy, especially thechange from an industrial to a services economy,have probably reduced the number of firms whichprovide apprenticeships at all. Indications regardinga cyclical dependence are quite different in the twoapproaches. Thus the coefficients for the businessexpectations do have the expected signs, at leastwhen they are significant. In contrast, however, thefact that an improvement in business expectationsdoes not influence the probability of opting for newtraining at all is surprising. The estimations indicatethat the extent of engagement in training dependson improvements and also on negative develop-ments in business expectations. Asymmetric behav-iour seems to occur in the logit case only. One possi-ble reason for this result is the fact that until nowwe have ignored possible dynamic aspects, which arequite plausible because reaction takes time. Thisconsideration is the subject of the next section.

4 Econometric analysis Ð Part 2:dynamic approach

4.1 Econometric model, data andvariables

In this section we model explicitly dynamic aspectsby estimating an “autoregressive distributed lag(ADL) model” with fixed effects. In our strategy wefollow Sheldon (2003). Adapting the model to ourquestion we use as explanatory variables not onlythe business expectations in t but also the businessexpectations lagged for several periods and thelagged information on new apprenticeship contracts.Fixed effects control for additional firm-specific fac-tors influencing the training decision such as thequalification level of the firm’s workforce, the indus-try of the enterprise or firm size. It should be noted

ZAF 2 und 3/2007 227

that Sheldon (2003) estimated an error correctionmodel (EC), but due to the fact that these two ap-proaches (ADL and EC) are structurally identical,the method to control for unobserved heterogeneityby ordinary within transformation can be applied toADL directly.7

If the residuals in an ADL model are independentand identically distributed (iid) and there is not aunit root in the variables, an OLS model usuallyleads to consistent parameter estimates (Verbeek2002; Beck 2005). In particular serially correlated er-rors cause serious problems in this context (Beck2005; Wolters 2003; Baltagi 1998). To avoid thisproblem we choose the appropriate lag structure,based on a Breusch-Godfrey test. Another problemcould, however, arise from the fact that the withintransformation induces another endogeneity prob-lem in a dynamic model by construction (Greene2003). As our model selection is based on a Breusch-Godfrey test, the resulting inconsistency problemshould be small.8

Following Carruth/Dickerson (2003), we define adummy variable and include an interaction term be-tween this dummy and the exogenous and endoge-nous variables in our estimates (the dummy has thevalue of one if business expectations are negative)to model a possible asymmetry.

Box 1 illustrates the model selection process in asimplified form. The application of this proceduresuggests the following model:

yi,t = αi + Θ1yi,tÐ1 + Θ2yi,tÐ2 + Θ3δi,tδi,tÐ1δi,tÐ2 yi,tÐ2

+ Θ4yi,tÐ3 + ø1xi,t + ø1xi,tÐ2 + λt + εi,t

With:

αi: Firm-specific effect (fixed effect)yi,t: Rate of change of the number of new

apprenticeship contracts against theprevious year, firm i in year t

xi,t: Business expectations, firm i in year tδi,t: Dummy which has the value of one if

the sign of the business expectations isnegative, firm i in year t

λt: Time-specific effectεi,t: Error term.

7 It would also be plausible to use a vector autoregressive modelwith fixed effects. As the endogeneity or causality is clear, how-ever, we preferred ADL (Hsiao 2004; Verbeek 2002). Up to nowdynamic panel models controlling for selectivity are not common;besides a lack of implemented standard statistics programmes, themain argument for not including selectivity in the models is theextent of complexity.8 Another indication in this direction is the fact that the differ-ence between the coefficients of the business expectations in t isnot statistically significant in the static and the dynamic models,which will be discussed later.


Our approach opens up the opportunity to identifydifferences in firms’ adjustment behaviour depend-ing on whether business expectations are good orbad for just one period or for a longer time. More-over it is possible to see asymmetries in the adjust-ment behaviour of the firms. One central assump-tion is that a firm immediately changes its adjust-ment scheme if the sign of business expectationschanges. What is quite interesting is the result thatWald tests do not find a significant influence of theinteraction term between the dummy and the exoge-nous variable, which is in line with the results in thestatic case.

The estimates are again based on the IAB Establish-ment Panel. The exogenous variable is again busi-ness expectations compared with the activity a yearearlier. As in the static model, the endogenous vari-able is the change in the number of new apprentice-ship contracts compared with the previous year Ðwe take the rate of change. Again we only take firmsinto account which have at least ten employees; thefact that the IAB Establishment Panel is “unbal-anced” is also ignored.

4.2 Results of the dynamic analysis

The results can be found in Table 3. As mentionedabove, firms’ responses to business expectations aresymmetrical with respect to the direction if we look

228 ZAF 2 und 3/2007

at the short run Ð in the case of a positive (negative)change in business expectations of one percentagepoint, the engagement in vocational training in-creases (decreases) by 0.35 percentage points. Itshould be mentioned that the result is really in linewith those obtained in the static case, the differenceis statistically insignificant. In the case that the signof business expectations does not vary for a longertime (for at least three periods) firms choose a lowerequilibrium rate of change in the negative case, al-though the asymmetry is not really significant (0.21percentage points in the positive case to 0.185 in thenegative case). Finally the strong quantitative corre-lation between the exogenous and the endogenousvariables is quite interesting. This indicates a strongcyclical dependence of the supply of apprentice-ships.

One possible economic explanation for the identi-fied (weak) asymmetry could be the fact that firms,irrespective of cyclical factors, train something likea supplement amount, which is not “touched” in the


case of short-term decreases in business expecta-tions. Such an explanation could be justified with thetheory of internal labour markets (Saint-Paul 1996;Doeringer/Piore 1971).

5 Application of the estimates

Increasing uncertainty regarding the economic fu-ture has been mentioned recently by various authorsas being an influential factor for a decreasing com-mitment of firms to training apprentices (Bellmann/Hartung 2005; DIHK 2006). Within this paper in-creasing uncertainty is defined as an increase in thechanges of the sign of business expectations, whilstthe mean of the individual firms’ business expecta-tions remains constant. This definition is in line withthe risk concept of Rothschild/Stiglitz (1970), whichdefines an increase in risk as a “mean preservingspread”. Moreover, it should be mentioned that,strictly speaking, uncertainty and risk are two differ-ent concepts, but it is usual to ignore this differenceif this distinction is not the subject of the paper (Mil-ler 2002).9

Taking a look at the estimates regarding the proba-bility of providing training, business expectationsonly induce firms to change from the status “newapprenticeships” to “no new apprenticeships”. Un-certainty as defined above would obviously lead toa reduction in the number of firms which providetraining.

Finally both the static and the dynamic models indi-cate that there is no asymmetry in firms’ behaviouras regards adjustment to business expectations inthe short run. Thus a negative reaction regarding en-gagement in vocational training, induced by poorbusiness expectations, reduces the basis (the level ofthe engagement) so considerably that a subsequentsymmetric positive reaction induced by good busi-ness expectations would be not strong enough tocompensate for the reduction. In the long run how-ever, the dynamic estimates give an indication thatnegative reactions are less sensitive, but it is clearlydoubtful that it makes sense to simulate increasinguncertainty by some long-run changes in the sign ofthe business expectations (which means the signdoes not change until the rate of change of the en-gagement in vocational training is in equilibrium), if

9 As an alternative to this purely experimental approach it wouldbe possible to include uncertainty directly into the econometricanalysis. One way to do so could be to include a dummy variablewhich indicates that the firm has no idea of their business expec-tations. Such a variable, however, has no statistical significant in-fluence in our static analysis.

ZAF 2 und 3/2007 229

it is taken into account that in the model contextlong term means around six years and the estima-tions are based on only seven years. Moreover somelong-run changes in the sign of business expectationsare usually not recognised as increasing uncertaintybut more or less as changes in profit expectations.

Hence the conclusion seems to be justified that in-creasing uncertainty regarding future prospects, forexample business expectations, could be a reason forthe decreasing amount of apprenticeship arrange-ments.

6 Conclusions and further options

The econometric estimations based on the IAB Es-tablishment Panel data provide some evidence thatfirms respond sensitively to changes in short-termbusiness expectations and adapt their training be-haviour correspondingly. These findings correspondwith a strong interrelation between training behav-iour and the business cycle.

In a short-term perspective, a decrease in businessexpectations affects the net cost assumptions offirms’ training decisions. From a perspective ori-ented towards long-term investment, a decrease inbusiness expectations affects the assumptions on fu-ture transaction and opportunity costs.

The empirical findings provide some evidence thatthe sectoral change from the manufacturing industryto the services industry explains some of the reduc-tion in the supply of apprenticeship places, as thedensity of apprenticeship places in the manufactur-ing sector is higher than in the services sector.

Furthermore there is some empirical evidence thatfirms’ departure from collective wage agreements(see Schnabel 2003) reduces both the decision toprovide training and the number of training places.But more research is needed to decide whether thetwo decisions are interrelated or not.

As assumed, the logit model provides some empiri-cal evidence of asymmetric behaviour of firms; thesefindings correspond with our assumptions on theloss-averse behaviour of firms. Again alternative hy-potheses have still to be tested: the costs of entryinto apprenticeship training are higher than exitcosts. How does this cost asymmetry affect firms’training behaviour? It is surprising, however, that inthe static model the decision to provide apprentice-ship training seems to be independent of business


expectations, in contrast to the decision against pro-viding apprenticeship training.

According to our assumptions, firms’ adjustment oftheir training capacities is positively correlated withthe direction of the change in business expectations.However, there is no empirical evidence for the ex-pected asymmetric adjustment behaviour in thestatic model and an inverse, but weak correlation inthe dynamic model.

The empirical findings are in line with our assump-tions about the correlation between increasing un-certainty regarding business expectations and thetraining behaviour. What needs to be done is to im-plement our assumptions directly in the econometricmodelling.

With respect to these limitations, however, it waspossible to show that our empirical results concern-ing the firms’ fundamental training decisions(whether or not to be a training firm and the num-ber of training places) support both our structuralexplanation and our assumptions on the relevanceof business expectations and their change over timeon these training decisions. Finally our results corre-spond with theoretical assumptions on uncertainty.

The formation of individual firms’ business expecta-tions by a given uncertainty over time seems, so far,to be an underestimated explanation but a relevantfactor in firms’ contribution to human capital forma-tion.

References

Ai, C./Norton, E. C. (2003): Interaction Terms in Logit andProbit Models. In: Economics Letters, Vol. 80, pp. 123Ð129.

Acemoglu, D./Pischke, J.-S. (1998): Why Do Firms Train?Theory and Evidence. In: The Quarterly Journal of Eco-nomics, Vol. 113, pp. 79Ð119.

Acemoglu, D./Pischke, J.-S. (1999a): Beyond Becker: Train-ing in Imperfect Labour Markets. In: Economic Journal,109, 112Ð142.

Acemoglu, D./Pischke, J.-S. (1999b): The Structure of Wa-ges and Investment in General Training. In: Journal ofPolitical Economy, Vol. 107, pp. 539Ð572.

Arellano, M. (2003): Panel Data Econometrics. New York(Oxford University Press).

Backes-Gellner, U. (1992): Berufsbildungssysteme und dieLogik betrieblicher Qualifizierungsstrategien im interna-tionalen Vergleich. Gibt es einen Wettbewerb der Sys-teme? In: Boettcher, E./Herder-Dornreich, P./Schenk, K.

230 ZAF 2 und 3/2007

(Hrsg.): Jahrbuch für neuere politische Ökonomie. Tü-bingen: 245Ð270.

Baltagi, B. H. (1998): Econometrics. Heidelberg (Springer).

Bank, M. (2003): Dispositionstheorie: Eine Theorie zurErklärung von Kauf- und Verkaufsentscheidungen vonInvestoren in risikobehafteten Assetmärkten. http://www.uibk.ac.at/c/c4/c409/kerschbamer/w04/BBLP-Bank.pdf.

Beck, N. (2005): Time Series Cross-Section Data Ð Dy-namics Ð Continuous. http://www.nyu.edu/classes/nbeck/q2/morning.short.pdf

Becker, G. S. (1962): Investment in Human Capital: A The-oretical Analysis. In: Journal of Political Economy, Vol.70, pp. 9Ð49.

Becker, G. S. (1964): Human Capital. A Theoretical andEmpirical Analysis, with Special Reference to Educa-tion. New York, London (NBER).

Beckmann, M. (2002): Firm-sponsored ApprenticeshipTraining in Germany. Empirical Evidence from Estab-lishment Data. In: Labour: Review of Labour Econom-ics and Industrial Relations Vol. 16 (2), pp. 287Ð310.

Beicht, U./Walden, G. (2004): Empirische Ergebnisse zumNutzen der betrieblichen Berufsausbildung: Ein Vergleichzwischen technischen und anderen Berufen. http://www.bmbf.de/pub/sdi_02_04.pdf.

Bellmann, L./Hartung, S. (2005): Betriebliche Ausbildung:Zu wenig Stellen und doch sind nicht alle besetzt. IAB-Kurzbericht Nr. 27.

Bellmann, L./Kohaut, S./Lahner, M. (2002): Das IAB Be-triebspanel Ð Ansatz und Analysepotenziale. In: Klein-henz, G. (ed.): IAB-Kompendium Arbeitsmarkt- undBerufsforschung. Nürnberg, pp. 13Ð20.

Büchel, F./Neubäumer, R. (2001): AusbildungsinadäquateBeschäftigung als Folge branchenspezifischer Ausbil-dungsstrategien. In: Mitteilungen aus der Arbeitsmarkt-und Berufsforschung, Vol. 34, pp. 269Ð285.

Carruth, A. A./Dickerson, A. P. (2003): An Asymmetric Er-ror Correction Model of UK Consumer Spending. In:Applied Economics, Vol. 35, pp. 619Ð630.

Chamberlain, G. (1980): Panel Data. In: Griliches, Z./Intri-ligator, M. D. (ed.): Handbook of Econometrics, Vol. 2,Amsterdam (North-Holland).

Clark, D./Fahr, R. (2002): Transferability, Mobility andYouth Training in Germany and Britain: A Simple Theo-retical Analysis. In: Konjunkturpolitik Vol. 48 (3Ð4),235Ð255.

Conway, M. R. (1990): A random effects model for binarydata. In: Biometrics, Vol. 46, pp. 317Ð328.

Dietrich, H. (2000): Betriebliches Ausbildungsverhaltenim Kontext der betrieblichen Ausbildungsstruktur. In:George, R./Struck, O. (eds.): Generationenaustausch imUnternehmen. München/Meiningen: pp. 159Ð176.


Dietrich, H./Gerner, H.-D. (2005): Theorie im Praxistest:Warum Betriebe in die Ausbildung investieren. In: IABForum, Vol. 2, pp. 56Ð61.

DIHK (2006): Impulse für mehr Ausbildung Ð Die Sichtder Unternehmen Ð Ergebnisse einer Online-Befragungvon 7500 Unternehmen.http://www.ausbildung-plus.de/allgemein/news/00469/index.html.

Doeringer, P. B./Piore, M. J. (1971): Internal Labor Mar-kets. Lexington.

Euwals, R./Winkelmann, R. (2004): Why Do Firms Train?Empirical Evidence on the First Labour Market Out-comes of Graduated Apprentices. In: International Jour-nal of Manpower, Vol. 25 (5), 447Ð463.

Filer, R. K./Hamermesh, D./Rees, A. (19966 Ed.): The Eco-nomics of Work and Pay. New York.

Fougere, D./Schwerdt, W. (2002): Are Apprentices Produc-tive?. In: Konjunkturpolitik Vol. 48 (3Ð4), 317Ð346.

Franz, W./Soskice, D. (1995): The German Apprentice Sys-tem. In: Buttler, F./Franz, W./Schettkat, R./Soskice, D.(eds.): Institutional Frameworks and Labour MarketPerformance. London, New York: pp. 208Ð234.

Franz, W./Steiner, V./Zimmermann, V. (2000): Die betrieb-liche Ausbildungsbereitschaft im technologischen unddemographischen Wandel. Baden-Baden (Nomos).

Freeman, R. B. (1972): Labor Economics. New Jersey.

Greene, W. H. (2003): Econometric Analysis. Upper SaddleRiver (Pearson Education).

Gul, F. (1991): A Theory of Disappointment Aversion. In:Econometrica, Vol. 59, pp. 667Ð686.

Harhoff, D./Kane, T. (1997): Is the German ApprenticeshipSystem a Panacea for the U.S. Labor Market?. In: Popu-lation Economics Vol. 10, pp. 171Ð196.

Hartmann, J./Kohaut, S. (2000): Analysen zu Ausfällen(Unit-Nonresponse) im IAB-Betriebspanel. In: Mittei-lungen aus der Arbeitsmarkt- und Berufsforschung, Vol.33, 609Ð618.

Holtbrügge, D. (2004): Personalmanagement. Berlin(Springer).

Hsiao, C. (2004): Analysis of Panel Data. Cambridge (Uni-versity Press).

Kölling, A. (2001): Ein „Schalter“ für die Forschung ÐDaten des IAB-Betriebspanels stehen externen For-schern seit 1999 zur Verfügung. In: IAB Werkstattbe-richt Number 9.

Lindley R.M. (1975): The Demand for Apprentice Recruitsby the Engineering Industry: 1951Ð1971. In: ScottishJournal of Political Economy, Vol. 22, Number 1, pp. 1Ð24.

Liu, Q./Pierce, D. A. (1994): A Note on Gauss-HermiteQuadrature. In: Biometrika, Vol. 81, pp. 624Ð629.

ZAF 2 und 3/2007 231

Miller, R. M. (2002): Experimental Economics Ð How WeCan Build Better Financial Markets. New Jersey.

Muehlemann, S./Schweri, J./Winkelmann, R./Wolter, S.(2007): An Empirical Analysis of the Decision to TrainApprentices. In: Labour Vol. 21, pp. 419Ð441.

Neubäumer, R./Bellmann, L. (1999): Ausbildungsintensitätund Ausbildungsbeteiligung von Betrieben. Theoret-ische Erklärungen und empirische Ergebnisse auf derBasis des IAB Betriebspanels 1997. In: Beer, D./Frick,B./Neubäumer, R./Sesselmeier, W. (eds.): Die wirtschaft-lichen Folgen von Aus- und Weiterbildung. München/Mehring (Hampp): 9Ð41.

Niederalt, M. (2004): Zur ökonomischen Analyse betrieb-licher Lehrstellenangebote in der BundesrepublikDeutschland. Frankfurt/Main (Lang).

Niederalt, M. (2005): Bestimmungsgründe des betriebli-chen Ausbildungsverhaltens in Deutschland. In: Lehr-stuhl für Arbeitsmarkt- und Regionalpolitik, Friedrich-Alexander Universität Erlangen-Nürnberg, Diskussions-papier, No 36.

Nijman, T./Verbeek, M. (1992): Nonresponse in Panel Data:The Impact on Estimates of a Life Cycle ConsumptionFunction. In: Journal of Applied Econometrics, Vol. 7,243Ð257.

Norton, E. C./Wang, H./Ai, C. (2004): Computing Interac-tion Effects and Standard Errors in Logit and ProbitModels. In: Stata Journal, Vol. 4, pp. 154Ð167.

Oehlert, G. W. (1992): A Note on the Delta Method. In:The American Statistican, Vol. 46, pp. 27Ð29.

Rothschild, M./Stiglitz, J. E. (1970): Increasing Risk: ADefinition. In: Journal of Economic Theory, Vol. 2, pp.225Ð243.

Sadowski, D. (1980): Berufliche Bildung und betrieblichesBildungsbudget. Stuttgart (Poeschel).

Saint-Paul, G. (1996): Dual Labor Markets Ð A Macroeco-nomic Perspective. Cambridge/London (MIT).

Schnabel, C. (2003): Tarifpolitik unter Reformdruck.Gütersloh (Bertelsmann).

Sheldon, G. (2003): Die Auswirkung der Ausländer-beschäftigung auf die Löhne und das Wirtschaftswachs-tum in der Schweiz. http://www.wwz.unibas.ch/fai/pages/arbeitspapiere/auslaenderbeschaeftigung.pdf.

Smits, W./Zwick, T. (2004): Why do business service firmsemploy fewer apprentices? A comparison between Ger-many and The Netherlands. In: International Journal ofManpower Vol. 25, pp. 36Ð54.

Stevens, M. (1994): An Investment Model for the Supplyof Training by Employers. In: The Economic Journal,Vol 104, pp. 556Ð570.

Timmermann, D. (1998): Nutzen aus der Sicht der Wissen-schaft. In: Bundesinstitut für Berufsbildung (ed.): Nut-


zen der beruflichen Bildung. Bielefeld: Bertelsmann,pp. 75Ð92.

Verbeek, M. (2002): A Guide to Modern Econometrics.Chichester (Wiley).

Werwatz, A. (1996): How Firm-Specific is German Appren-ticeship Training? Discussion paper 96Ð12, Humbold-Universität Berlin.

Williamson, O. E. (1990): Die ökonomischen Institutionendes Kapitalismus. Tübingen (Mohr).

Wolters, J. (2003): Neuere Entwicklungen in der ökonomet-rischen Analyse aggregierter Zeitreihen. In: Franz, W./

232 ZAF 2 und 3/2007

Ramser, H.J./Stadler, M. (eds.): Empirische Wirtschafts-forschung Ð Methoden und Anwendungen. Tübingen(Ottobeuren).

Xu, J./Long, S. (2005): Using the Delta Method to Con-struct Confidence Intervals for Predicted Probabilities,Rates, and Discrete Changes. Indiana University,Bloomington (Mimeo).

Zwick, T. (2007): Apprenticeship Training in Germany ÐInvestment or Productivity Driven? ZEW-DiscussionPaper No. 07Ð023.


Appendix

ZAF 2 und 3/2007 233

Documents

The determinants of apprenticeship training with …doku.iab.de/zaf/2007/2007_2-3_zaf_dietrich_gerner.pdfThe determinants of apprenticeship training with particular reference to business