18
 North American containerport productivity: 1984–1997 Hugh Turner  * , Robert Windle, Martin Dresner Robert H. Smith School of Business, University of Maryland, College Park, MD 20742, USA Received 31 January 2003; received in revised form 31 March 2003; accepted 1 June 2003 Abstract This paper undertakes two tasks: measurement of seaport infrastructure productivity growth in North America from 1984 to 1997, and exploration of several theorized causal relationships between  infrastructure  productivity  and industry  structure  and  conduct. A methodology is presented, data envelopment analysis (DEA), for measuring infrastructure productivity. Tobit regression is presented as a means of examining the determinan ts of infra struct ure productiv ity in seapo rts. The stud y suppo rts the presence of economies of scale at the containerport and terminal level. Among other factors, the longstanding relationship be- tween seaports and the rail industry appears to remain a critical determinant of containerport infrastructure productivity.  2003 Elsevier Ltd. All rights reserved. Keywords:  Containerport; Containerport productivity; Infrastructure; Data envelopment analysis; Tobit 1. Introduction The exi stence of sys temic unprodu cti ve inf ras tructur e in Nor th Ame ric an sea por ts ser vin g containerized trade, i.e.  containerports, as well as the root causes and magnitude should such a problem exist, is a question that has been debated for over two decades.  1 Despite this debate, empirical research addressing this issue is extremely limited. Theorized causes of over invest- ment in seaport capacity are based upon either industry structure, in particular the presence of * Corresponding author. E-mail addresses:  [email protected]  (H. Turne r), rwindl e@rhs mith.umd.edu (R. Windle ), mdres ner@ rhsmith.umd.edu (M. Dresner). 1 Hershmann et al. (1978), De Neufville and Tsunokawa (1981), National Research Council (1986), Hayuth (1988), Heikki la (1990) , James (1991), Slack (1993), MARAD (1994), Heaver (1995), Burke (1996), Mongelluzz o (1996, 1997, 1998), Fleming (1997), and USDOT (1998). 1366-5545/$ - see front matter    2003 Elsevier Ltd. All rights reserved. doi:10.1016/j.tre.2003.06.001 Transportation Research Part E 40 (2004) 339–356 www.elsevier.com/locate/tre

1881

  • Upload
    mita000

  • View
    212

  • Download
    0

Embed Size (px)

DESCRIPTION

hhg

Citation preview

  • 1. Introduction

    * Corresponding author.

    E-mail addresses: [email protected] (H. Turner), [email protected] (R. Windle), mdresner@

    rhsmith.umd.edu (M. Dresner).1 986), Hayuth (1988),

    gelluzzo (1996, 1997,

    Transportation Research Part E 40 (2004) 339356www.elsevier.com/locate/treHershmann et al. (1978), De Neufville and Tsunokawa (1981), National Research Council (1

    Heikkila (1990), James (1991), Slack (1993), MARAD (1994), Heaver (1995), Burke (1996), MonThe existence of systemic unproductive infrastructure in North American seaports servingcontainerized trade, i.e. containerports, as well as the root causes and magnitude should such aproblem exist, is a question that has been debated for over two decades. 1 Despite this debate,empirical research addressing this issue is extremely limited. Theorized causes of over invest-ment in seaport capacity are based upon either industry structure, in particular the presence ofNorth American containerport productivity: 19841997

    Hugh Turner *, Robert Windle, Martin Dresner

    Robert H. Smith School of Business, University of Maryland, College Park, MD 20742, USA

    Received 31 January 2003; received in revised form 31 March 2003; accepted 1 June 2003

    Abstract

    This paper undertakes two tasks: measurement of seaport infrastructure productivity growth in North

    America from 1984 to 1997, and exploration of several theorized causal relationships between infrastructure

    productivity and industry structure and conduct. A methodology is presented, data envelopment analysis

    (DEA), for measuring infrastructure productivity. Tobit regression is presented as a means of examining

    the determinants of infrastructure productivity in seaports. The study supports the presence of economies

    of scale at the containerport and terminal level. Among other factors, the longstanding relationship be-tween seaports and the rail industry appears to remain a critical determinant of containerport infrastructure

    productivity.

    2003 Elsevier Ltd. All rights reserved.

    Keywords: Containerport; Containerport productivity; Infrastructure; Data envelopment analysis; Tobit1998), Fleming (1997), and USDOT (1998).

    1366-5545/$ - see front matter 2003 Elsevier Ltd. All rights reserved.doi:10.1016/j.tre.2003.06.001

  • economies of scale and/or location and the lumpiness of capital provision or the conduct ofpublic port authorities and ocean carriers. 2

    We have two goals for this paper. The rst is to measure the growth in productivity of seaportinfrastructure in North America from 1984 to 1997 in order to assess trends in productivity. Amethodology is presented for measuring infrastructure productivity, data envelopment analysis

    Waterborne cargo ows at a seaport are often broken down into various categories based on

    340 H. Turner et al. / Transportation Research Part E 40 (2004) 339356the characteristics of the commodities being transported. General cargo is essentially all cargo notdened as bulk cargo. 3 While manufactured goods are the dominant general cargo, virtually anycargo can be carried as general cargo. Prior to 1957, nearly all general cargo was transported inbreak-bulk vessels, however, since the mid-1970s, most general cargo has been transported incontainers. Containerization refers to the unitization or aggregation of freight into standardizedmetal shipping containers. Containerization dramatically reduces labor costs and is considered toreduce damage rates. In addition, it greatly reduces the time these specially congured generalcargo vessels spend in the seaport as well, as the time containerized freight spends in transit. Thelatter savings result from the relative ease of exchange between modes as well as the reduction inhandling of the actual freight. However these savings come only through substantial seaport andvessel capital investment. In addition, ineciencies and congestion at any point in the system cangreatly reduce the benets.Marine container terminals may be under the direct control of the port authority or operated as

    a franchise granted by the port authority. In the case of the franchise, the operator is either an

    2 Regarding structure see: Walters (1976), Bennathan and Walters (1979), De Neufville and Tsunokawa (1981).

    Regarding conduct see: De Neufville and Tsunokawa (1981), Slack (1993), Burke (1996), and Fleming (1997).3 Stopford (1988, p. 185) observes that the range of items transported as general cargo is almost limitless. General

    cargo can be carried in the cargo holds of break-bulk vessels or in cargo containers stowed on their decks. However, for(DEA), that addresses several problems noted in previous seaport performance research. Thesecond task of this paper is to explore factors that impact infrastructure productivity. A causalrelationship between infrastructure productivity and industry structure and conduct is theorizedand an appropriate methodology, Tobit regression, is presented as a means of examining thedeterminants of infrastructure productivity in seaports.This study oers signicant contributions for both public policy leaders and a broad range of

    managerial decision makers concerned with the productivity of the North American seaportindustry. From a public policy perspective, employing infrastructure productivity as a perfor-mance measure addresses allocation of scarce resources and social welfare considerations. In anindustry that receives substantial public nancial support, ecient use of scarce public funds is ofinterest. From a managerial perspective, the many public and private managers whose organi-zations interact at North American seaports clearly would benet from an understanding oftrends in productivity and factors that inuence that productivity. Of particular individual interestare the factors under which their own organizations may exert as measures of control.

    2. Backgroundcontainers, the dominant form of ocean carriage is the fully-cellular containership.

  • independent for-prot marine terminal operator (MTO) or an ocean carrier-controlled marineterminal operator; the latter frequently given the option of operating the assigned marine terminalas a franchise or as an exclusive service to the controlling carrier. Regardless of which organi-

    proxies for all other inputs including labor, noting that labor is thought to be supplied in pro-portion to the number of cranes employed. Similarly, quay length is argued to be proportional to

    H. Turner et al. / Transportation Research Part E 40 (2004) 339356 341the amount of land dedicated to the marshalling and storage of containers.Beyond North America, Kim and Sachish (1986) conducted a time-series study of the Port of

    Ashdod (Israel) employing a translog cost function and total factor productivity (TFP) measures.The time period chosen (19661983) straddles the introduction of container handling technology.The inputs are annual direct labor and seaport capital expenditures with metric tons of generalcargo as the output. The authors nd a signicant contribution to productivity as a result of the

    4 Stevedore refers to the organization whose function is to manage the cargo operations of the terminal. This iszation manages these operations, longshore labor accomplishes the required work. In most in-stances, the marine container terminals within the seaport become the foci of the carrier-portauthority relationship (Heaver, 1995). Together these entities comprise a system designed, at leastin theory, to eciently and eectively serve the needs of international shippers. This study willlook exclusively at containerport productivity in North America by focusing on the inputs andoutputs associated with the containerport segment of the maritime industry.

    3. Literature review

    Despite the signicance of the seaport industry worldwide, relatively few empirical studies havebeen conducted, owing in part to the diculty in obtaining reliable comparable data.Tongzon (1995) conducts a short-term analysis of terminal eciency. His performance measure

    is container throughput (twenty-foot equivalent units/year). The author (Tongzon, 1995, p. 248)nds a signicant correlation between throughput and terminal eciency as the latter is dened asaverage number of containers per berth hour. Others have observed this relationship betweenoutput and eciency, including Caves and Christensen (1988) and De Neufville and Tsunokawa(1981).Chang (1978) employs multiple regression to estimate a CobbDouglas production func-

    tion using annual data over a 21-year period for the port of Mobile, Alabama. Gross portauthority earnings are the output employed by Chang, while the port authoritys net assetsand direct labor are the inputs. Chang does not consider stevedore and longshore labor aninput, so that the labor included represents only the direct labor employed by the port authorityitself. 4 Thus the result is more indicative of port authority productivity than overall seaportproductivity.In a longitudinal study of East Coast port authorities, De Neufville and Tsunokawa (1981) nd

    support for the presence of economies of scale in containerports. The authors consider metric tonsof cargo as the seaports output. Inputs are quay length and the number of cranes employed forthe ve US East Coast seaports comprising their sample. The authors argue that these inputs aredistinct from the longshore labor that is responsible for the physical work of handling cargo.

  • adoption of container handling technology at Ashdod. In addition to containerizations impact onTFP growth, Kim and Sachish (1986) nd support for long run returns to scale.Sachish (1996), in his study of productivity and its determinants at the Israeli seaports of Haifa

    inuence of six explanatory factors, or families, on both partial and total productivity values.Several of these families are latent variables (constructs). In addition to production volume and

    342 H. Turner et al. / Transportation Research Part E 40 (2004) 339356actual capital, Sachish (1996) suggests the number of workers, technological levels,management quality, and exogenous factors aect partial and total productivity.Jara-Daz et al. (2002) estimate a multi-output cost function using a exible form and a sample

    of 26 Spanish seaports over an 11 year period. Outputs are containerized general cargo, break-bulk general cargo, liquid bulk, dry bulk, and total rent received for leases of port space. Inputprices indices for labor, capital, and an index for other prices incurred in the provision of services.Capital employed and total dock length are used as measures of seaport size. The authors resultssupport the presence of economies of scale and scope.De Neufville and Tsunokawa (1981) describe a saw-tooth production function. Demand for

    a containerports services increase eventually reaching the limits of existing capacity. While 100%utilization of the port authoritys infrastructure is ecient from the ports standpoint, any furtherincrease in demand results in costly queues for vessels and cargo. In a competitive situation, theport authority is obliged, if possible, to add capacity. Under such circumstances, a middle-of-the-data approach to estimating a production function may not be as appropriate as a methodologythat estimates the ecient frontier of the production surface such as data envelopment analysis.Such a methodology has been suggested by Roll and Hayuth (1993) as appropriate in relation toseaports.A general conclusion of the literature is that the seaport industry, and the containerport seg-

    ment in particular, is characterized by economies of scale. This results from the capital intensivenature of the industry, the presence of indivisibilities in the provision of terminals, and the costsavings resulting from pooling demand in a system characterized by queues and congestion. 5

    4. Methodology

    Containerport infrastructure productivity is a key performance measure and is inuenced byindustry structure, conduct and demand. For this reason, this paper has two distinct objectives:measurement of the trend in infrastructure productivity during the study period; and examinationof the factors that determine infrastructure productivity.

    5 Chang (1978), Bennathan and Walters (1979), De Neufville and Tsunokawa (1981), Bobrovitch (1982), Jansson andand Ashdod, nds that the volume of activity and capital investment are main inuences on totalproductivity. To reach this conclusion, he develops engineering standards that are then used toderive partial productivity values for each of his 23 observations. These partial productivitymeasures are then weighted to allow the calculation of a total productivity value. In a uniqueapproach, the author employs data envelopment analysis to develop the appropriate weights forhis productivity function. Sachish follows with a linear programming model to determine theShneerson (1982), Kim and Sachish (1986), Varaprasad (1986), Heaver (1995), and Turner (2000).

  • Sachish (1996) identies several methodologies for measuring seaport productivity: econo-metric methods; partial productivity indices; accounting methods; data envelopment analysis(DEA); and engineering approaches. In the words of Charnes et al. (1981) whose methodology,

    H. Turner et al. / Transportation Research Part E 40 (2004) 339356 343DEA, is considered: The objective is to measure the eciency of resource utilization in whatevercombinations are present (loose or tight) in the organizations as well as the technologies utilized.As De Neufville and Tsunokawa (1981) note, short-term productivity analyses of container

    terminals may be biased by the impact of short-term uctuations in demand. Bennathan andWalters (1979, p. 58) add that cross-sectional analyses suer from the lack of comparable portsituations in respect to both trac and geography. It is therefore advisable to adopt a long-termapproach controlling for the impact of changes in industry output and allowing for dynamicadjustments to conditions of supply and demand. This paper will use a pooled time series/crosssection approach, examining a number of North American ports over a 14 year period.Oum et al. (1992) review methods employed in recent productivity research contrasting the two

    broad methodological approaches to productivity measurement: the nonparametric index numberapproach, of which data envelopment analysis is an example; and parametric statistical ap-proaches, i.e. econometrics. Oum et al. (1992) conclude statistical approaches, and specically theestimation of a parametric cost function based on the translog model, are more common today.However the authors state that data availability and reliability can inuence the choice ofmethodology. When physical, or quantity data is more readily available or deemed morereliable in comparison to cost data, they note DEA is an appropriate methodology. Oum et al.(1992, p. 497) observe no major disadvantages come from the nature of the DEA method al-though they do caution outliers can have a signicant impact on DEA scores.Thanassoulis (1993) compares regression analysis to data envelopment analysis when perfor-

    mance assessment is the objective. He concludes that DEA is a more accurate method for e-ciency assessment in that it is a boundary, or frontier, methodology but cautions that it is moreprone to extreme inaccuracies with respect to individual decision making units (DMUs) incomparison to regression. He notes this is due to the sensitivity of DEA scores to data swings atthe individual DMU level.Further support for the methodology can be found in Windle and Dresner (1995) who nd data

    envelopment analysis used in conjunction with Tobit regression may be as useful as parametricmodels, including total factor productivity, for measuring productivity and determining thesources of gains. The authors recommend Tobit regression to decompose DEA scores into thevarious sources of eciency.Based on the above issues and literature, this paper follows Roll and Hayuth (1993) and

    suggests date envelopment analysis be employed to evaluate container port infrastructure pro-ductivity during the study period. The data envelopment method, initially presented by Charneset al. (1978), encompasses the ecient frontier approach of Farrell (1957). DEA structures theproduction process as a constrained optimization problem and solves it using a linear pro-gramming approach. The methodology identies those decision making units (DMUs) that aremost ecient and thus species the shape of the ecient frontier as delineated by these units. 6 All

    6 For this eort the DMU is a containerport-year; e.g. Boston-1990.

  • leasindelayeectesize

    344 H. Turner et al. / Transportation Research Part E 40 (2004) 339356Ocean carrier conduct. Based on queuing theory, Jansson and Shneerson (1982) note that thetotal cost minimizing berth occupancy rate decreases with increasing vessel size. This has beensupported by other researchers including Talley (1990). Given this, increasing vessel size will leadto decreasing berth utilization. As Caves and Christensen (1988) note, utilization levels inuenceproductivity measurement. For a given level of infrastructure, reduced berth occupancy rates, i.e.idle capacity, will translate into a gross reduction in infrastructure productivity, all else equal.However, the relationship between vessel size and productivity is complex in the case of seaports.

    As vessel size increases, investment in port facilities could also be expected to increase as man-agement seeks to address the needs of larger vessels. The model employed below controls for bothvessel size and seaport size and thus attempts to disentangle these confounding eects. Thus it ishypothesized that as average container vessel size increases at a containerport, berth occupancyg to ocean carriers can lead to reduced utilization and productivity and increased totals to carriers and cargo. Dedicated leasing also reduces the impact of returns to scale byively creating smaller ports out of the larger whole. Therefore dedicated leasing is hypoth-d to be negatively correlated with infrastructure productivity.DMUs not on the ecient frontier are scored with reference to the hyperplane dened by thosethat are located on the frontier. The result is a relative measure of eciency.When attempting to explain dierences in data envelopment analysis scores through regression

    analysis, the dependent variable is continuous but truncated at 1.0. As a result, ordinary leastsquares regression is not appropriate, as its use will lead to inconsistent estimates. In such situ-ations, a Tobit regression (Tobin, 1958) is suggested as an appropriate methodology (Maddala,1983). The base Tobit model is similar to ordinary least squares regression but assumes a trun-cated normal distribution in place of the normal distribution and employs the maximum likeli-hood-ratio estimation method.The general form of the Tobit model to be estimated is:

    Infrastructure Productivity

    f seaport industry structure; port authority conduct;ocean carrier conduct;situational factors; control variables:

    The dependent variable measuring infrastructure productivity is the DEA score of each port in agiven year.Seaport industry structure. One possibility for explaining dierences in productivity is the

    presence of returns to scale and density. In order to measure returns to scale the sum of twenty-foot equivalent units handled at all container terminals within the seaport during the year ofobservation is included in the model. Also included is a second measure of scale: the averagecontainer terminal size at the seaport measured as total containerport twenty-foot equivalent units(TEU) divided by the number of container terminals at the seaport.Port authority conduct. It has been noted by Verhoe (1981) that port authorities often pursue

    economic development objectives. In the highly competitive environment of the North Americanseaport industry, large carriers often seek dedicated terminal leases requiring substantial portauthority investment in terminal capacity (Slack, 1993). This capacity reduces costs for the oceancarrier, but may increase the ports costs by requiring investments in terminals in excess of eco-nomically ecient level. As Varaprasad (1986) and Turner (2000) show, dedicated terminal

  • rates decline with infrastructure productivity declines following occupancy rates. Given that therelationship may not be linear, we include the squared vessel size as an additional variable.Situational factors. Intermediacy is dened as the ability to serve as an intermediary between

    regions (Fleming and Hayuth, 1994). Intermediacy is therefore related to the cost, quality andcapabilities of intermodal services, particularly rail carrier services, at the seaport.The eect of intermediacy on seaport infrastructure productivity is captured by three variables

    H. Turner et al. / Transportation Research Part E 40 (2004) 339356 345dening the quality and capability of each seaports connection to the North American intermodalnetwork. The rst variable is the existence of sucient overhead clearance to allow use of double-stack railcars on the railroad lines entering the terminal area of the seaport. The second variable isthe sum of class I rail carriers serving the seaport. 7 In the case of terminal railroads interveningbetween the seaport and the class I carrier, or carriers, as is the case at several study seaports, theseaport is still considered to be served by the class I carrier or carriers. The third variable is thesum of terminal hectares with immediate access to on-dock rail connections divided by totalterminal hectares. For a terminal to be considered as having on-dock rail, the data source orsources must specically identify the intermodal facility as on-dock.A nal situational variable included in the model is the harbor approach channel and berth

    depth. This is a measure of the maximum draft of vessels entering the harbor. 8 The deeper thechannel, the larger the vessels that can utilize the port.Control variables. These include other factors that may inuence seaport productivity. One such

    factor is longshore labor actions during the study period. Such work stoppages reduce outputduring the year of occurrence and may have an impact on carrier and shipper perceptions of theports labor. Longshore labor actions are measured as the duration of work stoppage in calendardays.Certain vessel types may also inuence the productivity of the seaport. For example, roll-on/

    roll-o vessel (ro/ro) operations consume large amounts of land for marshalling trailers andchassis-mounted containers as well as use of berthing space for vessels, but they make no use ofquayside container gantry cranes as the units are driven on and o the vessels similar to a tra-ditional ferry. Failing to recognize and control for ro/ro operations this could bias results. For thesame reason, failing to account for container barge operations could bias results. In NorthAmerica, container barges are commonly employed in feeder services due to their relatively lowoperating cost. 9 As a result the percentage of ro/ro arrivals in comparison to total arrivals and thepercentage of feeder services arrivals out of total arrivals are included as control variables.Another factor that could inuence port productivity is the presence of newer quayside gantry

    cranes (QSG) and their capability to serve large vessels. In order to control for this technology themodel includes the average outboard reach of all QSG cranes for each port-year observation. Thisis measured in meters from the shoreside to the seaside rail. Additional measures of crane tech-nology such as average lifting capacity in tons, average lift height in meters, average trolley speed

    7 Based on Interstate Commerce Commission/Surface Transportation Board classication.8 Draft is dened as the distance from the waterline to the lowest point of the vessel; i.e. how deep the water must be

    to prevent the vessels grounding on the bottom.9 Cabotage laws in the US and Canada restrict purely domestic waterborne cargo movements to vessels registered in

    the respective country. This has resulted in the low cost but low speed container barge dominating as the vessel of choicefor short-haul feeder services.

  • Containerport structure SSIZE Total containerport size (annual twenty-foot

    equivalent units)

    Por

    Oce

    Oce

    Situ

    346 H. Turner et al. / Transportation Research Part E 40 (2004) 339356Situation (intermediacy) DS Double-stack capable (binary variable for DS

    clearance into seaport area)

    Situation (intermediacy) CI Class I railroads serving seaport

    Situation DFT Draft (mean draft of entering vessels at or above

    90th percentile)

    Situation LBRRA Labor relations (Sum of labor (ILA/ILWU) strike

    days)

    Control FDSVC Feeder services (container carrying barge arrivals/total

    arrivals)

    Control RRSVC Ro/ro services (ro/ro vessel arrivals/total arrivals

    Control QSGREACH Mean QSG reach (m)in mthe aFi

    timeprovthe pare nD

    Tabl

    wher

    5. Da

    TaCana

    Ind

    Indt authority conduct DEDQUAY Dedicated terminal infrastructure (dedicate quay/total

    quay)

    an carrier conduct VSIZE Mean vessel size (TEU slots)

    an carrier conduct VSIZE2 Mean vessel size squared (TEU slots)

    ation (intermediacy) ODR ODR (terminal hectares/total terminal hectares)Containerport structure TSIZE Average container terminal size (annual TEU/number

    of container terminals)Table 1

    Variable names and denitions

    Type Variable name Denition (measure)eters per second and average hoist speed in meters per second are all highly correlated withverage maximum outboard reach.rm and time dummies are also included in the regression to account for unobserved port andeects. Assuming all other relevant variables are included in the model, time dummieside a means of assessing the impact of technological change not already specied. Inclusion ofort dummies control for port specic factors that inuence infrastructure productivity thatot already specied in the model.escriptions of independent variables employed in the estimation of this model are presented ine 1. The nal model to be estimated is:

    DEAxi b0C b1SSIZExi b2TSIZExi b3DEDQUAYxi b4ODRxi b5VSIZExi b6VSIZE2 b7DSxi b8CIxi b9DFTxi b10LBRRAxi b11FDSVCxi b12RRSVCxi b13QSGREACHxi

    XbxPORTx

    XbiYRi Exi

    e x designates seaport; i designates year.

    ta

    ble 2 presents the sample seaports. The selected seaports are the top 26 continental US anddian containerports for 1984 according to American Association of Port Authorities

    icator PORT Series of binary variable for seaport of observation

    icator YR Series of binary variables for year of observation

  • Table 2

    Study seaports (state/province)

    Baltimore (MD) Galveston (TX)

    Boston (MA) Houston (TX)

    H. Turner et al. / Transportation Research Part E 40 (2004) 339356 347(AAPA) data. Four of the study seaports are Canadian and the remaining 22 are US seaports. In1984 these 26 containerports accounted 94.1% of the total North American continentalthroughput measured in twenty-foot equivalent units for containerized waterborne commerce,and in 1997 for 90.7%. The Canadian share of TEU output ranged from a low of 7.9% in 1992 to ahigh of 9.4% in 1988.The period chosen for this investigation (19841997) is driven not by convenience but by

    regulatory changes in the environment. It lies between two signicant regulatory acts; the Ship-ping Act of 1984 and the Ocean Shipping Reform Act of 1998. 10 These regulatory reforms, incombination with US surface freight transportation deregulation, the introduction of newintermodal railcar technology, the increasing emphasis on logistics cost control, and the liberal-ization of trade and transport barriers in North America have signicantly altered the competitiveenvironment faced by North American seaports (USDOT, 1990; Slack, 1993).

    Charleston (SC) Miami (FL)

    Halifax (NS) New Orleans (LA)

    Hampton Roadsa (VA) Port Everglades (FL)

    Jacksonville (FL) Long Beach (CA)

    Montreal (QC) Los Angeles (CA)

    NY/NJ Oakland (CA)

    Philadelphia (PA) Portland (OR)

    Saint John (NB) San Francisco (CA)

    Savannah (GA) Seattle (WA)

    Wilmington (DE) Tacoma (WA)

    Wilmington (NC) Vancouver (BC)a Includes Newport News, Norfolk, and Portsmouth, VA.The data envelopment analysis model requires selection of inputs and outputs appropriate tothe research question being investigated. For this eort, inputs were restricted to physical mea-sures of containerport infrastructure and outputs to those produced by this infrastructure. Themost notable factor omitted is longshore labor. However, following De Neufville and Tsunokawa(1981), longshore labor hours are excluded as an input under the assumption that dierences inlabor productivity between North American seaports are minimal, owing to standardized col-lective bargaining agreements that establish longshore labor gang size and related work rules.For each port-year, inputs to the data envelopment analysis model were total terminal land

    dedicated to container operations, total quayside container gantry cranes, and total containerberth length. Total twenty-foot equivalent units handled was used as the measure of output. Whileconsideration was given to dening two outputs (TEU and short-tons), preliminary analysisfound that these two outputs were signicantly and highly correlated. TEU was considered to be

    10 Formally known as Public Law No. 98-237An Act to Improve Ocean Commerce Transportation Systems of the

    United States.

  • Table 3

    DEA variable descriptive statistics

    Number of cases (N ) Minimum Maximum Mean Standard deviation

    Output (TEU) 360 5553 3,504,803 622,376 644,943

    Quay length (m) 360 254 9050 2624 1950

    348 H. Turner et al. / Transportation Research Part E 40 (2004) 339356superior to short-tons because short-tons would capture cargo activity but not the additional useof resources required to handle empty containers. Given this and the fact that a larger sample sizewas possible when TEU was the only output, a single-output (TEU) DEA model was chosen. 11

    Table 3 contains descriptive statistics on both input and output variables used in the dataenvelopment analysis model. In order to evaluate the DEA model, individual containerport inputand output data as described above were required for the study seaports. Annual twenty-footequivalent unit output totals were obtained from the American Association of Port Authorities(AAPA) for all study containerports throughout the study period. These data are self-reported forall major North American containerports and published by the AAPA on its web site. 12 Inaddition to the output measure, AAPA data were used to construct seaport size and, in con-junction with the terminal data described below, average terminal size.Data on terminal inputs (hectares and meters of quay) needed to calculate data envelop-

    ment analysis scores were obtained from annual editions of Containerization International Year-book (CIY). 13 This data set has been used in previous published research (Fleming andHayuth, 1994; Fleming, 1989, 1997; Vandeveer, 1998) and has been shown to be accurate andreliable.The major data sources were supplemented by a number of other sources. Detailed data on

    quayside gantry cranes, used to construct the QSG reach variable were obtained directly from theresearcher supplying CIY. 14 These data were collected through annual surveys and interviewswith port authority administrators. Regarding the double stack and class I rail variables, LandsideAccess to US Ports (National Research Council, Transportation Research Board, 1993) andvarious editions of the US Corp of Engineers Port Series were used in addition to CIY. Wherespecic terminal or rail service questions remained unresolved, a structured content analysissearch of an automated database services was employed. The seaport industry was well coveredduring the study period by a handful of trade publications (Trac World, The Journal of Com-merce, and American Shipper) that were accessed through this service. This approach was par-ticularly useful in identifying double-stack services. Archives of these searches were created and

    Terminal land (ha) 360 20 572 150 129

    Container cranes (number) 360 1 52 12 11are available for reference.In addition to these data, the Tobit model estimation required data on ocean carrier operations

    including container vessel, roll-on/roll-o, and barge arrivals as well as slot capacity of the

    11 Containerized cargo data before 1990 was not available.12 www.aapa-ports.org.13 National Magazine Co., Ltd., London.14 Andrew Foxcroft, London.

  • container vessels. The US Maritime Administration (MARAD) provides detailed data on vesselarrivals in US seaports, including the vessel name, the date of entry, the type of vessel (container,

    vessel arrivals were obtained for the entire study period. Thus average vessel size, draft, feederservices, and roll-on/roll-o services variables had missing values for these four seaports reducing

    H. Turner et al. / Transportation Research Part E 40 (2004) 339356 349the sample that could be employed in the Tobit regression model by 56. Similarly, for US sea-ports, data on vessel arrivals was not available for years prior to 1987. This further reduced theTobit sample by 66. Thus the maximum sample for the Tobit model was 242. Given that there wasno DEA score for Wilmington (DE) in 1987 as noted, the actual sample size for the Tobitregression was 241.

    6. Results

    Fig. 1 provides a graphical representation of the growth of output and the three inputs forNorth American ports. Looking at the totals, it is clear that output has grown at a faster rate thanany of the three containerport inputs. Clearly productivity has improved over this time period andresulted in a reduction of the excess capacity problem.As Fig. 2 shows, the results for the three regional groupings varies. Productivity at the Gulf

    Cost ports rose steeply from 1992 to 1997. The West Coast ports showed a steady improvement inproductivity across the entire sample period. The East Cost ports were the worst performingsubgroup with growth in output falling below the growth in all three inputs from 1984 to 1993. Itwas only in the last four years of the sample period that growth in output accelerated for the EastCoast ports to a pace faster than input growth.With this evidence, we can state unequivocally that during the study period gross infrastructure

    productivity rose on average for North American containerports. This growth is likely a combinedeect from both improved capacity management (i.e. fewer unused inputs) and improvements

    15ro/ro, etc.), draft, registered net and gross tonnages, and indicators identifying the vessels lastport of call (domestic or foreign) including the country if the arrival is from a foreign seaport orthe US Port District if the last port was domestic. 15 Various annual issues of ContainerizationInternational Yearbook were used to identify the slot capacities of vessels based on the names ofthe vessel provided in the MARAD data set. With these data, the average container vessel size inslots was constructed by matching each port-years list of arriving vessels with the vessel slotcapacity data yielding a vessel size for each port-year. The draft variable was also constructedfrom these data by capturing each vessels draft as reported at the time of arrival and, based onthese data, calculating the draft at the 90th percentile for each port-year.Given 26 seaports studied over a 14-year period, the maximum sample size would be 364

    observations. However, data availability aected the actual sample size. For the model, and anyvariations on the model, missing data within an observation resulted in the exclusion of thatobservation from the sample. For data envelopment analysis scores, Wilmington (DE) was notincluded for the years 19841987 as the seaport had no quayside gantry cranes during this period.For the four Canadian seaports (Halifax, Saint John, Montreal, and Vancouver), no data onRegistered net and gross tonnage are measures of cargo carrying capacity and vessel size respectively.

  • 350 H. Turner et al. / Transportation Research Part E 40 (2004) 3393560.80

    1.00

    1.20

    1.40

    1.60

    1.80

    2.00

    2.20

    1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997

    TEU Hectares QSG Cranes Quay (m)in operational eciency (i.e. more ecient use of inputs). Since it is not possible to separate theseimpacts using trend analysis, a regression analysis is warrented.The parameter estimates from the Tobit regression are presented in Tables 4 and 5. Table 4

    includes the primary model variables and Table 5 presents the parameter estimates for the yearand port binary variables. The year variable for 1987 and the port variable for Baltimore havebeen omitted from the model to prevent perfect collinearity in the model estimation.Seaport industry structure. The parameters associated with containerport size and terminal size

    are both signicant (p < 0:001) and positively correlated to the dependent variable as hypothesized.Containerports, therefore, exhibit returns to scale. Larger containerports are more ecient pro-ducers, supporting the ndings of DeNeufville and Tsunokawa (1981) andKim and Sachish (1986).Port authority conduct. The parameter estimate associated with dedicated terminal capacity

    (DEDQUAY) is insignicant (p 0:999) for the sample. Based on the sample, port authorityconduct with respect to this aspect of leasing policy is not a signicant inuence on infrastructureproductivity. This lack of signicance could indicate that dedicated terminals are operated inmuch the same manner as other terminals. It is also plausible that dedicated terminals benetfrom more ecient scheduling resulting from a higher degree of operational control exercised bythe leasing carrier thus osetting the negative relationship hypothesized above. Given the pro-liferation of carrier alliances, particularly toward the end of the study period, it is possible that thecontrolling carrier further exploits its ability to coordinate schedules with alliance partners in aterminal sharing arrangement. This is particularly likely given the use of minmax leases thatencourage maximization of throughput by the carrier.

    Fig. 1. Input and output trends, North America.

  • Table 4

    Parameter estimates

    Variable Coecient

    estimate

    Standard

    error

    t-Statistic P -value

    Constant )0.199 0.249 )0.800 [0.424]Container port size (millions of TEU) 0.196 0.030 6.520 [0.000]

    Terminal size (millions of TEU) 0.900 0.072 12.444 [0.000]

    Dedicated container port quay/total quay )0.000 0.106 )0.001 [0.999]On-dock rail (hectares with access/total terminal

    hectares)

    )0.070 0.027 )2.607 [0.009]

    Vessel size (thousands of TEU slots) 0.271 0.076 3.585 [0.000]

    Vessel size squared (millions of TEU slots) )0.050 0.019 )2.584 [0.010]Double-stack clearance )0.025 0.022 )1.156 [0.248]Class I railroads (number) 0.137 0.017 8.330 [0.000]

    Draft (mean of vessel at or above 90th percentile

    in feet)

    )0.005 0.003 )1.531 [0.126]

    Labor strikes (duration in days) 0.001 0.003 0.303 [0.762]

    Feeder services (container carrying barges/total

    arrivals)

    0.279 0.253 1.104 [0.270]

    Ro/ro service arrivals/total arrivals )0.533 0.343 )1.553 [0.120]Mean container crane reach (m) )0.028 0.005 )5.170 [0.000]

    0.4000

    0.5000

    0.6000

    0.7000

    0.8000

    0.9000

    1.0000

    1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997

    NA East Gulf West

    Fig. 2. DEA trends.

    H. Turner et al. / Transportation Research Part E 40 (2004) 339356 351

  • 352 H. Turner et al. / Transportation Research Part E 40 (2004) 339356Table 5

    Parameter estimates for year and port dummies

    Variable Coecient estimate Standard error t-Statistic P -value

    YR88 )0.013 0.019 )0.677 [0.498]YR89 )0.026 0.020 )1.326 [0.185]YR90 )0.022 0.020 )1.095 [0.273]YR91 )0.013 0.020 )0.628 [0.530]YR92 0.009 0.021 0.457 [0.648]

    YR93 0.013 0.021 0.625 [0.532]

    YR94 0.001 0.022 0.024 [0.981]

    YR95 0.003 0.023 0.133 [0.894]

    YR96 )0.015 0.025 )0.589 [0.559]YR97 0.014 0.027 0.525 [0.599]Ocean carrier conduct. The mean capacity of container vessels calling on the port is associatedwith increased infrastructure productivity. The relationship is signicant (p < 0:001). This ndingis surprising and contradicts the literature based on queuing theory (Jansson and Shneerson,1982) which holds that the optimal (total cost minimizing) berth occupancy rate declines as vesselsize increases. In reality, more ecient containerports may attract larger vessels owing to thediseconomies of scale these large vessels experience in port (Talley, 1990). This is intuitivelyappealing and must be considered as a valid explanation. It is also possible that there are botheconomies and diseconomies to the port of dealing with larger vessels. The economies occur as aresult of dealing with fewer ships, but diseconomies occur as the result of an increase in capacityto handle larger ships. It may also be the case that economies of vessel size are nonlinear, so thatdiseconomies set in for very large ships. This result is supported by the signicant (p 0:01) andnegative sign for the squared vessel size variable.

    Boston 0.086 0.045 1.905 [0.057]

    Charleston 0.046 0.036 1.261 [0.207]

    Galveston )0.503 0.062 )8.054 [0.000]Houston )0.189 0.063 )3.008 [0.003]Hampton Roads 0.048 0.034 1.419 [0.156]

    Jacksonville 0.101 0.035 2.861 [0.004]

    Los Angeles )0.145 0.071 )2.046 [0.041]Long Beach )0.212 0.062 )3.422 [0.001]Miami 0.251 0.049 5.117 [0.000]

    New Orleans )0.386 0.081 )4.746 [0.000]New York, NJ )0.361 0.066 )5.463 [0.000]Oakland )0.128 0.064 )2.008 [0.045]Port Everglades 0.602 0.049 12.398 [0.000]

    Philadelphia )0.025 0.047 )0.545 [0.586]Portland )0.124 0.043 )2.897 [0.004]Savannah )0.098 0.045 )2.205 [0.027]Seattle 0.067 0.057 1.171 [0.242]

    San Francisco 0.006 0.044 0.137 [0.891]

    Tacoma 0.337 0.053 6.404 [0.000]

    Wilmington, DE 0.553 0.065 8.521 [0.000]

    Wilmington, NC )0.116 0.039 3.005 [0.003]

  • H. Turner et al. / Transportation Research Part E 40 (2004) 339356 353Situational factors. The parameter associated with double-stack capabilities is insignicant(p 0:248). It can be concluded that while double-stack clearance into the terminal area is a keymarketing tool for containerports, a direct impact on containerport infrastructure productivity inNorth America was not realized during the study period.The parameter associated with the number of class I railroads serving the port is both positive

    and signicant (p < 0:001). For the sample, the greater the number of class I railroads serving theseaport, the greater the productivity of the containerport infrastructure. This is clear support forthe importance of rail service quality, perhaps including frequency of service, and rail servicecompetition, to the success of containerports. A point that port authority management frequentlystresses.Contrary to the hypothesis stated above, the coecient associated with on-dock rail (ODR) is

    negative and signicant (p 0:009). In the sample, the greater the share of terminal capacityserved by on-dock rail, the less productive the containerport infrastructure. On-dock rail facilitiesconsume terminal land for staging cars and assembling/breaking trains. Based on the results of themodel, it is possible that any productivity gains from shiprail exchange within terminals are notsucient to oset the investment in the additional land required to support these facilities.The parameter associated with draft is not signicant (p 0:126), although the sign is unex-

    pectedly negative. Further investigation of the data suggest draft could be inuenced by anoutlier. The port of Philadelphia exhibited extreme values on the high end of the range in 1990,1991, 1995, 1996, and 1997. It is possible that data on vessel draft for the port of Philadelphia, andother ports as well, includes the draft of liquid and dry bulk cargo vessels arriving at anchorageand partially discharging onto barges at that anchorage (a process known as lightering) beforeproceeding to the berth.The parameter associated with longshore labor relations is insignicant (p 0:762). Labor

    actions have had no systemic eect on infrastructure productivity in the sample. Given the relativeinfrequency and short duration of wildcat strikes and the fact that the few multi-port strikes thatdid occur were of short duration, this is not surprising. Whatever impact labor relations have onproductivity, the impact is likely local and captured in the port binary variable.Control variables. The ratio of container carrying barge arrivals to total arrivals is included

    in the model to control for the potential but uncertain impact of feeder operations on inputsand outputs. Since feeder services most often use the same containerport assets as containervessels themselves, and their output is included in that of the containerport, omission wouldclearly lead to a bias. However the coecient for feeder services is insignicant (p 0:270)suggesting such operations neither contribute nor detract from infrastructure productivity in thesample.Unlike feeder services, an a priori assumption regarding the impact of roll-on/roll-o services

    on infrastructure productivity is possible. For containerports having a high proportion of totalarrivals attributed to roll-on/roll-o vessels, it was speculated that data envelopment analysisscores would be higher. The output of ro/ro services are often included in the twenty-footequivalent unit output for the containerport but these vessels do not require quayside gantrycranes to load and discharge. Thus one of the inputs employed in the date envelopment models isnot required to produce the output. Failure to control for this would surely bias the results infavor of containerports with high ro/ro volumes. Although positive, the coecient for ro/ro

    services, was not signicant (p 0:12).

  • As with many other research eorts, this study concludes size matters. The implications are

    354 H. Turner et al. / Transportation Research Part E 40 (2004) 339356clear for policy makers. It may not be wise to invest public funds in small facilities at smallseaports without a clear commitment from carriers and shippers to utilize the facility andencourage expansion. Without such commitments, investments are unlikely to attain sucientvolumes to recover costs, particularly in the face of competition from larger seaports.Regarding rail service, the longstanding relationship between seaports and the rail industry

    appears to remain a critical determinant of containerport infrastructure productivity. This studyfound greater numbers of railroads serving a port are correlated with increased port productivity.Unfortunately, rail service is a factor seemingly beyond the control of seaport management.Eorts to increase rail competition may be challenging but the eort is well warranted from theperspective of the port authority and those concerned with the economies they serve. Further,while investment in on-dock rail facilities may attract large carriers to the containerport, there isno evidence in this study that these facilities make productive use of the land required to supportthem.By employing data envelopment analysis as its measure of infrastructure productivity, this

    paper has addressed numerous methodological hurdles presented by industry characteristics anddata limitations. The addition of Tobit regression has allowed investigation and identication ofkey factors impacting containerport infrastructure productivity during the study period. The ef-fort supports the presence of economies of scale as has been observed in other eorts using avariety of approaches to productivity measurement. By examining the inuence of port authority,The reach of quayside gantry cranes, a proxy for improvements in QSG crane technology, wasexpected to be positively correlated with the dependent variable. However, the coecient isnegative and signicant (p < 0:001). This seems to be the result of a correlation between the reachof QSG cranes and the number of cranes at containerports. The number of cranes represents anincrease in the inputs of a containerport and therefore would be associated with a decline inproductivity, holding output constant. Since virtually all new cranes have a larger reach than oldcranes, and since old cranes do not disappear when new cranes are purchased, the increase inreach is positively associated with the number of cranes. A model that includes the number ofcranes in the specication results in an insignicant coecient in the reach variable.The year binary variables are all insignicant. Recall that 1987 is the base year. This suggests

    that the other variables in the model have adequately explained the observed positive trend in dataenvelopment analysis scores during the period 19871997 and that no particular events associatedwith specic years during this period have signicantly inuenced the trend in the dependentvariable.Many of the port specic binary variables are signicant. The fact that both positive and

    negative signs were observed suggests that port specic factors, not already specied in the model,inuence infrastructure productivity. These factors probably include the productivity of locallabor and the eectiveness of port authority administration and policies, including decisionsregarding the timing of additions to capacity.

    7. Conclusionocean carrier and rail carrier conduct on containerport productivity, valuable guidance is pro-

  • Mongelluzzo, B., 1998. West coast ports push for even larger terminals. Journal of Commerce (11 December), 1A.

    National Research Council, Marine Board, 1986. Improving Productivity in US Marine Container Terminals. National

    H. Turner et al. / Transportation Research Part E 40 (2004) 339356 355Academy Press, Washington, DC.

    National Research Council, Transportation Research Board, 1993. Landside Access to US Ports. US Department ofvided for those with managerial and policy interests directed at improving the performance of thiscomplex and critical system.

    References

    Bennathan, E., Walters, A., 1979. Port Pricing and Investment Policy for Developing Countries. Oxford University

    Press, Oxford.

    Bobrovitch, D., 1982. Decentralized planning and competition in a national multi-port system. Journal of Transport

    Economics and Policy (January), 3142.

    Burke, J., 1996. Field of dreams or prudent investment: just how much port capacity is enough? Trac World (March

    25), 31.

    Caves, D., Christensen, L., 1988. The importance of economies of scale, capacity utilization, and density in explaining

    interindustry dierences in productivity growth. Transportation and Logistics Review 24 (1), 332.

    Chang, S., 1978. Production function, productivities, and capacity utilization of the port of mobile. Maritime Policy

    and Management 5, 297305.

    Charnes, A., Cooper, W., Rhodes, E., 1978. Measuring the eciency of decision making units. European Journal of

    Operational Research 2, 429444.

    Charnes, A., Cooper, W., Rhodes, E., 1981. Evaluating program and managerial eciency: an application of data

    envelopment analysis to program follow through. Management Science 27 (6), 668697.

    De Neufville, R., Tsunokawa, K., 1981. Productivity and returns to scale in container ports. Maritime Policy and

    Management 8 (2), 121129.

    Farrell, M., 1957. The measurement of productive eciency. Journal of the Royal Statistical Society 120 (3), 25281.

    Fleming, D., 1989. On the beaten track: a view of US West-Coast container port competition. Maritime Policy and

    Management 16 (2), 93107.

    Fleming, D., 1997. The meaning of port competition. Paper presented at the Plenary Session of the International

    Association of Maritime Economists conference, London, September 22, 1997.

    Fleming, D., Hayuth, Y., 1994. Spatial characteristics of transportation hubs: centrality and intermediacy. Journal of

    Transport Geography 2 (1), 318.

    Hayuth, Y., 1988. Rationalization and deconcentration of the US container port system. The Professional Geographer

    40 (3), 279288.

    Heaver, T., 1995. The implications of increased competition among ports for port policy and management. Maritime

    Policy and Management 22 (2), 125133.

    Heikkila, E., 1990. Structuring a national system of ports. Portus, 1923.

    Hershmann, M., Goodwin, R., Rootstalk, A., McCrea, M., Hayuth, Y., 1978. Under New Management. Division of

    Marine Resources, University of Washington, Seattle.

    James, R., 1991. Privatization and consolidation seen as answers for ailing ports. Trac World (September 30), 27.

    Jansson, J., Shneerson, D., 1982. Port Economics. MIT Press, Cambridge, MA.

    Jara-Daz, S., Martnez-Budra, E., Cortes, C., Basso, L., 2002. A multioutput cost function for the services of Spanishports infrastucture. Transportation 29, 419437.

    Kim, M., Sachish, A., 1986. The structure of production, technical change and productivity in a port. The Journal of

    Industrial Economics 35 (2), 209223.

    Maddala, G., 1983. Limited-dependent and Qualitative Variables in Econometrics. Cambridge University Press.

    MARAD, 1994. Public Port Financing in the United States. US Maritime Administration, Washington, USDOT.

    Mongelluzzo, B., 1996. Whispers of overcapacity dog US ports. Journal of Commerce (14 February), 1A.

    Mongelluzzo, B., 1997. Ports, lines seen building too big. Journal of Commerce (11 February), 1B.Transportation, Maritime Administration, Washington, DC.

  • Oum, T., Tretheway, M., Waters II, W.G., 1992. Concepts, methods and purposes of productivity measurement in

    transportation. Transportation Research A 26A (6), 305493.

    Roll, Y., Hayuth, Y., 1993. Port performance comparison applying data envelopment analysis (DEA). Maritime Policy

    and Management 20 (2), 153161.

    Sachish, A., 1996. Productivity functions as a managerial tool in Israeli ports. Maritime Policy and Management 23 (4),

    341369.

    Slack, B., 1993. Pawns in the game: ports in a global transportation system. Growth and Change 24 (Fall), 579588.

    Stopford, M., 1988. Maritime Economics. Unwin Hyman, London.

    Talley, W., 1990. Optimal containership size. Maritime Policy and Management 17 (3), 165175.

    Thanassoulis, E., 1993. A comparison of regression analysis and data envelopment analysis as alternative methods for

    performance assessments. Journal of Operational Research Society 44 (11), 11291144.

    Tobin, J., 1958. Estimation of relationships for limited dependent variables. Econometrica 26 (1), 2436.

    Tongzon, J.L., 1995. Determinants of port performance and eciency. Transportation Research A 29a (3), 245252.

    Turner, H., 2000. Evaluating seaport policy alternatives: a simulation study of terminal leasing policy and system

    performance. Maritime Policy and Management 27 (3), 283301.

    US Department of Transportation, 1990. Double stack container systems: implications for US railroads and ports,

    Task I report double-stack status. June 1980.

    US Department of Transportation, 1998. The impacts of changes in ship design on transportation infrastructure and

    operations. Oce of Intermodalism. February 1998.

    Vandeveer, D., 1998. Port productivity standards for long-term planning. Ports 98, ASCE.Varaprasad, N., 1986. Optimum port capacity and operating policies: a simulation study. Transport Policy and

    Decision Making 3, 297312.

    Verhoe, J., 1981. Seaport competition: some fundamental and political aspects. Maritime Policy and Management 8

    (1), 4960.

    356 H. Turner et al. / Transportation Research Part E 40 (2004) 339356Walters, A., 1976. Marginal cost pricing in ports. The Logistics and Transportation Review 12 (3), 99144.

    Windle, R., Dresner, M., 1995. A note on productivity comparisons between air carriers. The Logistics and

    Transportation Review 31 (2), 125134.

    North American containerport productivity: 1984-1997IntroductionBackgroundLiterature reviewMethodologyDataResultsConclusionReferences