Unemployment and employment offices’ efficiency: What can be done?

ARTICLE IN PRESS

Socio-Economic Planning Sciences 40 (2006) 169–186 www.elsevier.com/locate/seps

Unemployment and employment offices’ efficiency: What can be done? Anatoli Vassiliev, Giovanni Ferro Luzzi, Yves Flu¨ckiger, Jose´ V. Ramirez Department of Economics, University of Geneva, Uni Mail, boulevard du Pont-d’Arve 40, CH-1211 Geneva 4, Switzerland Available online 23 May 2005

Abstract Regional employment offices provide placement services to job-seekers and employers and organize active labor market programs. In this paper, we carry out a quantitative evaluation of the employment offices’ performance in Switzerland based on production efficiency measures. We use Data Envelopment Analysis (DEA) to estimate the performance of all employment offices and then account for factors in the local operating environment that are outside managerial control. This approach, and the ranking of employment offices, may easily be interpreted by policymakers and provides guidelines for raising the efficiency of the public employment service. Our findings suggest that there is considerable room for improved efficiency in employment service, which could lead to a lower level of structural unemployment. We also find that differences in the external operating environment have a significant influence upon the efficiency of employment offices. r 2005 Elsevier Ltd. All rights reserved. JEL classification: C14; C61; J68 Keywords: Public employment service; Technical efficiency; Data Envelopment Analysis; Labor market conditions

1. Introduction The unemployment problem is most often tackled by pointing to business cycle conditions or excessive labor market regulations, at least at the macroeconomic level. More recently, Corresponding author. Fax: +41 22 705 82 93.

E-mail address: yves.fl[email protected] (Y. Flu¨ckiger). 0038-0121/$ - see front matter r 2005 Elsevier Ltd. All rights reserved. doi:10.1016/j.seps.2004.10.007

ARTICLE IN PRESS 170

A. Vassiliev et al. / Socio-Economic Planning Sciences 40 (2006) 169–186

employment and training programs, and other active labor market measures, are being evaluated to gauge their effectiveness in reducing unemployment duration, or guaranteeing new jobs with acceptable earnings [1]. In this paper, we take another route and look at how the number of jobless people, or the mean duration of unemployment, can be affected by the ‘‘technical’’ performance of employment offices, which typically aim at finding people jobs and act as a crucial intermediary between firms and job-seekers. We consider employment offices as production units which use inputs to generate some form of output, be it exits from unemployment or a lower duration of unemployment spells. The purpose of this paper is to examine the technical efficiency of the Swiss public employment service using a two-stage procedure. In the first stage, data envelopment analysis (DEA) is used to compute technical efficiency for all regional employment offices (REOs, hereafter) operating in Switzerland during the period 1998–1999. In the second stage, a regression model is used to analyze the impact of external factors of the operating environment on the variation in technical efficiency scores across employment offices. Our results provide a relative ranking of employment offices with respect to their ability to meet pre-defined targets, which can then be used as a guideline for greater efficiency amongst the offices. Our study was motivated by enforcement, in year 2000, of a new agreement between the Swiss federal authorities and cantons regarding the introduction of performance-based budgeting of regional employment offices [2]. This agreement was struck as Swiss unemployment policy was shifted in 1996 from a passive income maintenance program to active labor market measures aimed at faster reintegration of the unemployed into the labor market. Performance-based budgeting of REOs is part of a set of active labor market measures providing them with an incentive to become more efficient in lowering structural unemployment. Evaluation of employment offices’ performance is a necessary condition for implementation of the budgeting scheme and, more importantly, it helps to identify best practices in their internal management. Such practices can then be replicated and, so, improve the overall efficiency of the Swiss public employment system. Although the economics of public employment services is a topical issue, Cavin and Stafford [3] have noted how little published literature is available on the performance standards of employment services. While the situation has not changed much, more recent literature includes [4–7]. Important production frontier literature on the matching efficiency of local labor markets includes [8–10]. This literature is based on the theoretical matching function framework of Blanchard and Diamond [11] and concentrates on the relationship between vacancies and unemployment. We argue that this framework is not suitable for assessment of REO efficiency in Switzerland, as vacancies are registered by the employment offices themselves. Swiss employment offices can thus easily appear more efficient simply by reporting a lower number of available vacancies. The evaluation of Swiss REO performance [5] is currently carried out by Corrected Ordinary Least Squares [12]. Another Swiss study [6] used DEA based on a matching approach between the unemployed and vacancies. Neither [5] or [12] consider REO production factor inputs such as personnel and capital. These studies provide benchmarks for comparing outcomes achieved across REOs. We contribute to the understanding of employment service delivery by proposing an alternative model of the economic activity of REOs. It accounts for the targets of employment

ARTICLE IN PRESS A. Vassiliev et al. / Socio-Economic Planning Sciences 40 (2006) 169–186

171

offices as specified by the Swiss State Secretariat for Economic Affairs (Seco), and includes the REOs’ inputs. The study unfolds as follows: Section 2 describes some organizational aspects of the Swiss public employment service. Section 3 discusses the model of efficiency measurement. Section 4 describes the data, while Section 5 reports the results. In Section 6, we summarize the findings and conclude.

2. Organization of public employment service Like many other European countries, Switzerland underwent a severe recession at the beginning of the 1990s, with unemployment jumping from below 1% to more than 6% of the labor force in some cantons. This new state of affairs has led the Swiss federal government to adjust its unemployment policy. Several active labor market programs (ALMPs) aimed at faster re-integration of job-seekers into the labor market were launched in the second half of the 1990s. The public employment service underwent a deep transformation: in 1996, more than 2000 local employment offices were merged into about 160 REOs. The objective of this reform was to provide more professional and efficient services to both job-seekers and employers in an attempt to lower structural unemployment. Currently, public employment services are a joint effort in which federal authorities supply the main part of operating funds, and the cantons retain administrative control of local operations. Funding depends on a canton’s size and the REO’s performance. The latter is monitored by federal authorities. The cantons decide whether the REOs’ set of activities and organization type are responsive to a given locality’s needs in seeking to lower structural unemployment in the canton. The number of job counselors working with REOs is also decided at the canton level. Employment services produced by cantons are financed by federal authorities up to a given amount; however, if considered necessary, cantons are free to add funds for supplementary labor market programs,1 or for engaging more REO staff. Accordingly, substantial differences were reported in the use of various active labor market programs among cantons, in the administrative organization of REOs, and in the ratio of the number of REOs’ job-seekers to the number of job counselors [13]. In this complex funding system, REO managers must obtain the approval of canton authorities when hiring or firing in their office, so that they do not have a full control of the human resource management function. Prior to 1996, the role of public employment offices in Switzerland was limited to passive income maintenance of those job-seekers receiving unemployment insurance (UI) benefits. Tasks of the offices were mainly administrative, limited to calculating unemployment insurance payments and making sure that the registered unemployed were really looking for employment. The efficiency of these offices was evaluated according to the resources used and the number of activities performed (e.g., number of cases handled).

1 For example, the canton of Geneva runs income maintenance programs for long-term unemployed workers who have exhausted their entitlement to federal unemployment insurance benefits.

ARTICLE IN PRESS 172

A. Vassiliev et al. / Socio-Economic Planning Sciences 40 (2006) 169–186

As part of the new unemployment policy, and in order to control and improve their efficiency, REOs must now report results of their activities to the Seco. As of 2001, a distinctive feature of the Swiss REO supervision system is that offices are no longer evaluated based on the number of tasks performed but, rather, on their ability to reach some pre-specified goals. Swiss authorities recently implemented a financial framework designed to encourage employment office efficiency: as of 2001, offices achieving the best results are rewarded. REO goals, as given in [2], are correlated with the number of persons in structural unemployment. Thus, when REO results improve, structural unemployment should decrease, and vice versa. Their four goals are aimed at reducing: (1) the mean REO’s duration of unemployment; (2) the number of persons entering long-term unemployment (i.e., unemployment spells longer than one year); (3) the number of persons losing entitlement to federal unemployment insurance (UI) benefits (i.e., those who are unemployed for more than two years); and finally, (4) the number of persons re-entering unemployment less than four months after having found a job.

3. The efficiency measurement model This section begins with a description of the conceptual framework of REOs’ economic activity. We follow with development of a quantitative model to evaluate the technical efficiency of REOs. 3.1. Economic activity of regional employment offices The ultimate goal of employment offices is to match job-seekers with potential employers. An office cannot force matching, but must limit itself to services that facilitate the final contracting by the parties themselves. By choosing the appropriate set of activities, e.g., contacting employers within the local labor market, job counseling sessions with job-seekers, organizing active labor market programs, and applying financial sanctions, offices can reduce the duration of, and reentry into, unemployment. The day-to-day activities of a REO are viewed as a means of achieving its goals in the matching process. The pool of unemployed persons is formed largely by entries into unemployment from employment. This flow depends on local economic conditions and may not be under the office’s control. Therefore, at least in the short run, the number of unemployed in the local labor market is exogenous to the office. The employment office may however influence transition flows within the pool of unemployed persons, as well as some transition flows in and out of unemployment. The most direct way for an employment office to reduce both the number of job-seekers and the mean duration of unemployment spells is to increase the number of hires. We thus take the number of hires as the first result variable of regional offices, which is consistent with Seco’s objective of reducing the mean duration of unemployment [2]. At the same time, as the number of hires increases, the offices seek to prevent the most disadvantaged job-seekers from staying unemployed for long durations. In order to insure that REOs do not concentrate on those job-seekers having the best individual hiring characteristics, Seco is also evaluating: (i) office performance in minimizing the number of entries into long-term

ARTICLE IN PRESS A. Vassiliev et al. / Socio-Economic Planning Sciences 40 (2006) 169–186

173

unemployment, and (ii) the number of unemployed reaching the end of unemployment insurance benefit entitlement. These two variables are, respectively, our second and third result variables. Finally, in order to promote durable labor market reintegration, the REOs seek to minimize the number of job-seekers who re-enter unemployment after a short spell of employment. This is our fourth result factor, and may be viewed as a matching quality indicator of an office. The public services provided by REOs are produced by labor and capital. Labor is the dominant cost component in placement service production. However, labor input is not homogeneous. Employees working in REOs may be specialized in administrative tasks, computer support, tasks relative to active labor market programs or in job counseling involving direct contact with job-seekers. Furthermore, some job counselors are given specialized training, possibly with a formal diploma (brevet fe´de´ral). Finally, the job tenure of counselors should be an important determinant of the knowledge of local labor market conditions and thus influences results of their activities. It was only possible to obtain data on the number of REO job counselors for the period analyzed. No information on the formal diploma was available. However, it may be argued that job counseling is the main activity of REOs. These employees offer job-seekers available positions on the local market, decide whether or not the unemployed will take part in labor market programs, and may apply sanctions (e.g., unemployment insurance benefit cuts) if job-seekers do not comply with requests made by the office. As job counseling is the central activity of REO, all other office tasks are generally intended to help the counselors work efficiently. Assuming that they were given equal working conditions, we view the number of counselors as a reasonable proxy for the labor input of REOs. For sake of simplicity, capital in the form of office space and computer terminals is assumed to be proportional to labor input and, due to common standards, varies very little across offices. It is therefore reasonable to assume that labor and capital are strictly complementary. Consequently, efficiency in the use of labor is considered the main factor underlying productivity differences across offices, with capital input not accounted for in the study. 3.2. Methodology By technical efficiency of a production unit, we mean a comparison between observed and optimal values of its outputs and inputs. Koopmans [14, p. 60] provided a formal definition of technical efficiency: a producer is technically efficient if an increase in any output requires a reduction in at least one other output or an increase in at least one input, and if a reduction in any input requires an increase in at least one other input or a reduction in at least one output. Thus, a technically efficient producer is located on the production frontier while a technically inefficient producer is located below this frontier. If we are interested in increasing outputs rather than saving inputs, then a radial measure of technical efficiency, as introduced by Farrell [15] is needed. It is defined as the maximum equiproportionate increase in all outputs that can be achieved with a given amount of input. This measure addresses the difference between observed and optimal values of outputs, or between observed values of outputs and the efficient production frontier. The goal of efficiency analysis is to estimate the production frontier, and to compute the distance between the optimal point on this frontier and the actual values of outputs.

ARTICLE IN PRESS A. Vassiliev et al. / Socio-Economic Planning Sciences 40 (2006) 169–186

174

To estimate the production frontier, several econometric and mathematical programming techniques are available [16]. We analyze the efficiency of employment offices on the basis of a widely-used, deterministic, non-parametric technique named Data Envelopment Analysis (DEA). DEA is a mathematical programming method created by Charnes et al. [17] and Seiford [18]. It has proved particularly useful when evaluating efficiency of public services since it requires no information on the prices of inputs and outputs (that are often unavailable for public goods), and it captures all relevant information in whatever metrics. DEA places no parametric structure on the production frontier, which is a valuable feature since little information is available on the regional offices’ ‘‘production technology’’. When the underlying production function is unknown, non-parametric techniques are more attractive than parametric ones, such as stochastic production frontier [16]. DEA estimates a multivariate production frontier by forming a piecewise-constant envelope of the cloud of data points in the inputs/outputs space. The efficiency of a producer is thus measured relative to the efficiency of all other producers. The assumption of variable returns to scale (VRS) allows the cloud of data points to be enveloped as tightly as possible. Suppose we have n REOs, each of which consumes p inputs and produces q outputs. The output increasing efficiency measure, E O , for office O equals the inverse of the optimum value of the linear programming problem below, i.e., E O ¼ l
1 O max lO flO ;gg

subject to lO yO
Y 0 gp0;
xO þ X 0 gp0;

(1)

i0n g ¼ 1; lO free; gX0; where X : ðn  pÞ matrix of the n observed input vectors, Y : ðn  qÞ matrix of the n observed output vectors, g ¼ ðg1 ; . . . ; gn Þ0 is an ðn  1Þ vector of coefficients called intensity variables,2 and in is an ðn  1Þ vector of 1’s. The vector of intensities is used to weight the observations in constructing the reference frontier. xO : ðp  1Þ and yO : ðq  1Þ denote the vectors of inputs and outputs, respectively, for office O. Since yO and the elements of Y are all non-negative, the first constraint forces lO to be nonnegative even though it is not explicitly constrained. The inequalities in the output and input constraints in (1) guarantee the strong disposability of inputs and outputs [19]. In other words, we assume that reductions (increases) in all outputs (inputs) are feasible for a given input (output) mix. The program should be solved n times thus leading to an optimal efficiency score for each employment office. 2

The solution value g ¼ ðg1 ; . . . ; gn Þ0 indicates whether office i (i ¼ 1; . . . ; n) serves as a peer for office O. Offices that are peers define that part of the frontier relevant to O and, hence, efficient production for O. Such production is the linear combination of its peers where the weights are elements of vector g. For example, if gi ¼ 0, then office i is not a peer for O. However, if gi 40, say gi ¼ 0:6, then office i is a peer office with 60% weight placed on deriving the target efficient output and input level for office O. An efficient office has no peers. The number of times it is a peer for other offices gives the number of dominated offices.

ARTICLE IN PRESS A. Vassiliev et al. / Socio-Economic Planning Sciences 40 (2006) 169–186

175

The efficiency measure (or score) E is equal to 1 if the producer is fully efficient, and is smaller than 1 otherwise. It may be given the following interpretation: a score of 0.8 means that, given the inputs consumed, the production unit produces only 80% of its maximum output level. Therefore, to be efficient with respect to at least one output, it must increase its actual outputs’ levels by a factor of 1=0:8 ¼ 1:25, i.e., by 25%, keeping inputs constant. The DEA model of (1) computes the Farrell [15] radial measure of efficiency and ignores the non-radial input savings (slacks) and output expansions (surpluses) as sources of inefficiencies. By so doing, we follow a large part of DEA literature which recognizes that aggregating radial and non-radial inefficiency components is a tricky issue.3 3.3. Empirical strategy We now turn to the issue of applying DEA to the data from regional employment offices. The efficiency of REOs is to be measured with respect to their achieving a large number of transitions from unemployment to employment; a small number of transitions into long-term unemployment; a small number of people reaching the end of UI benefit entitlement; and a small number of reentries into unemployment. Therefore, the REOs have to manage four transition flows: one of these is to be maximized and the other three minimized. DEA requires the inputs to be conserved and the outputs to be increased. Within REO setting, we cannot consider the transition flows to be minimized as outputs. To be efficient, an office must decrease these variables. Instead of developing the reciprocals or proxies for these variables (which can lead to non-linear transformations of the original data), we enter them in (1) as inputs rather than as outputs. The proposed DEA comprises one output to be maximized, i.e., the number of transitions from unemployment to employment, and five inputs. There are two types of inputs. The first are the result variables of REOs—the number of transitions into long-term unemployment, the number of unemployed loosing federal UI benefit entitlement, and the number of re-entries into unemployment. The inputs of the second type are the ‘‘resources’’ of the REOs (labor). An employment office may report a large number of hires simply because a large number of job-seekers is registered at the office. In this case, the possibility of a low placement rate per job seeker is not ruled out. To avoid having large REOs automatically appearing efficient, we included the number of job-seekers entitled to unemployment insurance and registered at the office as the additional input variable. By including the number of hires as an output, and the number of job-seekers registered at an REO as an input to the production frontier, we realize an efficiency measurement model that credits, with a large efficiency score, those employment offices registering many hires for a given number of job-seekers. These offices obtain a high exit rate to employment. An increasing exit rate implies a shorter average unemployment duration, which is the main objective of an employment office. The result and resource variables of employment offices, and DEA program inputs and outputs are summarized in Table 1. An REO’s objectives should be set in terms of variables that can be influenced by its management. We chose to apply the output-oriented DEA since the DEA output (number of hires) is controllable, while some DEA inputs are not. In particular, the number of job-seekers 3

For a comprehensive discussion of models incorporating all inefficiency components, see [20].

ARTICLE IN PRESS 176

A. Vassiliev et al. / Socio-Economic Planning Sciences 40 (2006) 169–186

Table 1 Result and resource variables of employment offices and DEA inputs and outputs Variable 1 2 3 4

Number of hires Number of entries into long-term unemployment Number of UI benefit exhaustees Number of re-entries into unemployment 4 months after having found a job Number of REO job counselors Number of registered job-seekers with UI benefit entitlement

5 6

Type

Should be:

Enters DEA as:

Result Result Result Result

Maximized Minimized Minimized Minimized

Output Input Input Input

Resource Resource

Minimized Minimized

Input Input

registered at an office generally depends on local economic conditions. A large number of jobseekers in a region may thus be due to a mismatch between local labor demand and supply, and caused by factors not under the office’s control, at least in the short run. Note that when included in the list of inputs, the number of job-seekers may even be treated as an exogenous environmental factor entering the DEA formulation as outlined in [21, p. 250].4 According to the Swiss federal authorities, the REO efficiency measure should take into account the external operating environment. After computing efficiency scores for all offices in the first step, we thus analyze how the efficiency of REOs is influenced by exogenous operating conditions beyond control of the office. An alternative to this two-step procedure is the all-in-one approach, which incorporates the external operating environment variables directly into the DEA along with traditional inputs and outputs [21,22]. Unfortunately, the single-step approach has some drawbacks. First, it requires that each exogenous variable be classified as an input or an output prior to efficiency estimation. According to [21, p. 250], ‘‘y this prior input or output classification is unsuitable if the objective is to test whether a particular operating environment is favorable or unfavorable.’’ Second, incorporating several inputs and outputs into a DEA analysis performed on small- or mediumsized samples leads to a large percentage of units being fully efficient. This lowers our ability to separate the successful, i.e., fully efficient group from the less successful group. We thus opt for the more traditional two-step procedure. The results of our estimations are presented below in Section 5.

4. Data Our data pertain to 156 regional employment offices that operated during the period April 1998 to March 1999 (i.e., 12 months). Most data were provided by Seco, which uses them for supervision purposes. They consist of monthly averages, which allows for the elimination of 4

An additional input enables production of more output, holding efficiency constant.

ARTICLE IN PRESS A. Vassiliev et al. / Socio-Economic Planning Sciences 40 (2006) 169–186

177

cyclical fluctuations. We also considered selected variables characterizing the heterogeneity of labor markets across different cantons. These data consist of cantonal variables and originate from different official sources. Some of the REOs were aggregated for different reasons, such as a close complementarity of their activities, or re-organization (see [5] for a detailed description of the reasons for aggregation). While rendering interpretation of the results more difficult, aggregation has the advantage of making them comparable to those obtained in [5]. Aggregation reduced the number of offices to 137. Five additional offices were deleted due to missing values in the result or resource variables, thus reducing the total number of observations to 132. The variables used in this study may be separated into four distinct groups. The first comprises four variables measuring the REO’s results (see Section 3). The second group consists of the REO’s ‘‘resources’’—i.e., the number of job counselors and the number of registered UI benefit recipients. Next are the REO’s activities. These include the number of days during which UI benefits were cut off, the number of counseling sessions, the number of unemployed in intermediate earnings programs, and the number of active labor market programs organized by the office.5 The final group of variables describes the office’s operating environment, i.e., the exogenous variables that influence results but cannot be controlled by the office itself (at least in the short term). This group consists of variables relating to both the office and the canton. Table A1 in Appendix summarizes descriptive statistics for these variables.

5. Assessing the performance of regional employment offices In this section, the model of performance measurement introduced in Section 3 is applied to the REO data described in Section 4. We first report evidence of considerable variation in performance across regional offices. We then examine the portion of measured inefficiency that is associated with selected environmental factors. 5.1. Results on efficiency The DEA problem presented in Section 3 was solved with the Efficiency Measurement System (EMS) software of Scheel [23]. The resulting mean efficiency score was 0.8459, with standard deviation of 0.0906. The minimum efficiency was 0.5725 with the number of efficient offices being 15 (i.e., 11.36% of 132 offices). Among the 15 efficient REOs, none was efficient by default.6 However, three efficient offices dominate only one, two and three other offices, respectively. Since these offices were compared to 5

Intermediate earnings programs consist of a wage subsidy for temporary jobs in the regular labor market that would otherwise not be taken up by the unemployed. The most frequently used ALMPs are courses that seek to improve the effectiveness of an individual’s job search and his/her self-esteem, language courses (mostly offered to non-Swiss unemployed), computer courses (basic word processing and spreadsheet usage), and employment programs in the public sector. These programs represent approximately 85% of the total ALMP supply. The remaining 15% are programs for young unemployed, incentives for self-employment, etc. 6 An office would be efficient by default if, given the level of inputs it uses, there is no other office that uses similar levels of inputs but produces a lower output. This is basically a finite sample problem. For example, an office whose size

ARTICLE IN PRESS 178

A. Vassiliev et al. / Socio-Economic Planning Sciences 40 (2006) 169–186

very few other offices, their performances should be checked further. Indeed, their efficiency score of 1 may reflect mainly the small sample problem rather than true efficiency. All other efficient offices dominate more than five inefficient offices each. In addition to the relative ranking of REOs, DEA provides guidance for improving the performance of a given REO. For example, the REO with ID 3 realized an efficiency score of 0.8384. This suggests that, to be fully efficient, this office should increase the number of hires (result 1) by a factor of 1=0:8384 ¼ 1:1928 (i.e., a 19.28% increase), keeping constant the number of registered UI benefit recipients, job counselors, entries into long-term unemployment (result 2), number of persons losing UI benefit entitlement (result 3), and re-entries into unemployment (result 4). To estimate the efficiency score E ¼ 0:8384, office 3 was compared to four efficient offices (its peers).7 It may be that the performance of this REO could be improved by examining these similar, but more efficient, offices in order to discover what they are doing differently. 5.2. Determinants of inefficiency 5.2.1. Estimation We now ascribe the variations in efficiency scores across REOs to selected exogenous variables. Our objective is twofold. First, we attempt to identify those exogenous factors that may explain part of the variation in efficiency scores. Next, an efficiency score that integrates the influences of exogenous factors is proposed. The need for this score stems from the fact that strong regional differences in REOs’ operating environments render direct comparisons of the previously reported efficiency scores difficult. A new ‘‘corrected’’ score allows a ceteris paribus comparison of office efficiencies under different external conditions, e.g. in different cantons and/or facing heterogeneous populations of job-seekers. The usual approach to analyzing effects of the operating environment on efficiency is to regress efficiency scores on the exogenous variables: E i ¼ X 0i b þ
i ;

i ¼ 1; . . . ; n,

(2)

where X i is the vector of exogenous variables associated with employment office i and to the canton where it is located (see Table A1 in Appendix), and
i is a normally distributed error term with zero mean. In order to prevent size effects, all variables relating to REOs have been divided by the number of registered UI benefit recipients. We go a step further and interpret the above equation as a decomposition of the efficiency score in terms of the operating environment, X 0i b, and managerial performance,
i . More precisely, the estimated residual
^i is the deviation from the mean managerial performance that would be observed in a given environment X i . If
^i 40 (
^i o0), the REO’s managerial efficiency is higher (lower) than the mean efficiency that would be observed in environment X i ; and, if
^i ¼ 0, the efficiency is equal to the mean efficiency in the given environment. The ranking of employment (footnote continued) is substantially larger or smaller than the size of all other offices will be identified as efficient by DEA. While being ascribed an efficiency score of 1, such an office could possibly improve its results. Therefore, care should be exercised in interpreting the efficiency scores for those REOs that dominate no other, or few other, offices. 7 The complete list of efficient peers for all inefficient offices is not reported here for reasons of space.

ARTICLE IN PRESS A. Vassiliev et al. / Socio-Economic Planning Sciences 40 (2006) 169–186

179

offices from lowest to highest estimated residual
^i suggests relative performance and takes into account, ceteris paribus, the heterogeneity in their operating environments. As the dependent variable is restricted to values between zero and one, Eq. (2) is usually estimated by means of a Tobit regression.8 While being satisfactory for explanatory purposes, Tobit regression is inappropriate for prediction purposes. Indeed, the estimated residual
^i obtained from a Tobit model is a biased estimate of the true residual
i . It is possible to overcome the problem of truncation of the dependent variable E i by applying a ranking procedure of efficient units (i.e., employment offices with unit efficiency scores) initially proposed by Andersen and Petersen [24]. They note that since efficiency scores for efficient firms equal one, only the inefficient units can be ranked on the basis of DEA scores. Andersen and Petersen rank the efficient units by measuring the distance from the efficient unit to a frontier, based on a set of observations that excludes the efficient unit in question (see [25] for an application). The most efficient unit is the one that can proportionally reduce outputs relatively the most without becoming inefficient. In the Andersen–Petersen DEA model, inefficient observations are assigned the same efficiency score as in the DEA of Section 3, i.e., E APi ¼ E i , where E APi denotes the Andersen–Petersen score. These scores conserve their standard interpretation. For efficient observation, the Andersen–Petersen scores may be equal to or larger than one. The interpretation of scores larger than one remains the same as for the standard Farrell measure: the office with a score e.g. E APi ¼ 1:2 may decrease its output vector proportionally up to a factor of 1=1:2 ¼ 0:8333 and remain efficient. However, it will be dominated by other offices if the proportional decrease in the output vector exceeds 0.8333.9 Since the scores, E APi , are no longer truncated to one, we may now estimate equation (2) by ordinary least squares (OLS) where we substitute E i by E APi . The resulting model is reported in the first three columns of Table 2. The last two columns of the same table show, for purposes of comparison, the coefficients for Eq. (2) estimated by the maximum likelihood Tobit technique with the dependent variable truncated at one.

5.2.2. Interpretations Four variables relative to REOs and two relative to cantons where these offices are located explain 37.93% of the variance in the dependent variable, according to the OLS regression. Interpretation of the factors affecting REO efficiency is as follows. The fractions of foreign workers with yearly B and permanent C work permits among registered job-seekers registered affect efficiency in opposite ways. Both groups have lower skill levels than the Swiss;10 however, the foreign work force is concentrated in different economic sectors, according to the type of work permit. Foreigners with yearly work permits are often employed in economic sectors that are subject to cyclical fluctuations of activity (such as construction, retail 8

This approach is admittedly somewhat artificial inasmuch as it is hard to picture relative efficiency scores as partially observable variables censored at one. 9 Wilson [26] observes that the Andersen–Petersen measure is also suitable for detecting outliers in DEA. Wilson [26, p. 33] also notes that it is possible E APi cannot be computed for some observations under variable returns to scale technology. For these units (three in our sample), we arbitrarily set the Andersen–Petersen score E APi to 1. 10 This is a well known feature of the Swiss labor market. It is due to past immigration policy when work permits were issued mainly for unskilled, or low skilled, jobs.

ARTICLE IN PRESS 180

A. Vassiliev et al. / Socio-Economic Planning Sciences 40 (2006) 169–186

Table 2 Determinants of inefficiency OLS Variable

Coef.

Foreign B Foreign C Women No UI benefit Women cant Part-time cant Constant Ancillary parameter n R2 Adjusted R2 F ð6; 122Þ statistic Pseudo R2 Log L LR w2 ð6Þ

0.5567
0.3574
0.7029 0.1252 1.9212
1.5517 0.6226

*** *** *** ** *** *** **

Tobit St. Dev.

Coef.

0.2061 0.1233 0.2204 0.0539 0.6950 0.3269 0.2492

0.4801
0.2783
0.6573 0.0993 1.9943
1.1369 0.4901 0.0844

129a 0.3793 0.3488 12.4327

St. Dev. ** ** *** * *** *** ** *** 132

0.1919 0.1162 0.2081 0.0510 0.6468 0.3157 0.2306 0.0057

0.2877
108.5912 48.5312

*/**/***: Statistically significant at the 10/5/1 percent level. a REOs with IDs 27, 45 and 112 excluded from analysis as outliers.

selling, hostelry and restaurants), while the established foreigners work in quite the same economic sectors as the Swiss. According to empirical literature explaining duration of unemployment, workers in cyclical sectors experience the shortest unemployment spells [27]. Their lower education thus seems to be more than compensated for by employment in economic activities with high turnover. Clearly, this improves REO efficiency where large numbers of permit B job-seekers are registered. The unemployment rate of foreign workers is usually higher than that of nationals [28,29], which is due either to lower average skills, or possibly discrimination by employers.11 Both effects can prove detrimental to reinserting foreign workers in the labor market. In the case of established foreigners, the C work permit suggests low education and produces negative effect on REO efficiency. Employment offices with a high fraction of women among the registered job-seekers were found to be less efficient. According to empirical literature on the duration of unemployment in Switzerland, women experience longer unemployment spells than do men [27]. Next, it is possible that REOs’ dedication to placing women is weaker than for men, as they might be considered the family ‘‘bread-winner’’ by job counselors, or employers. Finally, womens’ motivation to reintegrate employment might be less than for their male counterparts.

11 For example, employers may give priority to national job-seekers when they seek to fill a vacant position, although a foreign unemployed worker could be available for the job with the required skills.

ARTICLE IN PRESS A. Vassiliev et al. / Socio-Economic Planning Sciences 40 (2006) 169–186

181

Table 3 Quantile regressions of score E APi 0.25 quantile Variable

Coef.

Foreign B Foreign C Women No UI benefit Women cant Part-time cant Constant n Pseudo R2

0.7017
0.2971
0.6310 0.1355 1.4004
0.6482 0.0470

** * ** * *

0.75 quantile St. Dev.

Coef.

0.3010 0.1727 0.2604 0.0710 0.9357 0.3835 0.3557

0.4363
0.3742
0.8714 0.1621 0.6948
1.1519 1.1166

129a 0.1134

St. Dev. *** *** *** *** *** *** 129a 0.2850

0.1595 0.1067 0.1919 0.0447 0.5718 0.3137 0.1862

*/**/***: Statistically significant at the 10/5/1 percent level. a REOs with IDs 27, 45 and 112 excluded from analysis as outliers.

REO efficiency improves as the share of registered job-seekers not entitled to federal unemployment insurance benefit increases. Hence, job-seekers without entitlement to UI benefits12 have, on average, better labor market characteristics and/or a stronger motivation than those unemployed entitled to UI benefits. We now turn to the cantonal variables characterizing the local labor market where the REOs operate. A higher fraction of women employed in the canton shows the opposite effect to that found for women registered at the office. Cantons with more women in the labor force may have greater market flexibility. Finally, a higher fraction of part-time workers in a canton’s labor force has a negative impact on the efficiency of REOs located in that canton. This result is not entirely new: in an analysis of Swiss micro-economic duration data, Sheldon [30] found that the unemployed searching for a part-time job are at higher risk of long-term unemployment. One could question whether the operating environment has the same impact on efficiency for REOs located in the lower and upper tails of the efficiency distribution. A straightforward way to answer this question is to split the sample of REOs by the mean efficiency and run an OLS regression for each of the two sub-samples. However, this procedure implies an important loss of degrees of freedom. We thus prefer to run the 0.25 and 0.75 quantile regressions for the entire sample. Results are reported in Table 3.

12 The reasons for not being entitled to UI benefits can be various: people who register as unemployed but were not working before (e.g., having just finished their education or are re-entering the labor force after a period of inactivity, or are returning from abroad) typically have not paid any UI contribution and thus do not receive UI benefits. Also, UI is not compulsory for some self-employed, as in most countries. People quitting their jobs have to wait a certain period of time before being entitled to benefits.

ARTICLE IN PRESS 182

A. Vassiliev et al. / Socio-Economic Planning Sciences 40 (2006) 169–186

Notice the difference in quality of fit (as evaluated by the pseudo R-square) between the two regressions. Accordingly, while five out of six explanatory variables remain statistically significant in the 0.75 quantile, only two conserve significance at the 5% level in the 0.25 quantile. Hence, impact of the operating environment is of more importance for the highly efficient offices than for the least efficient ones. This result is quite sensible: the highly efficient REOs may have already reaped all the productivity gains from improving their managerial efficiency. Efforts to further improve performance may thus be limited by the operating environment, which highlights why exogenous variables are of such importance. On the other hand, it could be that low efficiency offices may accomplish important productivity gains without reaching limits imposed by external factors. Hence, these factors play a less important role in such situations.

5.3. Efficiency scores corrected for the operating environment We now compute a new efficiency score that takes the effects of exogenous variables into account. This corrected score, E Ci , equals the estimated residual,
^i , of the OLS regression model in Table 2. Compared to the DEA scores, E i , the corrected scores have zero mean. In accounting for the operating environment, the standard DEA interpretation is lost. Thus, peers may not be identified nor is the score interpretation (percentage of the maximum output level) conserved. However, as noted above, the corrected scores allow for a ranking of REO efficiency while considering differences in exogenous conditions. Hence, a direct comparison of REO efficiency servicing different segments of the Swiss labor market becomes easier. In our research, accounting for the exogenous environment did not induce dramatic changes in the efficiency rankings of employment offices. Indeed, the correlation coefficient between the ordinary DEA and corrected efficiency scores was 0.92 (0.79 if three outliers are deleted). Further, Spearman coefficient and Kendall’s t are positive and significant (0.80 and 0.62, respectively). Finally, let us consider the question of how the activities of employment offices, which include both disciplining measures and supportive actions, influence the corrected efficiency scores. We thus ran an OLS of corrected efficiency scores on the four REOs’ activity variables. None of the resulting coefficients was statistically significant. The following interpretation may be offered for this finding: employment offices operate under very different exogenous conditions. As mentioned in Section 2, substantial differences were reported in the use of supportive actions and disciplining measures among various REOs. We may suggest that in adapting their activities to local market conditions, REOs have selected very specific sets of activities that would be inappropriate in other operating environments. It is thus possible that no systematic relationship exists between efficiency and activities that would be common to all offices. We can further conjecture that all REO activities are at optimal levels, so that either an increase or decrease would deteriorate efficiency. In such situation, the relation between explanatory and dependent variables should exhibit a quadratic pattern allowing for a maximum. We tested this hypothesis by regressing the corrected scores on squared values of the activity

ARTICLE IN PRESS A. Vassiliev et al. / Socio-Economic Planning Sciences 40 (2006) 169–186

183

variables. Since the regression coefficients remained insignificant, the hypothesis was, as expected, rejected.

6. Conclusions The Swiss authorities are currently implementing performance-based budgeting for public employment service in order to lower structural unemployment. Performance measurement of regional employment offices is thus necessary to reap the full rewards of performance-based budgeting. Efficiency measures encourage REOs to improve performance, as this information makes them more accountable to government authorities. To estimate the relative technical efficiency of REOs in Switzerland between April 1998 and March 1999, we used the non-parametric DEA technique. We found mean inefficiency on the order of 15% of best observed performance, which offers guidance on possible improvement in the number of hires. We also established a relative ranking of REOs based on efficiency. We then sought to relate variations in efficiency scores to the external operating environment and to variables describing REOs activities. We found that selected socio-economic characteristics of the local labor market explained nearly one third of the variation in REO performance. Further, a new ranking of employment offices, taking into account differences in their operating environments, was obtained. No relationship, however, was found between the activity variables of regional offices and their technical efficiency when accounting for the operating environment. Our evaluation approach, and the ranking of regional employment offices, may be easily interpreted by policymakers and managers of REOs. Indeed, it provides guidelines for raising the efficiency of the public employment service. In our analysis, for each inefficient employment office identified, a set of similar, but more efficient offices, was found. The performance of inefficient offices could thus be improved by discovering what these more efficient offices are doing differently.

Acknowledgements We wish to express our gratitude to Philippe Vanden Eeckaut for useful discussions from which this paper has benefited substantially. We appreciate the very helpful comments by two anonymous referees of this journal and by the Editor. All remaining errors and omissions should be solely attributed to the authors. The work underlying this paper stems from National Research Programme 45 (Problems of the Welfare State) financed by the Swiss National Science Foundation (NSF). Financial support from the NSF is gratefully acknowledged.

Appendix For the descriptive statistic see Table A1.

ARTICLE IN PRESS A. Vassiliev et al. / Socio-Economic Planning Sciences 40 (2006) 169–186

184 Table A1 Descriptive statistics Variable

Description

A. REO result Result 1 Result 2 Result 3 Result 4

variables Nb of exits from unemployment (hires) Nb of entries in long unemployment Nb of UI benefit exhaustees Nb of re-entries into unemployment in 4 months after having found a job

B. REO activity variables UIB cut Nb of days when UI benefits were cut off Sessions Nb of counseling sessions Int earnings Nb of unempl. in intermed. earnings pgm ALMP Nb of ALMPs organized by REO C. REO resource variables Recipients Nb of UI benefit recipients Counselors Nb of job counselors D1. Exogenous variables relative to REOa Foreign B Foreign UI benefit recipients with yearly residence permit Foreign C Established foreign UI benefit recipients Women Women No UI Job-seekers including those not entitled to UI benefit benefit D2. Exogenous variables relative to the labor force of cantonb Women cant Fraction of women Part-time Fraction of part-time workers cant

Mean

St. Dev.

Min.

Max.

150.70 26.05 26.35 12.16

121.36 28.29 24.40 11.24

15.92 1.33 2.08 0.75

897.33 254.00 152.42 83.25

332.94 238.19 286.18 260.37

272.14 244.95 232.83 191.07

8.33 11.83 2.58 7.50

1348.75 2047.83 1610.50 994.67

1048.07 15.45

1027.70 12.97

74.00 0.42

8945.25 95.17

0.13

0.04

0.04

0.26

0.27 0.45 1.45

0.08 0.05 0.16

0.07 0.26 1.05

0.41 0.59 1.96

0.41 0.26

0.01 0.03

0.38 0.18

0.43 0.30

a

These variables were divided by the number of UI benefit recipients registered at REO. For example, variable ‘‘women’’ equals the ratio of the number of women registered at REO to the number of UI benefit recipients registered at the same REO. b Cantonal variables pertain to the year 1998 and do not vary over the REOs located in the same canton. Source: [31].

References [1] Heckman JJ, LaLonde RJ, Smith JA. The economics and econometrics of active labor market programs. In: Ashenfelter OC, Card D, editors. Handbook of labor economics, vol. 3A. Amsterdam: Elsevier; 1999. [2] Accord ORP/LMMT/autorite´ cantonale. Swiss State Secretariat for Economic Affairs, Bern, 2000. [3] Cavin E, Stafford F. Efficient provision of employment service outputs: a production frontier analysis. Journal of Human Resources 1985;20:484–503. [4] Althin R, Behrenz L. A technical efficiency analysis of Swedish employment offices. In: Tronti L, editor. Benchmarking employment performance and labour market policies, final report of the Employment Observatory Research Network, Berlin: IAS; 1998. p. 228–36.

ARTICLE IN PRESS A. Vassiliev et al. / Socio-Economic Planning Sciences 40 (2006) 169–186

185

[5] ATAG Ernst and Young. Calcul des re´sultats des ORP compte tenu des facteurs d’influence exoge`nes. Final report, Swiss State Secretariat for Economic Affairs, Bern, 1999. [6] Sheldon G. The efficiency of public employment services: a nonparametric matching function analysis for Switzerland. Journal of Productivity Analysis 2003;20:49–70. [7] Torgersen A, Førsund F, Kittelsen S. Slack-adjusted efficiency measures and ranking of efficient units. Journal of Productivity Analysis 1996;7:379–98. [8] Lindeboom M, van Ours JC, Renes G. Matching employers and workers: an empirical analysis of the effectiveness of search. Oxford Economic Papers 1994;46:45–67. [9] van Ours JC. Matching unemployed and vacancies at the public employment office. Emprircal Economics 1994;19:37–54. [10] Ibourk A, Perelman S. Frontie`res d’e´fficacite´ et processus d’appariement sur le marche´ du travail au Maroc. Economie et Pre´vision 2001(150–151):33–45. [11] Blanchard OP, Diamond P. The Beveridge curve. Brookings Papers on Economic Activity 1989;1:1–76. [12] Greene WH. The econometric approach to efficiency analysis. In: Fried H, Lovell CAK, Schmidt S, editors. The measurement of productive efficiency: techniques and applications. Oxford: Oxford University Press; 1993. [13] Curti M, Meins E. Politique suisse du marche´ du travail: diffe´rences re´sultant de sa mise en application au niveau du service de l’emploi. La Vie Economique—Revue de Politique Economique 1999;2:64–70. [14] Koopmans TC. An analysis of production as an efficient combination of activities. In: Koopmans TC, editor. Activity analysis of production and allocation, Cowles commission for research in economics, Monograph, vol. 13. New York: Wiley; 1951. [15] Farrell MJ. The measurement of productive efficiency. Journal of the Royal Statistical Society, Series A 1957;CXX:253–81. [16] Lovell CAK. Production frontiers and productive efficiency. In: Fried H, Lovell CAK, Schmidt S, editors. The measurement of productive efficiency: techniques and applications. Oxford: Oxford University Press; 1993. [17] Charnes A, Cooper WW, Rhodes E. Measuring the efficiency of decision making units. European Journal of Operational Research 1978;2:429–44. [18] Seiford LM. Data envelopment analysis: the evolution of the state of the art (1978–1996). Journal of Productivity Analysis 1996;7:99–137. [19] Fa¨re R, Grosskopf S, Lovell CAK. Production frontiers. Cambridge: Cambridge University Press; 1994. [20] Cooper WW, Park KS, Pasto JT. RAM: a range adjusted measure of inefficiency for use with additive models, and relations to other models and measures in DEA. Journal of Productivity Analysis 1999;11:5–42. [21] Fried HO, Schmidt SS, Yaisawarng S. Incorporating the operating environment into a nonparametric measure of technical efficiency. Journal of Productivity Analysis 1999;12:249–67. [22] Banker RD, Morey RC. Efficiency analysis for exogenously fixed inputs and outputs. Operations Research 1986;34(4):513–21. [23] Scheel H. EMS: efficiency measurement system user’s manual. Mimeo, Operations Research und Wirtschaftsinformatik, University of Dortmund (available from http://www.wiso.uni-dortmund.de/lsfg/or/scheel/ems/), 2000. [24] Andersen P, Petersen NC. A procedure for ranking efficient units in Data Envelopment Analysis. Management Science 1993;39:1261–4. [25] Lovell CAK, Walters LC, Wood LL. Stratified models of education production using DEA and regression analysis. In: Charnes A, Cooper WW, Lewin AY, Seiford LM, editors. Data envelopment analysis: theory, methods and applications. New York: Quorum Books; 1994. [26] Wilson PW. Detecting influential observations in data envelopment analysis. Journal of Productivity Analysis 1995;6:27–45. [27] Flu¨ckiger Y, Vassiliev A. Les raisons de la diffe´rence de choˆmage entre Gene`ve et le reste de la Suisse. Swiss Journal of Economics and Statistics 2002;138(4):387–410. [28] de Coulon A. Four essays on the labor market assimilation of immigrants in Switzerland. PhD Thesis No. 494, Department of Economics, University of Geneva, 1999. [29] Flu¨ckiger Y, Ramirez JV. Analyse des salaires entre les hommes et les femmes sur la base de la LSE 1994 et 1996. Report No. 10, Observatoire universitaire del’emploi, University of Geneva, 2000.

ARTICLE IN PRESS 186

A. Vassiliev et al. / Socio-Economic Planning Sciences 40 (2006) 169–186

[30] Sheldon G. Die Langzeitarbeitslosigkeit in der Schweiz: Diagnose und Therapie. Bern: P. Haupt; 1999. [31] Recensement Fe´de´ral des Entreprises 1998, Swiss Federal Statistical Office, Neuchaˆtel, 1998.

Anatoli Vassiliev is Research Affiliate at the Center for Employment Research, University of Geneva. He holds an M.A. and a Ph.D. in economics and finance from the University of Geneva, Switzerland, and an M.A. in econometrics from the Universite´ Catholique de Louvain, Belgium. His research interests focus on evaluation of active labor market policies, measurement of efficiency of public services, and, more recently, on the application of production economics to environmental issues. He has co-authored two studies for the Swiss National Science Foundation, and his research has appeared in the Swiss Journal of Economics and Statistics. The current paper draws on his Ph.D. dissertation.

Giovanni Ferro Luzzi holds an M.A. in economics from University of Manchester, UK, and a Ph.D. in economics from the University of Geneva, Switzerland, where he now has a position of Assistant Professor from the IRIS research program. His research interests are in labor economics, earnings distribution analysis, and applied microeconomics. Professor Luzzi’s research has been published in the Swiss Journal of Economics and Statistics, International Journal of Manpower, and Pacific Economic Review. He also works as a consultant for the International Labor Organization (ILO) and the Organization for Economic Co-operation and Development (OECD). He contributed to a report on the social impact of globalization for the former, and currently writes reports on the employability of older workers in Switzerland and Italy for the latter.

Yves Flu¨ckiger is Professor, Department of Economics, University of Geneva, Switzerland. He is a member of the Board of the Applied Economics Center (LEA). He teaches labor economics, industrial organization and public finance. He received his Ph.D. in economics from the University of Geneva. Professor Flu¨ckiger has been Research Associate at the Universities of Harvard (USA) and Oxford (UK) and Invited Professor at the Universities of Lausanne, Fribourg (Switzerland) and Deakin (Australia). He has written numerous books and articles published in international reviews including The International Trade Journal, Economics Letters, Journal of Econometrics, Journal of Income Distribution, The Oxford Bulletin of Economic and Statistics, Swiss Journal of Economics and Statistics, and Economie Applique´e. His recent research involves migrations, gender wage differentials, gender segregation and new forms of employment. Professor Flu¨ckiger is a member of the Swiss National Science Foundation, and Vice-President of the Swiss Competition Commission.

Jose´ V. Ramirez is Professor, Geneva School of Business Administration, University of Applied Sciences of Western Switzerland, and Research Affiliate at the Center for Employment Research, University of Geneva, Swizerland. His research interests focus on wage formation and gender issues at the firm level, measurement of efficiency of public and private services, and, more recently, on the impact of environmental characteristics on rents. His research has been published in Research in Labor Economics, Swiss Journal of Economics and Statistics, International Journal of Manpower, and Urban Studies.