ArticlePDF Available

Unemployed and Their Caseworkers: Should They Be Friends or Foes?

Authors:
  • Swiss National Bank
Article

Unemployed and Their Caseworkers: Should They Be Friends or Foes?

Abstract and Figures

In many countries, caseworkers in public employment offices have dual roles of counselling and monitoring unemployed people. These roles often conflict, which results in important caseworker heterogeneity: some consider providing services to their clients and satisfying their demands as their primary task. However, others may pursue their own strategies, even against the will of the unemployed person. They may assign jobs and labour market programmes without the consent of the unemployed person. On the basis of a very detailed "linked jobseeker-caseworker" data set for Switzerland, we investigate the effects of caseworkers' co-operativeness on the probabilities of employment of their clients. Modified statistical matching methods reveal that caseworkers who place less emphasis on a co-operative and harmonic relationship with their clients increase their chances of employment in the short and medium term. Copyright (c) 2009 Royal Statistical Society.
Content may be subject to copyright.
DISCUSSION PAPER SERIES
ABCD
www.cepr.org
Available online at:
www.cepr.org/pubs/dps/DP6558.asp
www.ssrn.com/xxx/xxx/xxx
No. 6558
UNEMPLOYED AND THEIR
CASEWORKERS: SHOULD THEY
BE FRIENDS OR FOES?
Stefanie Behncke, Markus Frölich and
Michael Lechner
LABOUR ECONOMICS
ISSN 0265-8003
UNEMPLOYED AND THEIR
CASEWORKERS: SHOULD THEY
BE FRIENDS OR FOES?
Stefanie Behncke, University of St.Gallen
Markus Frölich, University of St.Gallen, IFAU and IZA
Michael Lechner, University of St.Gallen, ZEW, IZA, PSI and CEPR
Discussion Paper No. 6558
November 2007
Centre for Economic Policy Research
90–98 Goswell Rd, London EC1V 7RR, UK
Tel: (44 20) 7878 2900, Fax: (44 20) 7878 2999
Email: cepr@cepr.org, Website: www.cepr.org
This Discussion Paper is issued under the auspices of the Centre’s research
programme in LABOUR ECONOMICS. Any opinions expressed here are
those of the author(s) and not those of the Centre for Economic Policy
Research. Research disseminated by CEPR may include views on policy, but
the Centre itself takes no institutional policy positions.
The Centre for Economic Policy Research was established in 1983 as a
private educational charity, to promote independent analysis and public
discussion of open economies and the relations among them. It is pluralist
and non-partisan, bringing economic research to bear on the analysis of
medium- and long-run policy questions. Institutional (core) finance for the
Centre has been provided through major grants from the Economic and
Social Research Council, under which an ESRC Resource Centre operates
within CEPR; the Esmée Fairbairn Charitable Trust; and the Bank of
England. These organizations do not give prior review to the Centre’s
publications, nor do they necessarily endorse the views expressed therein.
These Discussion Papers often represent preliminary or incomplete work,
circulated to encourage discussion and comment. Citation and use of such a
paper should take account of its provisional character.
Copyright: Stefanie Behncke, Markus Frölich and Michael Lechner
CEPR Discussion Paper No. 6558
November 2007
ABSTRACT
Unemployed and Their Caseworkers:
Should They Be Friends or Foes?*
In many countries, caseworkers in a public employment office have the dual
roles of counselling and monitoring unemployed persons. These roles often
conflict with each other leading to important case-worker heterogeneity: Some
consider providing services to their clients and satisfying their demands as
their primary task. Others may however pursue their strategies even against
the will of the unemployed person. They may assign job assignments and
labour market programmes without consent of the unemployed person. Based
on a very detailed linked jobseeker-caseworker dataset, we investigate the
effects of caseworkers' cooperativeness on the employment probabilities of
their clients. Modified statistical matching methods reveal that caseworkers
who place less emphasis on a cooperative and harmonic relationship with
their clients increase their employment chances in the short and medium term.
JEL Classification: C31 and J68
Keywords: public employment services, statistical matching methods and
unemployment
Stefanie Behncke
Swiss Institute for International
Economics and Applied Research
University of St.Gallen
Bodanstr. 8
CH-9000 St.Gallen
SWITZERLAND
Email: Stefanie.Behncke@unisg.ch
For further Discussion Papers by this author see:
www.cepr.org/pubs/new-dps/dplist.asp?authorid=167250
Markus Frölich
Swiss Institute for International
Economics and Applied Research
University of St.Gallen
Bodanstr. 8
CH-9000 St.Gallen
SWITZERLAND
Email: markus.froelich@unisg.ch
For further Discussion Papers by this author see:
www.cepr.org/pubs/new-dps/dplist.asp?authorid=167239
Michael Lechner
Swiss Institute for International
Economics and Applied Research
University of St.Gallen
Bodanstr. 8
CH-9000 St.Gallen
SWITZERLAND
Email: Michael.Lechner@unisg.ch
For further Discussion Papers by this author see:
www.cepr.org/pubs/new-dps/dplist.asp?authorid=139325
* We are very grateful in particular to Heidi Steiger and Stephan Werner for
help with the data and to Stephan Hammer for fruitful cooperation in this
project. We also thank Martin Grubb and Thomas Ragni as well as seminar
participants in Mannheim (ZEW), Nürnberg (Federal Employment Agency),
St.Gallen (COST meeting), Oslo (EALE), and Munich (German Economic
Association) for helpful comments and suggest. All remaining errors are our
own. We are grateful to the research fund of the Swiss unemployment
insurance system (at the seco) for providing the administrative database as
well as substantial financial support for this project.
Submitted 31 October 2007
1 Introduction*
We use a very informative linked jobseeker-caseworker dataset augmented by detailed information
on caseworker characteristics in combination with matching estimation to examine a question that is
important in implementing the counselling processes of unemployment insurance systems. The
question is about the impact of different levels of caseworkers' cooperativeness concerning their
unemployed clients on the future employment chances of their clients. Observing the employment
outcomes up to 3 years after registration, we find that a caseworker who is more demanding (i.e.
less co-operative) vis-à-vis the unemployed person achieves higher re-employment probabilities.
In most countries, caseworkers are assigned the dual role of counselling and monitoring of unem-
ployed persons. These two roles often conflict with each other: On the one hand, caseworkers need
to establish a trustful and empathetic relationship with their clients for providing effective counsel-
ling. On the other hand, they have to police job search behaviour and initiate and enforce sanctions
if it falls short of the requirements mandated by the unemployment insurance law. Since the legal
rules typically leave some leeway to the caseworker on how to weight these two potentially con-
flicting roles, it is not surprising that individual caseworkers weight them differently. Some case-
workers pursue a more dominating and demanding stance vis-à-vis the unemployed, while others
aim at a more cooperative relationship devoid of conflicts. Caseworkers perform this dual task by
setting certain rules and initiating certain actions for their clients (e.g. sending clients to specific

* Markus Frölich is also affiliated with IFAU, Uppsala, and IZA, Bonn. Michael Lechner has further affiliations with
ZEW, Mannheim, CEPR, London, IZA, Bonn, and PSI, London. We are very grateful in particular to Heidi Steiger
and Stephan Werner for help with the data and to Stephan Hammer for fruitful cooperation in this project. We also
thank Martin Grubb and Thomas Ragni as well as seminar participants in Mannheim (ZEW), Nürnberg (Federal Em-
ployment Agency), St.Gallen (COST meeting), Oslo (EALE), and Munich (German Economic Association) for help-
ful comments and suggest. All remaining errors are our own. We are grateful to the research fund of the Swiss un-
employment insurance system (at the seco) for providing the administrative database as well as substantial financial
support for this project.
3
programmes of the active labour market policies or imposing sanctions), in addition to more per-
sonal channels such as counselling style, personal relationships, empathy or sympathy.
Substantial research interest has been devoted to the question on how to improve the public unem-
ployment system. Various aspects or instruments of the relationship between jobseeker and em-
ployment office have been considered. The impact of explicit rules and incentives are examined in
Fredriksson and Holmlund (2003), who considered 'optimal' unemployment insurance (UI) systems.
They examine three different means of improving the efficiency of UI via passive labour market
policy: the duration of benefit payments, monitoring in conjunction with sanctions, and workfare,
and find that a system with monitoring and sanctions improves search incentives. Pavoni and
Violante (2007) and Wunsch (2007) provide a theoretical economic framework to determine the
'optimal' choice between different passive and active labour market policy instruments. In their
models, the typical optimal sequence of policies is unmonitored job search, followed by monitored
job search and social assistance as an absorbing state.
So far, the role of increased monitoring has been examined in Meyer (1995), Gorter and Kalb
(1996), and Dolton and O'Neill (1996), who found significant positive effects of monitoring in the
USA, Netherlands, and the UK, respectively. Ashenfelter, Ashmore, and Deschenes (2000), and
Bloom, Hill, and Riccio (2003) found no significant effects of increased monitoring in the USA,
though. The effects of sanctioning as one of the instruments of the employment offices have been
examined in Lalive, van Ours, and Zweimüller (2005), who found significant positive effects of
sanctions for Swiss unemployment recipients. Van den Berg, van der Klaauw, and van Ours (2004)
found positive effects of sanctions for Dutch welfare recipients, and Abbring, van den Berg and van
Ours (2005) found positive effects of sanctions on Dutch unemployment benefit recipients. Svarer
(2007) also finds positive effects of sanctions on the job-finding rate in Denmark. Black, Smith,
Berger, and Noel (2003), and Graversen and van Ours (2006) find evidence for threat effects of
4
employment and training programmes: mandatory assignment to a programme increases job finding
rates before participation starts, because unemployed want to avoid time-consuming programmes.
We add to this literature in that we do not consider only the effects of one single instrument (such as
monitoring or sanctioning) but rather the relationship between caseworkers and their unemployed
clients as a whole.1 In addition to explicit rules and incentives, the personal relationship between
caseworker and unemployed could have an important effect on motivation, job search intensity, and
job acceptance. A more demanding caseworker may use certain instruments more often but the dif-
ferent counselling style in itself can have important effects. These personal or behavioural factors
had received less attention in the economics literature until recently.
Principal-agent theory suggests that caseworkers could increase the unemployed person's job-
finding and job-taking rates through a less cooperative behaviour: Unemployed persons have an
incentive to avoid costly job search or to wait for better job offers, and caseworkers are required to
set appropriate incentives through compensation and supervisory schemes.2 Caseworkers should
thus demand more job search effort and treat unemployed accordingly. Confirming this line of ar-
gument, several studies found that monitoring and sanctions increase employment probabilities. On
the other hand, effective counselling by the caseworker may require a trustful atmosphere. An un-
employed person who expects a stiff caseworker will obviously provide distorted information about
his preferences, needs, skills, aspirations, job search efforts, and the like. In such an atmosphere,
counselling by caseworkers may be useless as well as any labour market training programmes. In

1 Using production frontier analysis Ramirez and Vassiliev (2007) have pointed out that substantial scope for improv-
ing the effectiveness of public employment services in Switzerland seems to exist, see also Ferro-Luzzi et al. (2005)
and Sheldon (2003). However, those analyses are not directly instructive in pointing out how to improve effective-
ness.
2 Shavell and Weiss (1979) argue that unemployment insurance lengthens unemployment duration because of its effect
on job search effort and the reservation wage. However, if caseworkers could monitor job search behaviour, no such
5
addition, recent experimental evidence indicates that individuals may have reciprocal preferences.
When being treated nicely by the caseworker, they may be more willing to behave nicely (Fehr and
Schmidt, 2001). Hence, caseworkers might achieve higher employment rates by cooperating with
their clients instead of potentially punishing them or by ignoring their requests. Caseworkers are
well aware of these two opposing views of the world and their trade-offs, but each caseworker may
be weighting the importance of these models differently.
This paper also helps to understand the determinants of exiting unemployment. One strand of the
literature has estimated the relationship between the unemployed person's characteristics (age, gen-
der, education, etc.) and the hazard rate for leaving unemployment, e.g. Machin and Manning
(1999). Another strand of the literature has evaluated how certain instruments such as labour market
programmes, monitoring or sanctions affect employment.3 Caseworker characteristics have received
less attention, though. This analysis is made possible by a unique linked jobseeker-caseworker data-
set with very detailed information on both jobseekers caseworkers. (To the best of our knowledge,
such data was not available before.) Focusing on the caseworker-client relationship enriches the
traditional evaluation literature since the imposition of sanctions or assignment of labour market
programmes could already be considered as an outcome of caseworker's behaviour.
We use semiparametric matching estimators that have been developed in the statistics literature and
successfully applied in labour economics since several years.4 To be more precise, we use statistical
propensity score matching with an extension to radius matching and incorporating regression-type
   
problem would exist. Unemployment insurance benefits could be withheld if effort or the reservation wage was un-
satisfactory.
3 See Heckman, Lalonde, and Smith (1999) for a survey of empirical findings of programme effects in the USA and
Europe, or Martin and Grubb (2001) for a survey of OECD experiences, or Wunsch (2005) for a survey of pro-
gramme effects in Germany. See the previous footnotes for the literature on sanctions and monitoring.
4 See e.g. Rosenbaum and Rubin (1983), Lechner (2002), Black and Smith (2004), Imbens (2004), or Ham and
LaLonde (2005).
6
adjustment following ideas by Rubin (1979), Abadie and Imbens (2006), and others. In contrast to a
conventional duration model analysis, which would focus only on the hazard rate out of unemploy-
ment, we estimate the employment probability at a certain point in time. This thus captures not only
the exits from unemployment to employment but includes also re-entry in unemployment, an im-
portant aspect from a public policy perspective.
Our estimation results indicate a positive effect of reduced cooperativeness on employment prob-
abilities of about 2 percentage points. Hence, pursuing a more demanding stance vis-à-vis unem-
ployed persons increases employment probabilities in the short and in the medium term (up to 3
years after the beginning of unemployment) by a non-negligible amount. This increased employ-
ment is not obtained at the cost of reduced stability of jobs. The effects on stable employment are
also positive and of similar magnitude. The sensitivity of these results is explored by examining
several alternative specifications, in particular concerning the choice of the control variables and
various definitions of the treatment variable. The results are rather stable, although often less pre-
cisely estimated.
The structure of the paper is as follows: The following section discusses the unemployment insur-
ance system in Switzerland. Section 3 describes the data, and Section 4 the statistical methodology.
Sections 5 and 6 give the main estimation results, and Section 7 concludes. Several appendices pro-
vide additional details.
2 The Swiss labour market and the role of caseworkers
2.1 Overview
Until the recession of the early 1990s, unemployment was extremely low in Switzerland, a small
country with 26 different administrative regions, called cantons. With the recession, the unemploy-
ment rate rose rapidly to 5% (see figure below) and triggered a comprehensive revision of the fed-
7
eral unemployment insurance act in 1996/1997. The about 3000 municipal unemployment offices
were consolidated to a smaller number of regional employment offices. Compared to the previous
municipal offices, which were largely concerned with administering unemployment benefits, these
regional offices, of which there were about one hundred operating in 2003, aimed at providing pro-
fessional services with respect to counselling, placement, activation, and training. A large number
of caseworkers were hired and further trained for these purposes.
Figure 1: Unemployment rate in Switzerland from January 1990 to August 2007
0
1
2
3
4
5
6
Jan 90
Jan 91
Jan 92
Jan 93
Jan 94
Jan 95
Jan 96
Jan 97
Jan 98
Jan 99
Jan 00
Jan 01
Jan 02
Jan 03
Jan 04
Jan 05
Jan 06
Jan 07
Note: Monthly unemployment rate, January 1990 – August 2007, Source: Swiss National Bank Monatshefte.
This was financed by the unemployment insurance. In 2003, the period we consider in our study,
both employers and employees were obliged to contribute a share of 2,5% of the salary. Benefits
amounted to 70-80% of the former salary depending on age, dependents and income. By July 2003
the rules for benefit entitlement were tightened for individuals younger than 56: the minimum con-
tribution time was raised from 6 to 12 months and the maximum benefit entitlement period was
reduced from 24 to 18,5 months.
8
2.2 Caseworkers' autonomy
Whereas the federal State Secretariat for Economic Affairs (seco) has a clear vision about the aims
the employment offices and caseworkers should pursue, with a strong focus on rapid re-
employment, the caseworkers generally enjoy substantial freedom in how they attempt to achieve
these goals and how they treat their clients. This freedom and heterogeneity arises from two factors.
First, the 26 cantons in Switzerland generally enjoy a large autonomy in their implementation of the
unemployment insurance law. Although none of them would violate clear legal provisions, such as
imposing stronger benefit sanctions than legally permitted, they have substantial leeway on many
other margins. The operational costs of the employment offices and their staff, as well as the costs
of labour market programmes and benefits payments are fully financed by the federal unemploy-
ment insurance funds.5 Therefore, it is tempting and without serious cost consequences for cantons
to pursue their own goals and philosophies to a limited extent. For example, one goal of such a local
strategy might be to avoid large numbers of people drawing on welfare benefits, which are financed
by the cantons (and their municipalities).6 A demanding stance vis-à-vis the unemployed may
quickly reduce the number of registered unemployed, but it may also lead to poor job matches, in-
stable jobs, and repeated unemployment, which could eventually lead to more people drawing on
welfare as they are no longer entitled to unemployment benefits. Being more lenient and trying to
satisfy the unemployed person's wishes may on the other hand lead to better job matches, and thus
more sustained employment, less job separation, and less turnover in the medium term, or at least
reduce the number of persons in need of welfare benefits.

5 In addition to the unemployment benefits, this includes the costs of maintaining and operating the employment of-
fices as well as active labour market programmes. Technically, the cantons bear the costs of the employment offices
and active labour market measures, but they are refunded by the federal unemployment insurance funds up to a fixed
ceiling that depends on the number of unemployed in the canton.
6 Rules for social assistance are set at the cantonal level. They vary widely with regard to cost distribution between
cantons and municipalities, as well as with regard to form and level of benefits and organisation.
9
The second source of caseworkers' autonomy arises within the cantons and particularly within the
employment offices. Many of the employment office managers consider it important to grant sub-
stantial autonomy to caseworkers such that they can develop their own personal counselling style
and react to the needs of their clients without being bound by many bureaucratic rules. This is also
confirmed by the caseworkers, who consider this freedom to be a very important aspect for their job
satisfaction.7
2.3 The relationship between caseworker and unemployed persons
The relationship between the caseworker and his clients is characterized by the two roles of the
caseworker: to help the unemployed person in searching and finding appropriate employment and to
monitor whether the unemployed person searches thoroughly enough and is indeed willing to take
up any job offer with acceptable pay and within acceptable commuting distance. Some caseworkers
put more emphasis on their role as a counsellor and aim for a trustful relationship, whereas other
caseworkers may see their policing role to be more important and may be more dominating and
demanding vis-à-vis the unemployed person.
To analyse the effects of the caseworker-client relationship a written questionnaire was adminis-
tered to all caseworkers and office managers in Switzerland about their aims, attitudes, behaviour,
etc.8 A key question asked the caseworker how important he/she considers the cooperation with the
unemployed person:

7 See e.g. the French and German versions of the interview protocols in the appendix to Frölich et al. (2007).
8 This data was collected as part of a large evaluation project for the Swiss State Secretariat for Economic Affairs and
is described in more detail later. Qualitative face-to-face interviews with the management and caseworkers were con-
ducted beforehand in 12 employment offices. Subsequently, all managers and caseworkers were surveyed with an
extensive written questionnaire. For details, see Frölich et al. (2007).
10
Table 1: Survey question on cooperativeness of the caseworker
How important do you consider the cooperation with the jobseeker, regarding placements in jobs, and as-
signment of active labour market programmes?
1 Cooperation is very important; the wishes of the unemployed person should be satisfied.
2 Cooperation is important, but placements in jobs and active labour market programmes should
sometimes be assigned or declined in spite of the unemployed person's wishes.
3 Cooperation is less important; I should assign placements in jobs and active labour market
programmes independent of the wishes of the unemployed person
Note: English translation of the respective question in the questionnaire. Questionnaires were in German, French, and Italian.
52% of the caseworkers chose option one, 39% of caseworkers chose option 2, and 9% of case-
workers chose option three. Only very few caseworkers did not respond to this question. In the
main empirical analysis, we will compare those caseworkers who chose option 1 (cooperative) with
those who chose option 2 or 3 (not so cooperative). In additional analyses, we will compare option
1 versus 3, leaving out those who answered with option 2.
When comparing these answers with the responses to other items of the questionnaire we observe
that the less cooperative caseworkers consider control and sanctions, job assignments, and employ-
ment programmes as instruments that are more important. Counselling meetings and interim jobs
(temporary wage subsidies) being less important. They also responded that they tended to assign
active labour market programmes to apply pressure and to control their clients' availability for jobs
and to give less emphasis to the wishes of the jobseeker.
The variation in the cooperativeness across caseworkers will be exploited to estimate the impact of
cooperativeness on the employment prospects of the unemployed. The cooperativeness of the case-
worker may be driven by several factors, which we have to take into account as these various char-
acteristics may themselves have an independent effect on the employment chances of their unem-
ployed. First, many characteristics of the caseworker himself may affect his attitude and behaviour.
Important factors could be age, gender, education, and, in particular, experience in the form of ten-
ure and participation in caseworker training programmes. Previous own experience of unemploy-
11
ment may also strongly affect the way the caseworker treats his unemployed clients. The case-
worker's attitude may also depend a lot on the average characteristics of his clients. For example, a
caseworker who is attending mainly unskilled foreigners with a poor employment history may be
treating them differently than a caseworker who deals mainly with highly skilled unemployed who
experienced unemployment for the first time in their life. Hence, characteristics such as gender, age,
nationality, qualification, and past unemployment experience of the clients are likely to affect case-
worker's behaviour. Furthermore, characteristics of the local labour market will also be relevant. If
the local unemployment rate is low, a caseworker may be less lenient with unemployed than other-
wise. As a last aspect, we will also consider certain aspects of the organisation of the employment
office, where the caseworker is employed at, as these may also determine the caseworker's behav-
iour. The data needed to control for those characteristics is presented in the next section.
3 Data
3.1 Data and sample selection
The population for the microeconometric analysis are all individuals who registered as unemployed
anytime during the year 2003, and their outcomes are followed up until the end of December 2006.
For these individuals very detailed individual information is available from the databases of the
unemployment insurance system (AVAM/ASAL) and the social security records (AHV). These
data sources contain socio-economic characteristics including nationality and type of work permit,
qualification, education, language skills (mother tongue, proficiency of foreign languages), experi-
ence, profession, position, and industry of last job, occupation and industry of desired job, an em-
ployability rating by the caseworker, etc. The data also contain detailed information on registration
and de-registration of unemployment, benefit payments and sanctions, participation in ALMP, and
12
the employment histories since January 1990 with monthly information on earnings and employ-
ment status (employed, unemployed, non-employed, self-employed).
The databases contain the population of all jobseekers including individuals employed but search-
ing for a job, unemployed without benefit entitlement, and unemployed entitled to benefits. In the
econometric analysis, we will focus on the last group since the first two groups are largely immune
to potential threats and sanctions of the caseworker.
In total, 239,004 persons registered as new jobseekers during the year 2003. Notice that we consider
only the first registration in 2003 for each person and subsume any further registrations within the
outcome variables, i.e. the analysis is person based and not spell based. As mentioned above, we
exclude jobseekers without benefit claim and individuals who applied for or claim disability insur-
ance. Furthermore, we exclude foreigners without permanent or yearly work permit, as they are not
entitled to most of the services of the unemployment insurance. For some unemployed their case-
worker is undefined, which may happen if an unemployed person de-registers before being assigned
to a caseworker. We also exclude a few employment offices that are not comparable to other offices
in 2003. In our main analysis we focus on the prime-age group (24 to 55 years old), with a final
sample size of 100,222. See Appendix A for further details.
3.2 Definition of outcomes and treatment variables
Each newly registered unemployed person in 2003 was linked to his first caseworker via the user
database of the unemployment insurance system (AVAM). This database contains basic informa-
tion about each caseworker, such as age and gender. To complement this information we conducted
an extensive survey of all caseworkers. A written questionnaire was sent to all caseworkers and
employment office managers who were employed at an employment office between 2001 and 2003
and were still active in December 2004, i.e. at the time the questionnaire was sent. The question-
13
naire contained questions about aims, strategies, processes, and organisation of the employment
office and the caseworkers. 1560 caseworkers and employment office managers returned the ques-
tionnaire, which is equivalent to a rate of return of 84% of the active caseworkers.9
The question most relevant to our analysis was shown in Table 1. Of the 1560 individuals who re-
turned the questionnaire, 159 office managers who did not counsel jobseekers during the year 2003
were not asked that question. 16 caseworkers did not answer to this question. 723 answered with
option 1, 540 answered with option 2, and 122 answered with option 3. In our main specification,
we consider those who answered with option 1 as cooperative caseworkers and those who chose
option 2 or 3 as less cooperative, which we also sometimes label as "non-cooperative" caseworkers.
Although the cooperation attitude of a caseworker may clearly vary between his clients, we expect
that a cooperative caseworker be on average more cooperative to all his clients than a less coopera-
tive caseworker is.
Combining options 2 and 3 to define non-cooperation is based on the presumption that both show a
deviation from the full cooperation attitude that more than half of all the caseworkers display. In our
robustness analysis, we will also consider other definitions. In particular, we compare option 1 di-
rectly with option 3.
We seek to estimate the effect of the cooperation attitude on the employment chances of the unem-
ployed person. The unit of measurement for the outcome variable is thus the unemployed person.
We consider an individual as employed in month t if she has de-registered with the employment
office because of having found an occupation and has not yet re-registered. Thus, we solely rely on
information from the employment office database to determine the employment situation. The em-
ployment status is measured with some error since a de-registered individual could have left the

9 Obviously, the questionnaire could not be conducted anonymously since we needed to link the answers of each case-
worker with their clients. However, caseworkers were guaranteed full confidentiality.
14
active labour force or could have found an occupation after de-registering without letting the em-
ployment office know. Nevertheless, a validation study for earlier years where employment data
was available from two sources (the unemployment database and the social security data) shows a
rather high reliability of our outcome variable, see Frölich et al. (2007).
To analyse the evolvement of the impacts of the caseworker's attitude on the employment probabili-
ties, the employment status, 0
,it
Y
τ
+, is measured until the end of 2006, relative to the time of first
registration, t0. Hence, for individuals who registered in January 2003, their employment status is
followed up for 47 months, whereas only 36 months are observed for those registering in December
2003. Observing employment for at least three years allows not only the estimation of short-term
effects, but also them medium-term effects of cooperation.
Figure 2 shows the employment rate for our sample relative to the time of registration at the em-
ployment office. The black line represents the employment rates for unemployed who were coun-
selled by a less cooperative caseworker, whereas the grey line refers to cooperative caseworkers.
The employment rates for both groups of unemployed are very similar. About 1% de-register one
month after registering because of having found an occupation. About 7% have found a new job
after two months and 44% have found employment one year after becoming unemployed.
15
Figure 2: Average employment rate in month t after registering as unemployed
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47
cooperative less cooperative
Note: Average employment rates are for prime age individuals in main sample. The grey (black) line shows averages for the
51923 (48299) unemployed who were counselled by a cooperative (less cooperative) caseworker.
The employment rates in Figure 2, however, do not account for the fact that the unemployed coun-
selled by cooperative and less cooperative caseworkers are quite distinct. Table 2 shows that these
two groups of unemployed differ in many characteristics. Foreigners, less qualified unemployed,
and unemployed with a poor employability rating are more likely to be counselled by less coopera-
tive caseworkers. In this sense, the less cooperative caseworkers face the more difficult cases,
which may be the reason for the small differences observed in Figure 2. A larger effect may be
masked by these differences in client's characteristics.
16
Table 2: Selected characteristics by cooperation attitude of the caseworker
Caseworker considers cooperation as
Very important
(option 1) Important, but
(option 2) Less important
(option 3)
Characteristics of the unemployed person
N (number of unemployed) 51923 39310 8989
Female 0.45 0.44 0.42
Age 36.5 36.6 36.6
Swiss 0.63 0.61 0.56
Foreigner with permanent work permit 0.24 0.25 0.29
Foreigner with yearly work permit 0.13 0.14 0.16
Qualification: unskilled 0.21 0.23 0.28
Qualification: semiskilled 0.16 0.16 0.14
Qualification: skilled without degree 0.04 0.04 0.05
Qualification: skilled with degree 0.58 0.57 0.53
Employability rating: low 0.13 0.14 0.17
Employability rating: medium 0.75 0.74 0.72
Employability rating: high 0.12 0.12 0.11
Characteristics of the caseworker
Number of case workers 723 540 122
Female 0.42 0.41 0.40
Age 45.2 43.7 42.8
Tenure in employment office in years 5.75 6.01 5.54
Own experience of unemployment 0.65 0.61 0.62
Education: vocational training 0.30 0.34 0.41
Education: above vocational training 0.46 0.41 0.40
Education: tertiary track (university or polytechnic) 0.24 0.25 0.19
Degree in vocational training for caseworkers 0.20 0.26 0.26
Allocation of unemployed to caseworkers a)
At random 0.22 0.23 0.22
By alphabet 0.03 0.04 0.07
By number of clients 0.42 0.44 0.44
By industry 0.52 0.57 0.56
By occupation 0.52 0.62 0.57
By age 0.03 0.04 0.01
By employability 0.07 0.06 0.07
By region 0.12 0.11 0.17
Other 0.08 0.08 0.04
Local labour market characteristics
German speaking employment office 0.69 0.71 0.60
French speaking employment office 0.26 0.22 0.19
Italian speaking employment office 0.05 0.07 0.20
Cantonal unemployment rate 3.70 3.75 3.79
Unemployment rate in industry 4.83 4.93 5.11
Note: The entries in the table are shares in %, means, or number of observations, by subgroup.
a) Multiple answers to this question were permitted. Hence, the means do not sum up to 1.
17
Cooperative and less cooperative caseworkers also differ in their own characteristics, which may be
related to their efficacy in counselling and placing unemployed. Table 2 shows, more of the non-
cooperative caseworkers have participated in the vocational training programme for caseworkers,
whereas more of the cooperative caseworkers have a university degree. There are also differences in
the organisation of the employment offices, with the non-cooperative caseworkers being more often
specialised towards counselling unemployed of a certain industry or occupational group. Lastly,
there are also differences in the local labour market situations the caseworkers face, e.g. a some-
what higher local unemployment rate for the non-cooperative caseworkers. We also observe clear
differences by language region. Unemployed who live in the Italian-speaking region are more often
confronted with less cooperative caseworkers compared to their counterparts in the German and
French-speaking regions. The language region will therefore be an important control variable, and
we will use interaction terms with language regions in our later regressions throughout.
4 Methodology
As pointed out in the last section, clear differences exist between the unemployed attended by a
cooperative or a less cooperative caseworker, which we need to control for. We will seek to find a
parsimonious specification that captures the most important factors without introducing too much
noise due to irrelevant variables. First, the semiparametric econometric methodology is described.
4.1 Conditional independence assumption as identification strategy
Consider an individual i who registers as unemployed at time t0 at the nearest regional employment
office. This person is then assigned to a caseworker of that office.10 The caseworker is of a particu-
lar type with respect to his willingness to cooperate with his client. Let Di denote the attitude of the

10 This may take a few weeks because the secretariat may require all relevant documents before assigning a counselling
meeting. They may also send the unemployed person first to a one-day information workshop.
18
caseworker who is counselling individual i. In most of the analyses, we will define Di as binary,
where Di = 1 represents a non- or less-cooperative caseworker whereas Di = 0 represents a coopera-
tive caseworker.
We are interested in the impact of a cooperative caseworker on the subsequent employment pros-
pects of this unemployed person, which we measure by the employment status, 0
,it
Y
τ
+, in the month
τ after registration. In particular, we would like to compare the employment status with the potential
employment status if the same unemployed person was counselled by a caseworker with a different
attitude. We base our analysis on the prototypical model of the statistical evaluation literature with a
binary treatment variable D (see Neyman, 1921, Fischer, 1935, Rubin, 1974, 1979). Let
0
,
d
it
Y
τ
+
(1)
be the potential outcome at some time τ after unemployment registration at time t0, if the case-
worker was of type d. In other words, 0
0
,it
Y
τ
+
is the employment outcome that would have been ob-
served had person i been counselled by a cooperative caseworker, whereas 0
1
,it
Y
τ
+ is the employment
outcome that would have been observed had person i been counselled by a non-cooperative case-
worker. To simplify the notation in the following we will always consider the outcomes relative to
the time of registration and treat the time of registration t0 as an additional covariate of person i. We
will therefore drop the subscripts and denote the potential outcomes simply as 0
i
Y and 1
i
Y. The av-
erage treatment effect for a person who has been counselled by a non-cooperative or by a coopera-
tive caseworker is
10
[|1]EY Y D−=
(ATET),
10
[|0]EY Y D−=
(ATEN).
19
We will often refer to these parameters as the average treatment effect on the treated (ATET) and
the average treatment effect on the non-treated (ATEN), respectively. The following discussion
focuses on the ATET, with obvious modifications for the ATEN. Identification of these treatment ef-
fects requires further assumptions.
For being able to estimate the expected potential outcomes for different values of d, we need to ob-
serve variation in Di that is exogenous with respect to the outcome variable. The observed type of
the caseworker D
i might be related to many factors that also have an impact on employment
chances, such that in general
[] [ | ]
dd
YEYDd
=. (2)
However if we were to condition on all variables X that determined the type of the caseworker and
the potential employment chances of the unemployed person, conditional on X the potential out-
comes would be identified:
[| ] [| , ]
dd
EY X x EY X xD d x χ== = = ∀∈ , (3)
where ()Supp Xχ.11 This assumption is referred to as the conditional independence assumption (CIA) in
the following. It is also called unconfoundedness in the statistical literature (e.g. Rubin, 1974). We assume
the CIA to hold for every value of x that lies in the support of X in the D=1 and the D=0 population, i.e.
(| 1) (| 0)Supp X D Supp X Dχ===. This common support restriction is discussed further be-
low.
The most crucial aspect of the identification strategy thus relies on being able to observe all con-
founding variables X. To do so, the very detailed linked caseworker-client dataset, described above,
is essential together with an understanding of the determinants of the cooperation attitude of the
caseworker. The cooperativeness Di of the caseworker depends on three processes: First, which

11 Supp(A) denotes the support of the random variable A.
20
types of caseworkers are hired, second, how caseworkers are allocated to the unemployed, and third
how their attitude develops after having been trained and gained experience on the job. Since atti-
tudes of caseworkers could be related to their general skills of finding jobs for their clients, we in-
clude caseworker characteristics such as their age, gender, education, work experience, and experi-
ence of own unemployment as covariates. We are also able to control for the allocation process of
unemployed to caseworkers since we know from the questionnaire according to which criterion
(occupation, alphabet, age, employability, and the like) allocation took place. A further aspect is
that caseworkers not only differ in their personalities, but they also react to the types of unemployed
they counsel and the labour market environment they face. If vacancies are scarce and rapid re-
employment appears difficult, caseworkers may be less demanding than in a more favourable envi-
ronment. Similarly, a caseworker who counsels mainly individuals with a low employability rating
may react differently than a caseworker responsible, e.g., mainly for youth. Therefore, we will in-
clude in the analysis also a large number of covariates on the unemployed person's employment
history, the local labour market, etc.12
4.2 Semiparametric matching estimation
The estimator used is a matching estimator as implemented in Lechner, Miquel, and Wunsch
(2006). The advantage of matching estimators is that they are essentially nonparametric and that
they allow for arbitrary individual effect heterogeneity.13 By the conditional independence assump-
tion, the average treatment effect is identified as

12 The available information is much richer than usually available in studies that rely on the conditional independence
assumption (e.g. Heckman and Smith, 1999; Brodaty, Crépon and Fougère, 2001; Larsson, 2003; Dorsett, 2006).
13 See Heckman, LaLonde, and Smith (1999), for matching with a binary treatment, and Imbens (2000), Lechner
(2001), and Gerfin and Lechner (2002) for multiple treatments. Imbens (2004) provides an excellent survey of the
recent advances in this field.
21
10 0
0
[ | 1] [ | 1] [ | 1]
[ | 1] [ [ | , 1] | 1]
[| 1] [[| , 0]| 1],
EYYD EYD EYD
EY D EEY X D D
EY D EEY X D D
−== = =
==− ==
==− ==
where the first term can be estimated by the sample mean in the D=1 population and the second
term by
0
ˆ()
ii
i
i
i
mX D
D
,
where 0
ˆ()mx
is a nonparametric estimator of [| , 0]EY X xD
=
=, e.g. a first-nearest-neighbour es-
timator. As we search for each individual of the D=1 population for the nearest neighbour in the
D=0 population, this is usually referred to as a “matching” estimator, which matches observations
from the one population to the other population. Rosenbaum and Rubin (1983) have shown that
instead of matching on the high-dimensional vector X, consistent estimates are also obtained by
matching on the one-dimensional propensity score, () Pr( 1| )
p
xDXx
=
==, or by matching on
p(X) and a subset of X that is suspected to be highly correlated with the outcome variable as well as
with D. Such combinations, which they also refer to as balancing scores, can help to ensure that a
misspecification of the functional form of the propensity score has only a minor impact. We there-
fore match on the propensity score and a number of additional covariates, where the propensity
score is given a larger weight in the Mahalanobis distance calculation. (The weight is five times as
large as for any of the additional covariates.) The small sample properties of matching estimators
have been well explored and appeared to be quite robust in different practical applications (e.g.
Larsson, 2003; Gerfin, Lechner, and Steiger, 2005). Moreover, it was subjected to several Monte
Carlo studies (e.g. Lechner, 2002) investigating small sample problems and sensitivity issues.
22
In this paper we use an extension of conventional matching estimation, similar to Lechner, Miquel,
and Wunsch (2006), which extends the first-nearest neighbour propensity score matching estimator
in several directions: First, as mentioned above, matching does not only proceed with respect to the
propensity score but also incorporates additionally some other covariates. Second, instead of using
first-nearest neighbour matching, all neighbours within a pre-specified radius are used.14 Third, the
matching quality is increased by exploiting the fact that appropriately weighted regressions that use
the sampling weights from matching have the so-called double robustness property. This property
implies that the estimator remains consistent if the matching step is based on a correctly specified
selection model or the regression model is correctly specified (e.g. Rubin, 1979; Joffe, Ten Have,
Feldman, and Kimmel, 2004). Moreover, this procedure should increase precision and may reduce
small sample as well as asymptotic bias of matching estimators, see Abadie and Imbens (2006),15
and thus increase robustness of the estimator in this dimension as well.
The motivation for radius matching is the possibility of efficiency gains without the risk of incur-
ring too much additional bias. The matching algorithm in Gerfin and Lechner (2002) used the first
nearest control observation for each treated. However, when there are other comparison observa-
tions that are similar to the matched comparison observation, there are straightforward efficiency
gains (without paying a high price in terms of additional bias) by considering these additional 'very
close' neighbours and forming an 'averaged matched comparison' observation. Of course, there are
many ways to do this in practice (and we note the similarity to the idea of kernel matching). Here,
our basic consideration is to be much more cautious with respect to additional bias that with respect
to additional variance, because the variance of the estimator is visible after the estimation, whereas
the bias generally is not. To be conservative, we consider only observations that have a distance to

14 This is thus similar to a kernel estimator with a uniform kernel function.
15 The results of Abadie and Imbens (2006) do not apply directly to propensity score matching, but since we also match
on additional variables there are some similarities with the estimators they consider.
23
'their' treated observation of no more than 90% (denoted by R in the following) of the worst match
that we had obtained by one-to-one matching (after enforcing common support; R=0 is the case of
one-to-one matching; R corresponds to a bandwidth choice in kernel weighting). To be even more
conservative, we weight the observations proportionally to their distance from the treated (corre-
sponding to a triangular kernel). The results are not very sensitive to the exact way the weighting is
implemented. When R is reduced the means change little, but the estimated variances increase
slightly.
In addition to incorporating all control observations within a certain radius, we also reduce bias by a
weighted regression. Here we note that Abadie and Imbens (2006) have shown that the usual one-
to-K matching estimators, where K is a fixed number, may exhibit an asymptotic bias, because
matches are not exact. Our weighted radius matching estimator does not necessary imply a fixed K
and is thus probably less subject to this problem. Nevertheless, we follow their proposal and im-
plement a weighted regression based bias removal procedure on top of the matching. The regression
is done in the comparison sample only. Outcomes are predicted for the attributes observed in treated
and control samples. Specifically, the outcome variable is regressed on the propensity score and the
additional variables with weights coming from the matching step (see Imbens, 2004). The differ-
ence between the mean of the predicted outcomes using the observed X of the treated and the
weighted X of the comparison observations gives an estimate of the bias (see Table B.1 for the ex-
act implementation). Without the theoretical justification given by Abadie and Imbens (2006), a
somewhat similar procedure has been used by Rubin (1979) and Lechner (2000).
We calculate standard errors as in Lechner, Miquel, and Wunsch (2006) conditional on the weights
for the comparison observations, because in Monte Carlo simulations they showed (e.g. Lechner,
2002) good performance in finite samples (their generalization to non-integer weights as used here
is trivial). A difference to their implementation is that we have to take into account that the treat-
24
ment variable is measured at the level of the caseworker such that unemployed persons counselled
by the same caseworker are unlikely to be independent observations. We account for this by com-
puting standard errors clustered at the caseworker level. Details are given in Appendix B.
The different steps of the estimator are described in Table B.1 in the appendix. In the first step, a
probit model is used to estimate the propensity score. Step 2 ensures that we estimate only effects in
the region of common support. For observations of the D=1 sample with propensity score p(x) very
close to one we would not be able to find a corresponding observation in the D=0 sample with
characteristics leading to similar values of p(x). However, it will be seen in the following section
that the common support is very large, and that the loss of observations due is negligible.
5 Analysis of the determinants of cooperativeness
5.1 Estimation of the propensity score
The first step of the empirical analysis consists in examining the determinants of cooperativeness.
This is done by regressing the cooperativeness of the caseworker on several of his own characteris-
tics and on characteristics of his clients, the local labour market, and the employment office. Since
these estimates also serve as estimates of the propensity score for the subsequent impact analysis, a
probit regression on the level of the unemployed person is used, with standard errors clustered at the
level of the caseworker.
To examine the robustness of our results, we will consider four different sets of control variables X.
For our main specification (denoted as Xset 1 with 65 regressors), we include a large number of
characteristics of the caseworker, the unemployed, the local labour market, and the allocation proc-
ess within the employment office.
Table 3 shows the estimation results for the probit model for all the regressors that are included in
the main specification (Xset 1 with 46 regressors) and indicates those variables that are part of a
25
more parsimonious specification denoted as Xset 4 (see below). As mentioned before, the dependent
variable is defined as being equal to one if the caseworker is less cooperative (option 2 or 3), and
zero otherwise (option 1).
A first observation is that many of the coefficients are insignificant (the standard errors take the
clustering at the caseworker level into account). This implies that caseworkers' attitudes and behav-
iour are more like a personal characteristic of the caseworker instead of merely being an adaptation
to the external environment. The finding that many variables are insignificant cautions against a
model with too many X variables as they might simply be adding noise to the propensity score
matching estimator. We further observe that caseworkers who face many unskilled unemployed or
who work in offices that internally specialize by occupation tend to be less cooperative, i.e. more
demanding. The latter may be because specialization by occupation will lead to a better knowledge
of the employment situation and vacancies in the particular industry. Another observation from Ta-
ble 3 is that many of the interaction terms with the language region are significant. This may be
particularly related to the language in which the written questionnaire was conducted since the
translations from German to French and Italian may not have been able to pick up all the nuances of
language. We therefore retain all these interaction terms as control variables as they are capturing
important differences between the language regions of Switzerland.
Some goodness-of-fit statistics of the probit estimates of the propensity score (Efron's R2 is 0.06)
indicate some overall descriptive power with a substantial amount of randomness remaining.
26
Table 3: Probit estimates for prime age population (age 24-55, Xset 1)
Binary dependent variable: being a less cooperative caseworker
N = 100222 coefficient std error in Xset 4
Constant -0.24 0.36
&
French speaking employment office * 1.39 0.73 &
Italian speaking employment office *** 4.75 1.28 &
Allocation of unemployed to caseworkers (reference: at random)
By industry 0.14 0.10
x French speaking region -0.06 0.20
x Italian speaking region -0.45 0.36
By occupation ** 0.24 0.10 &
x French speaking region 0.16 0.21
x Italian speaking region -0.03 0.33
By age 0.12 0.22
By employability -0.09 0.17
By region 0.06 0.13
Other -0.05 0.15
Characteristics of the caseworker
Age -0.01 0.01
&
x French speaking region * -0.02 0.01 &
x Italian speaking region *** -0.06 0.02 &
Female -0.04 0.10
&
x French speaking region 0.07 0.20
x Italian speaking region 0.02 0.35
Experience in employment office (tenure in years) 0.02 0.02 &
x French speaking region -0.03 0.03
x Italian speaking region -0.07 0.05
Own experience of unemployment -0.04 0.10
x French speaking region -0.15 0.21
x Italian speaking region 0.03 0.38
Indicator for missing caseworker characteristics -0.10 0.25
Education: above vocational training * -0.20 0.11 &
x French speaking region 0.35 0.25
x Italian speaking region -0.39 0.41
Education: tertiary track (university or polytechnic) -0.20 0.14
x French speaking region 0.32 0.27 &
x Italian speaking region -0.35 0.48
Special vocational training of caseworker 0.09 0.12 &
x French speaking region 0.29 0.36
x Italian speaking region 0.42 0.35
Table 3 to be continued.
27
Table 3 continued
Coefficient std error in Xset 4
Characteristics of the unemployed person
Female -0.04 0.03
&
x French speaking region -0.10 0.07
x Italian speaking region 0.03 0.08
Mother tongue other than German, French, Italian -0.03 0.04 &
x French speaking region * 0.10 0.06
x Italian speaking region 0.05 0.07
Qualification: unskilled ** 0.10 0.04 &
x French speaking region -0.13 0.08
x Italian speaking region -0.04 0.09
Qualification: semiskilled 0.04 0.05 &
x French speaking region 0.00 0.08
x Italian speaking region -0.07 0.17
Qualification: skilled without degree 0.02 0.05 &
x French speaking region ** 0.19 0.09
x Italian speaking region * -0.28 0.17
Number of unemployment spells in last two years 0.01 0.01
x French speaking region -0.01 0.02
x Italian speaking region ** 0.05 0.02
Fraction of time employed in last years 0.00 0.03
x French speaking region ** -0.13 0.06
x Italian speaking region 0.03 0.09
Employability low 0.02 0.11
x French speaking region 0.15 0.17
x Italian speaking region 0.15 0.20
Employability medium 0.00 0.10
x French speaking region 0.02 0.14
x Italian speaking region 0.04 0.19
Local labour market characteristics
Unemployment rate in canton 0.06 0.06
x French speaking region -0.18 0.12
x Italian speaking region ** -0.27 0.14
Note: Standard errors are clustered at the caseworker level. Most variables are interacted with French and Italian language
region. (German is the reference language region.) The last column indicates variables included in Xset 4 with &.
A crucial aspect of our identification strategy is the conditional independence assumption and thus
the selection of the set of control variables. In the main specification, Xset 1, we included a large
number of caseworker and jobseeker characteristics as well as some indicators of the local labour
market and of the employment office, which we deemed important after several interviews with
caseworkers and employment office managers. A concern might be that Xset 1 contained too few
covariates to make the conditional independence assumption plausible. Furthermore, additional
variables that are related to the outcome variable could increase precision. On the other hand, in-
28
cluding too many variables also runs the risk of including endogenous control variables (i.e. those
already been affected by the treatment variable) and/or reducing the common support region. They
may also introduce more noise into the estimation of the propensity score.
In Xset 2 (= 232 regressors) we added a large number of additional covariates. The additional vari-
ables of the unemployed person are age, civil status, children, and earnings in the last job. Further-
more, there are three dummies for education, three dummies for foreign language knowledge, and
two dummies for the types of foreigners' work permit. To approximate the unemployed person's
labour market history, it contains variables capturing the duration of unemployment in the last two
years, the average wage in last ten years, the total number of employment spells in last ten years,
the number of employment spells in last five years, an indicator of having been out of labour force
in last five years, the fraction of time being employed and unemployed in last ten years, and a
dummy for having a zero contribution time to the unemployment insurance. Furthermore, it con-
tains 16 occupation dummies, six industry dummies, and a dummy for looking for a part-time job.
With regard to local labour market characteristics, additional variables are municipality size, and
the cantonal unemployment rate. All these variables are interacted with the French and Italian lan-
guage regions. The pension data also indicate the first month of contribution (since 1990), if ever
contributed. We also include this variable together with interaction terms with being young/old, and
foreigner/Swiss. These interaction terms roughly pick up in which year a foreigner migrated to
Switzerland.
Most of these variables turned out to be insignificant in the estimation of the propensity score. De-
spite of being insignificant, they still affect the calculation of the propensity score and can thus in-
troduce a lot of noise into the matching estimator. By sequentially deleting insignificant variables in
the probit model, we generated another Xset 3 (= 94 regressors), in an attempt to reduce noise due
to insignificant variables. In a general to specific approach, we eliminated covariates whose F-test
29
did not suggest any explanatory power at the 5% level. (One should note that this sequential statisti-
cal variable selection is subject to the pre-test problem such that the size of these repeated tests is
not exactly 5%. It should rather be considered as an algorithm for selecting the probably most im-
portant variables.) However, we retained all variables of Xset 1 here, even if insignificant. Hence,
Xset 1 is a strict subset of Xset 3. Further eliminating sequentially all variables with insignificant F-
test leads us to Xset 4 with 46 regressors, which is our most parsimonious specification.
Table 4 shows various goodness-of-fit statistics of the probit regression for these different sets of X
variables. The two parsimonious sets Xset 1 (obtained by deliberate choice) and Xset 4 (obtained by
statistical variable choice) appear to be hardly worse than the two most complex specifications.
Comparing Xset 2 with Xset 4, adding almost 200 regressors increases Efron's R2 by less than 2
percentage points and reduces the number of wrong predictions by less than 800 from more than
40,000. The additional variables thus mainly introduce noise.
Table 4: Goodness of fit measures for different Xsets for prime age population (age 24-55)
Regressors Number of covariates Log-Likelihood Efron's R2 NWP SSR WSSR
Xset 1 65 -66303 0.058 41496 23560.19 100167.38
Xset 2 232 -65408 0.074 40097 23167.07 100098.15
Xset 3 94 -65807 0.067 41031 23346.20 100183.13
Xset 4 46 -66478 0.056 40832 23626.93 100745.01
Note: The number of observations for each Xset is 100222. Efron's R2 (Efron, 1978) is a measure for residual variation. NWP is
the number of wrong predictions, SSR is the sum of squared residuals, and WSSR is the sum of weighted squared residu-
als.
5.2 Common support
The nonparametric identification strategy relies on estimating the expected counterfactual outcome
0
[|]
E
YX
for every D=1 observation. This is possible only if for every value of X that is observed
in the D=1 population also at least one individual with very similar values of X can be found in the
D=0 population. For values of X where Pr( 1| )DX
=
is equal to one this is impossible by defini-
30
tion, and it is very difficult do find comparison observations if Pr( 1| )DX
=
is very large. Figure 3
shows histograms of the estimated propensity scores for the D=1 and D=0 subsamples.
Figure 3: Distribution and common support of the propensity score
Note: Left graph histogram of the estimated propensity scores in the D=0 sample. Right graph histogram of the estimated propensity
scores in the D=1 sample. Propensity score with Xset 1.
Partly due to availability of a very large sample, the region of common support appears to be very
large as well: Even very large values of the propensity score are observed in the D=0 sample and
also very small values are observed in the D=1 sample. Observations that appear to be outside of
the common support are deleted for the matching estimator. For estimating ATET, all D=1 observa-
tions with a propensity score larger than the largest propensity score among all D=0 observations
are deleted. For estimating ATEN, all D=0 observations with a propensity score smaller than the
smallest propensity score among all D=1 observations are deleted. This leads to a loss of 312
treated observations (= 0.003%) for estimating ATET in our main specification with Xset 1. When
estimating ATEN we lose 57 control observations (= 0.0006 %).
5.3 Matching quality
An advantage of matching compared to conventional regression is that one can model the propen-
sity score before examining the outcome variable Y. Hence, the propensity score model can be re-
specified until a reasonable fit is obtained without having the researchers' decisions being affected
by the resulting estimates of the treatment effects. In this sense, this approach is immune to the re-
31
specification and pre-testing problem of conventional regression. In addition, the region of common
support of the X regressors can also be examined before the outcome variable Y is incorporated. As
suggested by Rosenbaum and Rubin (1983), matching on the propensity score leads to a balancing
of the covariates X in the D=1 and D=0 population. Hence, after matching treated and control, the
joint distribution of X and thereby the marginal distributions of X should be identical in both
matched subsamples. Thus, a simple way to validate the specification is to test for equality of
means in the two subsamples. However, since we use a more complex estimation procedure than
simple matching involving radius matching and regression, a more appropriate balancing test is to
the estimate the effect of the treatment on the covariates. If the model is correctly specified, this
effect must be zero asymptotically and should not be significantly different from zero in finite sam-
ples.
For both groups, Table 5 shows the estimated means, their difference, and t-stats for the hypothesis
that their means are equal. We find that the matching quality with respect to almost all covariates is
very good. The only exception is the number of unemployment spells in the past two years: the
weighting procedure over-adjusts for this variable by raising the mean in the comparison group
from 0.56 to 0.68. In one of our later specifications, we therefore include this variable as an addi-
tional covariate on which matching is conducted.
32
Table 5: Matching quality Xset 1, prime age population (age 24-55)
Control variables used in propensity score
Predicted
mean for
control
Mean of
treated Predicted mean differ-
ence (treated-control)
t-value for test
that difference
is zero
Observations (after imposition of common support) 51923 47987
French speaking employment office 0.21 0.21 0.00 0.00
Italian speaking employment office 0.09 0.09 0.00 0.00
Allocation of unemployed to caseworkers (reference: at random)
By industry 0.57 0.57 0.00 -0.08
x French speaking region 0.09 0.08 0.00 0.02
x Italian speaking region 0.04 0.04 0.00 0.01
By occupation 0.60 0.61 0.01 0.36
x French speaking region 0.16 0.16 0.00 0.03
x Italian speaking region 0.06 0.06 0.00 0.01
By age 0.03 0.03 0.00 0.12
By employability 0.06 0.06 0.00 0.23
By region 0.11 0.12 0.01 0.44
Other 0.08 0.07 0.00 -0.17
Characteristics of the caseworker
Age 43.46 43.63 0.17 0.22
x French speaking region 9.25 9.30 0.05 0.04
x Italian speaking region 3.71 3.75 0.04 0.05
Female 0.42 0.41 -0.01 -0.37
x French speaking region 0.08 0.08 0.00 0.00
x Italian speaking region 0.04 0.03 -0.01 -0.88
Tenure in employment office (in years) 5.98 5.91 -0.07 -0.29
x French speaking region 1.35 1.34 -0.02 -0.09
x Italian speaking region 0.67 0.65 -0.02 -0.16
Own experience of unemployment 0.61 0.61 0.00 0.01
x French speaking region 0.14 0.14 0.00 0.05
x Italian speaking region 0.05 0.05 0.01 0.65
Indicator for missing caseworker characteristics 0.05 0.04 0.00 -0.06
Education: above vocational training 0.43 0.41 -0.02 -0.52
x French speaking region 0.09 0.09 0.00 0.02
x Italian speaking region 0.04 0.03 -0.01 0.01
Education: tertiary track (university or polytechnic) 0.23 0.24 0.01 0.44
x French speaking region 0.08 0.09 0.00 0.02
x Italian speaking region 0.01 0.02 0.01 0.01
Special vocational training of caseworker 0.26 0.25 -0.01 -0.30
x French speaking region 0.01 0.02 0.00 0.01
x Italian speaking region 0.06 0.05 -0.01 0.01
Table 5 to be continued.
33
Table 5 continued
Predicted
mean for
control
Mean of
treated Predicted mean differ-
ence (treated-control)
t-value for test
that difference
is zero
Characteristics of the unemployed person
Female 0.44 0.43 0.00 -0.25
x French speaking region 0.09 0.09 0.00 -0.03
x Italian speaking region 0.04 0.04 0.00 0.36
Mother tongue other than German, French, Italian 0.34 0.32 -0.02 -1.17
x French speaking region 0.07 0.07 0.07 0.07
x Italian speaking region 0.03 0.03 0.03 0.03
Qualification: unskilled 0.26 0.26 0.26 0.26
x French speaking region 0.05 0.05 0.05 0.05
x Italian speaking region 0.03 0.03 0.03 0.03
Qualification: semiskilled 0.16 0.16 0.16 0.16
x French speaking region 0.04 0.04 0.04 0.04
x Italian speaking region 0.01 0.01 0.01 0.01
Qualification: skilled without degree 0.05 0.05 0.05 0.05
x French speaking region 0.02 0.02 0.02 0.02
x Italian speaking region 0.01 0.01 0.01 0.01
Number of unemployment spells in last two years 0.68 0.59 -0.09 -2.34
x French speaking region 0.16 0.16 0.16 0.16
x Italian speaking region 0.09 0.09 0.09 0.09
Fraction of time employed in last years 0.78 0.78 0.78 0.78
x French speaking region 0.16 0.16 0.16 0.16
x Italian speaking region 0.07 0.07 0.07 0.07
Employability low 0.16 0.16 0.16 0.16
x French speaking region 0.02 0.02 0.02 0.02
x Italian speaking region 0.02 0.02 0.02 0.02
Employability medium 0.72 0.72 0.72 0.72
x French speaking region 0.16 0.16 0.16 0.16
x Italian speaking region 0.05 0.05 0.05 0.05
Local labour market characteristics
Unemployment rate in canton 3.76 3.75 0.00 -0.06
x French speaking region 0.88 0.89 0.00 0.02
x Italian speaking region 0.38 0.38 0.00 -0.05
Note: Matching quality for estimation of ATET based on full estimator and after imposition of common support.
6 Estimated treatment effects
Section 6.1 gives the main empirical results of the propensity score matching. To examine the ro-
bustness of our results in Section 6.2, we consider four different sets of control variables X, differ-
ent outcome variables, as well as different definitions of the treatment variable D. For each of these
34
different combinations, we re-estimated the propensity score and applied the common support re-
strictions.
6.1 Impact of a less cooperative caseworker
The following figures show the matching estimates when the propensity score is estimated with
Xset 1 and treatment is defined as cooperation not very important versus very important. That is, D
is defined as one if the caseworker selected option 2 or option 3, and is defined as zero if the case-
worker selected option 1 (see Table 1). In this specification, we isolate those caseworkers who place
very much emphasis on cooperation versus all the others.
Figure 4: Impact of having a less cooperative caseworker on employment in %-points
Note: Average treatment effect on the treated (ATET) on employment, in percentage points. Prime age unemployed (24 to 55
years). Abscissa: Month after registration of unemployment. Ordinate: Treatment effect on employment in %-points. Dots
(triangles) indicate significance at the 5% (10%) level. The dashed line represents the pointwise 95% confidence interval.
Figure 4 gives the estimates for the outcome variable employment for the subsequent 36 months
after registration. (Triangles indicate point-wise significance at the 10% level, dots at the 5% level.)
These results indicate that having a less cooperative caseworker increases the employment probabil-
ity by about 2%-points. The effect sets in about five months after registration and is relatively stable
until month 36. From month 24 onwards, it is significant only at the 10% level, though.
35
An increase in employment probability by 2%-points is a non-negligible effect, given that it only
requires a change in the caseworkers' attitudes and behaviour towards their clients. At the same
time, it is also likely to lead to additional cost savings for the unemployment insurance system since
more demanding caseworkers may also often impose more sanctions in the form of suspension of
benefits. Unfortunately, reliable data about potential costs of such a policy shift are not available.
Figure 5: Impact of having a less cooperative caseworker on stable employment
Figure 5.1: Six months stability
Month after registration
Figure 5.2: Twelve months stability
Month after registration
Note: Average treatment effect on the treated (ATET) on stable employment, in %-points. In left graph, employment is only con-
sidered as stable if the employment spell has a duration of at least 6 months. In right graph, a duration of at least 12
months is required. Prime age unemployed (24 to 55 years). Abscissa: Month after registration of unemployment. Ordinate:
Treatment effect on employment. Dots (triangles) indicate significance at the 5% (10%) level. Dashed line represents
pointwise 95% confidence interval.
We also examined the average treatment effects separately for four subgroups: qualified, unquali-
fied, older than 55 years, and younger than 24 years. Most of these results, however, turned out to
be insignificant, mainly due to the smaller sample size. (This was already expected from Figure 4
where statistical precision was at the margin of detecting a significant effect.) However, estimates
from other specifications (available on request from the authors) suggest that the qualified, the un-
qualified, and the older unemployed tend to benefit from having a less cooperative caseworker,
while the effects for the young are always insignificant.
A general concern with tougher caseworkers is that they might push jobseekers into precarious or
unstable jobs, which due to the poor match quality might lead to higher separation rates soon after.
36
To examine the stability of jobs we define an individual as being in stable employment in a given
month if the employment spell is of at least six months duration. Alternatively, we require at least
12 months duration. Figure 5 gives the treatment effects on stable employment (analogously to Fig-
ure 4), which are positive throughout and only slightly smaller than in Figure 4. Although the statis-
tical precision is insufficient to draw very strong conclusions, it does not appear that non-
cooperativeness of the caseworkers would lead to unstable jobs.
6.2 Sensitivity analysis
The previous section showed a positive impact of non-cooperativeness of the caseworker on em-
ployment outcomes. Albeit being statistically significantly different from zero, the confidence
bands were nevertheless rather wide. In this section, we examine alternative specifications, which
all point towards a positive impact of non-cooperativeness, although with different degrees of statis-
tical confidence.
Table 6 shows the estimation results for various alternative specifications. For conciseness, we do
not show the entire graphs but display the effects only for 6, 12, 18, and 30 months after registra-
tion. The first three rows refer to alternative definitions of employment and correspond to Figures 4
and 5. They show that the positive impact does not seem to have been arisen at the cost of instabil-
ity of jobs. The magnitude of the effects tends to decrease when we look at stable employment but
these differences are not significant. The following rows show the results for various subgroups,
where the results are mostly positive but very imprecise throughout.
The subsequent rows examine alternative specifications of the estimator and of the regressor set.
First, we reduce the radius from 0.9 to 0.1 in the propensity score matching estimator. This is simi-
lar to a reduction of the number of neighbours in k-nearest neighbour matching. Coefficients and
standard errors are not much affected.
37
As an alternative specification, we exclude the regressor 'employability rating of the unemployed',
which might potentially be endogenous as it is a subjective assessment made by the caseworker. For
instance, more demanding caseworkers might consider the same type of client as easier to place
than the more lenient caseworkers might. The results do not change substantially. Second, we add
the number of clients that caseworkers report to counsel on average to Xset 1, because one may
want to control for caseworker's workload if more (or less) cooperative caseworkers are more suc-
cessful in placing clients and hence more clients are assigned to them, which might negatively af-
fect their efficiency. Again, the results are hardly affected. Third, we include a dummy for register-
ing as unemployed in the second semester of 2003, since rules for benefit entitlement were tight-
ened in July 2003. Although it has no significant impact on cooperation behaviour, it reduces stan-
dard errors of employment effects by increasing estimation efficiency. Finally, we include the num-
ber of unemployment spells in the past two years as an additional covariate on which to match ex-
actly, because the matching quality was imperfect in Table 5 with respect to this variable. The esti-
mated effects increase for month 6, but slightly decrease for the later months and become less sig-
nificant.
As discussed above, the choice of control variables is an important aspect for the credibility of the
conditional independence assumption. Table 6 shows estimates of propensity score matching with
Xsets 2, 3, and 4, respectively, which all show positive effects but with different degrees of preci-
sion. It seems that the two very large regressor sets 2 and 3, with many insignificant variables, lead
to noisy estimates, whereas the results with Xset 4, where the insignificant variables have been
purged, are much more precise. Hence, including too many control variables that are not related to
the treatment variable can introduce substantial noise into propensity score matching.
38
As a further robustness check, we examined the logit estimator as a parametric alternative to pro-
pensity score matching. The effects remain positive but smaller compared to the matching esti-
mates. This could be due to the functional form assumption imposed.
Finally, we estimate the ATEN to complement the estimates of the ATET. The effects remain posi-
tive but much smaller and clearly insignificant. A possible interpretation of this finding is that those
caseworkers who decided or happened to be more demanding were right in doing so, whereas the
gains from being more demanding are smaller or even zero for those caseworkers who decided
against this strategy. This is what we would expect if caseworkers adapt to their environment. This
also explains the previously mentioned small effects of the parametric logit estimator because the
logit model permits only very limited effect heterogeneity and, more or less, measures some kind of
average between ATET and ATEN.
In some further analysis, we also consider alternative definitions of the treatment variable. First, we
discard caseworkers of the intermediate type and only consider an attitude as being less cooperative
if the caseworker has explicitly chosen option 3, see Table 1. (In other words, we eliminate all
caseworkers who chose option 2.) In our main specification, estimates are still positive, but less
precise due to the smaller sample size. When using the large Xset 2, estimates remain positive but
less precise. The estimates of the parametric logit model are now similar to the matching estimates.
Hence, all these estimates remain positive, but tend to decrease over time. The results for ATEN are
partly negative, on the other hand, although not statistically significantly so. We therefore restrict
ourselves to interpreting the ATET estimates as overall positive, whereas we cannot say much about
ATEN.
39
Table 6: The impact of non-cooperativeness on employment, robustness analysis
Effect at month after registration
Obs. Month 6 Month 12 Month 18 Month 30
Coeff. t-Stat Coeff. t-Stat Coeff. t-Stat Coeff t-Stat
Non-cooperativeness (option 2 and 3) versus Cooperativeness (option 1)
Alternative definitions of the outcome variables (Pscore matching with Xset 1, age 24-55)
Employment 100222 0.021 2.360 0.018 1.955 0.019 1.917 0.017 1.725
Six-months stable employment 100222 0.020 2.316 0.018 1.908 0.019 1.959 0.017 1.785
Twelve-months stable employment 100222 0.012 1.614 0.017 1.825 0.016 1.656 0.015 1.532
Effects on employment for subgroups (Pscore matching with Xset 1, age 24-55)
Qualified, age 24-55 61191 0.007 0.669 0.008 0.710 0.011 0.994 0.007 0.637
Unqualified, age 24-55 39031 0.017 1.384 0.007 0.552 0.003 0.229 0.000 -0.033
Older than 55 years 8580 -0.010 -0.626 0.012 0.690 -0.002 -0.115 0.009 0.437
Younger than 24 years 28980 -0.006 -0.408 0.005 0.310 0.016 1.091 0.009 0.615
Alternative specifications (age 24-55)
PSM with radius 0.1 (Xset 1) 100222 0.022 2.333 0.021 2.128 0.020 1.978 0.018 1.819
PSM without employability (Xset 1) 100222 0.018 2.047 0.020 2.247 0.016 1.658 0.016 1.614
PSM with number of clients (Xset 1) 100222 0.020 2.319 0.018 1.944 0.022 2.271 0.014 1.475
PSM with dummy for 2nd semester
(Xset 1) 100222 0.023 2.557 0.021 2.224 0.025 2.466 0.022 2.280
PSM exact on number of unemploy-
ment spells (Xset1) 100222 0.023 2.611 0.016 1.784 0.018 1.922 0.014 1.514
Pscore matching (Xset 2) 100222 0.011 1.229 0.001 0.151 0.003 0.325 0.001 0.059
Pscore matching (Xset 3) 100222 0.010 1.137 0.001 0.083 0.004 0.368 -0.001 -0.056
Pscore matching (Xset 4) 100222 0.032 3.474 0.023 2.508 0.026 2.696 0.020 1.996
Logit estimates (Xset 1) 100222 0.008 1.251 0.005 1.098 0.002 0.388 -0.002 -0.322
ATEN using PSM (Xset 1) 100222 0.002 0.028 0.002 0.209 0.008 0.813 0.016 1.680
Non-cooperativeness (option 3) versus Cooperativeness (option 1), eliminating caseworkers with option 2
Pscore matching (Xset 1, age 24-55) 60912 0.036 1.852 0.001 0.044 0.010 0.507 0.011 0.530
Qualified Age 24-55 37346 0.028 1.226 0.015 0.670 0.024 1.058 0.014 0.623
Unqualified Age 24-55 23566 0.056 2.171 0.003 0.124 0.021 0.755 0.011 0.397
Old (> 55 years) 5201 0.003 0.096 0.001 0.042 0.016 0.413 0.008 0.195
Young (< 24 years) 18059 -0.017 -0.670 0.009 0.342 -0.003 -0.114 -0.007 -0.270
Pscore matching (Xset 2, age 24-55) 60912 0.032 1.534 0.004 0.232 0.009 0.438 0.011 0.530
Logit estimates (Xset 1, age 24-55) 60912 0.028 2.583 0.008 1.182 0.012 1.290 0.010 1.010
ATEN using PSM (Xset 1, age 24-55) 60912 -0.038 -1.529 -0.012 -0.609 0.003 0.129 0.004 0.197
Intermediate cooperativeness (option 2) versus no or full cooperativeness (option 1 or 3)
Pscore matching (Xset 1, age 24-55) 100222 0.010 1.159 0.009 0.971 0.009 0.873 0.004 0.442
Note: Standard errors are clustered at the caseworker level. Pscore matching and PSM are abbreviations for propensity score
matching.
In the last row of Table 6, we tested for possible non-linearities in the treatment variable. One could
imagine that a caseworker with intermediate cooperation behaviour might perform better compared
to a caseworker with very low or high cooperativeness. Therefore, as a final check we define D=1 if
the caseworker chose option 2 and D=0 if the caseworker chose option 1 or 3. The estimates are
close to zero and insignificant, thereby not confirming this hypothesis.
40
7 Conclusions
In most countries, caseworkers have substantial autonomy in the extent to which they cooperate
with their clients. Some place more emphasis on counselling, whereas others also consider monitor-
ing of job search as a very important part of their work. Using large and informative administrative
data on unemployed persons merged with data on caseworkers and their employment offices, ob-
tained from a detailed questionnaire, we investigate which attitude towards unemployed is more
successful for their subsequent employment chances. These data allow us to control for potential
selection bias by semiparametric matching estimators and to account for treatment effect heteroge-
neity. Estimates are obtained up to the first three years after unemployment registration.
More than half of the caseworkers responded that they considered cooperation with unemployed as
very important and that the wishes of the unemployed are of key importance for their decisions.
However, the estimates suggest that the employment probabilities of those unemployed persons
who were counselled by less cooperative caseworkers were higher because of their less cooperative
attitude. Such unemployed persons had about 2 percentage point higher employment probabilities
during the first three years after registration than similar unemployed persons who were counselled
by (somewhat) less demanding caseworkers. The most plausible explanation for our finding is that
caseworkers indeed influence their clients' behaviour to search for jobs and accept job offers. In an
extensive sensitivity analysis, almost all results confirmed the sign of the effect, but in several
cases, the effect was insignificant.
41
A Data Appendix
The population for the microeconometric analysis are all individuals who registered as unemployed
anytime during the year 2003 at one of the 103 employment offices under study. In total 239004
persons registered as new jobseekers during the year 2003. Notice that we consider only the first
registration in 2003 for each person and subsume any further registrations within the outcome vari-
ables, i.e. the analysis is person based and not spell based.
We restrict our analysis to the 103 regional employment offices that were independently operating
agencies responsible for a specific geographic area.16 We do not include the canton Geneva in our
study since in this canton the employment offices are functionally specialized according to profes-
sions and employability of the jobseekers, which is in striking contrast to all other cantons, which
largely follow a geographic structuring. We further exclude five employment offices from the
analysis: three offices that were newly established, split or re-organized during the year 2003, one
employment office which specialized on the difficult cases in Solothurn, and the tiny employment
office in Appenzell-Innerrhoden, which did not participate in the survey.
After excluding those offices, 219540 persons remain who registered in one of the 103 offices. For
215251 persons the first caseworker was well defined, whereas for the other 4289 no caseworker
was (yet) assigned. The reason for this is that it may often take several weeks until the first counsel-
ling meeting with a caseworker takes place, e.g. after having participated in a one-day course that
explains the duties and rights of an unemployed person. In total, 1891 different caseworkers were
identified in the data.

16 These employment offices had their own staff, a chief officer, and some flexibility in implementing the federal and
cantonal policies. Some employment offices operate a number of smaller branches e.g. in remote areas, or separate
between short- and longer-term unemployed. These employment offices usually swap staff between these branches
and pursue a common strategy. Thus, we consider them as a single entity.
42
We exclude foreigners without yearly or permanent work permit, as they are not fully entitled to all
services of the employment services. We also exclude individuals on disability or applying for it,
and for the main analyses restrict the sample to the prime-age population. Finally, we lose about
25% of observations whose caseworker had either not responded to the questionnaire in general or
to the cooperativeness question in particular. Comparing the samples before and after dropping
these observations, we do not find any differences neither in their characteristics, X, nor in their
observed outcomes, Y.
Table A.1: Sample selection criteria for empirical analysis
Number of individuals
deleted remaining
Population: all new jobseekers during the year 2003 239,004
Exclude Geneva and five other employment offices -19'464 219,540
Exclude jobseekers not (yet) assigned to a caseworker -4'289 215,251
Exclude foreigners without yearly or permanent work permit -5'399 209,852
Exclude jobseekers without unemployment benefit claim -18'434 191,418
Exclude jobseekers who applied for or claim disability insurance -3'163 188,255
Restrict to prime-age population (24 to 55 years old) -51'649 136,606
Exclude unemployed whose caseworker did not respond to the questionnaire -31'469 105,137
Exclude unemployed whose caseworker did not respond to the cooperativeness question -4'915 100,222
43
B Further details on the estimator
B.1 Matching protocol
Table B.1: A matching protocol for the estimation of ATET
Step 1 Estimate a probit model to obtain the choice probabilities: ˆPr( 1| )
ii
p
DXX
=
==
Step 2 Restrict sample to common support: Delete all D=1 observations with ˆi
p
larger than the largest estimated pro-
pensity score among the D=0 observations.
Step 3 Estimate the counterfactual expectation of the outcome variable 0
[| 1]EY D
=
Standard propensity score matching step (binary treatment)
a-1) Choose one observation from the D = 1 subsample and delete it from that pool.
b-1) Find an observation from the D = 0 subsample that is as close as possible to the one chosen in step a-1) in
terms of ˆ(),Px x
⎡⎤
⎣⎦
, with respect to the Mahalanobis distance. Do not remove that observation, so that
it can be used again.
c-1) Repeat a-1) and b-1) until no participant in D = 1 is left.
Exploit thick support of X to increase efficiency (radius matching step)
d-1) Compute the maximum distance (δ) obtained for any comparison between treated and matched comparison
observations.
a-2) Repeat a-1).
b-2) Repeat b-1). If possible, find other observations of the D = 0 subsample that are at least as close as R x
δ
to the one chosen in step a-2); R is fixed to 90% in this application but different values are examined in
the sensitivity analysis. Do not remove these observations, so that they can be used again. Compute
weights for all chosen comparisons observations such that these weights are proportional to their dis-
tance (calculated in b-1). Normalise the weights such that they add to one.
c-2) Repeat a-2) and b-2) until no participant in D = 1 is left.
d-2) For every D=0 observation, add the weights obtained in b-2).
Exploit double robustness property to adjust small mismatches by regression
e) Using the weights ()
i
wx obtained in d-2), run a weighted linear regression of the outcome variable on the
variables used to define the distance (and an intercept).
f-1) Predict the potential outcome 0()
i
yx of every observation in D = 0 and D = 1 using the coefficients of this
regression: 0
ˆ()
i
yx.
f-2) Estimate the bias of the matching estimator for 0
[1]EY D
=
as:
00
111
1ˆˆ
1( 1) ( ) 1( 0) ( ) ( )
NN
ii iii
ii
Dyx Dwxyx
N==
=−=
∑∑ .
g) Using the weights obtained by weighted matching in d-2), compute a weighted mean of the outcome variables
in D = 0. Subtract the bias from this estimate.
Final estimate
h) Compute the treatment effect by subtracting the weighted mean of the outcomes in the comparison group (D =
0) from the mean in the treatment group (D = 1).
Note: The table refers to the estimation of ATET. The modifications for ATEN are obvious.
x
includes the two dummy variables
French speaking and Italian speaking employment office.
x
is included to ensure a high match quality with respect to
these critical variables.
44
B.2 Standard errors for clustered matching
Lechner (2001) suggested an estimator of the asymptotic standard errors for ATET conditional on
the estimated weights. Since the treatment variable {0,1}D
is measured at the level of the case-
worker but the outcome variable is measured at the level of the jobseeker, for the computation of
the standard errors we have to take into account that the outcomes across the jobseekers counselled
by the same caseworker may be correlated. The calculation of the clustered standard errors is de-
scribed in the following.
The matching estimator of the potential outcome has the general form:
1
ˆ1( )
N
ll
iii
i
YDlw
y
=
==
where i =1,…,N indexes the unemployed persons and where the sum of the weights is one:
11( ) 1
Nl
ii
iDlw
=
=
=
.
To introduce the cluster structure we can re-write the matching estimator using a double sum
11
ˆ1( )1( )
JN
ll
iiii
ji
YDlCjwy
==
===
∑∑ ,
where i indexes the unemployed persons and j =1,…,J indexes the J caseworkers. The variable
{1, . ., }
i
CJ indicates the caseworker who is in charge of the unemployed i. The number of clients
of caseworker j in the D=l group and weighted by l
i
w is thus given by
1
:1( )1( )
N
j
l
iii
i
NDlCjw
=
===
.
We can compute the variance allowing that the outcomes across unemployed persons counselled by
the same caseworker are dependent, but assume that observations across caseworkers are independ-
ent:
45
11
2
11
ˆ
() 1( )1( )
11( )1( )
JN
ll
iiii
ji
JN
jl
iiii
j
ji
Var Y Var D l C j w y
NVar D l C jwy
N
==
==
⎡⎤
===
⎢⎥
⎣⎦
===
∑∑
∑∑
Hence, the variance is obtained by summing over the caseworkers the variance of the expression Aj
which is defined as
1
11( )1( )
Nl
j
iiii
ji
A
DlC
j
w
y
N=
===
.
Since the Aj are independent across the caseworkers, we can estimate
(
)
j
Var A as
m
()
2
11
11
JJ
jj
jj
Var A A A
JJ
==
=−
∑∑
,
which we now plug into the formula for ˆ
()
l
Var Y .
In the implementation, we ignore the regression step in the matching estimator. The justification is
given by Abadie and Imbens (2006) showing that a nonparametric regression step after the match-
ing does remove the bias in the asymptotic distribution without affecting its variance. Although our
estimator differs in some respects from the fixed-number-of-neighbours estimator they consider, the
general set-up is very similar. It may be conjectured that since we use a parametric instead of a non-
parametric regression the variance is indeed reduced that would lead to conservative inference.
References
Abadie, A., and G. W. Imbens (2006): "Large Sample Properties of Matching Estimators for Average Treatment Ef-
fects", Econometrica, 74, 235-267.
Abbring, J., G. van den Berg, and J. van Ours (2005): "The Effects of Unemployment Insurance Sanctions on the Tran-
sition Rate from Unemployment to Employment", Economic Journal, 115, 602-630.
46
Ashenfelter, O., D. Ashmore, and O. Deschenes (2000): "Do Unemployment Insurance Recipients Actively Seek
Work? Evidence From Randomized Trials in Four U.S. States", Journal of Econometrics, 125, 53-75.
Black D. A., and J. A. Smith (2004): "How Robust is the Evidence on the Effects of College Quality? Evidence from
Matching", Journal of Econometrics, 121, 99-124.
Black D. A., J. A. Smith, M. Berger, and B.J. Noel (2003): "Is the threat of reemployment services more effective than
the services themselves? Evidence from random assignment in the UI system," American Economic Review, 93,
1313-1327.
Bloom, H., C. Hill, and J. Riccio (2003): "Linking program implementation and effectiveness: lessons from a pooled
sample of welfare-to-work experiments", Journal of Policy Analysis and Management, 22, 551-575.
Brodaty T., B. Crépon, and D. Fougère (2001): "Using Kernel Matching Estimators to Evaluate Alternative Youth Em-
ployment Programs: Evidence from France, 1986-1988", in M. Lechner and F. Pfeiffer (eds.): Econometric Evalua-
tions of Labour Market Policies, Heidelberg: Physica Verlag, 85-124.
Dolton, P., and D. O'Neill (1996): "Unemployment duration and the restart effect: some experimental evidence", Eco-
nomic Journal, 106, 387-400.
Dorsett, R. (2006): "The New Deal for Young People: Effect on the Labour Market Status of Young Men", Labour
Economics, 13, 405-422.
Efron, B. (1978): "Regression and ANOVA with Zero-One Data: Measures of Residual Variation", Journal of the
American Statistical Association, 73, 113-212.
Fehr, E., and K. Schmidt (2001): "Theories of Fairness and Reciprocity - Evidence and Economic Applications", Work-
ing Paper, Institute of Empirical Research in Economics, University of Zurich.
Ferro-Luzzi, G., Y. Flückiger, J. Ramirez, and A. Vassiliev (2005): "Swiss unemployment policy: An evaluation of the
public employment service", Swiss Journal of Sociology, 30, 319-337.
Fisher, R. A. (1935), Design of Experiments, Oliver and Boyd: Edinburgh.
Fredriksson, P., and B. Holmlund (2003): "Improving incentives in unemployment insurance: A review of recent re-
search", Discussion Paper, IFAU.
Frölich, M., M. Lechner, S. Behncke, S. Hammer, N. Schmidt, S. Menegale, A. Lehmann, and R. Iten (2007), Einfluss
der Rav auf die Wiedereingliederung von Stellensuchenden, Schweizerisches Staatssekretariat für Wirtschaft (seco),
47
SECO Publikation, Arbeitsmarktpolitik No 20,
http://www.seco.admin.ch/dokumentation/publikation/00008/02015/index.html?lang=de.
Gerfin M., and M. Lechner (2002): "Microeconometric Evaluation of the Active Labour Market Policy in Switzerland",
Economic Journal, 112, 854-893.
Gerfin, M., M. Lechner, and H. Steiger (2005): "Does subsidised temporary employment get the unemployed back to
work? An econometric analysis of two different schemes", Labour Economics, 12, 807-835.
Gorter, C., and G. R. J. Kalb (1996): "Estimating the effect of counselling and monitoring the unemployed using a job
search model", Journal of Human Resources, 31, 590-610.
Graversen, B., and J. van Ours (2006): "How to help unemployed find jobs quickly; experimental evidence from a man-
datory activation program, Discussion Paper, Center, Tilburg University.
Ham J., and R. LaLonde (2005): "Special Issue on Experimental and Non-Experimental Evaluation of Economic Policy
and Models (Introduction)", Journal of Econometrics, 125, 1-13.
Heckman J., H. Ichimura, J. Smith, and P. Todd (1998): "Characterizing Selection Bias Using Experimental Data",
Econometrica, 66, 1017-1098.
Heckman, J., R. LaLonde, and J. Smith (1999): "The Economics and Econometrics of Active Labor Market programs,"
in O. Ashenfelter and D. Card (eds.), Handbook of Labour Economics, Vol. 3, 1865-2097.
Heckman, J., and J. Smith (1999): "The Pre-Program Earnings Dip and the Determinants of Participation in a Social
Program: Implications for Simple Program Evaluation Strategies", Economic Journal, 109, 313-348.
Imbens, G. W. (2000): "The Role of the Propensity Score in Estimating Dose-Response Functions", Biometrika, 87,
706-710.
Imbens, G. W. (2004): "Nonparametric Estimation of Average Treatment Effects under Exogeneity: A Review", Review
of Economics and Statistics, 86, 4-29.
Joffe, M., T. R. Ten Have, H. I. Feldman, and S. Kimmel (2004): "Model Selection, Confounder Control, and Marginal
Structural Models", American Statistician, 58-4, 272-279.
Larsson, L. (2003): "Evaluation of Swedish Youth Labor Market Programs", Journal of Human Resources 38(4), 891-
927.
Lalive, R., J. van Ours, and J. Zweimüller (2005): "The Effect of Benefit Sanctions on the Duration of Unemployment",
Journal of European Economic Association, 3, 1386-1407.
48
Lechner, M. (2000): "An Evaluation of Public Sector Sponsored Continuous Vocational Training Programs in East
Germany", Journal of Human Resources, 35, 347-375.
Lechner, M. (2001): "Identification and Estimation of Causal Effects of Multiple Treatments under the Conditional
Independence Assumption", in: M. Lechner and F. Pfeiffer (eds.), Econometric Evaluation of Active Labour Market
Policies, 43-58, Heidelberg: Physica.
Lechner M. (2002): "Some Practical Issues in the Evaluation of Heterogeneous Labour Market Programmes by Match-
ing Methods", Journal of the Royal Statistical Society, Series A, 165, 59-82.
Lechner, M., R. Miquel, and C. Wunsch (2006): "Long-run effects of Public Sector Sponsored Training in West Ger-
many", revised version of Discussion Paper 2004-19, Department of Economics, University of St. Gallen.
Machin, S., and A. Manning (1999): "The causes and consequences of long-term unemployment in Europe," in: O.
Ashenfelter and D. Card (eds.), Handbook of Labor Economics, Vol. 3c, 3085-3139, Amsterdam: North-Holland.
Martin, J., and D. Grubb (2001): "What Works and for Whom: A Review of OECD countries' experiences with active
labour market policies," Swedish Economic Policy Review, 8, 9-56.
Meyer, B. D. (1995): "Lessons from the U.S. unemployment insurance experiments", Journal of Economic Literature,
33, 91-131
Neyman, J. (1923): "On the Application of Probability Theory to Agricultural Experiments. Essay on Principles", Sta-
tistical Science, Reprint, 5, 463-480.
Pavoni, N., and G. Violante (2007): "Optimal Welfare-to-Work Programs," Review of Economic Studies, 74, 283-318.
Ramirez, J., and A. Vassiliev (2007): "An efficiency comparison of regional employment offices operating under dif-
ferent exogenous conditions", Swiss Journal of Economics and Statistics, 2007, 31-48.
Rosenbaum P., and D. Rubin (1983): "The Central Role of the Propensity Score in Observational Studies for Causal
Effects", Biometrika, 70, 41-55.
Rubin, D. (1974): "Estimating Causal Effects of Treatments in randomized and nonrandomized Studies," Journal of
Educational Psychology, 66, 688-701.
Rubin, D. (1979): "Using Multivariate Matched Sampling and Regression Adjustment to Control Bias in Observational
Studies", Journal of the American Statistical Association, 74, 318-328.
Shavell, S., and L. Weiss (1979): "The optimal payment of unemployment insurance benefits over time", Journal of
Political Economy, 87, 1347-1362.
49
Sheldon, G. (2003): "The efficiency of public employment services: a nonparametric matching function analysis for
Switzerland", Journal of Productivity Analysis, 20, 49-70.
Svarer, M. (2007): "The effect of sanctions on the job finding rate: evidence from Denmark", IZA discussion paper
3015.
Van den Berg, G. J., B. van der Klaauw, and J. C. van Ours (2004): "Punitive sanctions and the transition rate from
welfare to work", Journal of Labor Economics, 22, 211-241.
Wunsch, C. (2005): "Labour Market Policy in Germany: Institutions, Instruments and Reforms since Unification," Dis-
cussion Paper, Department of Economics, University of St. Gallen.
Wunsch, C. (2007): "Optimal Use of Labour Market Policies," Discussion Paper, Department of Economics, University
of St. Gallen.
... En Suisse, Behncke et al. (2010b) constatent que les conseillers moins coopératifs ont de meilleurs taux de placement et Huber et al. (2017) approfondit cette question en montrant que cet effet est susceptible d'être une conséquence des dimensions du conseil telles que les menaces de sanctions et la pression pour accepter des emplois. ...
... In Switzerland Behncke et al. (2010b) finds that less cooperative counselors have better placement rates and Huber et al. (2017) further explore this effect to show that it is likely to be driven by counseling dimensions such as threats of sanctions and pressure to accept jobs. ...
Thesis
Since the 1970s, the massive rise in the rates and duration of unemployment has led public authorities to orient their action towards so-called "active" labor market policies. Among these policies, job-search assistance proved to be particularly effective. Additionally, technological changes have significantly altered job searches and hiring practices in recent decades. The use of the internet has become a new norm in the job searching process for both employers and job seekers. The objectives of this thesis are twofold: understand the mechanisms behind the success of job search assistance and determine the role of digital technology in this type of program. It is presented in three chapters. The first chapter shows that a digital support program targeting a rarely studied population in the scientific literature - relatively autonomous job seekers - can increase their chances to find a job. The second chapter explores the role of job counselors. It measures the considerable influence they have on the quantity and the quality of exits to employment, showing that some counselors seem to favor one over the other, particularly through the services they recommend. Finally, in an experiment building upon an online platform, the last chapter simultaneously analyses the reaction of both jobseekers and firms in response to a job search stimulation. It concludes that providing targeted match recommendations increases job finding rates among women and firm hires on indefinite duration contracts.
... It aims to take as many components of the DGP as possible from real data. We build this EMCS on 96,298 observations of Swiss administrative social security data that is already used in previous evaluation studies (Behncke, Frölich, & Lechner, 2010a, 2010bHuber, Lechner, & Mellace, 2017). In particular, the EMCS mimics an evaluation of job search programs as in Knaus et al. (2017). ...
... The availability of extensive caseworker information and their subjective assessment of the employability of their clients distinguishes our data. Swiss caseworkers employed in the period [2003][2004] were surveyed based on a written questionnaire in December 2004 (see Behncke et al., 2010aBehncke et al., , 2010b. The questionnaire contained questions about aims and strategies of the caseworker and the regional employment agency. ...
Article
Full-text available
We investigate the finite sample performance of causal machine learning estimators for heterogeneous causal effects at different aggregation levels. We employ an Empirical Monte Carlo Study that relies on arguably realistic data generation processes (DGPs) based on actual data in an observational setting. We consider 24 different DGPs, eleven different causal machine learning estimators, and three aggregation levels of the estimated effects. Four of the considered estimators perform consistently well across all DGPs and aggregation levels. These estimators have multiple steps to account for the selection into the treatment and the outcome process.
... Therefore, it was excluded from further analyses. Probably the unemployed do not experience their advisor as a resource, as the advisor not only supports the unemployed, but is also allowed to sanction (Behncke, Frölich, & Lechner, 2010). At all points in time, modeling reemployment crafting this way showed a good fit to the data. ...
... As in Study 1, the item 'Today I've asked my advisor at the unemployment agency for advice" did not load well on the factor seeking resources. Probably because the unemployed do not see their advisor as a resource, as (s)he is allowed to sanction as well (Behncke et al., 2010). Moreover, if they have contact with their advisor, this probably is not on a daily basis. ...
Article
This article introduces the concept of reemployment crafting: the proactive, self-initiated behaviors undertaken by the unemployed to shape the environmental conditions of their job search in a way that enhances the person-environment (P-E) fit during the job search process. Using 2 longitudinal studies (Study 1: 3-wave study over a 3-month period, N = 153; Study 2: 4-day diary study, N = 189, days = 627), we investigated whether the manner in which the unemployed craft their job search is similar to the way employees craft their job. We examined whether reemployment crafting was positively related to job search performance (i.e., environmental exploration and networking behavior) and reemployment chances. Moreover, we examined whether contingency factors (i.e., social support and subjective goal attainment) affected the effectiveness of reemployment crafting. Results from both samples confirmed that the way the unemployed craft their job searches is similar to the way that employees craft their jobs. Reemployment crafting was positively related to job search performance, both within a 3-month period and within days. Moreover, reemployment crafting was especially beneficial for environmental exploration on days when social support and goal attainment were low. Last, environmental exploration was related to networking behavior, which in turn was predictive of reemployment chances. Specifically, in the diary study networking quality was related to reemployment status, while within the 3-month period, networking intensity seemed more effective. We conclude that reemployment crafting seems a promising way to enhance job search performance and ultimately the chances of finding reemployment. (PsycINFO Database Record (c) 2019 APA, all rights reserved).
... Nezaměstnaní představují velmi heterogenní cílovou skupinu, je velký rozdíl mezi nově příchozími osobami do evidence nezaměstnaných, kteří disponují pracovními zkušenostmi, vzděláním a ochotou pracovat, a těmi nezaměstnanými, kteří mají z různých důvodů velké obtíže se znovu integrovat na trh práce, jsou bez práce opakovaně a dlouhodobě. Studie ze Švýcarska ukázala, že "work-first" přístup (hrozba sankcí) a méně spolupracující přístup zprostředkovatelů vede u "čerstvě" nezaměstnaných k rychlejšímu nalezení zaměstnání (Behncke, Frὄlich, Lechner, 2007). Zkušenosti z Dánska ukazují, že v případě skupiny dlouhodobě nezaměstnaných osob odkázaných na dávky sociální pomoci ekonomické sankce nevedou ke zvýšení jejich zaměstnanosti, ale že mají velmi negativní vliv na sociální situaci sankcionovaných jedinců a jejich dětí (Caswell et al., 2011citováno v: Ravn, Nielsen, 2019. ...
Book
Full-text available
Cílem výzkumu bylo identifikovat roli poskytovatelů sociálních služeb (dlouhodobě) nezaměstnaným osobám v procesu hledání zaměstnání a zjistit, jak poskytovatelé služeb v tomto procesu spolupracují s Úřadem práce ČR. Vzhledem k heterogenitě jak poskytovatelů sociálních služeb, tak dlouhodobě nezaměstnaných osob, byla cílová skupina zúžená na poskytovatele přístřeší osobám bez domova (tj. azylové domy, noclehárny, domy na půl cesty atp.), které jsou osobami v hmotné nouzi. Byly použity kvantitativní i kvalitativní metody sběru dat. Hlavní zjištění našeho výzkumu potvrzuje závěry již dříve realizovaných výzkumů, že spolupráce mezi úřadem práce a sociálními pracovníky různých zařízení poskytujících sociální služby je funkční tam, kde na lokální úrovni existují dobré, často neformální vztahy mezi zúčastněnými aktéry. Pro efektivní realizaci "skutečné" sociální práce s osobami bez domova je však nutné zajistit koordinaci a v některých případech také funkční spolupráci mezi poskytovateli přístřeší osobám bez domova a zaměstnanci ÚP, zejména oddělením nepojistných sociálních dávek a útvarem zprostředkování. Aims of our research were to identify a role of social services providers to long-term unemployed people to find a job and determine how social service providers cooperate with labour officers in the process of looking for a job. Because of great heterogeneity of social service providers as well as long-term unemployed people, we narrowed a target group to providers of a shelter to homeless people whom social assistance benefits are payed by labour office. Examples of the social service providers included in our research were different types of asylum homes or dosshouses according to the Czech social service legislation. We used quantitative and qualitative methods of data collection. Main conclusions of our research confirm results of previously realised researches that cooperation between labour offices and social service providers in general is functioning well if at the local levels informal and friendly relations between them exist. To provide effective (and real) social work with homeless people is necessary to ensure coordination and, in some cases also functional cooperation between subjects providing a shelter to homeless persons and labour officers, who administer social benefits as well as job brokerage to them.
... Exploiting detailed Swiss data sets, Behncke et al. (2010a) found that the chance of re-employment increases if the counsellor and his client belong to the same social group. Behncke et al. (2010b), and later Huber et al. (2017) showed that "tough" caseworkers, who are less cooperative, are able to produce better employment results. Schiprowski (2017) used unplanned caseworker absences to investigate the effect of caseworker meetings. ...
Article
Full-text available
In a Public Employment Service reform implemented in 2013, sixty Finnish municipalities experienced an involuntary employment office closure. The Government’s objective was to replace traditional face-to-face employment counselling with modern online counselling and simultaneously generate savings in outlays. The reform created natural experiment circumstances that allowed us to estimate the aggregate causal effects of face-to-face counselling and advice. We estimated the effects of the reform on the unemployment rate and the average unemployment duration using municipality-level panel data and various panel data estimators. We found that while the reform had a barely discernible effect on municipal unemployment rates, it increased average unemployment durations by 2–3 weeks. Hence, face-to-face counselling and online counselling are not perfect substitutes in decreasing the length of unemployment spells. Consequently, the fiscal costs of the reform outweigh the fiscal benefits by a large margin.
... In a similar vein, a Swiss study by Behncke, Frölich, and Lechner (2010) -that focused on newly unemployed unemployment benefit recipients -found that similarity between caseworker and client (defined in terms of age, education, and nationality) improves the job chances of clients by three percentage points. Focusing also on newly unemployed, the same authors found that caseworkers who strictly emphasise the importance of finding work and use a less harmonious and less collaborative approach to the unemployed increase the clients' job chances (Behncke, Frölich and Lechner, 2007b). In other words, using a disciplining, work-first approach is more effective. ...
Article
Across the OECD countries, there is a growing consensus in favour of targeting active labour market policies (ALMP) on the disadvantaged unemployed and persons outside the labour force to increase their employment prospects. Despite increased efforts, little is known about what works for getting persons with physical, mental, and social problems into employment. Using difference-in-differences regressions and propensity score matching on longitudinal population register data from Denmark, we investigate the effects of investment in public employment services for disadvantaged social assistance recipients, where social worker caseloads have been severely reduced and active employment measures for the target group have intensified. We find significant and robust positive effects of intervention on subsequent employment outcomes for disadvantaged, hard-to-employ social assistance recipients, suggesting the need for an increased focus on this target group in future research and in the design and implementation of ALMPs.
... previous evaluation studies (Behncke, Frölich, & Lechner, 2010a, 2010bHuber, Lechner, & Mellace, 2017). In particular, the EMCS mimics an evaluation of job search programs as in Knaus et al. (2017). ...
Preprint
Full-text available
We investigate the finite sample performance of causal machine learning estimators for heterogeneous causal effects at different aggregation levels. We employ an Empirical Monte Carlo Study that relies on arguably realistic data generation processes (DGPs) based on actual data. We consider 24 different DGPs, eleven different causal machine learning estimators, and three aggregation levels of the estimated effects. In the main DGPs, we allow for selection into treatment based on a rich set of observable covariates. We provide evidence that the estimators can be categorized into three groups. The first group performs consistently well across all DGPs and aggregation levels. These estimators have multiple steps to account for the selection into the treatment and the outcome process. The second group shows competitive performance only for particular DGPs. The third group is clearly outperformed by the other estimators.
Article
Research about drivers of trust of the long‐term unemployed in their caseworkers is a white spot in the literature. The paper closes this gap using a unique data set. Embedded in a theoretical model at the organizational level a trust game with real long‐term unemployed and caseworkers is evaluated. The results support the social identity theory, i.e. trust in members of the ‘own’ group is higher than trust in members of the ‘other’ group, as well as more traditional explanations of trust. Thus, policy can raise trust using the concept of incentive ethics of the theoretical model.
Article
While there has been much research on welfare exit and entry into employment, less research has looked at return to government assistance. Applying survival analysis on data from a national government assistance programme in Singapore, we found two important factors of welfare return to which activation programmes need to pay greater attention. First, return was more likely if former beneficiaries accumulated a higher number of types of arrears rather than higher dollar values of arrears. This new finding contributes to the emerging literature on bandwidth tax, and suggests the importance of designing programmes that relieve mental accounting due to debt and poverty. Second, return was more likely if respondents had an infant or toddler child. This points to the importance of a range of support policies including affordable and accessible childcare, exemption from work requirement in receipt of welfare, and family leave for low-wage workers.
Article
We question whether accessibility to local public employment agencies impacts exits from unemployment. We deal with the potential endogeneity of the residential location of jobseekers by using the unanticipated creation of a new agency in the French region of Lyon as a quasi-natural experiment. We use exhaustive and geo-located individual data on jobseekers and local public employment agencies. Contrary to past evidence based on aggregated data, we find no evidence that jobseekers with improved accessibility to the local public employment services experience an improvement of their probability of exiting unemployment. We however find evidence of transitory organizational effects. These findings strongly question the costly strategy of a fine distribution of local public employment agencies across the territory while suggesting that institutional issues are key.
Article
Full-text available
The propensity score is the conditional probability of assignment to a particular treatment given a vector of observed covariates. Both large and small sample theory show that adjustment for the scalar propensity score is sufficient to remove bias due to all observed covariates. Applications include: (i) matched sampling on the univariate propensity score, which is a generalization of discriminant matching, (ii) multivariate adjustment by subclassification on the propensity score where the same subclasses are used to estimate treatment effects for all outcome variables and in all subpopulations, and (iii) visual representation of multivariate covariance adjustment by a two- dimensional plot.
Article
We consider regression situations for which the response variable is dichotomous. The most common analysis fits successively richer linear logistic models and measures the residual variation from the model by minus twice the maximized log likelihood. General measures of residual variation are considered here, including ordinary squared error and prediction error as well as the log likelihood. All of these are shown to be satisfactory in a certain primitive sense, unlike quantitative regression theory where only squared error is logically satisfactory. The relation of Goodman and Kruskal’s measures of categorical association to the theory of penalty functions and probability elicitation is demonstrated.
Article
This study develops a methodology for assessing the efficiency of job placement services based on the matching function. Unlike most previous estimates of the matching function, based on aggregate time-series regressions and parametric functional forms, this study employs micro cross-sectional data and uses nonparametric frontier estimation techniques (DEA). The methodology is applied to 126 regional placement offices operating in Switzerland in the period 1997–98. In contrast to time-series regressions, our results point to sizable increasing returns-to-scale. Our findings also suggest that counseling is more effective than other active labor market measures or disciplining actions in increasing matching efficiency.
Article
In traditional regression modeling, to control for confounding by a variable one must include it in the structural part of the statistical model. Marginal structural models are a flexible new set of causal models. The estimation methods used to estimate model parameters use weighting to control for confounding; this allows more flexibility in choosing covariates for inclusion in the structural model and allows the model to more precisely reflect the scientific questions of interest. An important example of this is in multicenter observational studies where there is confounding by cluster. We illustrate these points with data from a study of surgery to provide vascular access for hemodialysis and a study comparing different timings for coronary angioplasty.
Article
In The Netherlands, the average exit rate out of welfare is dramatically low. Most welfare recipients have to comply with guidelines on job search effort that are imposed by the welfare agency. If they do not, then a sanction in the form of a temporarily benefit reduction can be imposed. This paper investigates the effect of such sanctions on the transition from welfare to work using a unique set of rich administrative data on welfare recipients in The Netherlands. We find that the imposition of sanctions substantially increases the individual transition rate from welfare to work. We also describe the other determinants of the transition from welfare to work.