Printer Friendly

Active labor market programs and regional mobility of labor: evidence from the Swedish recession, 1994-1995.


This paper addresses the impact of participation in Active Labor Market Programs (ALMPs) on migration in Sweden. its purpose is to determine the extent to which programs that provide training and labor market assistance to jobless individuals induce them to migrate.

While most developed nations have programs to assist the unemployed, the Nordic countries are particularly active in this regard, spending as much as 2% of gross domestic product (GDP) to execute programs that offer direction, training, and placement to displaced workers. Consequently, they are of particular interest to researchers and policy makers. This paper focuses on the Swedish model. Sweden has a history of policy initiatives to facilitate transition of jobless individuals to the workforce. The highly developed nature of the Swedish program, along with availability of data registers that track labor market outcomes of individuals, provide opportunities for research about policy outcomes. Most recent studies have been concerned primarily with the programs' efficacy in returning individuals to employment and removing them from the ranks of unemployment benefit recipients. Examples include Sianesi (2002. 2004, 2005); Delander and Mansson (2007): Carling and Richardson (2004); and Larsson (2003). Other researchers have examined the impact of program participation on subsequent earnings (Andren and Gustafsson 2004). Additional studies of Swedish programs, as well as other programs in Europe, are summarized in the study by Kluve (2006).

One issue that has received less attention is the impact of program interventions on the mobility of participants. Regional migration is a potentially critical outcome for labor market programs. Mobility of workers provides a market-clearing mechanism that equates supply and demand across labor markets. From the standpoint of program participants, migration is a form of job search and investment in human capital that complements program-based training, experience, and job preparation. Research concerning the extent to which labor market programs enhance or impede mobility can be valuable in formulation and refinement of policy.

Fredriksson and Johansson (2003) suggest that participants in Swedish programs are as much as 3 percentage points less likely to migrate than a comparison group of nonparticipants. Lindgren and Westerlund (2003) estimate that program participants who receive employment training are less likely to migrate than individuals who remain in open unemployment, which is defined as jobless but not participating in labor market programs. Within programs, however, their results indicate that participants who receive employment training are more likely to migrate than those who are assigned to relief work or work experience schemes. Hamalainen (2002) finds mixed evidence from Finnish data. The estimates indicate that program participation may either deter or induce migration, depending on the type of program, ages of participants, and general unemployment conditions.

This paper adds to the literature by exploiting data registers compiled in 1994 and 1995 by Statistics Sweden and the Labor Market Board of Sweden. The longitudinal nature of the data allows us to follow individuals through time. We use that feature to estimate a recursive two-period model that encompasses program participation in the first step, followed by the subsequent propensity to migrate as a function of program participation.

A second advantage of the data is that they delineate the spatial dimension of migration in terms of labor market areas. The labor market area is conducive to migration modeling, as it is based on economic activity and commuting patterns of workers. This makes it more amenable to the human capital framework, as opposed to regions that are established historically or politically such as those found in the United States or Canadian provinces.

The average population in the labor market areas (LMAs) was approximately 98,000 in 1995. The largest, Stockholm, has more than 2 million and a population density of 135 inhabitants per square kilometer. At the other extreme, Sorsele is located in the northern inland and has a population of approximately 3,400 with a. density of 45 inhabitants per square kilometer. Generally, the LMAs have smaller populations and lower densities in the northern regions and in parts of the inland, middle, and southern regions. Unemployment rates are higher than the national average in those regions. In order to capture these disparities, in the empirical model we include dummy variables indicating regions as well as an additional control for regional unemployment.

A cursory examination of the data, which are described in Section V, reveals motivation for this study. Among men who were program participants in 1994, the proportion of migrants by 1995 was 0.048. By contrast, for nonparticipants the proportion of migrants was 0.023. For women, the corresponding proportions are 0,046 and 0.019. These discrepancies might reflect the absence of controls for measured covariates such as age and education. A possibility also exists that participants differ from the rest of the population in unmeasured attributes that simultaneously affect participation and mobility. In Section IV, we establish an empirical framework to address these issues.

The period encompassed by this study is useful for policy analysis. From the peak of the business cycle in the late 1980s, Swedish industrial production declined nearly 20% by 1993. Manufacturing capacity was reduced by approximately 7%, and open unemployment increased from 2% to more than 6% in 1993. Faced with a budget deficit in the public sector of 11.2% of GDP and impending integration into the European Community, the government adopted a package of crisis measures that included spending reductions, tax increases, an austere monetary policy, and intervention in the banking system. Thus the years used for this study provide an appealing context for analysis of unemployment policy, with potential implications for Sweden as well as the OECD countries experiencing the persistent recession of 2008-2010.


Swedish labor market policy is characterized by a set of ALMPs and a two-tiered system of unemployment insurance (UI). The ALMP component consists of six policy initiatives into which unemployed persons can be placed: (1) labor market training to equip individuals with job search skills: (2) work placement, primarily in the nonprofit and public sectors: (3) workplace introduction and vocational training: (4) relief work in temporary positions-the majority in public entities and subsidized by the central government; (5) temporary replacement of employees who are granted leaves of absence for training; and (6) employer subsidies for permanent positions. Placement of candidates, resulting from choices by participants and caseworkers, depends on local conditions and the available mix of programs.

The UI component is characterized by two tiers of compensation. Those who are UI-eligiblc, based on their work records, are entitled to replacement rates up to 80% of their previous wages for a period of as many as 60 weeks. The second tier, directed primarily to new labor market entrants, provides cash assistance for up to 30 weeks. The ALMP and UI benefit systems have been linked by policies governing renewal of UI assistance. Until 2001, individuals were eligible for the renewal of UT benefits if they became ALMP participants.1

A noteworthy feature of ALMP policy is that individuals have a degree of choice in the timing of entry into programs. Until the time of entry, they remain in open unemployment. Thus, at any point in time, the latter group is not necessarily to be regarded as nonparticipants in ALMPs, but rather as having postponed participation to a future date. This imparts a degree of endogeneity in the participation decision.

Researchers have addressed the efficacy of programs in returning individuals to employment. Sianesi (2002, 2004, 2005) finds that, although participants exhibit higher rates of employment, they are more likely to receive UI benefits over time than if they had remained in open unemployment. The largest impetus to employment is in programs that entail subsidized work and receipt of firm-provided training (see also Carling and Richardson 2004). Larsson (2003), focusing on outcomes among workers aged 20-24, finds no significant gains in participants' subsequent employment and earnings. Andren and Gustafsson (2004) find that ALMP impacts on earnings are ambiguous, with variation among three period-based cohort groups and with contrasts between native Swedes and foreign-born workers. Delander and Mansson (2007) focus on the Swedish Activity Guarantee, targeting workers at risk for extended joblessness. Their results indicate that the program increases the likelihood of returning to work and reduces the duration of unemployment.

These studies do not address the potential for migration of participants. There are several reasons to anticipate that program participation might increase individuals" propensities to migrate. To the extent that participants receive job search training and acquire knowledge about labor market opportunities, they are more inclined to expand their job searches across larger areas. To the extent that they experience general training and acquire enhanced work skills, they become attractive to employers in all regions. In addition, their choice of program participation serves as a signal to employers about their potential as workers.

On the other hand program participation might reduce mobility if participants are motivated by a desire to renew their eligibility for UI benefit receipt. Those individuals might be characterized by a general aversion to investment in human capital, including migration. Even in the absence of preference for UI benefits, participation restricts time available for job search, which in turn tends to diminish mobility. In addition, Sianesi (2002, 16) points out that changes in funding of social assistance in Sweden since the 1990s might have had the effect on reducing mobility. The shift has been toward decentralization of financial responsibility. Local governments, which assume responsibility for social assistance, have incentives to move unemployed persons to location-specific relief work, which is funded by the central government and thereby relieves the fiscal burden on local government. The indirect consequence might be to decrease participants' propensities to relocate.

The implications of ALMPs for participants' mobility are important for policy purposes, and additional research is warranted. In the following sections we outline an approach that addresses this issue.


Economists have traditionally viewed migration as an investment in human capital (Sjaastad 1962). In this formulation, individuals discount expected future gains in earnings from moving between the current region m to the next alternative n, net of the regional disparity in living costs and the one-time cost of moving;


where the E's represent expected earnings in regions m and n during period t; the L's denote suitably scaled living costs in each region; [C.sub.m,n] is the cost of moving between regions; T is the investment horizon, and r is the discount rate. The expected net gain from migration is [C.sub.m,n] > 0.

Individuals form expectations of wage offers in each region, along with a reservation wage. Hours worked are positive if the wage offer wq exceeds the reservation wage [w.sub.r]:

[H.sub.j] > 0 if [w.sub.0j] > [H.sub.j], j = m, n.

Accordingly, earnings are the product of the (accepted) wage offer and hours worked:

[E.sub.j] = [w.sub.0j] x [H.sub.j], j = m, n.

This framework is consistent with elementary models of job search, in which potential migrants sample from a known distribution of wage offers. Denoting the cumulative distribution function by F x [w.sub.0] and the marginal cost of sampling by M x [C.sub.w], the individual behaves as if he maximizes the expected return from search. This implies the formation of a reservation wage that equates the cost of search with expected wage returns at the margin:


For the case in which the individual possesses perfect knowledge of F x [w.sub.0] it can be optimal to use migration as part of the search strategy, either migrating prior to search or searching one region and migrating to another in the event of an unsuccessful outcome (Maier 1987, 191-94). In this model, the reservation wage remains constant over time (see, e.g., Maier 1987, 191; Pickles and Rogerson 1984, 133).

If the assumption of perfect information is relaxed, the reservation wage changes as the individual learns about wage offers, and participation in ALMPs provides an important means of job market learning. Program participants experience a variety of treatments intended to improve their skills and strengthen their labor market attachments. In addition, they acquire general knowledge about the world of work and available employment opportunities.

These outcomes have implications for reservation wages and wage offers. With respect to the latter, the consequence of improved skill is a favorable shift in the distribution of offers. The individual is induced to expand the geographic scope of search and thereby to consider migration as an investment strategy. Thus an outcome of participation is to enhance wage offers:

[w.sub.0j] = [f.sub.1] x ([??], Weeks, ALMP),

[??] x [[omega].sub.0j] / [??], Weeks, ALMP > 0, j = m, n,

where [bar.X] is a set of control variables and the second argument is annual weeks of participation.

Similarly, the reservation wage depends on participation and a vector of control variables. However, the impact of participation on reservation wages is ambiguous:

[w.sub.r] = [f.sub.2] x ([bar.Z], Weeks, ALMP),

[partial derivative] x [w.sub.r] / [partial derivative], Weeks, ALMP [less than or equal to] 0 or [greater than or equal to] 0.

The ambiguity arises from opposing effects of program participation. As participants gain skill and learn of wage offers, their reservation wages increase. On the other hand, as the duration and geographic scope of search increase, and with them the number of unacceptable offers, the reservation wage declines (Maier 1987, 199-200). In addition, the policy design of ALMPs influences the reservation wage. Some individuals enroll in ALMPs to assure the receipt and generosity of UI benefits, which in turn tend to increase the reservation wage.

Because the model is based on the expectation of wage offers, it is not necessary for the individuals to have received an acceptable offer before the move. Rather, perceiving that their wage offers have improved, some will move in order to conduct more efficient search and solicitation of offers at a desired destination. Hence, what is sufficient for migration is the anticipation of wage offers in excess of the reservation wage, and migrants will be drawn from the population of those "perspicacious peregrinators" (Polachek and Horvath 1977). On the other hand, as noted earlier, some ALMP participants increase their reservation wages and hence are less likely to migrate. Even in the latter group, however, migration is possible if wage offers improve to a greater extent than the increase in the reservation wage.

These opposing behaviors result in a theoretically ambiguous effect of ALMPs on migration. An additional complication is the potential for latent personal traits that affect both outcomes. Some individuals possess unmeasured attributes such as personal ambition and a propensity for risk taking in the form of human capital investment, or what Pickles and Rogerson (1984, 138) describe as "aspiration levels" and "willingness to move" (see also Jaeger et al. 2007). Their choices regarding the program participation and migration might be simultaneously driven by those attributes.

Taken together, these considerations suggest that resolution of the ALMP impact on migration is an empirical issue. In the following section, we outline an econometric approach that isolates the direct impact of participation on mobility while controlling for the presence of confounding unmeasured attributes.


The model is based on the observations of individual i at two points in time. In the first period we observe the extent, if any, of his participation in labor market programs, along with an associated set of covariates. Program engagement is modeled as annual weeks of participation:

(3) Weeks_ALM[P.sub.i,t] = f x ([beta]' x [x.sub.i,t],

where [x.sub.i,t] is a vector of covariates in period t and [beta] is a conformable vector of unknown coefficient parameters. In the second period we observe the individual's migration decision:

(4) [Migrate.sub.i,t + 1] = g x ([delta] x [z.sub.i,t] + [gamma] x Weeks_ALM[P.sub.i,t]), where [z.sub.i,t] is a vector of covariates that determine the migration outcome, [delta] is a vector of coefficients, and [gamma] is a scalar that captures the effect of program participation on migration.

The dependent variable in Equation (3) is measured as nonnegative integer counts. This occasions the use of a member of the Poisson family of regression models. The familiar problem that arises is potential for over-dispersion; the Poisson model restricts the variance of the dependent variable to be equal to its conditional mean. A common approach to address over-dispersion is to specify the model on the basis of a negative binomial-II distribution (Cameron and Trivedi 1986). The probability mass function (pmf) is given by:


where [GAMMA] (*) denotes the gamma function, [[lambda].sub.i,t] = exp x ([beta]' x [x.sub.i,t] is the conditional mean of annual weeks participation, and the conditional variance is given by [[lambda].sub.i,t] x [1 + (1 / [theta] x [[lambda].sub.i,t]]. The null hypothesis for the absence of over-dispersion is

equivalent to [theta] = 0.

We assume the decision to move is embodied in a latent migration propensity;

(6) [Migrate*.sub.i, t+1] = [delta]' [z.sub,i, t] + [gamma] [Weeks_ALMP.sub.i, t] - [[epsilon].sub.i, t].

where [[epsilon].sub.i,t] is a random disturbance term assumed to be normally distributed with zero mean and unit variance. The individual chooses to migrate if [Migrate*.sub.i, t+1] > 0.

The latent index is not observed. Instead we observe a dichotomous indicator:

[Migrate*.sub.i, t+1] = 1 if [Migrate*.sub.i, t+1] > 0

[Migrate*.sub.i, t+1] =0 if [Migrate*.sub.i, t+1] [less than or equal to] 0.

Then the probability of migration can be expressed as

(7) Pr (Migrat[e.sub.i,t] = 1) = [PHI]([sigma][z.sub.i,t] + [gamma][Weeks_ALMP.sub.i,t]). where [PHI] denotes the cumulative distribution function of a standard normal distribution.

Equation (7) is a probit model of migration. In principle, its parameters can be estimated by the method of maximum likelihood. A complication arises because of the presence of Weeks_ALMP on the right-hand side. As an endogenous choice variable, program participation has potential to impart bias in the estimated coefficients of Equation (7), in particular [gamma], the coefficient of participation. The estimation procedures described in the following text explicitly account for the potential endogeneity of program participation.

A second complication arises in that the migration decision is potentially a composite of unmeasured characteristics that are simultaneously associated with program participation and mobility. To accommodate that possibility, we amend the equations to include a common factor that captures the latent propensity. The conditional mean of weeks participation becomes

[[lambda].sub.i,t] = exp([beta]' [x.sub.i,t] + [[alpha].sub.w][u.sub.i]).

and the amended probit equation is

P([Migrate.sub.i,t+i = l)

= [PHI]([delta]'[z.sub.i,t] + [gamma][Weeks_ALMP.sub.i,t] + [[alpha].sub.m][u.sub.i])

where [u.sub.i] captures individual unobserved heterogeneity.

If [u.sub.i] could be observed, then the estimated coefficient y would be interpreted as the causal effect of program participation on migration after controlling for unmeasured heterogeneity. However, as [u.sub.i] is not observed, estimation of the model in its absence produces biased estimates. Our approach is to use the method of maximum simulated likelihood (MSL), which generates synthetic draws of ut based on the assumption that it arises from a standard normal distribution (Gourieroux and Monfort 1996). The MSL procedure entails forming the joint pmf of the observed dependent variables. The pmf embodies the average of a set of S random draws of ii[. For individual i, the pmf is given by


In this notation, [~.h.sub.s] embodies a modification of the conditional mean in Equation (5) to incorporate the simulated values of [u.sub.i], denoted [~.u.sub.i]: [~.[lambda].sub.i,t] = exp ([beta]'[x.sub.i,t] + [[alpha].sub.w] [~.u.sub.i,s]). Similarly, the notation [MATHEMATICAL EXPRESSION NOTREPRODUCIBLE IN ASCII] recognizes simulation of the heterogeneity term in Equation (7'):

[~.[PHI].sub.s] [[Migrate*.sub.i, t+1]] = [PHI]([delta]' [z.sub,i, t] + [gamma] [Weeks_ALMP.sub.i,t] + [[alpha].sub.m] [~.u.sub.i,s]).

The log likelihood function is formed from the logarithm of the product of Equation (8) over all observations. MSL produces estimates of the coefficient parameters by numerically integrating the [u.sub.i] terms out of the joint pmf and averaging across the S draws. Further exposition of these techniques is found by Train (2003). Other applications include Deb and Trivedi (2006) and Deb et al. (2006).(2) Gourieroux and Monfort (1996) show that the simulated function is a consistent estimator of the true log likelihood function if S increases faster than the square root of the sample size. Estimates reported in this paper are based on 1,100 simulations.(3)

Under the assumption that the latent factors are distributed standard normal, the fit of the model is invariant to location and scale. As the model must also be invariant to labeling of the latent factors, the first coefficient, [[alpha].sub.w], is restricted to unity, leaving [[alpha].sub.] as a free parameter.


A. Data and Sample

We extracted the sample from registers compiled by Statistics Sweden and the Labor Market Board of Sweden. The base sample comprises approximately 60,000 individuals aged 18-62 in 1994. As noted in Section I, this period is of particular interest for labor market programs, as it witnessed a dramatic increase in joblessness in Sweden. The sample is restricted to individuals who experienced at least 1 day of registered unemployment or ALMP participation or earned positive income from labor in 1994. In order to avoid concerns with spouses who might be tied movers, we further restrict attention to unmarried individuals. Because similar "spouse.'" constraints are possible in the population of unmarried cohabitants, we also exclude cohabiting couples from the sample. We do not, however, exclude individuals with children. Single parenthood is common in Sweden, and while some lone parents might be tied by the interests of their children, we do not expect those ties to be as binding as in the case of spouses, and excluding them from the sample is not warranted.

Because the nature of migration might differ between genders, we estimate the model separately for men and women. This is in contrast to most previous research, which tends to focus on men. That has been necessitated by the low-rate of labor force participation of women relative to men. However, in the ease of Sweden, the female participation rate exceeds 75% (compared to 80% for males), which ranks among the highest in the world. Consequently, it is appropriate to include a female sample in the empirical analysis. Our final samples consist of 22,091 men (653 migrants from 1994 to 1995) and 19,583 women (550 migrants).

For each individual, we measure weeks of ALMP participation in 1994. This is defined as the sum of commenced weeks participation, where 1-7 days (including weekends and holidays) is counted as I week, 8-14 days is counted as 2 weeks, and so on. During the recession-afflicted period under study, a large share of the unemployed in a particular program participated in at least one other program within a brief time interval. As a result, identification of effects of any program option is problematic. Consequently, following other studies of the Swedish program (Andren and Gustafsson 2004; Fredriksson 1999; Sianesi 2002, 2004, 2005; Westerlund 1997), we consider participation per se and do not distinguish among program options. Thus, our estimates reflect the average effect of participation on mobility.

The data allow us to identify the labor market area of residence in both 1994 and 1995; migration is defined to occur when the labor market area changes between years. Created by Statistics Sweden, labor market areas are based on an algorithm that exploits commuting flows within clusters of municipalities. The algorithm creates a set of regions in which commuting flows across borders are minimized. Because, by definition, commuting options are limited between labor market areas, migration is the only viable option for the workers to take advantage of employment opportunities in other areas. Consequently, movements of workers are more likely motivated by economic considerations and disparate employment opportunities than would be the case if regions were defined on the basis of historical or political considerations. Because of increases in frequencies and distances of commuting, the number of labor market areas has declined over time. We use the partition that was in effect in 1994, in which the country's 286 municipalities were aggregated into 108 labor-market areas.

In addition to program participation and region of residence, the data file contains background information for each person, including age education, and origin status of foreign born individuals, parent status, disability status, and the regional unemployment rate.

We use dichotomous indicators for region of residence in the origin year of 1994 to capture variation in economic conditions not fully reflected in the region-specific unemployment rate. The region indicators separate Stockholm, with more than 20% of the population and which serves as the reference category from the West (including Goteborg) the South (including MalmO), and the more sparsely populated regions in the middle and north of Sweden. An additional regional consideration is the potential for variation in administration of labor market programs (see, e.g., Delander and Mansson 2007). To capture the program supply phenomenon, we constructed a variable that measures program enrollment as a proportion of the population aged 16-64 in the individual's municipality of residence in 1994 (denoted ALMP_Supp).

In specifying the migration equation, we include an indicator of the individual's recent (1992-1994) history of migration. Similar variables have appeared in other models of the decision to migrate (Nakosteen and Westerlund 2004; Tunali 2000).

Using these variables, we present estimates for the following specifications:
Equation (5) Equation (7)

Dependent variable = Weeks_ALMP Dependent variable = Migrate
Age Weeks_ALMP
Age Squared Age
Education Education
Immigrant Origin Region Regional Unemployment Rate
Region Parent Status
ALMP_ Supp Migrant 1992-1994
Disability Status

B. Exclusion Restrictions

The model includes immigrant origin, disability status, the region indicators, and program supply in the participation equation but excludes them from the migration equation. The latter equation, on the other hand, includes the regional unemployment rate, parent status, and the migration history variable, but restricts them from the program weeks equation.

Regarding the exclusions from the migration equation, measures of program supply at the regional or program jurisdiction level have previously been used as identifying instruments in research that examines labor market outcomes of job-preparation programs. Examples include Aakvik, Heckman, and Vytlacil (2005) and Mclnnes, Ozturk, and McDermott (2010). The remaining exclusions deserve additional consideration. The migration literature makes no unequivocal statement about the effect of health-related disability on mobility. On one hand, individuals with disabilities might be inclined to move to large cities, where they have access to medical care and convenient transportation. At the same time, however, their costs of migration (monetary and psychic) might offset the urban amenities. Immigrants are by definition individuals who demonstrate propensities to move. Theory does not establish, however, that they are necessarily prone to absorb the costs of repeated relocations in their adopted countries. Moreover, as stated earlier, the migration model contains a control for past migration, so it is not clear that immigrant status per se leads to internal migration. Finally, the region indicators might be candidates for inclusion in the migration equation; even though the model includes the regional rate of unemployment, which is expected to induce migration, regions might possess disparate amenities that are not related to job opportunities but nonetheless affect labor mobility.

Because these exclusions are not necessarily well established on an a priori basis, in Section VI, we present alternative estimates of the migration equation from several specifications that relax the restrictions. As the estimates indicate, the essential conclusions of this study are robust across specifications.

C. Description of the Data

Variable definitions and sample means are provided in Table h which partitions the sample by gender. The sample means reveal an average age of 37 years for males and 38 years for females, with the highest sample density of educational attainment at 2 years upper secondary for both genders. Approximately 3% of the sample migrated between 1994 and 1995, and 5% had previously migrated between 1992 and 1994. Coincidentally, this is consistent with the data that have been reported for year-to-year interstate migration in the United States (see, e.g., Borjas, Bronars, and Trejo 1992, 159). The immigrant population comprises 4%-5% from Nordic countries and 5%-6% from non-Nordic countries. The sample proportions also reveal a greater incidence of parent status among women. The unemployment statistics reflect labor market turmoil in the early 1990s: males and females experienced 7 and 5 weeks of open unemployment, respectively, in 1994 (not shown in Table 1), and they resided in regions with rates of unemployment that approached 8%. Males participated in labor market programs for an average of 8 weeks; for females average participation exceeded 9 weeks. These averages conceal the frequency distribution of participation weeks. Inspection of the data reveals the following frequencies:
Weeks Proportion: Males Proportion: Females

0 0.708 0.697
1 - 10 0.063 0.052
11 - 20 0.077 0.065
21 - 30 0.063 0.059
31 - 40 0.036 0.044
41 - 52 0.051 0.083
Variable Definitions and Sample Means: 1994

Variable Definition Males Females

Weeks_ALMP Weeks in active labor market 8.281 9.518
 programs 1994

Migrate Dummy variable = 1 if individual 0.030 0.028
 migrated to labor market areas.

Age Years 36.714 38.332

Education: Reference category is less than 9
 years primary.

Educ 2US Dummy variable = 1 if educational 0.385 0.403
 attainment is 2 years upper
 secondary school.

Educ 3US Dummy variable = 1 if educational 0.146 0.136
 attainment is 3 years upper
 secondary school.

Educ PS Dummy variable -- I if educational 0.113 0.141
 attainment is post-secondary

than 4 vears.
Educ Univ Dummy variable = I if educational 0.097 0.114

 attainment university degree or

Region: Reference category is Stockholm
East Mid Dummy variable = 1 if residence is 0.169 0.163
 in Uppsala. Sodermanlaiid.

 Ostergotland. or Viistmanland
Smaland Dummy variable = 1 if residence is 0.074 0.069

 in Jonkoping, Kronoberg. Kalmar. of
 Gotland counties.

South Dummy variable = 1 if residence is 0.133 0.135
 in Blckinge or Skane counties.
West Dummy variable = 1 if residence is 0.190 0.189

 in Halland or Viistra Gotaland
North Mid Dummy variable = 1 if residence is 0.101 0.094

 in Varmland. Dalarna or Gavleborg
Mid-Norrland Dummy variable = 1 if residence is 0.045 0.043

 in Vasternorrland or Jamtland
Upper Dummy variable = 1 if residence is 0.060 0.054

Norrland in Vast.erbolten or Norrbolten

Stockholm Dummy = 1 if residence is in 0.228 0.253
 Stockholm county-

Immigrant Reference category is native
Origin: Swedes.
Nordic Dummy variable = 1 if individual 0.040 0.052

 was born in a Nordic country:
 Norwav, Denmark, Finland. Iceland.

Non-Nordic Dummy variable = 1 if individual 0.057 0.051
 was born in a non-Nordic country.
Disabled Dummy variable = 1 if individual 0.085 0.079

 has a health impairment thai

 or prevents work.
Parent Dummy variable = 1 if individual 0.041 0.238

 has children under 18 residing at

Reg Unem Unemploymenl rate in the region of 7.725 7.663
ALMP_Supp Proportion of ALMF participants in 0.312 0.304

 municipality of residence to
 population aged 18-64.
Migration: Dummy variable = 1 individual 0.049 0.053

1992- 1994 migrated to labor market areas

N 22.091 19.583

For those with positive weeks, there is modest concentration through 30 weeks, with similar frequencies between genders. There is slight evidence of censoring at year's end (greater length of the final weeks interval notwithstanding), particularly for women.

The migration rates in Table 1 do not distinguish persons on the basis of ALMP participation. As noted in Section L when we make the distinction, the proportion of migrants for participants of each gender is larger than for nonparticipants. While this might be indicative of a direct program impact, the possibility also exists that there is unmeasured heterogeneity at work, such that individuals with unmeasured tendencies to participate are likewise prone to migrate. Results in the next section address this issue in the context of the model described in Section IV.


Estimates of Equations (5) and (7) are presented in Tables 2-4. Table 2 presents estimates of Equation (5), the weeks participation model, for men and women. Estimates of the migration model, Equation (7), are presented in Tables 3 and 4 for men and women, respectively.
Weeks in Active Labor Market Programs 1994: Maximum Simulated
Likelihood Estimates

 Males Females
Variable Coefficient t Statistic Coefficient t Statistic

Intercept 5.179 19.65 4.282 15.58
Age -0.213 -14.29 -0.116 -7.56
Age Sq/100 0.207 10.49 0.070 3.39
Educ 2US 0.092 1.75 -0.116 -1.86
Educ 3US -0.137 -1.90 -0.161 -2.14
Educ PS -0,341 -3.80 -0.431 -4.84
Educ Univ -0.798 -6.70 -0.496 -4.81
Nordic 0.361 3.41 0.212 1.85
Non-Nordic 1.297 16.13 0.846 9.82
Disabled 2.191 44.63 1.612 28.27
East Mid 0.249 2.72 0.437 4.55
Smaland 0.472 4.69 0.829 8.22
South 0.413 4.45 0.571 5.86
West 0.394 4.70 0.609 7.18
North Mid 0.382 3.51 0.641 5.49
Mid-Norrland 0.555 4.74 0.750 5.95
Upper Noniand 0.440 3.39 0.753 6.07s
ALMP.Supp 0.208 7.58 0.136 4.52
[^.[theta]] 10.108 62.01 10.432 60.06
N 22.091 19,583

A. ALMP Participation

Estimates in Table 2 correspond to the conditional mean of weeks participation embedded in the negative binomial model. The estimates indicate a U-shaped age profile for both genders. The education coefficients reveal a reduced incidence of participation for individuals with more than 2 years upper secondary education relative to the reference category of fewer than 9 years primary schooling. The magnitudes of the coefficients indicate that the deterrent effect of education increases with the level of schooling.

A clear pattern of regional disparities emerges for both genders. Participation is higher for all regions than in the reference region of Stockholm. For men, the highest regional participation is in the mid-Norrland counties, while for women the largest difference is in Smaland. The estimates also indicate greater participation among immigrants relative to native Swedes, particularly those who originate outside the Nordic countries. After controlling for personal characteristics and regional effects, the ALMP participant supply variable is positive and significant. This is consistent with abundant anecdotal evidence of regional variation in the intensity with which ALMPs are administered. (4) Estimates of the over-dispersion parameter easily reject the null hypothesis of an absence of over-dispersion.

B. Migration: Estimates for Men

Table 3 presents the estimates of the migration equation for men. As a point of departure and a base for comparison, the first column represents a single equation model, labeled Base Model, that treats Weeks_ALMP as an exogenous variable. Column 2 estimates the same model by MSL. As described in Section IV, this approach treats Weeks_ALMP as endogenous and controls for unobserved heterogeneity. The estimated heterogeneity parameter is denoted [[^.[alpha]].sub.m] in the table. (Recall from Section III that we impose the restriction [[alpha].sub.w] = 1.) Estimates in the first column indicate reduced migration among older men and a propensity to migrate from regions with high rates of unemployment, two findings that are common in the literature. Also consistent with previous studies (Tunali 2000; Nakosteen and Westerlund 2004), there is evidence of greater mobility among men with recent histories of migration. Migration is also more likely among those with postsecondary and university schooling.
Estimates of the Migration Model: Men (a)

 Probit: MSL MSL MSL
Variable Base Model 1 Model 2 Model 3

Inlercepl -1.871 -1.888 -1.887 1.817
 (14.24) (13.07) (13.03) (8.16)

Age 0.014 -0.015 -0.015 0.015
 (7.86) (7.69) (7.45) (7.50)

Educ 2US -0.086 -0.088 -0.087 -0.087
 (1.76) (1-76) (1.71) (1.66)

Educ 3US 0.028 0.023 0.026 0.049
 (0.47) (0.39) (0.41) (0.78)

Educ PS 0.247 0.241 0.244 0.270
 (4.19) (3.98) (4.00) (4.29)

Educ Univ 0.356 0.348 0.350 0.395
 (5.83) (5.59) (5.58) (6.05)

Pnrcni 0.041 0.041 0.041 0.035
 (0.45) (0.45) (0.45) (0.37)

Migration: 1992 0.749 0.755 0.752 0.780
 (14.24) (13.50) (13.40) (12.66)

Reg Unemp 0.037 0.039 0.040 -0.014
 (2.65) (2.73) (2.78) (0.45)

Weeks_ALMP 0.008 0.006 0.006 0.004
 (6.71) (4.14) (3.42) (2.95)

[[^.[alpha]].sub.m] 0.165 0.172 0.280
 (1.53) (1.53) (2.14)

Nordic 0.037 0.067
 (0.39) (0.69)

Non-Nordic 0.071 0.138
 (1.00) (1.88)

Disabled 0.012

Hast Mid 0.430

Snuiland 0.461

South 0.287

West 0.320

North Mid 0.466

Mid-Norrland 0.446

Upper Norriand 0.489

Variable Model 4 Model 5

Inlercepl -1.796 -1.856
 (8.07) (7.82)

Age -0.015 -0.016
 (7.13) (6.78)

Educ 2US -0.089 -0.087
 (1.70) (1.62)

Educ 3US 0.052 0.053
 (0.83) (0.83)

Educ PS 0.270 0.279
 (4.31) (4.29)

Educ Univ 0.399 0.406
 (6.08) (5.93)

Pnrcni 0.036 0.034
 (0.38) (0.35)

Migration: 1992 0.785 0.780
 (12.61) (11.26)

Reg Unemp -0.01 -0.014
 (0.48) (0.44)

Weeks_ALMP 0.004 0.004
 (2.53) (1.98)

[[^.[alpha]].sub.m] 0.279 0.376
 (1.99) (2.30)

Nordic 0.070

Non-Nordic 0.147

Disabled 0.035 0.053
 (0.45) (0.65)

Hast Mid 0.422 0.443
 (4.61) (4.60)

Snuiland 0.448 0.476
 (5.51) (5.45)

South 0.282 0.295
 (2.66) (2.68)

West 0.313 0.330
 (4.06) (4.06)

North Mid 0.454 0.482
 (3.89) (3.93)

Mid-Norrland 0.433 0.461
 (4.07) (4.11)

Upper Norriand 0.478 0.503
 (3.34) (3.37)

Note: Figures in parentheses are absolute t statistics.
(a.) n = 22.091. (b.) The base model is estimated by
single equation probit maximum likelihood. The variable
Weeks_ALMP is trealed as oogenous.

The principal result of interest is the coefficient of Weeks_ALMP. For men, the estimate is positive ([^.[gamma]] = 0.008) and highly significant ([t.sub.[^.[gamma]]] = 6,71); holding other factors constant, the effect of participation is to significantly increase mobility for men.

These estimates should be viewed with caution, because the Base Model treats participation weeks as an exogenous variable and fails to correct for selective participation based on unmeasured heterogeneity in the population. The MSL procedure described in Section IV addresses both issues. The resulting estimates are presented in the second column of Table 3, which is labeled Model 1. For the control variables, the estimates are similar to their single equation counterparts in the first column. For the weeks participation variable, the estimate declines ([^.[gamma]] = 0.006) but remains significant ([t.sub. [^.[gamma]]] = 4.14). The estimated heterogeneity parameter is 0.165, which points to positive selection of migrants: individuals who for unmeasured reasons experience longer durations of participation tend to possess unmeasured propensities to migrate. However, that inference is supported only at the 0.13 level of significance.

The specification of Model 1 reflects several exclusions for purposes of identification. As an informal lest of instrument validity and robustness of the main results, we estimated additional migration models that relax several combinations of restrictions in Model I. Results of those experiments are presented in the remaining columns of Table 4. Model 2 includes the immigrant origin dummies and the disability indicator in the migration equation. Model 3 includes the immigrant dummies and region identifiers. Model 4 includes disability and the regions, and Model 5 includes all three sets of variables. In each case, the weeks participation equation remains as specified in Section V and Table 2.
Estimates of the Migration Model: Women (a)

Variable Base Model 1 Model 2 Model 3 Model Model
 Model(b) 4 5

Intercept -1.555 -1.577 -1.588 -1.575 -1.384 -1.525
 (10.45) (9.36) (9.21) (2.76) (4.44) (3.91)

Age -0.020 -0.022 -0.023 -0.031 -0.027 -0.030
 (10.62) (8.83) (9.19) (2.70) (6.66) (4.79)

Educ 2US -0.042 -0.047 -0.041 -0.074 -0.062 -0.063
 (0.74) (0.80) (0.70) (0.90) (0.92) (0.83)

Educ 3US 0.015 0.009 0.019 0.020 0.032 0.037
 (0.21) (0.12) (0.26) (0.21) (0.39) (0.42)

Educ PS 0.021 0.008 0.018 -0.0003 0.015 0.018
 (0.30) (0.1 1) (0.25) (0.001) (0.18) (0.20)

Educ Univ 0.157 0.148 0.161 0.208 0.198 0.225
 (2.18) (1.98) (2.12) (1.84) (2.25) (2.20)

Parent -0.086 -0.092 -0.094 -0.126 -0.107 -0.122
 (1.78) (1.85) (1.87) (1.59) (1.85) (1.83)

Migration: 0.659 0.681 0.688 0.872 0.761 0.842
 (11.66) (10.41) (10.36) (3.06) (7.62) (5.40)

Reg Unemp 0.032 0.036 0.036 -0.066 -0.058 -0.065
 (2.02) (2.16) (2.13) (1.31) (1.45) (1.42)

Weeks_ALMP 0.005 0.002 0.002 -0.001 -0.001 -0.002
 (3.97) (1.44) (1.09) (0.56) (0.42) (0.85)

[[alpha].sub.m] 0.296 0.347 0.929 0.638 0.866
 (1.67) (2.29) (1.37) (2.66) (2.41)

Nordic -0.022 0.032 0.028
 (0.20) (0.23) (0.21)

Non-Nordic 0.017 0.114 0.109
 (0.18) (0.86) (0.86)

Disabled 0.127 0.140 0.169
 (1.62) (1.58) (1.63)

Easl Mid 0.708 0.598 0.681
 (2.58) (4.74) (3.91)

Smaland 0.572 0.477 0.545
 (2.40) (4.08) (3.54)

South 0.373 0.307 0.354
 (1.81) (2.23) (2.16)

West 0.561 0.473 0.540
 (2.48) (4.43) (3.72)

North Mid 0.803 0.675 0.767
 (2.46) (4.24) (3.60)

Mid-Norrland 0.777 0.652 0.744
 (2.44) (4.32) (3.69)

Upper Norrland 0.895 0.761 0.862
 (2.43) (4.11) (3.50)

Note: Figures in parentheses are absolute t statistics.
(a.) n = 19,583.
(b.) The base model is estimated by single equation probit
maximum likelihood. The variable Weeks_ALMP is treated as

Results from this exercise indicate that the estimates for the control variables from Model 1 are robust across specifications, with the exception of regional unemployment. Inclusion of the region indicators in Models 3-5 renders the unemployment coefficient insignificant. Inclusion of both regional unemployment and the region dummies would be justified if the region effects captured geographic amenities, apart from employment opportunities, that induce or dissuade migration. The fact that the unemployment coefficient loses precision when the region indicators enter the model suggests that the indicators are a proxy for regional employment conditions. In any case, the results in Table 4 speak to a Model 1 that is largely robust with respect to alternative specifications.

The other supplemental variables entered in Models 2-5, indicators of nativity for immigrants and disability status, are insignificant in nearly every case. Taken together with their corresponding estimates in the participation equation (Table 2), where they are significant for both genders, they appear to be effective instruments.

Turning to the coefficients associated with participation weeks, the results in Models 1 -5 indicate that, the estimates are consistent across specifications. In particular, the results support the proposition that the effect of duration in ALMP participation is to increase mobility at the margin, holding constant measured factors such as age and schooling and after controlling for unobserved heterogeneity. There is also evidence of positive self-selection of migrants. as indicated by the heterogeneity coefficient, which is positive and significant in three of the four alternative specifications.

An issue of interest concerns the marginal effect of program participation on migration. It is well known that the estimated probit coefficients are not marginal effects of explanatory variables with respect to the probability of migration. For individual i, the marginal effect of Weeks_ALMP is obtained by the expression [^.[gamma]][[PHI]([[^.[delta]]'.sub.z.i.t] + [^.[gamma]][Weeks_ALMP.sub.i,j])], where [PHI] denotes the density function of the standard normal distribution and [^.[delta]] and [^.[gamma]] are estimated coefficients. In principle, this can be estimated for each individual in the sample. Using sample averages from Table 1 rather than individual observations, we obtain an average marginal effect of 0.0003 for the male sample. Recalling that this is the effect, ceteris paribus, of 1 week of participation at the margin, for the average participant (8.3 weeks: Table 1) this extrapolates to a marginal effect of 0.0025. Further extrapolation to 30 weeks of participation (the maximum duration of second-tier UI cash assistance) results in a marginal effect of 0.009. This suggests that the program impacts, while statistically significant, are not large in magnitude. At the same time, recognizing that migration is a somewhat rare event in the sample, this represents a nontrivial increment relative to the sample proportion of migrants. In Table 1, the proportion of migrants is 0.03, and the marginal effect estimated here at 8.3 weeks is equal to 8% of the base estimate. For the more extreme projection to 30 weeks, the marginal effect is 30% of the base.

As these estimates are based on data from 1994 and 1995, this is an effect on migration within 1 year of participation. Because migration is a decision that materializes for some individuals over a more extended period of time, the estimated marginal effect serves as a lower bound of eventual migration effects.

C. Migration: Estimates for Women

Estimates of Equations (5) and (7) for females are presented in Tables 2 and 4. respectively. Estimates for weeks participation in Table 2 are similar in direction and significance. but in some cases differ in magnitude, relative to their counterparts for males. As with males, female participation is U shaped with respect to age, is greater among immigrants, disabled individuals, and residents outside the Stockholm region. After controlling for personal characteristics and regional effects, the ALMP supply variable is again positive and significant.

Table 4 presents estimates of the migration model for females. Its format is identical to Table 3, with a single equation model followed by Model 1 and four extended specifications. The estimates for the control variables are similar in nearly every respect to the males1 estimates in Table 3, with the exception of parent status. For women, presence of children tends to reduce mobility, at levels of significance generally below 10%. For men (Table 3), there was no evidence of a deterrent effect of children.

An important contrast is in the estimated program impact. For women, there is no discernible effect of weeks participation on migration. This attests to the importance of modeling the genders separately. There is also a contrast within the female estimates that illustrates the importance of the econometric approach used here. Treating program participation as exogenous in the Base Model produces a positive estimate of program-induced mobility ([[^.[gamma]].sub.w] = 0.005; [t.sub. [gamma]] = 3.97). With endogenous participation and accounting for unmeasured heterogeneity, the estimate declines in magnitude and is not significant. That remains the case for the alternative specifications represented by Models 2-5.

It is noteworthy that the heterogeneity coefficient attains strong significance in three of the extended models and marginal significance in Model 1. Similar to the estimates from the male sample, the evidence is suggestive of positive migrant selection.

D. Alternative Approaches to Estimation

The estimates in Tables 3 and 4 are reasonably robust across alternative specifications. An additional question concerns robustness across estimation methods. One alternative is recursive two-stage residual inclusion (denoted 2SRI: Terza, Baus, and Rathouz 2008). In the first stage, we estimate a probit model in which the dependent variable is a dichotomous indicator equal to one for program participants and zero otherwise:

(9) Pr([Weeks_ALMP.sub.i,t] > 0) = [PHI]([beta]'[x.sub.i,t]).

From estimates of Equation (9) we obtain the residuals, which are then inserted as an auxiliary variable on the right-hand side of the migration probit equation:

(10) Pr([Migrate.sub.i,(t+1)] = l)

= [PHI]{[delta]'[z.sub.i,t] + [[lambda].sub.1][I(Weeks_ALMP > 0)]

+ [[lambda].sub.2][ALMP_res.sub.i,t]},

where I(Weeks_ALMP > 0) is an indicator that assigns a value of one to program participants and zero to nonparticipants, and ALMP_res is the residual obtained from Equation (9). (5)

A second alternative, which also measures program participation as a dichotomous variable, is a recursive bivariate probit model (Greene 2003, 714-19):

(11) Pr([Migrate.sub.i,(t+1)] = l) = Biv_ [PHI]{[[beta]'[x.sub.i,t], [delta]'[z.sub.i,t]

+ [[lambda].sub.1][I(Weeks_ALMP > 0)]; [rho]},

where Biv_[PHI] denotes the cumulative distribution function of a bivariate normal distribution with correlation parameter [rho].

In order to estimate the alternative models, we constructed a dichotomous variable equal to one for individuals with positive program weeks and zero for nonparticipants. This variable replaces the integer count of weeks participation and appears in the model as shown in Equations (10) and (11). Estimates of the migration equation for both models and both genders are presented in Table 5. (6) The first two columns present the 2SRI estimates, which reveal that the estimates for the control variables are generally consistent with their counterparts for the model based on Equations (5) and (7) and presented in Tables 3 and 4.
Migration and Active Labor Market Programs: Alternative
Estimates (a)

 Two-Stage Residual Recursive Probit
 Inclusion Maximum Likelihood

Variable Males Females Males Females

Intercept 2.014 -1.925 -2.004 -1.747
 (13.81) (1 1.26) (14.50) (10.91)

Age -0.012 -0.013 -0.012 -0.016
 (6.06) (5.41) (5.93) (7.06)

Edue 2US 0.077 -0.008 -0.077 -0.025
 (1.42) (0.15) (1.57) (0.43)

Educ 3US 0.053 0.067 0.050 0.043
 (0.88) (0.96) (0.83) (0.62)

Educ PS 0.296 0.107 0.292 0.061
 (4.74) (1.49) (4.76) (0.85)

Educ Univ 0.416 0.255 0.410 0.205
 (6.33) (3.34) (6.39) (2.77)

Parent 0.036 -0.098 0.036 -0.092
 (0.38) (2.00) (0.40) (1.89)

Migration: 0.731 0.634 0.730 0.629
 (13.00) (10.62) (13.85) (1 1.08)

Reg Unem 0.031 0.014 0.031 0.023
 (2.17) (0.81) (2.18) (142)

Weeks_ALMP 0.459 0.750 0.432 0.445
 (4.27) (5.05) (4.18) (3.85)

Weeks_res -0.167 -0.564
 (1.47) (3.53)

[^.[rho]] -0.080 -0.140
 (1.28) (2.00)

N 22.091 19.583 22.091 19.583

 (a.) Figures in parentheses are absolute t statistics.

For men, the estimated coefficient of program participation is positive and significant, thus corroborating this study's principal finding that ALMP participation induces labor mobility. For women, the coefficient is likewise positive and significant, which we did not find for the negative binomial-based model in Table 4.

For men, the coefficient of the residual from the participation prohit is negative but not significant. This is in contrast to the general pattern of positive heterogeneity coefficients from the model based on weeks participation (Table 3). For women, the residual coefficient is significant but negative, which again is in contrast to the heterogeneity coefficients in the negative binomial-based model (Table 4).

Columns 3 and 4 present estimates of Equation (11), the recursive bivariate probit model. The results for both Renders are consistent with estimates obtained by 2SRI. In particular, the coefficients of ALMP_Weeks again point to a significant migration-inducing effect of program participation. The estimated correlation parameter is not significant for men but negative and significant for women. This indicates an absence of migrant selection on unobservables for men and negative selection for women. Thus, both models indicate women who possess latent propensities to enroll in ALMPs are characterized by unmeasured attributes that reduce their subsequent mobility.

In order to assess the robustness of these estimates, we extended the specifications of Equations (10) and (II) to match Models (2) through (5) for the negative binomial-based model, as shown in Table 3 for men and Table 4 for women. Proceeding in this manner, we estimated four extended specifications for each migration model, resulting in a total of eight additional sets of estimates for each gender. The results, not shown in Table 5 but available from the authors on request, corroborate our principal finding with respect to ALMP participation and mobility. For men, the effect of participation is positive and significant in all but one ease. For women, the participation effect is positive and significant in three cases but not significant in the others. For the selection parameters associated with unobservables, [[^.[lambda]].sub.2] in Equation (10) for the 2SR1 approach and [^.[rho]] for the recursive probit approach, the results remain mixed. For men, the estimates are not significant (similar to Table 5. columns 1 and 3) in six cases and negative in both cases where they are significant. The same pattern is true for females.

Taken together, the estimates in Tables 3-5 support the central conclusion of this paper. For men, ALMP participation induces migration whether participation is measured by weeks duration or a dichotomous indicator. For women, on the other hand, mobility in response to participation appears only in a model that incorporates enrollment but ignores the information conveyed by weeks participation.

With respect to the issue of migrant self-selection, the estimates are sensitive to the statistical formulation. While the mixed results do not permit an unequivocal conclusion, they suggest that the issue of migrant selection is important for policy purposes and a useful item for additional research.


This study contributes to a still-developing literature on ALMP. Past studies of the Swedish experience have addressed the efficacy of programs in returning workers to the job market and restoring their wages. The general picture that emerges from existing research is that the programs have achieved mixed success in terms of increasing the probability of reemployment and reducing the duration of unemployment benefit receipt.

Previous studies have been less concerned with the extent to which program participation induces human capital investment in the form of migration. Participants gain skills that translate to more plentiful and attractive job offers. In addition, they are exposed to information about opportunities in the labor market. Moreover, employers might see an individual's program participation as a signal of greater future productivity relative to those who remain in open employment. On the other hand, participation might alter individuals' reservation wages in such a way that they refrain from the type of extensive job search that often triggers migration.

Simple descriptive statistics suggest a higher rate of migration among program participants. While it is plausible that the ALMP experience directly induces mobility, the possibility also exists that migration is a statistical manifestation of measured and unobserved traits that are associated with both behaviors. Focusing on the recession-affected years of 1994 and 1995, this paper uses a two-equation recursive model that explicitly incorporates measured covariates and unmeasured heterogeneity, treating annual weeks of program participation and migration as a sequence of choices. A second formulation ignores the duration of program attachment, instead using a dichotomous indicator for positive weeks participation. Estimates of the models suggest that program participation directly induces migration among men, even after controlling for covariates that include the individual's own history of migration. We find mixed results for women, which points to the usefulness of partitioning the sample by gender. In particular, estimates for women are sensitive to the statistical formulation of the model. When we measure program participation by weeks duration, the program impact is not significant, but we find a significant positive impact when participation is treated as a dichotomous enrollment choice. In any case, results for both genders cast doubt on ALMP participation as a deterrent to migration.

For men, we find some indications of positive association of unmeasured heterogeneity between migration and program participation as measured in weeks. These results are sensitive to specification of the migration equation and the statistical formulation of the participation equation. In particular, with participation modeled as a dichotomous state, the selection effect is negative but marginally insignificant. These discrepancies notwithstanding, the estimates suggest to future researchers the importance of caution in interpreting sample differences in mobility between participants and nonparticipants, namely that they arise in part because of latent migration propensities among participants and not solely the direct effect of participation per se.

For women, the evidence regarding self-selection is likewise mixed. When we measure participation weeks, we find positive selection in some specifications of the migration equation. However, when participation is framed as a dichotomous choice, there is evidence of negative selection, and unlike the case for men, the estimates differ significantly from zero. For both genders, the difference is likely because of the contrasting approaches to measuring program attachment. Stated differently, evidence suggesting negative selection appears when participants are identified at the intensive margin, in contrast to positive selection when program weeks are used to measure participation at the extensive margin.

These approaches refine the interpretation of "selection." When ALMP behavior encompasses the duration of participation, longer program tenure itself might be viewed as evidence of negative selection. In that context, if longer tenure is associated with migration due in part to unmeasured factors, that is not necessarily indicative of positive selection in the customary sense. In any case, for men migration is positively impacted by participation at the extensive margin, even after accounting for unobserved heterogeneity. For women, there are indications of negative migrant selection into participation at the intensive margin, while the duration of participation does not induce migration. (7)

The magnitudes of direct program effects for males reported here are not large, but it bears repeating that they convey effects only in the first year following participation. Like all investments under uncertainty, migration entails a time frame in which returns do not materialize from the outset, and thus a migration interval longer than I year might be a useful basis for future research. The results further suggest that research attempting to assess implications of labor market programs for employment and earnings should take account of worker mobility as an additional avenue through which the programs impact the market.

(1.) As a result of additional modifications in 2007, the first 7 days of unemployment are not compensated, followed by replacement rates of 80% for the next 200 days, 70% for days 201-300. and 65% thereafter. Unemployed individuals with children receive 70% replacement for a period of as many as 450 days.

(2.) In principle, one could assume a distribution for the latent factor and integrate it out of the joint probability expression that forms the likelihood function. However, for the specification used in this model, the integral does not possess a closed form solution. Consequently, estimation by means of conventional maximum likelihood methods is problematic (see. e.g., Deb and Trivedi 2006).

(3.) The model in this paper, entailing two equations and two endogenous variables, requires a large number of simulation draws. To address the resulting computational requirements, recent advances in numerical methods have been used to expedite convergence. The use of Halton sequences (Train 2003) exploits "intelligent" systematic draws instead of pseudo-random numbers. In our estimations, we initially used 1.000 Halton draws. As a check on the stability of convergence, we then increased the number of draws to 1,100. The resulting estimates were virtually unchanged from the initial version. The estimates reported in Tables 2-4 are from the latter trial.

(4.) An analogous positive impact of program supply has been reported in studies of job programs for individuals with physical and health impairments (Aakvik, Heckman, and Vytlacil 2005) and persons with developmental disabilities (McInnes, Ozturk, and McDermott 2010).

(5.) The authors thank an anonymous referee for this suggestion. Standard errors for the two-stage residual inclusion model are obtained by bootstrapping. Terza, Baus, and Rathouz (2008) establish the consistency of the two-stage probit estimator with residual inclusion.

(6.) Estimates of the probit equation for program participation are available from the authors on request.

(7.) The authors thank an anonymous referee for suggesting this idea.


Aakvik, A., J. J. Heckman. and E. J. Vytlacil. "Estimating Treatment Effects for Discrete Outcomes When Responses to Treatment Vary: An Application to Norwegian Vocational Rehabilitation Programs." Journal of Econometrics, 125,2005. 15-51.

Andren, T., and B. Gustafsson. "Income Effects and Labor Market Training Programs in Sweden during the 8(Ts and 90's." International Journal of Manpower. 25(8), 2004.688-713.

Borjas, G. J., S. G. Bronars, and S. J. Trejo. "Self Selection and Internal Migration in the United States." Journal of Urban Economics, 32(2), 1992, 159-85.

Cameron. A. C, and P. Trivedi. "Econometric Models Based on Count Data: Comparisons and Applications of Sonic Estimators and Tests." Journal of Applied Econometrics, 1(1), 1986.29-53.

Curling, K.. and K. Richardson. "The Relative Efficiency of Labor Market Programs: Swedish Experience from the 1990's Labour Economics, 11(3). 2004. 335-54.

Deb, P., C. Li, P. K. Trivedi, and D. M. Zimmer. "The Effects of Managed Care on Use of Health Care Services: Results from Two Contemporaneous Household Surveys." Health Economics. 15, 2006, 743-60.

--. "'Managed Care, Self Selection, and Health Care Utilization." Econometrics Journal. 9(2), 2006, 307-31.

Delander. L.. and J. Mansson. Eorensk: Evaluation: A Strategy for and Results of an Impact Evaluation of a Universal Labor Market Program; The Swedish Activity Guarantee, Berlin: Social Science Research Center. 2007.

Fredriksson. P. "The Dynamics of Regional Labor Markets and Active Labor Market Policy: Swedish Evidence." Oxford Economic Papers. 1. 1999.623-48.

Fredriksson, P., and P. Johansson. "Employment. Mobility, and Active Labor Markel Programs." Institute for Labour Market Policv Evaluation Working Paper 2003:3. Uppsala. 2003.

Gourieroux, C and A. Monfort. Simulation Econometric Methods. New York: Oxford University Press, 1996.

Greene, W. Econometric Analysis. Upper Saddle River, NJ: Prentice Hall. 2003.

Hamalainen, K. "Unemployment. Selective Employment Measures and Inter-regional Mobility of Labor." Papers in Regional Science, 81. 2002. 423-41.

Jaeger, D, H. Bonon. T. Dohmen, A. Falk. D. Huffman, and U. Sunde. "Direct Evidence on Risk Attitudes and Migration.'1 1ZA Discussion Paper 2655, 2007.

Kluve, J. "The Effectiveness of European Active Labor Market Policy." IZA Discussion Paper 2018, 2006.

Larsson. L. "Evaluation of Swedish Youth Labor Market Programs." Journal of Human Resources.. 38(4). 2003, 891-927.

Lindgren, U., and O. Westerlund. "Labour Markel Programmes and Geographical Mobility: Migration and Commuting among Programme Participants and Openly Unemployed." Institute for Labour Market Policy Evaluation Working Paper 6. Uppsala, 2003.

Maier. G. "Job Search and Migration." in Regional Labour Markets: Analytical Contributions and Cross-National Comparisons, edited by M. Fischer and P. Nijkamp. Amsterdam: North Holland Publishing Company. 1987.

Mclnncs, M. M., O. D. Ozturk. and S. McDermott. "Does Supported Employment Work?" Journal of Policy Analysis and Management, 29(3). 2010. 506-25.

Nakosteen, R., and O. Westerlund. "The Effects of Regional Migration on Gross Income of Labour in Sweden." Papers in Regional Science, 83, 2004, 581-95.

Pickles, A., and P. Rogerson. "Wage Distributions and Spatial Preferences in Competitive Job Search and Migrated Regional Studies, 18(2). 1984. 131-42.

Polachek. S., and F. Horvath. "A Life Cycle Approach to Migration: Analysis of the Perspicacious Perigrinator," in Research in Labor Economics, edited by R. Ehrenberg. Greenwich. CT: JAI Press. 1977.

Sianesi, B. Swedish Active Labour Market Programmes in the l99(Vs: Overall Effectiveness and Differential Performance, London: Institute for Fiscal Studies WP 02-03, 2002.

--. "An Evaluation of the Swedish System of Active Labor Market Programs in the 1990's." The Review of Economics and Statistics, 86(1). 2004. 133-55.

--. Differential Effects of Active Labour Market Programs for the Unemployed. London: Institute for Fiscal Studies WP0I-25. 2005.

Sjaastad. L. "The Costs and Returns of Human Migration." Journal of Political Economy, 370, 1962, 80 93.

Terza, J., A. Basu. and P. Rathouz. "Two Stage Residual Inclusion Estimation: Addressing Endogeneity in Health Econometric Modeling." Journal of Health Economics, 27(3), 2008, 531-43.

Train. K. Discrete Choice Methods with Simulation. New York: Cambridge University Press, 2003.

Tunali. I. "Rationality of Migration." International Economic Review. 41(4), 2000, 893-920.

Westerlund, O. "Employment Opportunities. Wages and Interregional Migration in Sweden 1970-1989." Journal of Regional Science, 37, 1997. 55-73.

* The authors wish to thank David Zimmer for his extensive assistance and advice regarding the programing and estimation issues, Earlier versions of this paper were presented at the Symposium on Population Dynamics in the Cireumpolar North, Umea University. Sweden: June 2008; Research Colloquium. Slatislics Denmark. Copenhagen: June 2008; and the 45th Conference of the Missouri Valley Economic Association: October 2008.

Nakosteett: Isenherg School of Management. University of Massachusetts, Amherst. MA 01003. Phone 413-545-5687. Fax 413-545-3858. E-mail

Weslcdund: Department of Hconomics, Umea Univcrsity. SK-901 87 Umea. Sweden. Phone +46-90-786 6148. Fax +46-90-77 23 02. E-mail

Zimmer: Schroeder Family School of Business Administration, University of Evansville. Evansville. IN 47722. Phone 812-279-2864. Fax 812-488-2872. E-mail


ALMP: Active Labor Market Program

GDP: Gross Domestic Product

LMA: Labor Market Area

MSL: Maximum Simulated Likelihood

PMF: Probability Mass Function

UI: Unemployment Insurance

COPYRIGHT 2012 Western Economic Association International
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 2012 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Nakosteen, Robert A.; Westerlund, Olle; Zimmer, Michael
Publication:Contemporary Economic Policy
Article Type:Report
Geographic Code:4EUSW
Date:Apr 1, 2012
Previous Article:Taxes, health insurance, and women's self-employment.
Next Article:The cost of land use regulation versus the value of individual exemption: Oregon Ballot measures 37 and 49.

Terms of use | Privacy policy | Copyright © 2020 Farlex, Inc. | Feedback | For webmasters