Abstract
Policymakers often struggle with addressing urban pockets of distress, and place-based policies are increasingly discussed as a potential remedy. This article evaluates the impact of such a policy: the Metropolitan Development Initiative implemented by the Swedish government. The initiative included various interventions such as job search assistance, career counseling, and skills development programs, aiming to increase the labor supply of targeted residents—a type of place-based policy that has been scarcely researched. To assess the effects of the initiative, this study employs canonical difference-in-differences, supplemented by interaction-weighted estimators. The analysis focuses on labor participation as the key outcome variable, comparing individuals from targeted neighborhoods to a control group. The findings reveal substantial, significant, and enduring effects on foreign-born males, who on average exhibit a 3.3 percentage point increase in their probability of labor participation, with estimates remaining significant across all post-treatment periods. However, the results for foreign-born females are less definitive.
Keywords
Introduction
Growing income inequalities have increasingly become a pronounced societal and political issue in countries across the world (Chancel et al. OECD 2024). Income can broadly be divided into two main types: income from capital and income from labor. In Western economies, the share of national income derived from capital has risen significantly since the 1980s, driven largely by the fact that capital returns have persistently outpaced economic growth. This trend, coupled with an increasing concentration of capital among the wealthiest segments of society, has contributed to rising income and wealth inequality (Nolan, Richiardi and Valenzuela 2019; Stiglitz 2015). Labor income inequality has also increased in most countries, though its evolution has been more heterogeneous. For example, in the United States and the United Kingdom, the share of national income accruing to top wage earners has grown substantially, reflecting a rise in the bargaining power of high-income individuals and a weakening of institutional mechanisms that once compressed wage disparities (Piketty 2020). In contrast, Sweden has seen a more modest increase in top wage shares, though labor income inequality has grown due to policy shifts, particularly during and after the country's economic crisis of the 1990s. Large cuts in public spending, tax reforms that reduced progressive elements, and the privatization of public enterprises contributed to a more unequal wage distribution (Therborn 2020). However, since the late 2000s, labor income inequality in Sweden has remained relatively stable, with wage growth occurring at a similar rate across different income groups. An important development, however, is the growing gap in income between those who are employed and those who are not, reflecting changes in welfare policies and labor market conditions (The Swedish Fiscal Policy Council 2024). The groups that face less stable employment and higher income volatility are immigrants and women, both of whom exhibit weaker labor market attachment and larger fluctuations in annual hours worked (Friedrich, Laun and Meghir 2022). Immigrants with weak labor market attachment and low education are also increasingly concentrated in specific neighborhoods, creating “urban pockets of distress” (DELMOS 2022).
This has exacerbated spatial inequalities, prompting policymakers to explore place-based interventions to improve labor market prospects for residents in disadvantaged neighborhoods (Redar et al. 2024). These types of policies are fundamentally regarded as instruments to address spatial inequalities and are increasingly prominent in international policy discussions, informing deliberations on their appropriate use (McCann 2023). A crucial aspect of their implementation involves identifying the suitable geographic scale to achieve their objectives. Active Labour Market Policies (ALMPs), such as skills enhancement and job-search assistance, are considered well-suited to be implemented at the local or neighborhood level (Green 2023). However, only a few studies have examined the effects of place-based policies aimed at improving local labor supply (Christensen 2015; Jump and Scavette 2022). Previous research on place-based labor market policies has mainly focused on demand-side interventions, such as tax cuts for firms (Bartik 2023; 2022; Neumark and Simpson 2015). Furthermore, there is reason to believe that the effectiveness of labor supply-oriented policies varies significantly depending on labor market conditions and the general policy environment. Sweden has a long history of using space-blind ALMPs (Kluve 2010; Sianesi 2001) and therefore, the Swedish Metropolitan Development Initiative (MDI) presents an opportunity to evaluate the effects of highly localized labor supply augmenting programs implemented amidst a national policy landscape saturated with universal ALMPs.
The purpose of this study is, therefore, to use the MDI-case to examine the impact of place-based labor supply-augmenting policies on the probability of labor participation among the targeted population. The MDI was the Swedish government response to growing concerns about the geographic concentration of socioeconomic issues and immigrant populations. The initiative was comprised of place-based programs aimed at vulnerable neighborhoods and marked a departure from Sweden's conventional reliance on space-blind policies to address unemployment and social exclusion (Andersson 2006; Andersson, Bråmå and Holmqvist 2010). Local development agreements (LDAs) between seven municipalities and the national government identified 24 disadvantaged neighborhoods for special interventions. The government allocated approximately two billion SEK over a four-year period (1999–2003), with substantial funding directed toward improving labor market outcomes for immigrants (Arbetsmarknadsdepartementet 2005). MDI projects focusing on labor market outcomes included job search assistance, career counseling, entrepreneurship promotion, language proficiency enhancement, workplace training, and skills development, all designed to improve the quality and quantity of the residents’ labor supply. Concurrent with the implementation of the MDI, several ALMPs were in place, including adult vocational training programs (Albrecht, Van den Berg and Vroman 2005), private sector subsidized employment schemes (Forslund, Johansson and Lindqvist 2004), trainee replacement schemes (Sacklén 2002), and job-search assistance, all targeting the labor force as a whole. As such, this context offers an opportunity to evaluate whether place-based policies that enhance labor supply can complement broader active labor market programs. Specifically, it is an ideal setting to investigate whether tailored interventions, such as those implemented under the MDI, can yield significant outcomes in an environment already offering comprehensive job-finding and training programs.
Previous studies on the MDI have rarely employed quantitative methods to measure outcomes for targeted populations and have not made causal claims. Quantitative studies have focused on aggregate neighborhood outcomes within the first year of the intervention, without long-term analysis (Arbetsmarknadsdepartementet 2005; Giertz 2004; Integrationsverket 2002). In addition to this, recent literature reviews indicate that quantitative assessments of the impacts of place-based policies remain a relatively underexplored field in Sweden (Gregorowicz-Kipszak, Bröchner and Hagson 2022; Roelofs and Salonen 2019).
This study aims to address these research gaps by tracking the labor market trajectories of individuals in the targeted neighborhoods over six years following the intervention, using a two-way fixed-effects difference-in-differences (DiD) methodology complemented by interaction-weighted estimators as recommended by Sun and Abraham (2021) for causal inference. As the Swedish policy debate on segregation increasingly revolves around implementing new place-based policies (e.g., Kopsch 2024; Redar et al. 2024) and the government has begun to allocate funds toward identifying geographical areas of socioeconomic vulnerability (Prop. 2023/24:1 2023), this research seeks to provide new, causal evidence on the medium to long-term effects of labor supply-augmenting place-based policies in an environment already saturated with space-blind active labor market programs. This constitutes the primary contribution of this study to both international research and policymaking.
The following section provides a brief overview of the literature on place-based policies and active labor market policies. Next, the Swedish policy context and the intervention of interest, the MDI, are described. The subsequent section outlines the data and methods used in this study. Following this, the results of the analysis are presented. Finally, the implications of this study are discussed, and conclusions are drawn.
Place-Based Policies and Active Labor Market Policies
Ladd (1994) categorizes policies for addressing pockets of urban distress into three overarching strategies: people-oriented strategies, place-based people strategies, and pure place-based strategies. People-oriented strategies involve space-blind assistance initiatives, such as welfare programs or universal ALMPs, which provide similar aid regardless of geographical location. In contrast, place-based people strategies tailor interventions to specific geographic areas, encompassing measures like local tax credits that incentivize employers to hire workers residing and working within designated zones, along with policies aimed at improving labor market access for residents in targeted locales. Pure place-based strategies focus primarily on the geographic areas themselves rather than the inhabitants therein, addressing both physical and economic attributes. The three types of strategies can vary in scale, spanning from interventions aimed at entire regions to those that operate at a much more localized level such as neighborhoods (Green 2023). The MDI falls under the category of place-based people strategies and was executed both at the level of metropolitan regions and at the neighborhood level. At the regional level, it was comprised of Regional Growth Agreements between the government, the regions, and private enterprise with the aim of increasing cooperation between them. At the neighborhood level, the initiative took the form of LDAs between the government and municipalities that focused on improving the socioeconomic outcomes of neighborhood residents.
Place-based people strategies aimed at increasing employment can simultaneously be defined as highly localized ALMPs and these can be classified into four primary types (Kluve 2010): (1) Training programs, which enhance human capital, thus increasing productivity and employability. Common examples include skill-enhancement through education and practical training. These types of policies improve labor supply. (2) Private sector incentive schemes, which typically involve subsidies to firms (augmentation of labor demand), financial incentives to workers (increasing labor supply), and grants to support unemployed individuals in establishing their own businesses (also increasing labor supply). The overarching goal is to stimulate employment by encouraging firms to hire workers or retain existing employees, as well as alleviate credit constraints faced by potential entrepreneurs. (3) Direct employment programs, which create jobs in the public sector and thereby increase labor demand. The objectives include sustaining labor market participation among disadvantaged groups and mitigating potential human capital depreciation resulting from prolonged periods of unemployment. (4) Services and sanctions, which are designed to enhance job search efficiency, such as job search assistance coupled with penalties for noncompliance with program requirements. These types of policies aim to increase labor supply.
Although ALMPs historically have been space-blind, it is essential to understand the magnitude of their effects to assess the potential of their place-based variants. In a meta-analysis by Card, Kluve and Weber (2018), which includes 207 studies and 807 estimates, the authors summarize the effects of various ALMP program types on employment probability. The findings indicate that training programs and private sector incentive schemes have modest positive effects in the short term, but these effects become more pronounced in the medium to long term. This temporal variability is likely due to “lock-in” effects, where participants are temporarily removed from active job search activities during the human capital augmentation phase of the program. In contrast, public direct employment programs show limited or even negative effects. Job search assistance programs demonstrate relatively constant positive effects over time, as they focus on matching individuals with employers rather than enhancing the job seekers’ human capital, as is the case with training programs. The average effect of ALMPs on employment probability ranges from 1 to 3 percentage points in the short run, 3 to 5 percentage points in the medium run, and 5 to 12 percentage points in the long run (Card, Kluve and Weber 2018). In the meta study, the effects vary across groups, with larger impacts observed for females and the long-term unemployed, and smaller effects for youths and older job seekers. A comparison between nonexperimental designs and randomized experiments reveals minimal differences in their estimated average effects.
The greater impact of ALMPs on women is particularly relevant given persistent gender disparities in labor market outcomes. Women generally exhibit lower employment rates and wages compared to men, driven by occupational sorting into lower-paid sectors, firm and industry selection, and structural constraints linked to family responsibilities. While the precise mechanisms behind these disparities remain subject to debate, societal norms and gendered expectations regarding caregiving responsibilities play a significant role. Countries with more conservative gender norms tend to have lower female employment rates. Additionally, gender differences in commuting patterns, particularly following the birth of a first child, suggest that women's employment opportunities are geographically more restricted due to preferences for shorter commutes and greater work-time flexibility (Petrongolo and Ronchi 2020). As for differences across age groups, reemployment probabilities after a job loss decline significantly with age, particularly beyond the age of 55, as older workers face greater barriers to labor market reintegration. Older individuals are also more likely to exit the labor force entirely following job loss, indicating structurally weaker employment prospects (Öylü, Motel-Klingebiel and Kelfve 2024). This pattern aligns with evidence on the effects of ALMPs on unemployed individuals over 50, who generally experience negligible or even negative impacts. However, effects depend on the type of intervention, with training programs producing possible positive effects after four to five years. The limited effectiveness of ALMPs for older job seekers highlights the distinct labor market challenges faced by this group and points toward the need for targeted interventions that address barriers such as skill obsolescence and employer discrimination (Petrongolo and Ronchi 2020).
The history of place-based people strategies has been the subject of extensive research (e.g., Bartik 2020; Neumark and Simpson 2015). The MDI shares similarities with several British initiatives, such as the region-wide City Deals and neighborhood-oriented interventions such as the New Deal for Communities and the Neighbourhood Renewal Fund, and with the Urban Committee program in Denmark. The City Deals scheme involved granting increased financial autonomy for certain metropolitan regions to foster conditions conducive to economic growth (Alonso and Andrews 2023). The New Deal for Communities entailed neighborhood committees formulating projects in the areas of employment, crime, education, health, and the environment. The employment programs focused on increasing labor supply and demonstrated positive effects on the probability of employment among jobless individuals, particularly those who had invested in education prior to the program's initiation (Romero 2009). The Neighbourhood Renewal Fund was a larger-scale successor to the New Deal for Communities, establishing Local Strategic Partnerships between municipalities, authorities, and civil society in England's most deprived districts. This program focused on worklessness, crime, health, and education, with employment-increasing projects aimed at boosting labor supply among targeted populations. Results indicate significant positive impacts on local employment rates and a reduction in the number of individuals receiving unemployment benefits, with faster reductions seen among younger demographics (Jump and Scavette 2022). Additionally, the intervention succeeded in reducing property and violent crimes in the targeted neighborhoods (Alonso, Andrews and Jorda 2019). Evaluations of the Danish Urban Committee program examine the effects of this intervention on individual labor market outcomes and neighborhood social mix. The program targeted distressed housing estates and aimed to improve residents’ skills and qualifications through networking and information about education and employment opportunities. The results show positive effects on employment among the targeted individuals, particularly men. However, the program did not significantly impact the social mix in the neighborhood. Instead, the study reveals significant selective migration, with individuals of higher socioeconomic status leaving the neighborhood, while those of lower socioeconomic status moved in (Christensen 2013; 2015).
In summary, previous studies indicate that standard, space-blind, ALMPs tend to have more pronounced effects on females and long-term unemployed individuals, with smaller impacts on both older and younger cohorts. However, research on place-based policies (PBPs) aimed at increasing labor supply presents slightly different findings. These studies highlight that males benefit the most from such interventions, suggesting differences in impact across sexes, so it remains unclear which sex is most affected overall. Regarding the effects across age cohorts, there appear to be differences between PBPs and space-blind ALMPs, particularly concerning the number of benefit claimants. PBPs show larger reductions in the number of benefit claimants among younger individuals. However, the studies on ALMPs do not examine the impact on benefit claimants but rather focus on employment outcomes. Thus, direct comparisons are not feasible. Nonetheless, the discrepancy is noteworthy, as the number of benefit claimants is likely negatively correlated with employment. Further evidence on the effects of PBPs indicate that those most helped by them are those who have previously invested in their skills through education, vocational training, or similar activities.
Apart from the studies mentioned above, there is a scarcity of research on the effects of place-based policies aimed at increasing labor supply. Bartik (2023) highlights this gap, noting that much of the existing literature focuses on place-based strategies targeting the demand side of the labor market, such as offering tax incentives to firms or subsidies for hiring residents of specific neighborhoods.
Regarding research on the MDI itself, numerous evaluations have explored various dimensions. These include examinations of labor force participation and reliance on welfare benefits (Axelsson 2004; Bevelander et al. 2004; Giertz 2004; Hosseini-Kaladjahi 2002; Kihlström and Birger 2004), enhancement of Swedish language proficiency (Axelsson, Lennartsson-Hokkanen and Sellgren 2002; Bak et al. 2004), interventions in educational institutions (Bak et al. 2004), neighborhood development and safety (Ländin 2004; Lindahl and Mattsson 2003), public health (Marttila and Tillgren 2002), democratic engagement (Andersson et al. 2004; Integrationsverket 2002), and the organizational dynamics and collaboration among different stakeholders involved in implementing the initiative (Bunar 2004; Hosseini-Kaladjahi 2002; Jensen 2004). Andersson and Bråmå (2004) examine the dynamics of resident turnover in the 16 neighborhoods within the Stockholm region that were the focus of the MDI from 1990 to 2000, the decade preceding the initiative. Their findings reveal a selective migratory pattern they describe as “middle-class leakage.” Consequently, the authors argue that “The intervention might be successful in terms of assisting residents in finding jobs and better education, but that might not improve the general position of the areas targeted, since people who make a socioeconomic career very often move out of the areas, to be replaced by poorer and less well-established residents” (Andersson and Bråmå 2004, 1). In essence, while the authors suggest that the interventions may benefit individuals, they do not assess whether these interventions actually had an impact or if they influenced individuals to relocate. They merely indicate that the selective migration patterns create unfavorable conditions for policies aimed at benefiting individuals to also enhance the overall quality of the targeted areas.
Among the MDI-evaluations focusing on labor force participation and welfare dependence, two endeavors aim to quantify the development of neighborhood employment rates (Giertz 2004; Integrationsverket 2002). However, these analyses primarily concentrate on changes at the neighborhood level, limit themselves to descriptive analysis, and solely examine developments within the initial year of project implementation (year 2000). In addition to this, recent literature reviews indicate that quantitative assessments of the impacts of place-based policies remain a relatively underexplored field in Sweden (Gregorowicz-Kipszak, Bröchner and Hagson 2022; Roelofs and Salonen 2019).
The Swedish Context
Sweden is a noteworthy case in the global landscape of ALMPs due to its extensive history of implementation across all four types (Training programs, Private sector incentive schemes, Direct employment, Services and sanctions) (Kluve 2010). Here, a brief overview of the policies in effect around the time of MDI is provided. One prominent initiative was the “Knowledge Lift” an adult education program operational from 1997 to 2002, targeting low-skilled adults with the aim of enhancing their general skills through classroom instruction, often coupled with vocational training opportunities. Although the program achieved widespread participation, its impact on average income or employment for women was negligible, while men experienced significant improvements in their transition to employment (Albrecht, Van den Berg and Vroman 2005). During the same period, a private sector incentive scheme targeted long-term unemployed individuals, offering wage subsidies covering half of total wage costs for up to six months. Evaluation by Forslund, Johansson and Lindqvist (2004) indicates positive effects on employment probabilities postprogram participation. Additional ALMPs during this period included vocational employment training programs focusing on skill development in sectors facing labor shortages (Richardson and van den Berg 2001), and trainee replacement schemes subsidizing training and hiring costs for employed workers and their substitutes (Sacklén 2002). Basic ALMPs, such as training courses provided by employment agencies, also formed part of the policy landscape (Andrén and Andrén 2002). This overview underscores that “In Sweden nobody is really left ‘untreated’” (Sianesi 2001, 9), demonstrating that the MDI was implemented in an environment saturated with space-blind ALMPs that targeted the labor force in general.
The Metropolitan Development Initiative
The inception of the MDI can be traced back to the economic turmoil of the 1990s in Sweden, characterized by escalating unemployment rates and a shift in post-war fiscal policy from prioritizing full employment and growth to prioritizing inflation control and debt reduction. Policy debates at the time placed much emphasis on rising social and ethnic segregation in the metropolitan regions and how increasing spatial social mix would be conducive to growth and better life-outcomes for citizens (Musterd and Andersson 2005). The MDI was the incarnation of the proposals outlined in the Metropolitan Proposition (Socialdepartementet 1998), which recognized major urban areas as pivotal for future economic advancement. Consequently, the MDI focused on two primary objectives: firstly, to foster favorable conditions for growth in major urban regions, thereby stimulating job creation within these regions and across the nation; and secondly, to address social and ethnic segregation in urban areas, striving for equitable living standards among residents (Arbetsmarknadsdepartementet 2005).
The implementation process of the MDI can be delineated as follows: the government introduced a mechanism called the “local development agreement” (LDA), forming contracts between the national government and willing municipalities. However, not all municipalities were eligible for LDAs; rather, the government identified municipalities most in need, primarily by pinpointing those with neighborhoods with high concentrations of immigrants, welfare recipients, and unemployed (Arbetsmarknadsdepartementet 2005; Integrationsverket 2002). These identified municipalities would then engage in negotiations with the government to sign LDAs. Agreements were established between 1999 and 2000, with end dates spanning from 2003 to 2005. In total, seven municipalities and 24 neighborhoods were impacted by LDAs and the contracts specified the allocation of funds according to project type in designated neighborhoods. The government allocated 2 billion SEK over the years 1999 to 2003 for MDI projects, with municipalities required to match these funds, bringing the total investment to around 4 billion SEK (Arbetsmarknadsdepartementet 2005; Integrationsverket 2002).
The MDI encompassed diverse project categories, including support for local development, cultural and sports activities, preschool initiatives emphasizing language development, as well as active labor market policies. Initially, the focus was on local development, but in 2001, the remaining government funds (approximately 600 million SEK) were redirected toward active labor market policies (Axelsson 2004; Bunar 2011; Hosseini-Kaladjahi 2002). Given this prioritization of growth and employment, analyzing the effects of such policies becomes particularly pertinent. The labor market measures within the MDI encompassed a range of initiatives including language training, motivational interviews, computer skills enhancement, job search training, career guidance, networking, business development, leadership training, apprenticeships, vocational training, and collaboration between employment offices and city districts. These efforts primarily targeted immigrants and fall under the ALMP categories Training programs, Private sector incentive schemes, and Services and Sanctions. They can thus be characterized as attempts to increase the labor supply of residents in the targeted neighborhoods. Project implementation commenced in 2000 in Stockholm, Huddinge, Haninge, Botkyrka, Södertälje, and Malmö, and expanded to Gothenburg in 2001 (Andersson et al. 2004; Arbetsmarknadsdepartementet 2005; Hosseini-Kaladjahi 2002; Integrationsverket 2002).
Data and Empirical Strategy
The study initiates with administrative register data encompassing the entire population aged 16 and above in the three Metropolitan Labour Markets (Stockholm, Gothenburg, and Malmö) seen in Figure 1. The map shows all Demographical Statistical Areas (DeSO) as defined by Statistics Sweden, with those pertaining to the Metropolitan Labour Markets marked in gray. Given the focus of this study on labor market outcomes among unemployed immigrants, foreign-born individuals registered as unemployed with the employment agency form the pool from which both treated and control individuals are selected. Treated individuals are those foreign-born persons registered as unemployed and residing in a treated area during the year when the MDI commenced in that specific location. The control group is made up of those foreign-born individuals who are unemployed and residing elsewhere in the Metropolitan Labour Markets. A key limitation of this study is the absence of administrative records identifying individuals who directly participated in the intervention. Instead, we can only infer the population that was highly likely to be offered participation based on their place of residence. Given that enrollment in the program was largely voluntary, we are able to identify the targeted group—namely, unemployed immigrants residing in designated neighborhoods—but not the subset that actually received the intervention. Consequently, the estimated effect in this study reflects an intent-to-treat (ITT) effect rather than the treatment effect on the treated (TOT). This distinction is important, as it implies that the true causal effect of the intervention on participants cannot be directly observed. Nonetheless, despite this limitation, the available data and methodology enable a meaningful contribution to the literature on place-based active labor market policies, providing new knowledge even if some questions remain unanswered.
At least one report provides data on participant numbers at a specific job search and training center, the Work and Development Centre (Sw. Arbete- och utvecklingscentra) in Malmö (Bevelander et al. 2004). Between 2000 and 2002, the center registered 7,606 participants, of whom 65 percent were foreign-born. During this period, the overall share of foreign-born individuals in Malmö's local labor market was 17 percent, indicating a substantial overrepresentation of foreign-born individuals among program participants. At the onset of the MDI in 2000, 2,671 individuals were registered at the center. Although disaggregated yearly data on the foreign-born share of participants is unavailable, assuming little variation across years, an estimated 1,740 foreign-born individuals were enrolled in the program in 2000. In comparison, the sample data indicate that 3,314 foreign-born individuals were registered as unemployed in the targeted neighborhoods of Malmö in the same year. Based on these figures, the program participation rate among foreign-born unemployed individuals in the targeted areas could be as high as 52 percent. However, this estimate relies on the strong assumption that participation was strictly limited to registered residents of the targeted areas, with no exceptions. If individuals from outside the designated neighborhoods were allowed to participate, the actual participation rate among the targeted population would be lower than this estimate suggests. Appendix Figures A2 to A4 show each metropolitan labor market, the treated neighborhoods, and the number of control individuals selected from each other neighborhood.
Map of DeSO Areas in Sweden with Metropolitan Labour Markets in Grey. Appendix Contains Three Additional Maps Focusing on Each Metropolitan Labour Market and the Targeted Neighborhoods therein.
Table 1 presents the group characteristics of treatment and control groups at the year of treatment initiation, along with the differences between them. The analytical sample consists of 16,083 individuals in the treated group and 48,871 individuals in the control group, yielding a total of 64,954 individuals. While the groups exhibit similarities in some characteristics, notable differences exist in key variables such as birthplace, years since immigration, and educational attainment. Among the treated individuals, 4.4 percent were born in Nordic countries (excluding Sweden), compared to 13.7 percent in the control group. Similarly, 12.3 percent of the treated group has resided in Sweden for more than twenty years, whereas the corresponding figure for the control group is 26.5 percent. Additionally, 31 percent of the treated group has received nine or fewer years of schooling, compared to 20.9 percent in the control group. These differences indicate that treated individuals are, on average, more likely to have been born further from Sweden, to have arrived recently, and to have lower educational attainment than their control counterparts. Given these differences, an important concern is whether the control group serves as a valid counterfactual for assessing the outcomes of the treated individuals. To evaluate the comparability of the groups, a common support analysis was conducted as recommended by Angrist and Pischke (2009) and Caliendo and Kopeinig (2008) among others. This involved estimating propensity scores using a logistic regression model, incorporating all observable covariates to predict treatment assignment. The distribution of propensity scores was then examined to identify individuals who lack comparable counterparts in the opposite group (see Figure A1 in Appendix). This analysis identified 43 individuals (40 from the control group and 3 from the treated group) as falling outside the region of common support. To assess the potential impact of these individuals on the estimates, an additional analysis was conducted using a trimmed sample (excluding these 43 individuals). The results from this alternative specification remained virtually unchanged in magnitude and statistical significance, indicating that their exclusion does not materially affect the findings. Consequently, the full sample is retained in the main analysis.
Characteristics of the Treated and the Control Group.
*A large portion of residents do not receive this type of income, mean value is therefore biased toward 0.
The staggered adoption of the MDI across municipalities necessitates the use of relative time to treatment rather than calendar time to ascertain the treatment status of treated individuals. Consequently, the initial sample is restricted to individuals observed throughout the entire period from 5 years prior to treatment to 5 years post-treatment. In total, there are 11 periods (−5, 0, 5) and while the final sample is balanced in relative time, it remains unbalanced in calendar time. The exclusion of individuals who are not observed during the five years preceding and following treatment may introduce some degree of sample selection bias. Specifically, individuals who are very young at the time of treatment (e.g., those aged 18) are systematically excluded, as the earliest age of inclusion in the dataset is 16. Consequently, the effective lower age boundary of the sample is 21 years at the time of treatment. Similarly, individuals who immigrated to Sweden shortly before the intervention (e.g., within three years of implementation) are also excluded, effectively imposing a minimum residency requirement of five years in Sweden. These sample restrictions may lead to the omission of potentially policy-relevant effects, particularly among recently arrived immigrants, a group that might be among the most in need of such interventions. However, these exclusions are necessary for methodological reasons. In order to establish causal inference, the key assumption of parallel trends must be validated, which requires observing the pretreatment trajectories of both treated and control groups. Since individuals who are not present in the dataset for a sufficient period cannot contribute to this validation, their exclusion is necessary to maintain the credibility of the empirical strategy. Numerous studies indicate that estimates derived from Two-Way Fixed Effects (TWFE) regressions may be compromised when encountering heterogeneous treatment effects or staggered implementation of treatment (Callaway and Sant'Anna 2021; Goodman-Bacon 2021; Sun and Abraham 2021). The conventional TWFE approach produces treatment effect estimates as a weighted average of all potential two-by-two DiD comparisons among groups treated at different times (Wing et al. 2024). Consequently, these estimates are reliable only under the assumption of homogeneous treatment effects and consistent treatment implementation. However, in the context of the MDI, the policy's implementation varied across municipalities, making it inappropriate to assume treatment effect homogeneity. The MDI encompassed diverse projects with distinct characteristics (e.g., job search assistance and skills development), each proven to yield different treatment effects (Card, Kluve and Weber 2018). Moreover, project initiation within municipalities exhibited variation, with some municipalities initiating projects early in the adoption year while others commenced later in the same year. Such discrepancies could yield apparent heterogeneous treatment effects.
To examinate potential contamination of event study estimates, canonical TWFE regressions are conducted and their estimates subsequently compared with interaction-weighted (IW) estimators following the approach proposed by Sun and Abraham (2021). The TWFE regression model employed in this study is defined as follows:
where
Results
This section presents the results from a DiD analysis that estimates the effect of a place-based policy on labor market participation among foreign-born individuals. The analysis relies on a quasi-experimental design, a widely used approach in the social sciences when randomized controlled trials are not feasible. By comparing changes in employment outcomes over time between treated and control groups, DiD allows for credible causal inference under the assumption that both groups would have followed similar trends in the absence of the intervention. Given that program participation was not randomly assigned and affected individuals may differ systematically from those in the control group, quasi-experimental methods are particularly well-suited for evaluating interventions of this kind.
The results reveal meaningful differences in how the intervention affected labor market participation across sex and age cohorts. Among men, the policy appears to have increased employment probabilities, particularly for those aged 30 to 39, while effects were more limited or statistically inconclusive for older cohorts. For women, the estimated effects were generally smaller, though significant improvements were observed for younger cohorts, especially those under 30. In both cases, causal interpretation is only possible where pretreatment trends were sufficiently parallel across groups. Robustness checks using interaction-weighted estimators confirm the main findings and help address concerns related to differential treatment timing and heterogeneity in effects. Taken together, the results suggest that the intervention had a positive but uneven impact on labor market integration, with the strongest effects concentrated among prime-aged males and younger females.
Table 2 presents the results of the TWFE regression analysis for foreign-born individuals by sex and age cohort. The regression estimates the probability of labor market participation across relative periods ranging from −4 to 5, with relative period −1 omitted as the base year and period −5 omitted due to multicollinearity. Throughout this section, pretrends refer to cases where pretreatment trajectories differ between treated and control groups, challenging the parallel trends assumption.
Regression Output, Two-Way Fixed Effects.
Standard errors are omitted in Appendix for full regression output.
*** p < .01, ** p < .05, * p < .1.
The model includes control variables identified in previous research as key determinants of employment probability, including age, marital status, number of children, educational level, and enrollment in education. To account for nonlinearities in age, age squared is included, reflecting the assumption that the probability of labor participation declines as individuals approach retirement. Additionally, several confounding factors are controlled for in the analysis. Registered unemployment, defined as a binary indicator for whether an individual was unemployed during the observed year, accounts for frequent transitions in and out of ALMPs, as noted in previous evaluations of the MDI (Hosseini-Kaladjahi 2002). Welfare receipt, also a binary indicator, captures the potential disincentives to work associated with welfare benefits, as well as unobserved characteristics affecting employability. Similarly, the inclusion of housing allowance receipt controls for the impact of this type of income support on work incentives. Finally, a lagged measure of labor participation accounts for the persistence of employment status over time.
The regression results indicate that males experienced slightly larger and more robust effects than females. However, two specifications for males (4 and 5) and four for females (6, 8, 9, and 10) violate the parallel trends assumption, complicating causal interpretation. Specification (1) suggests a significant increase in the probability of labor participation for males across all post-treatment periods, with effects ranging from 4.2 to 4.7 percentage points in periods 0 and 1, gradually decreasing to 2.4 percentage points in period 5. The average post-treatment effect is estimated at 3.52 percentage points. While males under 30 exhibit significant effects in periods 0 and 1, the results in specification (2) suggests that their probability of labor participation converges with that of the control group from period 2 onwards.
Among age cohorts, males aged 30 to 39 presented in specification (3), exhibit the largest and most sustained effects, with coefficients ranging between 2.7 and 6.6 percentage points and an average post-treatment effect of 4.02 percentage points. Males aged 40 to 49 (specification 4) also show significant differences in labor participation (ranging from 2 to 6.4 percentage points), yet the presence of pretrend violations in periods −3 and −2 precludes causal inference. While the post-treatment estimates indicate a 4.8 percentage point increase in labor participation probability for treated males in this age group, these effects cannot be solely attributed to the intervention. For males aged 50 to 59, specification (5) reveals no significant effects, suggesting that either the intervention was primarily effective for younger individuals or that older workers faced greater barriers to reemployment.
For females, the effects of the intervention are less straightforward. Specification (6) detects a significant pretrend in period −4, which complicates causal interpretation. Nevertheless, labor participation increases across all post-treatment periods. Comparing specification (6) with specification (1) reveals that the average post-treatment coefficients for females (2.62 percentage points) is lower than for males (3.52 percentage points). Further disaggregation by age group suggests that younger women benefited the most from the intervention. Specification (7) indicates significant positive effects for females under 30 in periods 0, 1, 2, and 5, with an average post-treatment effect of 2.6 percentage points, making this the only female cohort for which the results allow a causal interpretation. Specification (8) suggests that females aged 30 to 39 also experienced improvements in labor participation, with an estimated average coefficient of 4.13 percentage points, comparable to that of males in the same age group (4.02 percentage points). However, the presence of a pretrend violation in period −2 complicates causal claims.
For older female cohorts, the results indicate a more limited impact. Specification (9) shows that females aged 40 to 49 had significantly lower labor participation rates than the control group before treatment, with an estimated pretreatment gap of 3.16 percentage points. While post-treatment estimates suggest some improvements, these changes cannot be attributed to the intervention. A similar pattern emerges for females aged 50 to 59, where pretreatment differences persist, albeit with weaker statistical significance.
Several covariate patterns also emerge from the analysis. The presence of young children appears to affect labor participation differently across sexes, with no discernible impact on males but a clear negative effect on females, particularly among those under 30. Marital and cohabitation status also exhibits a gendered relationship with labor participation: while married or cohabiting males are more likely to participate in the labor market, females in similar household arrangements exhibit lower probabilities of employment. Finally, educational attainment is positively correlated with labor participation across all age cohorts and both sexes. Even among those receiving the intervention, higher education levels are associated with a greater likelihood of employment, suggesting that individual human capital plays a critical role in shaping labor market outcomes.
Table 3 presents the results from the IW estimator following Sun and Abraham (2021). These estimates corroborate the main findings from the TWFE regressions, though with slightly smaller coefficient magnitudes. Specification (1) confirms lasting effects on male employment across all post-treatment periods (on average 3.3 percentage points), while specifications (2) and (3) reaffirm that males under 40—particularly those aged 30 to 39—experienced the strongest and most sustained effects. Specification (4) reiterates significant pretrends for males aged 40 to 49, reinforcing the conclusion that the positive post-treatment estimates for this cohort cannot be causally attributed to the intervention. Specification (5) confirms the absence of significant effects for males aged 50 and above.
Regression Output, Two-Way Fixed Effects With Interaction Weighted Estimators.
Standard errors are in parentheses.
*** p < .01, ** p < .05, * p < .1.
For females, the IW estimates largely align with the TWFE results. Specification (6) detects a significant pretrend in period −4 but not in other pretreatment periods, suggesting that treated females exhibited a more favorable development in labor participation probability over time. Specifications (7) through (9) confirm the presence of intervention effects for females under 30 and suggest improvements for those aged 30 to 39, although the violation of the parallel trends assumption in period −2 complicates causal interpretation. Specification (9) provides weak evidence of post-treatment gains for females aged 40 to 49, though these improvements must be interpreted cautiously due to pretreatment differences between treated and control individuals.
Figure 2 presents the IW estimates and 95 percent confidence intervals for specifications (1) and (6), illustrating that post-treatment confidence intervals for males and females overlap, except in the 40 to 49 age cohort. The appendix (Figures A5–A8) further visualizes these estimates, demonstrating that the main results hold across sex and age groups. Although the study does not provide definitive evidence of sex differences in treatment effects, it suggests that the probability of identical effects across sexes remains relatively low.
Specification (1) and (6) IW Estimates and 95 Confidence Intervals by Sex, All Ages.
Discussion
This study contributes to the literature on Place-Based Active Labour Market Policies by demonstrating that their effectiveness varies significantly by gender and age. While such policies can enhance labor market participation, the findings suggest that uniform interventions may fail to address the distinct constraints faced by different demographic groups.
The stronger and more persistent effects among men, particularly those aged 30 to 39, align with research showing that prime-aged males are more responsive to ALMPs (Albrecht, Van den Berg and Vroman 2005; Christensen 2013). In contrast, the weaker and less apparent effects for women diverge from prior findings that ALMPs tend to yield greater benefits for females (Card, Kluve and Weber 2018). This difference does not necessarily indicate a contradiction. Existing evidence shows that program outcomes vary considerably depending on both design and target group. Card, Kluve and Weber (2018) find that women are more responsive to training programs and private sector incentive schemes than to job-search assistance (JSA). If the MDI relied heavily on JSA measures, this could help account for the weaker effects observed among women in this study. A more detailed investigation into the program's composition may help clarify this pattern. Another explanation may lie in the sample composition: as the study focuses exclusively on foreign-born individuals, a substantial share likely originates from societies with more conservative gender norms, which may restrict female labor market participation (Petrongolo and Ronchi 2020). Structural labor market barriers, including occupational gender segregation and disproportionate caregiving responsibilities, may further limit the effectiveness of the MDI for women. These results point towards the need for gender-specific policy adaptations, such as subsidized childcare, targeted skills training, and private sector incentive schemes aligned with female-dominated sectors. The results also highlight a clear age difference in policy impact. While the intervention significantly increased labor participation among men aged 30 to 39, it had little to no effect on those aged 50 to 59. This pattern is consistent with evidence that job displacement is more detrimental for older workers due to skill obsolescence, employer biases, and a greater likelihood of permanent labor market exit (Card, Kluve and Weber 2018; Orfao and Malo 2023; Öylü, Motel-Klingebiel and Kelfve 2024). These findings suggest that the intervention may have been inadequate for this group, necessitating interventions such as retraining programs and incentives for employers to retain and hire older workers.
Additionally, there is a significant and positive correlation between education level and labor participation across almost all age cohorts and for both sexes. This indicates that treated individuals with higher education are more likely to be gainfully employed than those with lower education levels. This finding aligns with previous research (Romero 2009)—suggesting that place-based policies aimed at increasing labor supply have a greater impact on individuals who have previously invested in their human capital.
The study's findings must be interpreted within Sweden's broader ALMP landscape, where concurrent space-blind interventions were likely influencing post-treatment outcomes in the control group. If these interventions increased employment probabilities among the control group, the estimated treatment effects of the MDI may be downwardly biased, meaning that the estimated effect could be smaller than the true effect of the intervention. Moreover, the interaction between space-blind and place-based ALMPs remains an open question. While prior research has treated these policy types as distinct, future studies should explore whether combining spatially targeted interventions with broader national ALMPs could enhance employment outcomes across different demographic groups. A promising approach could be to exploit heterogeneity in exposure to ALMPs across different labor market subgroups, allowing for an estimation of whether these programs complement or substitute each other. Overall, the findings suggest that one-size-fits-all interventions are insufficient to address heterogeneous employment constraints. While the MDI effectively increased labor participation among certain subgroups, its limited impact on women and older workers highlights the need for differentiated policy approaches that account for gendered labor supply constraints and age-related employment challenges.
Finally, this study complements the findings of Andersson and Bråmå (2004) by explicitly estimating the causal effect of the MDI. It demonstrates that the intervention not only potentially impacts targeted individuals but indeed has a measurable effect. This advances our understanding of the efficacy of place-based policies. However, the study does not address whether individuals who experienced improved labor market outcomes remained in the neighborhood or relocated. This issue presents a valuable avenue for future research, as it could help determine the potential of place-based policies to benefit not only individuals but also entire neighborhoods and communities. A similar methodological approach such as in the present study could be useful in shedding light on this.
Conclusion
This study examines the labor market effects of a place-based active labor market policy implemented in Sweden under the MDI. While most research on place-based policies focuses on stimulating labor demand, this analysis contributes to the growing but limited literature on supply-side interventions. Employing a DiD framework, the analysis reveals positive but uneven treatment effects. The intervention appears to have significantly improved labor participation among men, especially those aged 30 to 39, while the estimated effects for women were smaller and statistically inconsistent. For older individuals, the intervention showed no measurable impact. These findings suggest that while place-based Active Labour Market Policies can be effective, their success depends on demographic characteristics, with varying responses possibly linked to factors such as caregiving responsibilities and age-related employment barriers.
The MDI emerged in a context of rising socioeconomic inequality and increasing residential segregation. It was introduced as a means to improve individual life outcomes and to counteract socioeconomic and ethnic segregation. This article contributes to the evaluation of the first objective, improving life outcomes, by assessing whether the policy increased the probability of employment among the targeted population. The results suggest that the MDI, and by extension the application of active labor market policies in a place-based format, can enhance employment outcomes for residents of disadvantaged areas. While targeted interventions may also prove more cost-effective than space-blind measures for certain groups, this remains to be formally assessed. Regarding the initiative's second objective—reducing spatial inequality—the extent to which place-based active labor market policies can improve neighborhood-level conditions remains an open empirical question. Previous research from other countries offers mixed evidence, with some studies indicating neighborhood improvements and others pointing to selective out-migration. Future research should examine whether individuals who experienced improved labor market outcomes subsequently relocated, and whether sustained place-based interventions can function as long-term institutional mechanisms for supporting labor market integration among both long-term residents and new arrivals in marginalized areas.
The results also raise methodological considerations. If national, nonspatial Active Labour Market Policies were simultaneously improving employment prospects in the control group, estimated treatment effects may understate the full impact of the local intervention. This potential downward bias calls for further investigation into the interactions between spatially targeted and universal labor market policies. Future work should also assess how program composition influences outcomes, particularly whether heavier reliance on job-search assistance, as opposed to training or employment subsidies, might explain weaker effects for some groups.
Taken together, these findings point towards the importance of designing place-based Active Labour Market Policies that reflect the diverse constraints and capabilities of targeted individuals. A more differentiated policy approach that accounts for gender, age, and household structure could improve the overall effectiveness and equity of place-based labor market programs.
Footnotes
Funding
The author disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by the Svenska Forskningsrådet Formas (grant number 2022-00656).
Declaration of Conflicting Interests
The author declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Author Biography
Appendix
Summary Statistics of Propensity Scores (PS) for Treated and Control Groups.
| Group | Observed | Mean PS | Standard. Deviation PS | Minimum PS | Maximum PS |
|---|---|---|---|---|---|
| Treated | 16083 | .316 | .128 | .026 | .767 |
| Control | 48871 | .225 | .124 | 0 | .746 |