This study aims to evaluate the impact of Internship Contracts, a youth-targeted active labor market policy implemented in Spain between 2016 and 2019.
Using administrative data from the Continuous Sample of Working Lives and propensity score matching, the authors estimate the Average Treatment on the Treated (ATT) for contract duration and monthly wages in the short term (one year) and medium term (two to three years). Logistic models estimate the probability of obtaining a permanent contract and remaining with the same employer after contract completion.
Relative to Temporary Contracts (TCs), Internship Contracts lead to higher wages, longer subsequent contracts and a greater likelihood of permanent employment, although with lower employer retention. Internship contract holders earn on average €186,42 more than comparable TC holders over the 2017–2019 period, corresponding to an earnings premium of 14,7%. Effects are broadly consistent across subsamples, with heterogeneity by gender, experience and education.
The results reveal a paradox: the strongest benefits of Internship Contracts accrue to individuals with prior work experience, despite the programme’s objective of facilitating labor-market entry for first-time participants, with prior experience shaping both the magnitude and timing of these effects.
1. Introduction
Labor markets have undergone profound transformations since the economic crises of the 1970s and 1980s, marked by increasing flexibility through the expansion of temporary employment (Úbeda et al., 2020). These shifts have contributed to persistently high unemployment, job instability and wage stagnation, particularly affecting individuals under 30 (García-López, 2011; González-García, 2013; O'Reilly et al., 2019).
By 2016, even as economies recovered from the 2008 financial crisis, youth unemployment remained disproportionately high across Europe. According to EUROSTAT, unemployment rates for individuals under 30 exceeded those of the general population, and Temporary Contracts (TCs) were heavily concentrated among young workers. This duality has entrenched a divide between workers with stable permanent contracts and those facing repeated short-term contracts and lower earnings.
In response, many governments have implemented Active Labor Market Policies (ALMPs), including training-oriented contracts designed to reduce hiring costs and improve youth employability. This study focuses on the Internship Contract (IC), a specific training contract in Spain originally designed to facilitate the transition of recently qualified young individuals into the labor market, particularly those without prior work experience. We evaluate its impact on four key early-career outcomes:
the duration of subsequent contracts;
monthly wages in the short and medium term (one to three years);
the likelihood of securing a permanent contract; and
the probability of remaining with the same employer.
While previous research has examined the regulatory framework and descriptive characteristics of ICs, empirical evaluations of their effectiveness remain scarce (e.g. De la Rica and Gorjón, 2022). This paper addresses this gap by providing a counterfactual impact evaluation of Internship Contracts in Spain using administrative microdata and Propensity Score Matching. Beyond contract duration and wages, it estimates their impact on job stability and examines treatment heterogeneity by gender, education and prior work experience.
Although De la Rica and Gorjón (2022) acknowledge that IC are often used by individuals with previous work experience, their analysis focuses exclusively on first-time entrants. This narrower scope does not reflect how the policy is implemented in practice. By analyzing all IC users, our study offers a more comprehensive view of the instrument’s real-world effects.
The remainder of the article is organized as follows: Section 2 reviews the relevant literature. Section 3 presents the institutional framework. Section 4 describes the data source and sample selection. Section 5 presents the descriptive analysis. Section 6 outlines the empirical strategy. Section 7 discusses the results, and the final section concludes.
2. Literature review
ALMPs targeting young workers are widely adopted to mitigate youth unemployment. These policies typically encompass hiring subsidies, skill development programs and contractual arrangements designed to facilitate entry into stable employment. Among these, subsidized contracts such as Apprenticeship, Training and Internship Contracts are frequently promoted as tools to enhance youth employability, though their long-term effectiveness remains contested (Mato-Díaz and Escudero-Castillo, 2024).
An expanding body of literature has examined the impact of these contracts on early-career trajectories, with mixed evidence. While some studies highlight positive effects on employment stability, others suggest they may reinforce labor market segmentation. For instance, De la Rica and Gorjón (2022) analyze the effectiveness of ICs in Spain. Their findings reveal that, although these contracts are associated with lower job stability and wages in the short term, they improve long-term employment prospects, particularly in terms of securing permanent contracts, relative to standard TCs. These results suggest that ICs may act a stepping stone toward more stable employment, albeit not uniformly across all groups.
Similar findings emerge from studies on Apprenticeship Contracts in Italy and Training Contracts in Spain, where long-term employment prospects improved under certain conditions (Troncoso-Ponce et al., 2016; Albanese et al., 2021). However, other research is more skeptical. García-Pérez et al. (2019) report that fixed-term contracts facilitate short-term entry into the labor market without necessarily improving long-term stability. Sciulli (2013), in an evaluation of Italy’s Treu Law, finds that Apprenticeship Contracts often delay transitions to stable employment, indicating that firms may value flexibility more than worker development. Cueto and Rodríguez-López (2019) further observe that reformed Training Contracts underperform TC in terms of duration, earnings and integration into stable jobs.
Cross-country comparisons provide additional insights. Picchio and Staffolani (2019) show that Italian apprentices are more likely to secure permanent contracts within the same firm, though mobility across firms remains limited. Kiersztyn (2021) argues that transitions from fixed-term to permanent contracts are influenced by individual and firm-level factors in segmented labor markets. In France, Roger and Zamora (2011) find that subsidized contracts improve short-term outcomes for low-skilled youth but often fall short of delivering sustainable employment.
In the vocational education and training (VET) literature, Bolli et al. (2021) demonstrate that VET can enhance youth labor market integration and job stability, provided there is alignment with employer needs. Bassanini and Garnero (2013) review broader employment protection frameworks, noting that the design of such institutions, including subsidized contracts, significantly shapes labor market performance. A meta-analysis by Card et al. (2010) similarly finds that while ALMPs such as hiring subsidies and training programs can yield positive effects, their success is highly context-dependent.
More recently, Vaquero-García et al. (2024) compare youth-focused ALMPs in Spain, France and Germany, concluding that targeted youth contracts have had limited success in reducing youth unemployment and may have contributed to labor market duality. The authors advocate for maintaining existing mechanisms while removing age restrictions, aligning with reforms in other countries.
Taken together, this literature underscores the complexities of designing effective youth employment interventions. While subsidized contracts can facilitate labor market entry, their long-term success hinges on institutional features, firm incentives and broader economic conditions.
3. The internship contract: Context
Spanish labor legislation has undergone multiple reforms to adapt to changing labor market conditions. To promote youth employment, Article 11 of the Workers Statute [1] introduced training contracts, distinguishing between the Training and Apprenticeship Contract – for individuals without professional qualifications – and the Internship Contract, aimed at those with a university degree, vocational training or equivalent. The regulatory framework for ICs was modified twice. The first reform, in 2010, extended eligibility to holders of professional certifications and increased the post-graduation eligibility period from four to five years. It also allowed interruptions in contract duration for cases such as temporary incapacity, maternity and paternity leave.
In response to the 2008 crisis, Law 11 / 2013 [2], introduced measures to promote employment. From July 28, 2013, individuals under 30 could sign ICs regardless of time since graduation, and employers hiring them (2014–2016) received Social Security reductions. These incentives were later repealed in 2018.
The second major reform came with the 2021 Labor Reform, which restructured training contracts as of March 31, 2022. The previous Internship Contract was replaced by the Training Contract for the Acquisition of Professional Practice [3]. While eligibility criteria remained largely unchanged, the post-graduation eligibility period was reduced from five to three years (five for individuals with disabilities), and contract duration was set between six months and one year. Employers were also required to develop individualized training plans and appoint mentors [4].
Despite these efforts, administrative data from the Ministry of Labor and Social Economy reveals that uptake of ICs remains low compared to TCs. Usage peaked briefly post-2008 crisis but declined sharply after 2019. De la Rica and Gorjón (2022) attribute this to structural disincentives: the six-month minimum duration conflicts with employer preferences for shorter, cyclical contracts. In addition, while reforms broadened access to vocational graduates and aimed to curb wage suppression, employers often prioritize cost flexibility over long-term training investments. This highlights a persistent mismatch between policy objectives – facilitating education-to-work transitions – and labor market realities.
4. Data and sample selection
4.1 Continuous sample of working lives
The data used for this study come from the Continuous Sample of Working Lives (CSWL; MCVL), which is an administrative record compiled by the Ministry of Inclusion, Social Security and Migration of the Spanish Government. It contains information from Social Security databases, supplemented with data from the Municipal Continuous Register (National Statistics Institute) and the annual summary of withholdings and income tax advances from the Tax Agency.
The CSWL provides highly relevant information about the Spanish labor market, as it consists of a random sample of 4% of individuals affiliated with Social Security in the reference year. It enables tracking of employment and unemployment spells from the start of each individual’s working life through to the last available year (Durán, 2007).
For this study, the variables of interest include contract type and duration (with exact start and end dates), economic sector and company size. The CSWL also provides demographic data, including gender, age, education, place of residence and nationality.
4.2 Sample selection
Multiple Treatment and Control Groups are defined [5]. The primary focus is on Group 1 (Treatment and Control), used in the Descriptive Statistics (Section 6) and Model Estimations (Section 7, excluding 7.3). The remaining groups are reserved for the Sensitivity Analysis in Section 7.3.
As the legal framework in 2016 (Workers’ Statute and Law 11 / 2013) established specific eligibility criteria for IC –though not always followed in practice-, group definitions are designed to adhere as closely as possible with these regulations.
Treatment Group 1 consists of individuals who signed an IC in 2016, were under 30, held at least a First- or Second-Degree Vocational Training, Industrial Official or Higher Vocational Training qualification, and whose contract lasted at least three months. Observations falling outside these criteria – about 16% of the sample – are excluded.
Control Group 1 follows the same age, education and minimum contract duration criteria as the treatment group but includes individuals who signed a TC lasting at least three months.
To examine the role of prior work experience, we define two additional groups. Group 2 includes individuals without any work experience before signing the Internship or TC, while Group 3 includes those with such experience.
Group 4 focuses on individuals with higher education (bachelor’s, master’s or doctoral degrees), corresponding to the population for whom the IC was originally intended.
Groups 5 and 6 test the sensitivity of the results from Groups 1 and 4 by raising the minimum contract duration from three to six months, aligning with the regulatory threshold established in 2016.
Estimating effects separately for first-time entrants and for the broader population of IC users allows us to assess both average programme effectiveness and the consequences of deviations from the intended targeting of the policy.
Accordingly, we distinguish throughout between estimates for the analysis sample of IC users complying with legal eligibility criteria (Group 1) and for first-time labor market entrants (Group 2), consistent with the existing literature.
5. Methodology
Accurately estimating the effects of public interventions is essential to provide solid evidence about their effectiveness or to identify areas for improvement. To evaluate the effects of IC, four outcome variables are used:
the average duration of subsequent contracts;
the income of beneficiaries in the short term (one year) and medium term (three years);
the probability of obtaining a permanent contract; and
remaining with the same company [6].
For this research, information from individuals included in the CSWL for the year 2016 and their labor trajectories up to 2019 is used. This choice ensures a sample not exposed to major shocks and allows for a three-year observation window. Although the CSWL provides more recent information, disruptive events occurring between 2019 and 2021 present obstacles to making robust estimates. For 2019, the removal of reductions on employer social security contributions can be cited as a significant legislative change that likely impacted the effectiveness of this tool. For 2020 and 2021, the COVID-19 pandemic is a major shock that substantially alters policy evaluations.
Finally, 2022, the most recent year for which information is available, also does not allow for feasible results, particularly considering the legislative change that occurred at the end of 2021 and came into effect from the second quarter of 2022, leaving insufficient time to estimate the impact of IC[7].
It is important to note that for this study, the effects of legislative changes prior to 2016, such as the 2010 reform or the creation of incentives under Law 11 / 2013, were not estimated because it was considered that these modifications did not alter the way ICs helped young people enter the labor market.
The methodology to estimate the effect of the IC on the variables of interest was divided into three basic steps. First, control groups comparable to treatment groups were created. Thus, individuals who did not use IC, but could have done so based on their characteristics were selected through Propensity Score Matching. This required estimating the Propensity Score (PS) for the treatment and control groups; (2) using a PS matching method to generate comparable Treatment and Control Groups; (3) choosing the best matching algorithm; and (4) conducting a balance diagnostic to verify the compatibility of both groups (Rosenbaum and Rubin, 1983; Rosenbaum, 1989; Heckman et al., 1997; Dehejia and Wahba, 1999).
Second, once both groups are obtained and verified to be similar, the next step is to estimate the Average Treatment Effect on the Treated (ATT) to assess the first and second outcomes of interest (Cueto and Mato-Díaz, 2009; Caliendo and Kopeinig, 2008; Abadie and Imbens, 2011; Diamond and Sekhon, 2013; Mahalanobis, 2018; Guo et al., 2020; Zhao et al., 2021), which are the difference in (1) days worked between the two groups and (2) wage differences. In addition, a Sensitivity Analysis was performed to ensure the most robust possible results.
To further assess the role of prior work experience, we also estimate treatment effects separately across subgroups defined by alternative thresholds of accumulated employment prior to treatment. In addition, we implement a reweighting procedure to align the full sample of IC users with the observable characteristics of first-time labor market entrants, allowing us to distinguish between compositional differences and structural treatment heterogeneity.
Inference after matching requires accounting for the dependence induced by the matching procedure. Because individuals matched within the same set share similar PS, observations are no longer independent. To address this issue, we identify each matched set using the subclass variable generated by the matching algorithm. Both ATT estimates and matched-sample regression models introduced below are estimated using robust standard errors clustered at the matched-set (subclass) level, allowing for arbitrary correlation within matched sets while maintaining independence across sets. This approach constitutes a standard analytic correction for post-matching inference and is conceptually related to the variance estimators proposed by Abadie and Imbens (2011). As an additional robustness check, we implement a matched-set bootstrap, in which resampling is conducted at the level of matched sets rather than individual observations. Results from this procedure are reported in AppendixTable A5 and lead to identical qualitative inference.
Third, using a binomial logistic model and assessing the relevance of its parameters (Long, 1997), the probability of transitioning to a permanent contract and of remaining in the same firm after the contract ends was calculated for both treatment and control groups.
6. Descriptive statistics
This section provides an empirical characterization of the beneficiary population of Internship and TCs in Spain, focusing on Group 1 as defined in Section 4 (Table 1) [8].
Descriptive statistics
| Characteristic | Control Group 1 (%) | Treatment Group 1 (%) |
|---|---|---|
| Sample (number of individuals) | 1592 | 491 |
| Experience | ||
| Prior experience | 67.4 | 70.3 |
| No prior experience | 32.6 | 29.7 |
| Gender | ||
| Female | 53.6 | 48.9 |
| Male | 46.4 | 51.1 |
| Age | ||
| ≤ 22 years | 17.3 | 14.1 |
| > 22 years | 82.7 | 85.9 |
| Contracts signed | ||
| Up to 2 contracts | 22.9 | 31.4 |
| 3–6 contracts | 41.8 | 49.3 |
| More than 7 contracts | 35.3 | 19.3 |
| Days worked | ||
| Up to 1 year | 10.2 | 4.3 |
| More than 1 and up to 2 years | 10.3 | 9.8 |
| Between 2 and 3 years | 20.3 | 14.7 |
| More than 3 years | 59.2 | 71.3 |
| Next contract type | ||
| Not permanent | 93.0 | 84.9 |
| Permanent | 7.0 | 15.1 |
| Next employer | ||
| Different company | 71.0 | 81.9 |
| Same company | 29.0 | 18.1 |
| Education | ||
| Vocational training | 52.1 | 33.8 |
| University degree | 47.9 | 66.2 |
| Average monthly salary (€) | ||
| 2017 | 1,147.5 | 1,332.2 |
| 2018 | 1,255.3 | 1,493.1 |
| 2019 | 1,417.1 | 1,701.0 |
| Nationality | ||
| Spanish | 80.1 | 93.9 |
| Non-Spanish | 19.9 | 6.10 |
| Average contract duration (days) | ||
| After the end of the initial temporary or internship contract | 553 | 683 |
1: Treatment Group 1 includes individuals under 30 who signed an Internship Contract in 2016, held at least a First/Second-Degree Vocational or Higher Vocational qualification, and whose contract lasted ≥ Three months. About 16% of observations falling outside these criteria were excluded. Control Group 1 applies the same age, education and duration filters to individuals with a Temporary Contract
2: Autonomous Community and Economic Activity are presented in Table S2 in the supplementary material
A key aspect of the sample is prior labor market experience: 70.3% of individuals with an IC in 2016 had previously held formal employment, compared to 67.4% for those with TCs. This suggests that IC may not strictly function as an initial entry point for recent graduates, but rather as an alternative employment pathway when permanent positions are unattainable.
Contract duration patterns differ markedly across groups. Among IC, 10% ended before the legal minimum of six-months, while 20% reached the two-year maximum. TCs, by contrast, exhibit pronounced instability: 53% lasted less than one month, and only 33% exceeded six months. Employment continuity is also captured by contract frequency: 35% of the Control Group signed more than seven contracts between 2016 and 2019, compared to 19% in the Treatment Group. Meanwhile, 31% of the latter group signed no more than two contracts, suggesting lower exposure to precarious employment. Average contract duration was also significantly longer among Internship beneficiaries.
IC are also associated with greater employment continuity and an increased probability of transitioning into permanent positions. Between 2017 and 2019, individuals who entered employment through an IC also reported systematically higher mean monthly earnings. Sectoral differences are notable: Internship recipients were concentrated in Professional, Scientific and Technical Activities (25.7%), while TC holders were more prevalent in Commerce (15.5%) and Hospitality (11.6%) –sectors generally linked to lower job stability.
Demographic differences between groups are notable. Women account for 53.6% of the Control Group, whereas men represent 51.1% of the Treatment Group. Spanish nationals are more prevalent among Internship recipients (93.9%) than among TC holders (80.1%). Educational attainment also differs: 66% of the Treatment Group hold a university degree, compared to 47% in the Control Group.
Overall, the descriptive evidence suggests that ICs are linked to better labor market outcomes than TCs, including greater job stability and higher earnings. Although not fully aligned with their original aim of supporting labor market entry, they appear to function as a pathway to more stable employment, particularly for individuals with higher education.
7. Results
7.1 Propensity score matching
The validity of treatment effect estimates depends on constructing comparable treatment and control groups across all observable covariates. To achieve this, we apply the optimal PS matching algorithm, selected after sensitivity analyses (detailed in AppendixTable A2) confirmed the robustness of results. This method optimizes PS distances without reducing sample size, unlike alternatives such as Nearest Neighbor, Coarsened Exact Matching or Exact Matching.
Each treated observation (IC in 2016) is matched to a unique counterfactual (TC in 2016) based on similarity in PS, estimated via logistic regression. The dependent variable is treatment status, and covariates include gender, age, education, Spanish nationality, sector of activity, Autonomous Community of residence and 2016 salary.
The final matched sample comprises 978 individuals (489 per group) [9], excluding 1,101 Control Group subjects and two Treatment Group subjects. AppendixTable A2 reports the sample, the standardized mean differences for each covariate, and confirms balance across all demographic characteristics (p-values < 0.05) [10].
7.2 Contract duration and wage effects (ATT 1)
This section presents the estimated Average Treatment Effect on the Treated (ATT 1) for individuals who signed an IC in 2016, compared to those with TCs in the same year. Table 2 reports results for outcome variables:
the average duration of contracts following the initial Internship/TC; and
average monthly salary in 2017, 2018 and 2019.
ATT 1 Estimates on contract duration (days) and monthly wages (€) for treatment group (internship contract)
| Complete sample and subgroups | |||||||
|---|---|---|---|---|---|---|---|
| Variable | Complete | Female | Male | ≤ 22 Years | > 22 years | Voc. Training | University |
| Contract duration | 106.72*** | 91.35*** | 120.78*** | 152.19*** | 87.24*** | 145.56*** | 87.15*** |
| Std. Error | (20.97) | (29.78) | (29.74) | (50.19) | (23.35) | (33.06) | (27.33) |
| Monthly Salary 2017 (€) | 174.67*** | 191.19*** | 160.17*** | 198.26*** | 153.27*** | 182.80*** | 172.16*** |
| Std. Error | (39.00) | (54.97) | (55.49) | (79.97) | (44.35) | (56.02) | (52.58) |
| Monthly Salary 2018 (€) | 189.86*** | 187.85*** | 191.28*** | 196.40** | 169.69*** | 235.12*** | 171.21*** |
| Std. Error | (46.12) | (63.59) | (67.01) | (100.25) | (52.21) | (68.16) | (61.55) |
| Monthly Salary 2019 (€) | 194.74*** | 167.15*** | 210.68*** | 164.63* | 185.48*** | 179.35*** | 194.47*** |
| Std. Error | (47.82) | (66.19) | (69.00) | (110.55) | (53.75) | (72.53) | (63.22) |
ATT = Average Treatment Effect on the Treated. Significance levels: ***p < 0.01, **p < 0.05, *p < 0.10
Overall, Internship recipients held subsequent contracts that were, on average, 106 days longer than those of their counterparts, a difference statistically significant at the 1% level. When disaggregated by gender, this effect was larger for men (120 days) than for women (91 days), with both estimates significant at the same confidence level. This pattern persists across age and education subgroups. The effect was stronger among individuals aged 22 or younger and those with vocational training, compared to those over 22 and university graduates. In all cases, IC are associated with longer employment spells, highlighting their contribution to medium-term labor market stability.
Regarding average monthly earnings, IC holders earned €186,42 more per month than TC holders, on average, over the 2017–2019 period, which correspond to an earnings premium of 14,7%. Gender-disaggregated results reveal diverging trends: for women, the wage premium declined over time, while for men it increased. In 2017, women in the treatment group earned €191.19 more, compared to €160.17 for men. By 2019, the differential fell to €167.15 for women but rose to €210.68 for men. All estimates are positive and statistically significant at the 1% level.
These results suggest that while ICs improve short-term earnings for both genders, the long-term wage benefits appear to favor men, indicating potential unobservable factors limiting women’s gains.
Subgroup analysis by age reveals diverging trends in wage effects. For individuals under 22, the largest differential is observed in the short term, while for those over 22 the strongest effect appears in 2019. After two years, the wage premium remains higher for the younger group. This pattern, further exposed in the Sensitivity Analysis, suggests that older individuals, likely with more work experience, benefit from higher wages not directly attributable to the treatment, but rather to accumulated experience.
Consistent with this interpretation, we further explore heterogeneity by prior employment experience within the treated group by splitting the sample using two alternative thresholds: fewer than 120 days versus 120 days or more of prior employment, and less than one year versus one year or more (see AppendixTable A3). The results reveal meaningful differences across groups: short-term wage gains are larger for individuals with limited prior experience, whereas medium- and longer-term effects tend to be stronger among those with more extensive employment histories. Contract duration effects remain positive across most groups but are weaker and not statistically significant for individuals with at least one year of prior experience. Overall, these patterns suggest that prior employment experience shapes both the magnitude and timing of treatment effects.
Educational subgroups show contrasting dynamics. Among those with Vocational Training, the wage effect grows in the first two years but declines in the third year to below its initial level. For university graduates, the trend is more stable: the coefficient slightly declines in 2018 before increasing again in 2019.
Overall, IC recipients in 2016 experienced sustained wage growth and longer subsequent contracts over the short and medium term. These effects appear persistent, and in some groups, they intensify over time. The observed positive wage differentials and extended contract duration suggests that IC contribute to improved employability through skill acquisition and work experience. While all subgroups outperform the Control Group over the three-year period, the gains are more modest for women and for individuals under 22, two groups with typically less prior work experience, despite being the primary target of this ALMP.
Wage effects are estimated conditional on being observed in formal employment, as earnings in the MCVL are recorded only when individuals have Social Security contributions. Consequently, reported wage effects capture intensive-margin changes among the formally employed and do not incorporate extensive-margin effects related to employment probabilities. We explicitly acknowledge this scope when interpreting the results and address related robustness concerns below.
To disentangle compositional effects from structural treatment heterogeneity, we reweight the full internship sample to replicate the observable characteristics of first-time labor market entrants. We estimate the probability of being a new entrant conditional on predetermined covariates (age, sex, education, region, sector, firm size and year) using a logit model. New entrants receive unit weights, while internships held by individuals with prior experience are weighted by wi=pi/(1−pi). We then re-estimate treatment effects using the reweighted full sample and compare magnitudes with baseline estimates for new entrants.
The reweighting exercise highlights meaningful differences between compositional and structural effects. When the full internship sample is reweighted to match the characteristics of first-time labor market entrants, the estimated impact on employment stability declines substantially, indicating that a non-negligible share of the average stability effect in the full sample is driven by individuals with prior experience. Short-term wage effects remain very similar across samples, while medium- and long-term wage gains are systematically smaller for first-time entrants, consistent with more limited inmediate returns to human capital accumulation at labor-market entry. In the longer term, wage effects become larger in the reweighted full sample, suggesting that individuals with prior experience translate accumulated labor-market attachment into delayed earnings gains. Overall, these results point to stronger measured benefits accruing to individuals with prior work experience and to reduced effectiveness, as well as altered timing of returns, when IC are applied to their intended target population (see AppendixTable A4).
7.3 Permanent employment and employer retention
Additional effects of IC confirm the positive impact on job stability outlined in the previous section. Descriptive evidence showed that IC recipients experienced more stable career trajectories – measured by longer total days worked, fewer contracts signed and greater average contract duration – relative to those with TCs. This section presents more robust evidence using two binomial logistic models estimated on the matched Treatment and Control Group 1 samples. Results are reported in rows 1 and 4 of Table 3.
Marginal effects: probability of obtaining a permanent contract and remaining with the same employer
| Treatment group: Internship contract (vs Temporary contract), matched sample | |||||||
|---|---|---|---|---|---|---|---|
| Variable | Complete | Female | Male | ≤ 22 Years | > 22 years | Voc. Training | University |
| Obtained permanent contract | 0.09** | 0.08** | 0.11** | 0.08* | 0.10** | 0.06* | 0.11** |
| Std. Error | (0.02) | (0.03) | (0.03) | (0.05) | (0.02) | (0.03) | (0.03) |
| Maximum likelihood p-value | 0.00 | 0.00 | 0.00 | 0.01 | 0.00 | 0.05 | 0.00 |
| Wald test p-value | 0.00 | 0.00 | 0.00 | 0.04 | 0.00 | 0.06 | 0.00 |
| Stayed at the same company | −0.09** | −0.12* | −0.05* | −0.15** | −0.08** | −0.18* | −0.04* |
| Std. Error | (0.03) | (0.04) | (0.04) | (0.07) | (0.03) | (0.04) | (0.03) |
| Maximum likelihood p-value | 0.00 | 0.00 | 0.01 | 0.04 | 0.01 | 0.00 | 0.20 |
| Wald test p-value | 0.00 | 0.00 | 0.01 | 0.03 | 0.00 | 0.00 | 0.00 |
The coefficients presented correspond to the marginal effects evaluated at the mean. Significance levels: ***p < 0.01, **p < 0.05, *p < 0.10
The models assess two outcomes:
whether the next contract was permanent; and
whether the individual remained with the same employer, regardless of contract type.
The first outcome reflects whether IC help beneficiaries accumulate experience that facilitates access to stable employment. The second indicates whether employers view ICs as a means to integrate and retain young workers, potentially reflecting investment in training and human capital.
Both models control for age, gender, education and economic activity. To avoid potential collider bias, the specifications do not condition on realized contract duration, which is itself affected by treatment assignment.
The results show that IC holders were 9.2 percentage points more likely to obtain a permanent contract than those with TCs. Gender-disaggregated effects reveal a higher impact for women (8.4 pp) than men (10.6 pp). Differences by age are also notable: individuals over 22 years were 9.9 pp more likely to transition to permanent employment, compared to 7.8 pp for younger individuals. No substantial variation was found by education level (6.4 pp for individuals with Vocational Training and 10.6 pp for university graduates).
However, results for employer retention are negative. IC recipients were 8.8 pp less likely to remain with the same employer than their TC counterparts. This effect varies considerably by subgroup. For women, the gap reaches 12.2 pp, compared to 5.4 pp for men. By age, the effect is stronger among younger individuals (−14.9 pp) than those over 22 (−8.1 pp). Differences by education are particularly stark: individuals with Vocational Training were 18.3 pp less likely to remain with the same employer, compared to only 3.9 pp among university graduates.
These findings suggest that some employers may use ICs as a substitute for TCs –potentially to reduce costs or circumvent regulatory constraints- without necessarily aiming to retain the worker. At the same time, the higher probability of securing permanent employment among IC holders –especially among men, older individuals, and those with higher education- points to their signaling value in the labor market. While TCs often lead to continued instability, ICs appear to provide a more stable stepping stone, particularly for individuals with greater prior experience or human capital.
To ensure robustness, both models are assessed using Likelihood Ratio and Wald tests. Results for the model estimating the probability of obtaining a permanent contract are shown in rows 3–4 of Table 3, while those for the probability of remaining with the same employer appear in the last two rows. In all cases, for the full sample and all subgroups, the tests confirm joint significance of the covariates at the 1% level, indicating that the explanatory variables are collectively relevant.
7.4 Sensitivity analysis
This section has two objectives. First, it tests whether the estimated ATT coefficients for contract duration, wages (short and medium term), and the probabilities of obtaining a permanent contract and remaining with the same employer are robust to changes in the PS matching method. Second, it evaluates whether the results generalize across different treatment and control groups representing distinct target populations.
Standard errors for matched estimates are clustered at the matched-set level. As a robustness check, AppendixTable A5 reports matched-set bootstrap standard errors, which lead to identical qualitative inference.
The first objective ensures that the observed effects are not artifacts of the matching technique. Since different algorithms may yield samples with varying characteristics, robustness across methods increases confidence in the internal validity of the results.
The second objective assesses external validity by testing whether the results hold when key sample features are modified –specifically, by imposing alternative minimum contract durations, restricting the sample to university graduates (the policy’s original target group), or distinguishing between individuals with and without work experience.
Table 4 presents the ATT estimates and predicted probabilities obtained using different PS matching algorithms: Optimal, Genetic and Nearest Neighbor (with and without replacement, and with caliper adjustment). Most methods yielded comparable results, both in terms of ATT and estimated probabilities. Matched sample sizes ranged from 467 to 489 observations, with most algorithms retaining the full treatment group.
ATT and estimated probabilities using different propensity score matching methods
| Variable | Optimal | Nearest neighbor with caliper | Nearest neighbor without replacement | Nearest neighbor with replacement | Exact | Coercion | Genetic |
|---|---|---|---|---|---|---|---|
| Matched observations | 489 | 467 | 489 | 489 | 360 | 243 | 489 |
| Covariate balance | Balanced | Balanced | Balanced | Balanced | Balanced | Balanced | Balanced |
| Average treatment effect on the treated | |||||||
| Average contract duration | 106.72*** (20.97) | 101.70*** (21.44) | 99.67*** (22.90) | 108.75*** (21.03) | 50.12 (22.51) | 39.77 (29.03) | 65.24*** (21.93) |
| Avg. Monthly Salary 2017 | 174.67*** (39.00) | 141.60*** (40.36) | 146.65*** (42.17) | 151.32*** (39.24) | 22.82 (42.94) | 32.81 (52.72) | 135.48* (38.16) |
| Avg. Monthly Salary 2018 | 189.86*** (46.12) | 164.95*** (46.99) | 140.56*** (49.53) | 167.37*** (45.92) | 34.86 (50.11) | 30.44 (62.93) | 150.42* (46.55) |
| Avg. Monthly Salary 2019 | 194.74*** (47.82) | 187.21*** (49.36) | 158.49*** (51.36) | 190.57*** (48.11) | 56.73 (52.74) | 67.53 (67.21) | 162.66* (48.12) |
| Probabilities | |||||||
| Permanent contract | 0.078*** (0.242) | 0.050** (0.227) | 0.053** (0.239) | 0.056** (0.223) | 0.080*** (0.225) | 0.087*** (0.295) | 0.069** (0.235) |
| Same employer | −0.086*** (0.167) | −0.125*** (0.167) | −0.126*** (0.174) | −0.121*** (0.164) | −0.130*** (0.177) | −0.147*** (0.242) | −0.092** (0.166) |
1. ATT estimates reflect average treatment effects on the treated for each outcome
2. Marginal effects are reported for the binary outcome variables
3. Standard errors are shown in parentheses
4. Statistical significance levels: ***p < 0.01, **p < 0.05, *p < 0.10
5. The caliper value chosen for Nearest Neighbor is 0.01
For contract duration, estimates varied moderately across methods, with the Genetic algorithm yielding the lowest effect and Nearest Neighbor (with replacement) the highest. Wage effects followed a similar pattern: Optimal matching produced the largest estimates, while Genetic marching yielded the smallest. Despite some variation in magnitude, almost all methods confirmed positive and statistically significant effects on both contract duration and monthly wages.
Regarding the probability of obtaining a permanent contract, estimates ranged from 5.0% (Nearest Neighbor with caliper) to 7.8% (Optimal), with standard errors between 22.7% and 24.2%. For the probability of staying with the same company, coefficients ranged from −12.6% to −8.6, with standard errors between 16.4% and 16.7%. In some specifications, these error margins allow for the possibility of coefficients crossing zero.
Except for contract duration, all outcome estimates across matching methods fall within the range of their respective standard deviations, reinforcing the robustness of results. The choice of the Optimal algorithm, used in the main analysis, was based on its ability to minimize PS distances without reducing sample size, unlike other methods. Although algorithms like Exact or Coercion offer better covariate balance, they entail substantial information loss (approximately 25 observations), which was deemed too costly given the small sample size.
The second part of this section presents the results of the Propensity Score Analysis, ATT estimation and probability calculations for additional subgroups, labeled ATT 2 through ATT 6, as reported in Table 5. The table is organized into two panels. Panel A reports the core estimates, comparing the analysis of IC users complying with legal eligibility criteria (ATT 1) with first-time labor market entrants (ATT 2), the population explicitly targeted by the policy; and with individuals with prior work experience (ATT 3). Panel B reports additional estimates used for sensitivity analysis: ATT 4 focuses on university graduates; and ATTs 5 and 6 examine the sensitivity of the results to increased minimum contract duration (based on ATT 1 and ATT 4, respectively).
Average treatment effects on the treated (ATT): full sample, experience groups and sensitivity analyses
| Panel A. Core estimates (full IC sample /First-Time entrants/prior work experience) | Panel B. Sensitivity analyses | |||||
|---|---|---|---|---|---|---|
| Variable | ATT 1 | ATT 2 | ATT 3 | ATT 4 | ATT 5 | ATT 6 |
| Average treatment effect on the treated | ||||||
| Avg. Contract duration | 106.72*** (20.97) | 136.38*** (40.60) | 102.80*** (24.59) | 80.32*** (27.20) | 68.87*** (22.45) | 65.83** (27.93) |
| Avg. Monthly Salary 2017 | 174.67*** (39.00) | 164.33** (73.06) | 174.46*** (45.50) | 164.18*** (49.52) | 141.61*** (38.71) | 154.69*** (50.26) |
| Avg. Monthly Salary 2018 | 189.86*** (46.12) | 153.85* (87.73) | 176.60*** (53.35) | 151.55** (59.38) | 177.39*** (46.41) | 184.27*** (60.99) |
| Avg. Monthly Salary 2019 | 194.74*** (47.82) | 166.51* (91.46) | 200.85*** (56.88) | 157.90** (62.12) | 185.80*** (47.89) | 197.14*** (63.06) |
| Probabilities | ||||||
| Permanent contract | 0.078*** (0.24) | 0.043 (0.447) | 0.077*** (0.269) | 0.104*** (0.29) | 0.068*** (0.229) | 0.069** (0.262) |
| Same employer | −0.086*** (0.17) | −0.136** (0.288) | −0.074*** (0.207) | −0.12*** (0.20) | −0.064** (0.167) | −0.029 (0.202) |
The coefficients presented in the “Probabilities” section represent marginal effects at the mean
1: The standard errors are in parentheses
2: Significance levels: ***p < 0.01, **p < 0.05, *p < 0.10
3: ATT 1–6 correspond to the following subgroups
– ATT 1: Full sample (baseline)
– ATT 2: No prior work experience
– ATT 3: Prior work experience
– ATT 4: University education only
– ATT 5: Minimum 6-month contract duration only
– ATT 6: University education and minimum 6-month contract duration
PS were estimated using the same model specification as in ATT 1 and matched using the Optimal algorithm. In all cases, matching substantially reduced standardized mean differences across observable covariates. Balance was confirmed at the 95% confidence level, as shown by p-values from Standardized Mean Difference tests, and visually illustrated in Panel 2 in the supplementary material.
Panel A of Table 5 highlights the comparison between the overall analysis sample (ATT 1) and first-time labor market entrants (ATT 2). The results for individuals without prior work experience are particularly significant. While all coefficients for contract duration and wages are positive and statistically significant –mirroring those of other groups- the wage gains are more limited over the medium term and do not display a clear upward trajectory. Moreover, this group shows a lower probability of remaining with the same company and no significant effect on the likelihood of transitioning to a permanent contract. These findings are especially relevant given that IC are intended for this profile, yet their benefits appear more limited and differently timed. By contrast, individuals with prior work experience (ATT 3), reported as part of the sensitivity analysis, exhibit outcomes similar to those observed in the original group. Wage effects are persistent and increase over time, reaching the highest levels among all groups. They also show a higher probability of securing a permanent contract and of remaining with the same employer. Taken together, these differences suggest that deviations from the intended targeting of the programme tend to benefit individuals with accumulated experience, consistent with the heterogeneity patterns discussed above, whereby prior experience amplifies medium- and longer-term returns.
Panel B reports sensitivity analyses. The coefficients for ATT 4 (university graduates) remain positive but are smaller than those in ATT 1, indicating modest gains in contract duration and earnings. However, the probability of transitioning to a permanent contract is higher for this group, while the likelihood of remaining with the same employer is comparatively lower. For ATTs 5 and 6, which assess whether the effects observed in ATT 1 and ATT 4 hold when the IC lasts at least six months (the legal minimum), results are broadly consistent across all outcomes.
In addition, the third column of Table A4 reports a sensitivity analysis in which the conditioning set is augmented with richer pretreatment labor-market information. Incorporating detailed indicators of prior labor-market attachment and firm characteristics leaves the estimated effects not only statistically significant but also slightly larger in magnitude. In particular, both contract duration and post-treatment wages exhibit modest increases relative to the baseline specification. This pattern suggests that the baseline estimates are not driven by omitted pretreatment heterogeneity and provides additional support for the credibility of the unconfoundedness assumption underlying the matching strategy.
Moreover, we assess the robustness of the results to state-of-the-art causal estimators based on observables. Implementing a doubly robust augmented inverse probability weighting (AIPW) estimator yields ATT estimates that are very close to the baseline results across all outcomes (Table A4, last column). Both contract stability and post-treatment wage effects remain statistically significant and of comparable magnitude, indicating that the findings are not sensitive to the choice of estimator. Taken together, these results further strengthen the credibility of the identification strategy.
Finally, we assess sensitivity to unobserved selection using Rosenbaum bounds. The results indicate that wage effects are robust to substantial hidden bias, while the estimated effect on job stability is more sensitive to relatively modest unobserved heterogeneity ( AppendixTable A6). Taken together, these results further strengthen the credibility of the identification strategy.
8. Conclusions
This study contributes to the literature on the evaluation of ALMPs by providing empirical evidence on the effects of IC –a relatively under-researched policy instrument in both Spain and Europe- on the labor market outcomes of qualified young individuals. Specifically, it assesses the impact of these contracts on wages and job stability between 2016 and 2019, comparing them to TCs, the second most common form of employment in Spain.
Using data from the CSWL, the analysis applies Propensity Score Matching to address selection biases and estimate ATT for contract duration and earnings over the short (one year) and medium term (two to three years). In addition, logistic models were used to estimate the probability of transitioning to a permanent contract and of remaining with the same employer. These outcomes were assessed for a baseline group as well as subgroups defined by gender, education and prior work experience.
Results consistently show that ICs, compared to TCs, are associated with higher wages, longer subsequent contracts and a greater likelihood of securing permanent employment. However, individuals with TCs were more likely to remain with the same employer. While these general patterns held across all subgroups, important differences emerged by gender, education level and previous work experience.
For instance, the average duration of subsequent contracts was 30 days longer for men than for women. Although women initially benefited more in terms of wage gains, this advantage diminished over time, while for men it increased –suggesting the presence of gender-related structural factors limiting long-term benefits for women. Furthermore, although both men and women were less likely to remain with the same employer, the probability was significantly lower for women.
Contract duration and education also proved relevant. For university graduates, the positive impact of ICs increased with contract length. Conversely, for those with vocational training, shorter internships had greater effects –indicating that the benefits of these contracts are not restricted to highly educated individuals.
Prior work experience significantly influenced outcomes. Individuals over 22 –likely with more experience- saw increasing wage gains over time and were nearly twice as likely to remain with the same employer or secure a permanent position compared to younger participants. Additional subgroup analyses confirmed that while those without prior experience experienced modest, stable wage gains, they had lower employer retention rates and no significant improvement in permanent job prospects. By contrast, participants with experience benefited the most from the IC, with the highest wage gains, greater employer retention and better transition to stable jobs.
These results reveal a central paradox: although ICs were originally designed for first-time entrants into the labor market, they are frequently used by individuals with prior experience, who also reap the greatest benefits. This raises questions about the policy’s implementation and targeting.
Finally, the groups that benefited most were men with university degrees and prior experience who completed contracts of at least the legal minimum. These patterns suggest employers may favor retaining men and experienced workers, while women may need to switch employers to improve their job prospects after an internship.
The strongest benefits of IC were concentrated among individuals with favorable labor market profiles although the effects for other groups were still significant and positive, supporting the relevance of these contracts as an ALMP tool. Better data, including the field of study and occupation, would help to further understand the impact of this type of hiring, as well as to propose changes to improve its results and encourage its use by companies.
A final note concerns the choice of control group. Ideally, the counterfactual would be individuals eligible for but not employed under an IC. TC holders were selected as the closest proxy, following practices in prior research. However, caution is warranted in interpreting results, given the distinct characteristics and volatility of TCs.
In conclusion, this study finds that ICs significantly enhance early labor market trajectories, particularly for young people with greater vulnerability. Their effects on wage growth, contract stability and transitions to permanent jobs support their continuation and expansion, especially for women, individuals with vocational training and those without prior experience. At the same time, the results indicate that the strongest measured benefits accrue to individuals with prior work experience, revealing a mismatch between the policy’s intended target and its realized impacts. This suggests that improving targeting and enforcement, rather than redefining the policy’s objectives, is key to enhancing its effectiveness.
An additional avenue for future research is the further exploration of heterogeneity in prior employment experience. While this paper provides initial evidence based on alternative experience thresholds, richer measures of labor-market attachment and larger samples would allow for a more detailed assessment of how prior experience shapes programme effectiveness.
Notes
For the full text of the original regulation, see: Link to boeLink to the website of boe. For legislative modifications introduced between 2015 and 2024, see: Link to boeLink to the website of boe.
For a detailed reading of the Law, see: Link to boeLink to the website of boe 1&p=20130727#a13.
For the full text of Royal Decree-Law 32/2021, of December 28, on urgent measures for labour reform, job stability guarantee and labor market transformation, see: Link to boeLink to the website of boe.
For detailed legislative changes to the Internship Contract, refer to Table S1 in the supplementary material.
The selected variables align with those examined in previous research: variables 1 and 2 have been analyzed by Cueto and Rodríguez-López (2019), and variables 3 and 4 by Picchio and Staffolani (2019) and De la Rica and Gorjón (2022).
The relevant legislative changes are detailed in Section 3 of this study and in Vaquero-García et al. (2024).
Descriptive statistics related to Autonomous Community and Economic Activity are displayed in Table S2 in the supplementary material.
This data set corresponds to the Treatment and Control Groups 1 with which the estimations of sections 7.2 and 7.3 were made.
For further clarity, Panel 1 in the supplementary material illustrates covariate distributions before and after matching, demonstrating significant reduction in selection bias.
References
Further reading
Appendix
Characteristics of the treatment and control groups
| Shared characteristics | Group | Previous experience | Education level | Contract duration |
|---|---|---|---|---|
| Individuals who signed an Internship Contract in 2016 under 30 years | Treatment Group 1 | No distinction | Higher vocational training and university | At least 3 months |
| Treatment Group 2 | No experience prior to signing the contract | Higher vocational training and university | At least 3 months | |
| Treatment Group 3 | With experience prior to signing the contract | Higher vocational training and university | At least 3 months | |
| Treatment Group 4 | No distinction | University | At least 3 months | |
| Treatment Group 5 | No distinction | Higher vocational training and university | At least 6 months | |
| Treatment Group 6 | No distinction | University | At least 6 months | |
| Individuals who signed a Temporary Contract in 2016 under 30 years | Control Group 1 | No distinction | Higher vocational training and university | At least 3 months |
| Control Group 2 | No experience prior to signing the contract | Higher vocational training and university | At least 3 months | |
| Control Group 3 | With experience prior to signing the contract | Higher vocational training and university | At least 3 months | |
| Control Group 4 | No distinction | University | At least 3 months | |
| Control Group 5 | No distinction | Higher vocational training and university | At least 6 months | |
| Control Group 6 | No distinction | University | At least 6 months |
Covariate equilibrium diagnostics
| Characteristic | Control group | Treatment group | p | Standardized mean difference | Condition |
|---|---|---|---|---|---|
| Sample | 489 | 489 | |||
| Gender | 0.701 | 0.029 | Balanced | ||
| Male | 256 (52.4) | 249 (50.9) | |||
| Age | 0.474 | 0.052 | Balanced | ||
| Older than 22 years | 411 (84.0) | 420 (85.9) | |||
| Education | 1 | 0.004 | Balanced | ||
| University | 322 (65.8) | 323 (66.1) | |||
| Nationality | 0.794 | 0.025 | Balanced | ||
| Spanish | 456 (93.3) | 459 (93.9) | |||
| Autonomous community | 0.074 | 0.369 | Balanced | ||
| Andalucía | 60 (12.3) | 69 (14.1) | |||
| Aragón | 17 (3.5) | 9 (1.8) | |||
| Asturias | 8 (1.6) | 24 (4.9) | |||
| Islas Baleares | 13 (2.7) | 4 (0.8) | |||
| Canarias | 22 (4.5) | 14 (2.9) | |||
| Cantabria | 7 (1.4) | 6 (1.2) | |||
| Castilla y león | 21 (4.3) | 18 (3.7) | |||
| Castilla-La mancha | 22 (4.5) | 20 (4.1) | |||
| Cataluña | 77 (15.7) | 100 (20.4) | |||
| Comunidad valenciana | 57 (11.7) | 50 (10.2) | |||
| Extremadura | 27 (5.5) | 11 (2.2) | |||
| Galicia | 30 (6.1) | 34 (7.0) | |||
| Comunidad de Madrid | 74 (15.1) | 75 (15.3) | |||
| Murcia | 15 (3.1) | 9 (1.8) | |||
| Navarra | 9 (1.8) | 13 (2.7) | |||
| País vasco | 28 (5.7) | 31 (6.3) | |||
| La rioja | 1 (0.2) | 2 (0.4) | |||
| Ceuta | 1 (0.2) | 0 (0.0) | |||
| Economic activity | 0.060 | 0.733 | Balanced | ||
| Primary sector | 17 (3.5) | 0 (0.0) | |||
| Manufacturing industry | 38 (7.8) | 60 (12.3) | |||
| Energy Supply \1 | 2 (0.4) | 1 (0.2) | |||
| Water Supply \2 | 0 (0.0) | 7 (1.4) | |||
| Construction | 16 (3.3) | 28 (5.7) | |||
| Commerce | 75 (15.3) | 46 (9.4) | |||
| Transportation and storage | 10 (2.0) | 10 (2.0) | |||
| Hospitality | 31 (6.3) | 8 (1.6) | |||
| Information and communications | 26 (5.3) | 30 (6.1) | |||
| Financial activities | 6 (1.2) | 15 (3.1) | |||
| Real estate activities | 2 (0.4) | 1 (0.2) | |||
| Professional, scientific, and technical activities | 55 (11.2) | 126 (25.8) | |||
| Administrative and support services | 83 (17.0) | 67 (13.7) | |||
| Education | 58 (11.9) | 26 (5.3) | |||
| Healthcare and social services | 40 (8.2) | 53 (10.8) | |||
| Artistic activities \3 | 17 (3.5) | 1 (0.2) | |||
| Other services | 13 (2.7) | 10 (2.0) |
ATT 1 Estimates on contract duration (days) and monthly wages (€) by prior employment experience
| Outcome | Complete | < 120 Days | ≥120 Days | < 1 Year | ≥ 1 Year |
|---|---|---|---|---|---|
| Matched observations | 489 | 80 | 265 | 192 | 154 |
| contract duration | 106.72 *** | 138.92 ** | 73.23 ** | 124.86 *** | 42.22 |
| Std. Error | (20.97) | (50.48) | (28.59) | (33.28) | (37.98) |
| Monthly Salary 2017 (€) | 174.67*** | 232.42** | 168.86 *** | 271.66 *** | 82.64 |
| Std. Error | (39.00) | (98.66) | (45.02) | (58.31) | (66.52) |
| Monthly Salary 2018 (€) | 189.86*** | 247.40** | 204.31*** | 256.43*** | 158.95* |
| Std. Error | (46.12) | (106.16) | (60.23) | (70.28) | (81.33) |
| Monthly Salary 2019 (€) | 194.74*** | 195.28* | 334.91*** | 302.70*** | 320.00*** |
| Std. Error | (47.82) | (116.80) | (63.21) | (76.38) | (82.63) |
ATT = Average Treatment Effect on the Treated. Significance levels: ***p < 0.01, **p < 0.05, * < 0.10. Reweighted estimates adjust the full IC sample to match the observable characteristics of first-time labor market entrants. Estimates are reported for subgroups defined by alternative thresholds of prior employment experience (accumulated days and years of prior work). While these measures allow for a transparent partition of the sample, they may not fully capture qualitative differences in prior labor-market attachment (e.g. job stability or relevance of experience). Further disaggregation would substantially reduce sample sizes and statistical power in some cells. Results should therefore be interpreted as indicative of broad differences in prior experience rather than precise categorizations
ATT Estimates: Baseline, reweighted estimates and sensitivity analyses (rich controls and AIPW)
| Baseline | Reweighted | Rich controls | AIPW | |
|---|---|---|---|---|
| Outcome | (ATT 1) | (ATT-RW) | (ATT-RC) | (ATT-AIPW) |
| Contract duration | 106.72*** | 107.48*** | 119.51*** | 114.89*** |
| Std. Error | (20.97) | (24.76) | (21.86) | (17.41) |
| Monthly Salary 2017 (€) | 174.67*** | 142.56*** | 186.07*** | 156.01*** |
| Std. Error | (39.00) | (41.97) | (33.13) | (26.09) |
| Monthly Salary 2018 (€) | 189.86*** | 160.15*** | 222.39*** | 191.27*** |
| Std. Error | (46.12) | (50.58) | (42.16) | (33.66) |
| Monthly Salary 2019 (€) | 194.74*** | 170.06*** | 223.89*** | 207.27*** |
| Std. Error | (47.82) | (55.79) | (45.92) | (35.99) |
ATT = Average Treatment Effect on the Treated. Significance levels: ***p < 0.01, **p < 0.05, *p < 0.10. Reweighted estimates adjust the full IC sample to match the observable characteristics of first-time labor market entrants. Estimates with rich controls extend the baseline specification by incorporating firm size (number of employees), number of prior contracts, and total days worked prior to treatment. Sector of activity is already controlled for, while measures capturing months previously worked or unemployment spells are excluded to limit multicollinearity. AIPW estimates are obtained using a doubly robust augmented inverse probability weighting (AIPW) estimator and include an extended specification including rich pretreatment labor-market histories (prior earnings, months worked, unemployment spells, number and duration of previous contracts, firm size and sectoral information, and a finer age structure). Robust standard errors are reported in parentheses
Inference after matching: clustered and matched-set boostrap standard errors
| Outcome | ATT_Cluster (1) | SE_Cluster (2) | Pvalue_Cluster (3) | ATT_boot (4) | SE_boot (5) | Pvalue_boot (6) |
|---|---|---|---|---|---|---|
| Contract duration | 119.5174 | 21.86555 | 0.00 | 120.3417 | 16.02326 | 0.00 |
| Monthly Salary 2017 (€) | 186.0692 | 33.13059 | 0.00 | 187.7505 | 26.10446 | 0.00 |
| Monthly Salary 2018 (€) | 222.3907 | 42.16642 | 0.00 | 225.0969 | 33.39868 | 0.00 |
| Monthly Salary 2019 (€) | 223.8998 | 45.92191 | 0.00 | 226.7438 | 34.36518 | 0.00 |
ATT estimates are obtained after propensity score matching. Columns (1)–(3) report ATT estimates, robust standard errors, and p-values computed using standard errors clustered at the matched-set (subclass) level, allowing for arbitrary correlation within each matching set. Columns (4)–(6) report ATT estimates, standard errors, and p-values obtained from a matched-set bootstrap, where resampling is performed at the level of matched sets rather than individual observations. This procedure preserves the dependence structure induced by the matching process
Sensitivity to hidden bias: Rosenbaum bounds
| Outcome | Critical Γ | Interpretation |
|---|---|---|
| Contract duration | 1.30 | A hidden confounder increasing the odds of treatment by 30% would render the effect statistically insignificant |
| Monthly Salary 2017 (€) | 3.00 | Conclusions remain unchanged unless an unobserved factor increases treatment odds by 200% |
| Monthly Salary 2018 (€) | 1.80 | Robust to moderate levels of unobserved selection |
| Monthly Salary 2019 (€) | 1.80 | Robust to moderate levels of unobserved selection |
Γ denotes the Rosenbaum sensitivity parameter. A value of Γ = 1\Gamma = 1Γ = 1 corresponds to no hidden bias. Reported values indicate the smallest Γ\GammaΓ at which the treatment effect loses statistical significance at the 5% level. Rosenbaum bounds are computed using the matched sample underlying the baseline estimate
Supplementary material
The supplementary material for this article can be found online.
