Does timing matter? Language course participation and language outcomes amongst new immigrants

The study investigates whether the timing of enrolment into formal host-country language courses affects subsequent credentialed language proficiency among new immigrants. Against a backdrop of substantial policy investment in language training in high-income countries and a historical contrast with earlier policy gaps, the paper notes that prior research has seldom examined the timing of formal training, instead treating participation as a static factor. Drawing on life-course theory (timing matters) and literature on delays in other integration processes (e.g., asylum procedures), the authors hypothesize that earlier enrolment may foster better outcomes by structuring initial learning, avoiding language learning plateaus, and boosting motivation and belonging. Using longitudinal data from Flanders (Belgium) for 2009–2021, the paper evaluates whether later enrolment is associated with lower attainment of Dutch language certificates and whether such associations persist across migrant legal categories when accounting for compositional differences.

The paper situates its inquiry within: (1) the Flemish policy context of civic integration and publicly funded language training for newcomers, aligned to CEFR levels (A1, A2, B1+), with documented variability in enrolment timing due to supply (fixed starts, waiting lists) and demand constraints (housing, mobility, childcare, procedures). (2) The three-pillar framework for migrants’ language acquisition (exposure, efficiency, incentives) by Chiswick and Miller, including pre- and post-migration exposure, individual efficiency factors (education, age, gender, linguistic distance), and incentives (expected returns and costs). (3) Mechanisms through which timing may matter: early formal training structures knowledge that can be reinforced through daily practice; prevents language learning plateaus (phonological, morphological, syntactic, semantic/pragmatic) formed during unguided informal learning; and enhances motivation/belonging, potentially avoiding negative cycles of low proficiency and disengagement. (4) Potential (self-)selection mechanisms that can confound timing-outcome associations: migration background (e.g., legal category, origin, arrival cohort, waiting lists), demographics (age, gender, children), human capital (education, learning capacity, prior language knowledge), labour market attachment, social integration resources (e.g., family networks), and ideational factors (valuation of host-language learning). Prior empirical work has rarely tested timing causally; one key study (Hoehne and Michalowski, 2016) focused on pre-1975 Turkish and Moroccan migrants and is not readily generalizable to recent cohorts.

Data: Longitudinal administrative microdata from the Crossroads Bank for Civic Integration (Flanders), covering all immigrants entering Belgium in 2009–2021 who subsequently resided in Flanders. From 117,818 immigrants who ever participated in language courses, the study excludes educational migrants (due to unobserved acquisition in education) and those holding B1+ at intake, yielding 114,022 immigrants and 6,899,353 monthly person-period observations. Intake testing assigns starting levels and learning tracks. Courses and credentials follow CEFR (A1, A2, B1+). During the observation period, courses were subsidized and free of charge. Outcome: Progression in credentialed Dutch proficiency, operationalized as ordered levels: (I) no certificates, (II) A1, (III) A2, (IV) B1 or higher. Analytic stratification by intake level: sub-A1 (N=88,845), A1 (N=14,793), and A2 (N=10,384). Model: Ordered logit models estimate the odds of progression over time as a function of timing of course participation. Key predictors include: time since course start (cubic specification) and timing of participation (scaled to 6-month shifts in the lag between arrival and enrolment). Nested model specifications: - Model 1: baseline with timing and time since enrolment. - Models 2–6 incrementally add compositional controls: migration background (legal migration type: labour, family, humanitarian, regularisation, other/unknown; region of origin: EU15, EU10, EU3, other European, North African, other African, North American, other American, Asian, Australian, unknown; year of arrival), demographics (gender, children, age at arrival categories), human capital (educational attainment; intake-based learning capacity track: basic, long adult, standard adult, short-track, unknown; self-reported knowledge of other languages at arrival with detailed categories), and labour market attachment (intake status: employed/studying vs not; time-varying first registration as jobseeker at the employment office). - Model 7: interaction between timing and legal migration type to assess heterogeneity by migrant category, controlling for all compositions. Estimation and interpretation: Exponentiated ordered logit coefficients reported as odds ratios for progressing k or more levels versus less than k. Average Marginal Effects (AMEs) are computed to illustrate timing effects on probabilities of specific credential outcomes over months since start. Robustness checks: (a) Full model stratified by legal migration category (reliable for sub-A1 only) showed timing estimates consistent with main interaction results. (b) Excluding the time-varying employment office contact yielded substantively similar timing effects, suggesting limited bias from this variable. All controls except employment office contact are time-constant or measured at arrival/intake, limiting reverse causality risks.

• Sub-A1 intake group (largest subgroup): A 6-month delay in enrolment associates with an 11.5% decrease in the odds of progression (OR=0.885) in the baseline; a 24-month delay corresponds to a 38.7% decrease. After controlling for migration background, this disadvantage reduces to ~6.3%; adding human capital further halves the association; in the full model (all controls), a smaller but still negative association remains. - A1 intake group: Similar negative association in baseline (6-month delay OR≈0.873–0.907 across models), with the effect roughly halving when controlling for migration background and human capital; about one-third of the baseline gap persists in the full model. - A2 intake group: Baseline association is weaker but negative (6-month delay OR=0.914). Notably, after adding migration background or human capital, the timing disadvantage can widen; in the full model, a persistent negative association remains (e.g., OR=0.821). - Heterogeneity by legal migration type (Model 7): For sub-A1 starters, a 6-month delay is associated with: • Labour migrants: 3.6% advantage in odds of progression (OR=1.036). • Family migrants: 5.3% decrease (combined OR=1.036×0.914). • Humanitarian migrants: 2.9% decrease (combined OR=1.036×0.937). A 24-month postponement (common in data) corresponds to ≈19.6% lower odds among family migrants and ≈11.1% among humanitarian migrants. Among higher intake levels, family migrants show the strongest negative timing penalties (e.g., 8.3% per 6 months from A1; 19.9% per 6 months from A2). - Composition effects: Female and younger migrants exhibit higher progression odds. Higher education and stronger intake-assessed learning capacity strongly predict progression; knowledge of certain other languages at arrival also differentiates outcomes. Employed/studying at intake associates with lower progression odds, whereas contact with the employment office positively associates with outcomes. - AMEs over time confirm that enrolment delays increase the probability of no progression and reduce probabilities of reaching A1, A2, and B1+; these gaps diminish notably after controlling for migration background and human capital but do not fully disappear for some groups. Overall: Much of the negative association between postponed enrolment and outcomes is explained by migration background and human capital, but a meaningful, persistent disadvantage remains for family and humanitarian migrants.

The results directly address whether delaying formal language course enrolment is linked to poorer credential outcomes. While initial associations suggest substantial penalties for postponement, much of the observed disadvantage is attributable to compositional differences—particularly migration background and human capital—indicating that simple cross-sectional interpretations can overstate timing effects. Nevertheless, the persistence of a negative association among family and humanitarian migrants implies that timing likely matters more for these groups, consistent with theoretical mechanisms: early structured learning may prevent plateaus formed during informal acquisition, sustain motivation and belonging, and create positive feedback loops between proficiency and exposure. In contrast, labour migrants may sometimes benefit from short delays, possibly due to job-related exposure and incentives. These findings highlight that policies promoting timely access to formal courses are particularly salient for family and humanitarian migrants, for whom delays may exacerbate vulnerabilities and constrain longer-term integration trajectories. However, given potential unobserved selection (e.g., social networks, motivational or ideational factors), the remaining associations should not be taken as definitive causal effects.

This study advances understanding of how the timing of participation in formal language training relates to credentialed proficiency among new immigrants. Leveraging rich longitudinal registers for Flanders (2009–2021), it shows that although delayed enrolment is associated with lower progression, especially for those starting from sub-A1 or A1, a substantial portion of this association reflects migration background and human capital. Crucially, a persistent negative linkage remains for family and humanitarian migrants, pointing to greater susceptibility to less effective informal trajectories and motivational setbacks when training is postponed. Policy implications include prioritizing early course access and supports (e.g., addressing childcare and administrative bottlenecks) for these groups. Future research should employ designs suited for causal inference (e.g., exploiting exogenous variation in waiting times) and test external validity in other contexts, given similarities across OECD language programs and the overall scarcity of timing-focused studies.

• Observational design: Despite extensive controls, causal effects cannot be established; unobserved factors (e.g., social networks, attitudes toward host-language learning, perceived returns) may confound timing-outcome links. - Measurement constraints: Only one time-varying control (employment office contact) could partially mediate or be affected by language progression; robustness checks mitigate but cannot eliminate concerns. - Heterogeneity analyses for A1/A2 starters by legal category were limited by small cell sizes, reducing precision. - Generalizability: Results pertain to Flanders’ context and period (2009–2021); although programs are similar across OECD countries, direct extrapolation requires caution. - Formal credentials focus: Outcomes are credential-based; informal skills gains not leading to certification may be undercaptured.