Language and civic education requirements as gatekeepers or tools for economic integration: a question of gender?

4.1 Cross-national studies

One of the main contributions to the rather limited and specific literature on mandatory LCE requirements and their effects is arguably Goodman and Wright (2015). They test whether mandatory language and civic integration requirements have any effect on the integration of immigrants in the EU-15 countries by using data from the European Social Survey (ESS) and applying CIVIX as a policy indicator of LCE requirements (see Footnote 2). The integration outcome is measured in political, social, and economic integration. They observe a positive and significant effect only for political integration, but not for economic or social integration. The authors stress the discovery of no effect at all, either positive or negative, leading them to conclude that civic integration policies may be imposed primarily for symbolic reasons.

Inspired by Goodman and Wright (2015), Neureiter (2019) conducts a similar study by also applying CIVIX to ESS data for 15 EU-countries but with a more advanced methodological strategy. With an econometric difference-in-differences approach, Neureiter finds, in contrast to Goodman and Wright, that mandatory LCE requirements have a strong and positive effect on the economic integration of non-EU born migrants, while no effect is observed for social and political integration. The conflicting results of these two studies underscore the need to further investigate the effects of LCE policies, potentially through alternative methods and policy indicators. The methodology of Neureiter is enhanced compared to that of Goodman and Wright, by including a larger set of control variables and addressing unobservable heterogeneity. Still, there is room for improvement regarding transparency and robustness, for example by including confidence intervals for the predicted probabilities.

By applying several integration indices, including MIPEX (see Footnote 2), Koopmans (2010) finds that restrictive integration policies produce better socio-economic integration of immigrants, particularly in contrast to laissez-faire multiculturalism. Koopmans shows that Sweden, Belgium and the Netherlands, countries characterized by generous welfare states, multicultural policies and easy access to equal rights without major incentives to learn the host language, display low levels of labor market participation among immigrants. In contrast, countries with more restrictive and assimilationist integration policies such as Germany, Austria, and France, or limited welfare states such as the UK, have produced better socio-economic integration (Koopmans 2010: 20–22).

Other studies have shown that language and integration requirements for legal statuses do not necessarily motivate migrants but instead exclude certain groups from a secure legal status. Böcker and Strik (2011) argue that standardized language and KoS tests limit the possibilities to tailor language and integration programs to the specific needs of immigrant groups. They highlight results from an interview study where many respondents perceived the required level of language proficiency as being too low to improve their labor market position, but too high to include all immigrants with a will to integrate (Böcker and Strik 2011:182). Van Oers (2021) reports that citizenship tests in Germany and the Netherlands prevent specific groups of immigrants from fully integrating into society, including the elderly, those with limited education, women and refugees (Van Oers 2021: 281). Reinforcing this critical stance, Guzi et al. (2023) demonstrate that designing less strict immigration and integration policies could help reduce existing labor market disadvantages for immigrants (Guzi et al. 2023: 4182). To summarize, the findings on the effects of LCE requirements and other integration policies remain rather mixed and inconclusive.

4.2 LCE requirements from an applied linguistics perspective

As stated in the introduction, the field of applied linguistics lacks empirical evaluation of these policies, with the exception of qualitative studies such as Pulinx and van Avermaet (2017) and Cvejnová and Sladkovská (2017). It should however be added that, although not investigating the direct effect of integration policies, scholars like Gujord (2023) and Carlsen and Hamidi (2023) have shown through the examination of a large number of language test results that low-literate learners have little chance of achieving levels required for citizenship, such as the B1 level. As such, these latter studies contribute to some extent to the empirical evaluation of these policies. The most distinctive feature of linguistic research on language testing is, nevertheless, the critical approach that is often taken (see Schildt and Deygers, 2024). This approach is found among language testing experts, but also among researchers with a sociolinguistic focus. Sociolinguistic researchers criticize the consequences of language tests for human rights and argue that the tests are discriminatory. In his analysis of policy documents and newspaper articles, Milani (2008) discusses language tests based on Bourdieu’s concept of “rites of institution” – an approach also advocated by Shohamy (2022). Milani argues that language, in the framework of civic integration, is intimately connected to culture, norms and values, and that by requiring language tests for citizenship, governments require linguistic and cultural assimilation. Horner (2015) makes a similar case to Milani (2008) and places great emphasis on the concept of “linguistic authority” to explain the “policy of demands” that has emerged in the field of migration in Europe during the 2000s.

Civic integration has also been analyzed within applied linguistics from the perspective of political philosophy. Oakes and Warren (2007) analyze how language policy has been used in Quebec to strengthen national identity, taking their starting point in the two most common conceptualizations of citizenship in political theory: liberal and republican citizenship. The liberal view of citizenship is primarily rights-based, while the republican definition of citizenship places greater emphasis on obligations, primarily in terms of expectations to participate in social debate and contribute to the common good. The latter definition consequently places relatively high demands on citizens in terms of identification with the state. Oakes and Warren (2007) argue that the model chosen in Quebec, a so-called interculturalist model, lies somewhere between the French, republican definition of citizenship and the English-speaking Canadian, “multiculturalist” and liberal definition. Oakes and Peled (2018) review this interculturalist model (proposed by sociologist Gérard Bouchard), which expresses the view that integration means that both the new arrivals and the majority population of the country they have migrated to adapt to each other in different ways, but is nevertheless fundamentally based on the fact that there are majority and minority relationships between different groups and languages. The authors further explain how Bouchard has reasoned regarding the importance of a common “cement”. They refer to the political philosophers Banting and Kymlicka (2015) who question whether it is possible to build solidarity between different religious and ethnic groups solely on the basis of a common view of human rights and democracy, unless there is also a non-political cultural community. Oakes and Peled (2018), however, believe that this position by Bouchard, regarding the non-negotiable status of the French language, can be seen as problematic from a traditional, liberal perspective. It thus appears that the choice of citizenship model has a great significance for how a country views language requirements. In sum, whether the research stems from critical sociolinguistics or uses concepts from political philosophy, linguists mostly concentrate on critical/normative perspectives of these tests and have not studied effects, as is the case in the social sciences. The present endeavor is transdisciplinary in that it uses an index developed in applied linguistics with methods from the social sciences to measure, regardless of normative stance, the effects of civic integration policies, thus responding to the call from Schildt and Deygers (2024).

4.3 Gender perspectives on language acquisition and labor market integration

To incorporate the gender aspect, we turn to literature identifying gender differences in language acquisition and language-related labor market outcomes. In an Italian context, Ghio et al. (2023) find that female immigrants are more disadvantaged than males by poor linguistic proficiency in labor force participation. They relate this to the Italian immigrant labor market where domestic work and work in elderly care is strongly characterized by the employment of female immigrants. The female domination in the domestic sector makes proficiency in Italian more important for female immigrants than for male immigrants, as the former are more likely to need language skills and interact with native employers and speakers (Ghio et al. 2023: 388).

The female domination in the care sector is not unique to Italy. The fact that language skills seem to matter more for immigrant women in terms of labor market outcomes can largely be explained by the gender-segregated labor market, present in most countries. According to the Gender Gap Report of 2023, healthcare and care services is still a female-dominated field with a share of females around 65 %. Women also outnumber men in occupations within education, mainly in primary school teaching. Meanwhile, men are heavily overrepresented in sectors like infrastructure and construction or oil, gas and mining (World Economic Forum 2023).

The female-dominated occupations listed in the report are characterized by communicative elements and require robust skills in the native language. This contrasts with traditional male-dominated occupations like construction, which may be more physically demanding but require less language skills (Ghio et al. 2023). Studies also imply that immigration plays a significant role in reinforcing/maintaining the gendered labor market, particularly in the care sector. Östbakken et al. (2023) find that the female dominance in nurturant care is maintained largely due to a growing share of immigrant women in professions like nursemaids or doctors’ secretaries (Östbakken et al. 2023:836).

Furthermore, Birger and Bar-Haim (2023) find that migrant men who use a different language than the host language at home are not prone to educational-occupational mismatch, once education is controlled for. On the contrary, migrant women seem to suffer more from such mismatch when using a foreign language at home (Birger and Bar-Haim 2023: 279). The researchers attribute the disadvantage of women using a foreign language at home to cultural gender norms around the roles of men and women within the household, which shape decisions regarding division of work and working hours (Birger and Bar-Haim 2023: 280). All in all, these findings suggest that men and women could be affected differentially by language policies with respect to economic integration.

6.1 European Social Survey

The European Social Survey (ESS) is a biennial social survey that is carried out across many European countries. It is well known for its strict methodological standards and focus on academic research (Platt et al. 2022; European Social Survey 2024). While it was not specifically designed to study immigrants, the ESS’s high-quality and consistent measurements across countries makes it suitable material for cross-national analyses of immigrant experiences (Goodman and Wright 2015; Neureiter 2019; Platt et al. 2022).

We use two recent ESS rounds: 9 and 10. Even with only two rounds, this provides a large starting sample of n=76,769 and a relatively large share of respondents who identify as foreign-born (n=6,691). In round 9, interviews were conducted between 2018 and early 2020. For round 10 the period of conduction was longer than usual due to the COVID-19 pandemic,^[4] and some countries opted for self-completed written questionnaires instead of traditional interviews. Therefore, the data for certain countries are based on interviews while others are based on self-completed written questionnaires. We used ESS data for the 18 countries that we had LAPIM scores for. These were: Sweden, Denmark, Finland, Norway, United Kingdom, Ireland, Germany, France, Spain, Italy, Austria, the Netherlands, Belgium, Czech Republic, Poland, Slovenia, Bulgaria and Greece. All these countries participated in both rounds except for Denmark, which only took part in round 9, and Greece, which only took part in round 10. The ESS data has been anonymized and does not contain any personal information (https://www.europeansocialsurvey.org/about/privacy-and-data-protection/survey-participants).^[5]

6.2 Methods

6.2.1 Variables

The main dependent variable used in the study was employment status. It is a dichotomous variable for which individuals were coded either 1 or 0 based on their main activity over the previous 7 days. We were interested in the active population, meaning that a respondent was considered unemployed if he or she was unemployed and actively looking for work. Hence, employment was coded 1 if the respondent did not mark themselves as unemployed and actively looking for a job during the previous 7 days (i.e. had a job). Individuals who responded that they were unemployed and actively looking for a job were coded 0.

The main independent variable at the country level was the categorical variable based on LAPIM scores (0–100). Low values represent strict requirements, and high values imply more relaxed requirements. The LAPIM scores of countries in our sample ranged from 34 for Austria to 99 for Sweden. LAPIM scores were divided and categorized into three broader categories: strict, moderately strict, and lenient. LAPIM scores of 35–55 were considered strict, while the label moderately strict was assigned to LAPIM-scores between 55 and 75. LAPIM scores above 75 were treated as lenient.

Figure 5:

European map showing categorization of LCE requirements for the included countries.

Since there are other country-level factors that may explain overall employment outcomes, we also controlled for national unemployment rates and GDP per capita as continuous variables. Depending on the country and when the respondent was interviewed, the appropriate GDP per capita and unemployment rate for that year was used. This data was retrieved from Eurostat and the UK Office for National Statistics (2024a, 2024b).

Since the main point of interest was to analyze the labor market outcomes of immigrants (economic integration), we distinguished individuals born in the country where the interview was conducted from those who were not. Moreover, within the group of foreign-born participants, we distinguished between immigrants born in another EU country and non-EU migrants. Hence, the main independent variable at the individual level was origin, with native-born respondents coded 0, immigrants born in another EU country coded 1, and immigrants born outside the EU coded 2. By including natives in our sample, we could compare the relative economic integration across LAPIM categories, as it allowed for a benchmark value in the native residents’ employment level. Although the primary focus of our study was on non-EU migrants, EU-born migrants were included to test if they are also impacted by LCE requirements. The other main independent variable at the individual level was gender, a binary variable with females coded 1 and males coded 0. We controlled for age, the highest level of education completed, marriage status, residence, and also included a dummy variable for interview years before and after the COVID-19 pandemic so as to control for any potential effects of the pandemic and the different data gathering methods used. For more details about the sample, see Appendix 1.

6.3 Data analysis

Since our data is cross-sectional with a hierarchical structure where individual-level data is nested within countries, we applied a multilevel random effects regression model. As the outcome of employment is binary, logistic regression was used accordingly. Multilevel models account for the clustering of observations within countries and allow us to examine how contextual factors like LCE policies influence employment outcomes, while the random effects account for unobserved heterogeneity across countries. Random-effects models are beneficial for cross-sectional data where the variability arises from contextual differences in policies and conditions across countries rather than changes over time (Li et al. 2010; Hox et al. 2017: 11–16).

The appropriate number of second-level entities (in this case countries) needed in a multilevel model is debated in the literature. Bryan and Jenkins (2016: 19–20) argue that the number of second-level units should be larger than 25 for hierarchical linear models and larger than 30 for logistic models with binary outcomes to not produce biased estimates for second-level covariates. Stegmueller (2013: 754) proposes that estimates become less biased when the second-level sample size approaches 15 and beyond. In our case, we had 18 countries for which both ESS data and LAPIM scores were available. While ideally, we would have had data for a larger number of countries, the range of 15–20 second-level units made for considerably less biased results compared to country samples smaller than 15, judging by Stegmueller’s (2013: 754) study. To address this issue we also ran several robustness checks to validate the results, for example by re-running all multilevel models as linear probability models (LPMs) with clustered standard errors, another viable option for binary outcomes.

Since logistic regressions are not suited to stepwise models where different independent variables are included successively (Mood 2010), we directly present the full models with all variables. In a first model, the regression included all control variables and a two-way-interaction term between LAPIM category and origin. As non-EU migrants were the main focus of interest, they were set as the reference group. For the results, predicted probabilities were plotted with 95 % confidence intervals for each combination of the interaction. Full regression tables with odds ratios can be found in Appendix 2. In a second model, we included a three-way interaction term between LAPIM category, origin, and gender, to show how the associations found in model 1 change when interacting with gender.

7.1 Descriptive statistics

For the descriptive statistics, weights were applied to ensure that the survey sample was more representative of the whole population by correcting for selection bias, nonresponse, and population size across countries (Solon et al. 2015). Tables 1–3 illustrate employment by origin, LAPIM category, and gender separately with weighted shares. Figure 6 shows how employment varies by combinations of gender, LAPIM categories, and origin.

First, it should be noted that the share of employed individuals is high for all categories. As expected, both EU-born migrants and non-EU migrants exhibit slightly higher unemployment rates compared to natives, and this is particularly pronounced for non-EU migrants.

Respondents from countries coded as having moderate LCE requirements clearly exhibit higher unemployment rates, compared to individuals in countries categorized as strict or lenient. Somewhat surprisingly, there is no major difference in employment between men and women overall. Small gender differences in unemployment of at most 1.6 percentage points remain even when sorted by origin (Table 3).

Finally, Figure 6 displays how unemployment varies by gender, origin and LAPIM category, highlighting a pronounced unemployment rate among non-EU migrant women under moderate requirements and a relatively large unemployment rate among non-EU migrant men under strict policies. The differences by gender are more apparent among non-EU migrants than among EU migrants and natives, particularly under strict and moderate settings.

For non-EU migrant men, unemployment is lower under moderate and lenient requirements compared to strict requirements. Among female non-EU migrants, however, the unemployment share is smallest under strict policies, and lower than for non-EU migrant men, while it is substantially higher under moderate requirements. With these patterns in mind, plotted results from the two logistic multilevel models are presented below. Model 1 is limited to focusing on the interaction between LAPIM category and origin, while Model 2 also accounts for the interaction with gender.

Figure 6:

Share (%) of unemployed by gender, origin, and LAPIM category.

7.2 Model 1

The first model estimates the associations between LCE requirements and employment status via the interaction between LAPIM categories and origin. For the full model with the odds ratios of all covariates, including the two-way interaction and all control variables, see Table 7 in Appendix 2. Figures 7 and 8 show the predicted probabilities of employment across LAPIM categories and origin. To determine any significant differences in predicted employment between these groups, one should look at whether the 95 % confidence intervals overlap or not.

Overall, the employment rate of respondents in this sample is rather high (see Section 8.1), and therefore the predicted probabilities only differ from 0.89 (non-EU migrants under lenient policies) to 0.96 (natives under strict policies).

7.2.1 Within groups, across LAPIM categories

Figure 7 shows the predicted probabilities of employment within each group of origin. The predicted probability of employment is higher for natives than for migrants across all LAPIM categories. Overall, EU-born migrants are estimated to have a higher likelihood of being employed compared to non-EU migrants, although the relatively wide confidence intervals imply some uncertainty. For both natives and EU-born migrants, the predicted probability of employment is lower under moderate requirements than under lenient and strict requirements. The same pattern is not seen for non-EU migrants. However, these differences across LAPIM categories are not significant, as the confidence intervals overlap.

Figure 7:

Predicted probabilities of employment by LAPIM category and origin with 95 % confidence intervals. Blue squares: native born, Green: EU-born immigrants, Yellow: non-EU migrants.

Across none of the three origin categories—natives (blue), EU-born migrants (green), and non-EU migrants (yellow)—does the predicted probability of employment differ significantly between LAPIM categories, as indicated by the overlapping confidence intervals.

Thus, neither EU-born nor non-EU migrants exhibit a statistically significant increase or decrease in the likelihood of employment under any specific LCE requirement setting.

7.2.2 Within LAPIM categories, between groups

In Figure 8, the same predicted probabilities are sorted differently to illustrate any significant differences between natives, EU-born migrants and non-EU migrants within the different LAPIM categories. A stepwise explanation is given for each LAPIM category.

Under strict requirements, the predicted probability for non-EU migrants is significantly lower compared to both natives and EU-born migrants under strict requirements (first yellow square compared to first blue and green square). Hence, both natives and EU-born migrants are substantially and significantly more likely to be employed under strict language policies compared to non-EU migrants. No significant difference between EU-born migrants and natives was found.

Figure 8:

Predicted probabilities of employment by LAPIM category and origin with 95 % confidence intervals.

Under moderate requirements, no significant differences were observed in predicted employment between natives, EU-born migrants, or non-EU migrants as all three confidence intervals overlap. This is also implied by the interaction term between LAPIM and origin (see Table 7 in Appendix 2), showing that the advantage native-born individuals have over non-EU migrants in terms of employment odds decreases when moving from a strict to a moderate LCE setting.

Under lenient requirements, non-EU migrants have a significantly lower predicted probability of employment compared to natives (third yellow square compared to third blue square) but not compared to EU-born migrants (third yellow compared to third green square).

Hence, the plot shows that the predicted gap in labor market outcomes between natives and non-EU migrants is larger (and significant) in countries with strict and lenient LCE requirements, while the likelihood of employment for the two groups is more similar under moderate policies. The predicted probabilities for EU-born migrants are not significantly different for any LAPIM category compared to natives (green compared to blue), implying that EU-born migrants may exhibit similar employment levels as natives.

Figure 9:

Predicted probability of employment by LAPIM category, origin, and gender, with 95 % confidence intervals. Natives in blue, EU-born immigrants in green, non-EU migrants in yellow. Squared nodes for males and diamond-shaped nodes for females.

7.3 Model 2

In the second model, the three-way interaction between LAPIM, origin, and gender is added. We show the predicted probability of employment by gender for natives, EU-born migrants, and non-EU migrants across LAPIM categories (see Figure 9).

Under strict requirements, male non-EU migrants have significantly lower predicted probabilities of employment compared to native men. Non-EU migrant men under strict policies also exhibit significantly lower probabilities of employment compared to male EU-born migrants. The same is not true for non-EU migrant women compared to native or EU-born women. Neither gender among EU-born migrants demonstrate significantly different probabilities of employment compared to natives.

Figure 10:

Predicted probability of employment by LAPIM category, origin, and gender with 95 % confidence intervals (sorted differently). Native-born in blue, EU-born immigrants in green, non-EU migrants in yellow. Squared nodes for males and diamond-shaped nodes for females.

Under moderate requirements, neither gender among non-EU migrants have significantly lower predicted probabilities compared to natives or EU-born migrants. The same is true for both genders among EU migrants compared to natives. The predicted probabilities for both male and female non-EU migrants are not significantly different from those of male and female EU-born migrants.

Under lenient requirements, non-EU migrant men do not have a significantly lower predicted probability of employment compared to native men. On the contrary, female non-EU migrants have a significantly lower likelihood of being employed compared to native women. This does not hold compared to EU-born migrants. Neither are any of the predicted probabilities for both genders among EU-born migrants significantly different from natives.

When examining the non-EU migrant group across LAPIM categories, the only statistically significant difference is a higher predicted probability of employment for non-EU migrant men under moderate requirements compared to strict requirements (second yellow square vs. first yellow square). Under moderate requirements, non-EU migrant men also have significantly higher estimated chances of employment compared to non-EU migrant women (second yellow diamond). Finally, even if tempered by overlapping confidence intervals, the pattern of predicted employment probabilities for female non-EU migrants suggests a potential positive association between stricter requirements and employment outcomes for this group.