The role of education in innovation–migration nexus in Europe

The channels through which immigration affects innovation concern the increase in the size and density of the population, in its share of immigrants, the skill composition of immigrants and their cultural diversity (Ozgen et al. 2012).

As to the effects on the population of receiving societies, it can stimulate higher levels and diversity of local production besides imports (Mazzolari and Neumark 2012) and, in the long run, product and process innovation, firm growth leading to further innovation (Freeman and Soete 1997). Increased population density by immigrants through agglomeration stimulates the exchange of ideas and spillovers of knowledge (for an overview, see Carlino and Kerr 2015).

As Borjas (1999) highlights, there is a process of self-selection guiding migration in terms of age, size and skill composition and out-selection or external selection on the base of specific characteristics of immigrants (Bertoli and Rapoport 2013). As to self-selection, skilled immigrants tend to be younger, more entrepreneurial and less risk averse, which also determine their higher mobility (Kloosterman and Rath 2003; Poot 2008). When changing their residence, they bring with them new ideas, knowledge and work practices with a spillover effect in host countries (Ozgen et al. 2012). Self-selection can also be driven by the effect of migration networks, as immigrants clustered in close geographic areas tend to develop social networks, which provide support also in terms of information and knowledge to newly arrived immigrants who transform them into tangible resources (Massey 1999). In this respect, large networks by reducing the moving costs tend to increase the flow of low-educated migrants (Beine et al. 2011), whereas with small networks the level of immigrants’ education raises (McKenzie and Rapoport 2010).

Out-selection of skilled (and unskilled) immigrants is determined by external factors like entry regulation based on observable characteristics like education, age and ability and integration policy (Bertoli et al. 2012 and Macaluso 2022, for an overview).

Cultural diversity (e.g., language, ethnicity, religion) may foster innovation and the creation of new knowledge mostly relying on talents and skills from a wider variety of cultural backgrounds (Alesina and La Ferrara 2005; Berliant and Fujita 2004). This influence depends on the production complementarity between different types of population (Alesina and La Ferrara 2005; Niebhur 2006). Complementarity is hampered by transaction costs due to high cultural distance caused by ethnic, linguistic and religious diversity, all features shaping the integration process and immigrants’ sense of belonging to host societies (Ozgen et al. 2012; Niebhur 2006). In fact, cultural diversity and polarization can cause difficult communication, less trust and potential conflicts among employees and eventually hamper innovation (Ozgen et al. 2013).

Overall, extant empirical research has found that skilled immigrants count for innovation in receiving countries. For instance, Hunt and Gauthier-Loiselle (2010) show a positive impact of the share of immigrant college graduates, and of scientists and engineers in the population on patents per capita in the USA. Niebur (2006) highlights that cultural diversity of highly qualified employees positively influences patent applications in Germany. Ozgen et al. (2012) assess the impact of the size, skills and diversity of migrants on the number of patent applications of European Regions (NUTS2 level) in 1991–1995 and 2001–2005. Since no information was available on the skill levels of immigrants, the authors used region-of-origin information as a proxy for the skills and the cultural values that were specific to the country of origin. While the size reveals no conclusive results, a sufficiently high ethnic diversity and average skill level are positively linked to innovation.

From a historical perspective, Hornung (2014) finds that Huguenot diaspora in Prussia led to a transfer of technological and managerial knowledge between the Huguenots and natives in the textile sector, whose effects lasted for nearly the whole eighteenth century. More recently, Akcigit et al. (2017) have demonstrated that in technological areas of the USA, the prevalence of foreign-born expertise raised patents between 1880 and 1940.

Nevertheless, few scholars have analyzed the impact of unskilled immigrants on innovation. Recent research has mainly focused on the causality running from skills and education to technology and to changes in capital-to-labor ratio for the US (Hanson and Slaughter 2002; Lewis 2005) and the Italian provinces (De Arcangelis et al. 2015). It acknowledges that the increase in labor supply driven by low-skilled immigrants leads firms to make production technique more complementary with the characteristics of the labor force, which entails the adoption of labor-intensive technique and the reduction in skill-complementary capital. Lately, some scholars have driven their attention to the influence of low-skilled immigrants on innovation. Ozgen et al. (2013) relying on Dutch micro-data show that when excluding sectors with the highest share of low-skilled foreign born from the sample, the positive influence of ethnic diversity on product innovation is stronger. Bratti and Conti (2018) use province-level data to assess the impact of immigrants on patent applications in Italy, which is a destination country mainly for low-skilled individuals. They address the likely endogeneity of the immigrant share of the population by using immigrant enclaves. The results underline the weak performance of science-based industries and the lower return to immigrants’ human capital with respect to natives as factors that attract lower-skilled immigrants in Italy, but they find no evidence of any effects of immigrants on innovation. Pinate et al. (2022) analyze the impact of internal and international migration in Italy at the province level in the period 2003 to 2012 and reveal a negative association between low-skilled migration and innovation.

We extend the existing contributions in several directions. We build a theoretical model from which the empirical part is derived. The empirical analysis is explicitly centered on the effect of low-skilled immigrants on innovation in the EU-15 group of countries and Switzerland between 2012 and 2020.¹ Low-skilled immigrants are defined as those with lower secondary education to the highest level, as in Bratti and Conti (2018) and Pinate et al. (2022). Formal education is the most used proxy for human capital since it is a signal of skills in the labor market and a comprehensive measure of knowledge, skills and competency needed to participate in society (OECD 2019).

The empirical analysis relies on three datasets. Data on patents registered by firms from the scoreboards have been sourced from the OECD, REGPAT database (August 2022). The shares of migrants and of low-skilled migrants at regional level (NUTS-2) were calculated using the microdata drawn from the European Labour Force Survey (2010–2018).

Finally, when we address the endogeneity of immigrant share in the population, we use as instrumental variables not only immigrant enclaves (Bratti and Conti 2018), but also measures of local attitudes toward immigrants at regional level (NUTS2) drawn from the European Social Survey in 2010 and of integration policies at country level (NUTS0) from the indicators of Migrant Integration Policy Index (MIPEX) (Giacomo and Thomas 2020). How receiving societies perceive immigrants and their policies to integrate migrants in fact are likely to shape out-selection of immigrants as this factor may bring them to move to specific geographical areas, where it can have a fallout in terms of innovation, depending on the skills of immigrants. The outlay of the paper is as follows: Sect. 2 reports some evidence about immigration flows in Europe, Sect. 3 describes our theoretical model, Sects. 4–5 focus on the empirical analysis and main findings, and Sect. 6 contains concluding remarks.

To better understand the reason of our interest in low-skilled immigrants, it helps a glance at some evidence in Europe. Around the end of the last decade, in Europe the share of immigrant population with low education was one-third with 39% of non-EU migrants and about 26% of EU-born migrants compared to 23% of the native-born. Poorly educated—over 35% of immigrants—were present especially in Southern Europe, Belgium and France, while highly educated immigrants corresponded to 29% (OECD 2018). Lately, migration flows have been characterized by higher level of education generally throughout Europe (Fig. 1) so that the share of highly educated has become larger.

Immigrants’ education characteristics are reflected in their labor market performance though they generally face a penalization in terms of employment against natives. In EU, in 2017 they were mainly concentrated in low-skilled jobs; for instance, 30% of immigrant workers in Southern Europe, except for Portugal, occupied these jobs, three times more than native-born individuals (OECD 2018). As one can notice in Fig. 2a throughout Europe, the percentage of low-skilled jobs held by immigrants was higher compared to natives, while the opposite occurred for high-skilled jobs (Fig. 2b).

Looking at high-skilled jobs, they were held by one-third of immigrants; this share was lower than the one of natives and even lower for non-EU migrants (OECD 2018). However, in most European countries in the decade from 2006–2007 to 2017, there was an increase in high-skilled jobs occupied by immigrants except for Belgium, Italy, Finland and Denmark (Fig. 3).

The above evidence shows that despite recent changes in immigrants’ education and type of employment, low-educated immigrants remain a relevant phenomenon, especially in some European countries. Furthermore, immigrants’ employment is concentrated in low-skilled jobs. These features mainly concern non-EU migrants.

Further inspection of immigrants from non-European countries by considering the Human Development Index (HDI) of their country of birth allows to assess immigrants’ characteristics that can help the innovation process in destination countries. In fact, the index includes not only life expectancy at birth, but also the mean of years of schooling for adults aged 25 years and more and expected years of schooling for children of school entering age, and standard of living captured by gross national income per capita (UNDP-Human Development Reports Index). One can observe that over the period 2013–2021, when the indicator is available, migrants from countries with a high HDI are mainly in Germany and Spain, followed by France and Italy. Germany, Italy and France are major destinations for migrants from countries with low HDI (Figs. 4 and 5).

In general, the data highlight that the percentage of poorly educated immigrants in Europe is still non-negligible and that the presence of low-skilled immigrants—primarily from non-EU countries—is heterogeneously distributed across countries. This could trigger lower innovation in those areas where low-skilled immigrants are prevailing and ultimately bring down economic growth. For this reason, we focus on low-skilled immigrants defined as those who attained lower secondary education to the highest level. We chose lower secondary education as cutoff point as it covers a more significant group of low educated individuals who left education after compulsory school. In all European countries, the full-time compulsory education/training period includes primary and lower secondary education (European Education and Culture Executive Agency, European Commission 2022). Importantly, we know that formal education does not entirely capture individual skills meant as “the ability or capacity of an agent to act appropriately in a given situation” involving the ability to use text-based information including mathematical information. This ability depends on understanding and elaborating such information (OECD 2013).²

The following section will explain how low-skilled migrants can affect innovation and growth.

In this section, we presume the existence of two countries: the destination and the origin one. As our focus is on the impact of immigration on knowledge exchange in the exercise of the diffusion of agents’ ideas during the innovation process in the destination economy, we will investigate more extensively the latter. The analysis of the source economy will be constrained to workers’ migration choice. In line with Acemoglu (1996), we consider a simple non-overlapping generation model where in both the economies each generation consists of a continuum of entrepreneurs and workers normalized to unity. All agents live for two periods and are assumed to be risk neutral with an intertemporal preference rate equal to zero. In the first period, entrepreneurs and workers of the host economy determine their investments in physical and human capital, respectively, while workers of the source country must opt for migration or not. In the second period, patents’ production arises in the form of a partnership of one firm and one worker, and then, consumption from the patent’s benefits occurs. Finally, all agents die leaving no bequests. Different costs of acquiring human capital are assumed in the two economies so here is why in the host country there are two types of workers—natives and immigrants—with different levels of human capital. In the host economy, the production of patents is assumed to have constant returns to scale and takes the following functional form:where \({P}_{i,j,t}\) represents patents of the host country, \({h}_{i,t}\) is the human capital of the ith worker while \({K}_{j,t}\) the physical capital of the jth entrepreneur. A captures technological and geographical proximity effects.

We assume that in host and source economies labor markets are characterized by the presence of a costly search activity and a matching technology function assumed to be constant return to scale in its arguments: job vacancies and unemployed workers. The randomness of the matching technology function will imply that:

The randomness of the matching function will imply anonymity of contracts as each worker does not discern the characteristics of firms they are going to meet. Therefore, their choices will be contingent on the total distribution of physical capital across all entrepreneurs.

We estimate the impact of migration flows (m) on innovation in the EU-15 nations⁶ and in Switzerland between 2012 and 2020 through the following regression model, in which innovation is proxied by the number of patents (p):where i, c and t are regions (NUTS2), countries and time subscripts, respectively, and \({\varepsilon }_{it}\) is the error term. The share of immigrants in the population (m) is lagged two periods to make it predetermined with respect to the dependent variable; the analysis focuses on the association between the percentage of low-skilled immigrants (mlow) and innovation.

The vector X includes working-age population (pop) and people with tertiary education (tertiary) as measured by the 2011 census. These variables have been added in a year preceding the estimation period because they could be affected by the emigration flows in the years 2012–2020 (see also Bratti and Conti 2018).

We rely on a parsimonious specification of the model, excluding potentially endogenous controls as local industrial sectors and level of R&D effort.⁷ However, country fixed effects are added to the regressors to control for national institutional and socio-economic factors.

Data on patents registered by firms from the scoreboards have been sourced from the OECD, REGPAT database (August 2022). REGPAT collects data on patents registered with PATSTAT (the EU patent office) and allocated to each country.

Using sample weights, the share of immigrants in the population at regional level (NUTS2),⁸ as well as the percentage of low-skilled immigrants, is computed using data from the Labor Force Survey (2010–2018). As already specified, low-skilled immigrants correspond to individuals who attained a lower secondary education to the highest level.

Taking into account that the same socioeconomic factors at regional level could both attract immigrants and increase local innovation, we investigate potential endogeneity issues. We propose several instruments to test for endogeneity of the local share of immigrants.

Firstly, we propose an instrument, largely used in previous studies,⁹ that relies on the tendency for immigrants to migrate to areas where communities of immigrants from the same country of origin have already settled. Previous findings also highlight that networks favor less-skilled more than skilled immigrants (Beine and Salomone 2013; Bratti and Conti 2018).

Therefore, the distribution among regions of immigrants according to their origins in the past should provide the required exogenous source of variation in the local share of immigrants.

Formally, the instrumental variable “immigrant enclaves” can be written as follows:where “i” denotes the NUTS2 region (in the EU country “c”) and “a” denotes immigrants’ geographical area of origin;¹⁰ hence, the first term represents the share of immigrants from “a” living in region i in 2001, whereas \({{\text{immigrants}}}_{ac2001-i}\) denotes the total number of immigrants coming from “a” to the European country “c” in 2001 minus the contribution of region i to this total, and \({i{\text{mmigrants}}}_{ae2001}\) denotes the total number of immigrants coming from “a” to Europe in 2001.

The identifying assumption is that local factors that attracted migrants by different nationalities in the past (2001) are uncorrelated with the state of innovation in the years 2012–2020, controlling for the correlates included in the final specification of the model (e.g., population in working age, people with tertiary education, year and country dummies accounting for unobserved socioeconomic conditions). Indeed, 10 years seems like enough time for the assumption to be considered fair. To test the validity of the chosen IV (by the Sargan test), we also rely on additional instruments, namely natives’ concern about immigration at a regional level (NUTS2) and an indicator of migration policies in receiving societies.

A large body of literature has been written about the correlation between migration flows and local population attitudes toward immigrants (Fetzer 2000; Dustmann and Preston 2007; Sobczak 2007; Jolly and DiGiusto 2014; Nese 2022). Immigrants may choose to live in regions where residents are less averse to foreigners; at the same time, the share of foreigners at regional level may increase hostility toward ethnic minorities, in particular, for economic reasons. The instrumental variable “ethnic threat” was developed at a regional level (NUTS-2) based on the following three questions included in the European Social Survey: (i) “Immigration is good or bad for the country’s economy,” (ii) “the country’s cultural life is undermined or enriched by immigrants,” and (iii) “immigrants make the country a worse or better place to live.” The answers were coded on a scale of 0 to 10 (lower values indicate lower propensity to accept immigration).¹¹ To make the instrumental variable exogenous with respect to the estimation period, we measure “ethnic threat” using the European Social Survey carried out in 2010.

The last instrument, Permanent Residence is another aspect of reception context at country level (NUTS0). It measures whether temporary legal residents have facilitated access to a long-term residence permit (e.g., like UE nationals) on a scale 0–100.¹²

Table 1 reports the descriptive statistics of the variables used in the empirical model. We observe that migrants are about 14% of the population, and about 34% of them are low skilled.

Firstly, Eq. (15) is estimated by OLS considering m and mlow as a exogenous variables (Table 2).

The findings in column 1 indicate a positive correlation between migration flows (m) and innovation. Moreover, we get a positive correlation between people with tertiary education and the number of patents, as well as a positive correlation between active people and innovation.

Consistently with our argumentations, when the share of low immigrants (mlow) is added to the regressors (column 2), we report a negative association with innovation.¹³

As we may observe in Table 3, the endogeneity tests are rejected at 5 percent level, suggesting that we may treat ‘m’ and “mlow” variables as endogenous. To evaluate whether our potential instruments are weak, opportune test is employed. Indeed, the relevance of the instrument is assessed by evaluating the F-test for the joint significance of the instruments in the first-stage regression. The first-stage regression is reduced-form regression of the endogenous variable on the instrument and other exogenous regressors. A rule of thumb states that an F-statistic below ten is indicative of a weak instrument problem (Staiger and Stock 1997; Stock et al. 2002). Table 4 clearly shows that we have an F-statistic largely above the threshold value of ten.

The validity of our instruments is confirmed by Sargan Test of over-identification in Table 5.

Since the instruments are relevant and valid, we rely on T-SLS estimation in Table 6.

The IV estimation highlights (i) a positive effect of migration flows and (ii) a negative effect of low-educated migrants on innovation, proxied by the number of patents. OLS coefficients on m and mlow are biased downwards,¹⁴ suggesting that same institutional and socioeconomic factors at regional level could both attract immigrants and promote local innovation.

Our findings display an increase of 0.83% in patenting as the result of raising the overall immigrants’ share in the population by 1 percentage point. The result is much lower than the estimates (12–15%) reported by Hunt and Gauthier-Louiselle (2010) though for college immigrants in the USA. This finding is different from the one of Ozgen et al. (2012), who show that in European regions the share of immigrants does not conclusively influence innovation.

An increase in the share of low-educated immigrants by 1% determines a 1.4–1.8% decrease in patents applications. This evidence is consistent with the result of Ozgen et al. (2013) from Dutch micro-data, which points to a reduction in the positive influence of ethnic diversity on product innovation due to a larger share of low-skilled immigrants. It seems to indirectly confirm that the increase in labor supply driven by low-skilled immigrants induces firms to adopt labor-intensive technique and to reduce skill-complementary capital (Hanson and Slaughter 2002; Lewis 2005; De Arcangelis et al. 2015). A major implication is scarce innovation. It is non-negligible the fact that low-educated immigrants may also face significant barriers to participating in the innovation process, such as language barriers, discrimination and lack of access to education and training (Ozgen et al. 2013).

Overall, while cultural diversity may contribute to innovation, the results reinforce the arguments that low-educated immigrants may hamper it. Hence, it is important to consider the complexities of this issue when designing policies allowing both high- and low-skilled immigrants to fully participate in the innovation process.

This paper focuses on the effects of overall and low-educated immigrants on innovation in most Western European regions (NUTS2). This issue is particularly important in Europe, where at the end of the last decade, the share of immigrant population with low education was one-third with 39% of non-EU migrants and about 26% of EU-born migrants compared to 23% of the native-born.

The important findings of our analysis are as follows: Firstly, the regions with relatively many immigrants have a positive impact on the production of patent applications; secondly, low-skilled immigrants have opposite effects on innovation. We addressed the potential endogeneity of migration flows with valid instruments, nominally immigrant enclaves, local attitudes toward immigrants and an indicator of migration policy.

Therefore, our empirical findings support the proposition in the theoretical section that lower-educated immigrants determine social decreasing returns in the economy. It is likely that the growth in labor supply due to low-skilled immigrants leads firms to make changes in the production techniques by making them more adapt to the education characteristics of labor force. The consequent adoption of labor-intensive techniques hampers innovation. This has considerable implications for Europe. As recent evidence shows, poorly educated immigrants are heterogeneously distributed across European countries and concentrated, especially in Southern Europe, Belgium and France. Hence, in these areas there could be innovation and growth at a slower rate. It may occur a certain disparity among countries with a higher level of human capital triggering a virtuous circle with social increasing returns and countries with low human capital with the opposite situation. In fact, low-educated immigrants tend to move where the demand for lower skilled jobs is higher, which is likely to further the shift toward labor-intensive techniques. On the other hand, immigrants may be trapped in low-skilled jobs with poor earnings and social mobility. In this respect, given that political instability and warfare are likely to feed growing waves of young and low-educated immigrants (Bratti et al. 2018), an in-depth understanding of the relationship among human capital formation, labor markets and migration in receiving countries would help. In general, integration programs providing support and training to new arrivals should aim to balance the needs of immigrants with the development needs of different European geographical areas.