Spaniards in the wider world: the role of education in the choice of destination country

In the 2013 Christmas period, almost all Spanish television (TV) channels broadcast an advertisement with the slogan “Become foreigner”.

In Spanish, the slogan of the advertisement was “Hazte Extranjero”. The company was Campofrío.

As explained by Bauer et al. (2005), when analyzing patterns of migration, it is not only the decision to emigrate that is relevant, so is the location choice. Several researchers have studied the mechanisms involved in the emigrant's selection of location, including economic incentives (Eliasson et al. 2003; Nifo and Vecchione 2014), native language of the destination country [see Chiswick and Miller (2014) for a review], immigration policy barriers (Clark et al. 2007), past colonial relationships (Mayda 2010), institutions (Nifo and Vecchione 2014), physical distance (Sjaastad 1962), religious affiliations (Neudörfer and Dresdner 2014), labor market factors (Clemente et al. 2016; Dotti et al. 2013), and cultural links (Belot and Ederveen 2012; Wang et al. 2016), among others. This work explores whether there is a relationship between the destination country and the education level of Spanish emigrants.

The existing literature examining the connection between education and migration generally finds that more-educated individuals are more likely to emigrate, and that those individuals tend to locate in countries with high rewards for skills (Grogger and Hanson 2011). The business cycle also appears to be relevant for the migration choices of more-educated individuals (Venhorst et al. 2011). Other papers studying the educational selectivity of emigrants conclude that cultural similarities and the physical distance between countries have more relevance in the educational selectivity process than wages or immigration policies (Belot and Hatton 2012). In addition, there is a growing literature that analyzes the migration of high-skilled populations from poor to rich countries, which, as researchers suggest, may affect the education incentives in developing countries (Beine et al. 2008; Docquier et al. 2007).

This paper extends prior research by using microdata from the Census related to 21 destination countries of Spanish emigrants in the period 2000–2007. The data come from the Integrated Public Use Microdata Series–International (Minnesota Population Center 2014). The use of Census data is quite common in the literature, but the majority of papers consider the analysis at the aggregate level. The microdata from the Census do not contain information on why individuals decided to migrate, but the Census does provide valuable information on Spanish emigrants, which allows us to study the relationship between their education level and several characteristics of the destination country: the language, the physical distance from Spain, and the economic circumstances measured by the unemployment level. We first analyze each of them individually, and then we consider them jointly. We add to the literature by studying – at the individual level – the dissimilarities in the location choice of Spanish emigrants depending on their level of education. The novel methodology proposed here allows us to jointly consider the location choice of individuals based on language, migration costs, and job search.

The native language of the host country is one of the characteristics that may be important in the location choice of emigrants. However, researchers have found conflicting results in the relationship between language and choice of destination country. Some conclude that language matters (Adsera and Pytlikova 2015; Beine et al. 2011; Belot and Hatton 2012; Belot and Ederveen 2012; Clark et al. 2007; Grogger and Hanson 2011), while others detect no relationship (Mayda 2010; Ortega and Giovanni 2009). Language skills can be seen as a form of human capital, since they satisfy the following requirements: they are productive (knowing the language of your place of residence facilitates access to a job), they are costly to produce (time and money are needed to acquire language proficiency), and they are embodied in the individual (language skills may not be separated from the person; Chiswick and Miller 2014). If it is assumed that the greater the human capital, the greater is the gift for languages, it would be expected that those Spanish emigrants with a higher education level would be more likely to settle in countries where Spanish is not the official language. However, those who are highly educated also have higher opportunity costs derived from migration to non-Spanish-speaking countries. Since learning a language can be costly, not only because of the direct costs of learning but also because of the lower earnings received in the destination country during the adjustment period before acquiring proficiency, it can be argued that highly educated ones are more likely to settle in countries in which they already know the language; for example, countries with the same language as the home country or countries where emigrants do not have to spend a long time to acquire proficiency, either because the native language of the destination country is a second-language in the schools of their country of origin or because it is a common language in business. Thus, a priori, it is not possible to infer which languages would be preferred for Spanish emigrants with a higher, or lower, level of education, and so, in theory, it is unclear what destination country they would choose.

The physical distance between the home and the host country can also play a role in the location choice of emigrants. It would be expected that low-education-level individuals, who normally have low wage-earning potential, would be less likely to move to distant areas because the travel costs between the home and the destination country are higher. Additionally, physical distance is important to the quality of information about the destination country (job opportunities, income differentials, job market characteristics, and so on), as noted in Sjaastad (1962). If highly educated individuals have greater access to such information, we would expect a less-significant impact of distance on highly educated individuals than on less-educated individuals. Proximity also facilitates illegal immigration, which may increase the migration of less-educated individuals. In this case, all these possible explanations indicate that, for highly educated individuals, the physical distance between the home and the host country is less important than for low-education-level emigrants, although we will check it here empirically.

Migration can be a consequence of job search (Herzog et al. 1993; Malamud and Wozniak 2012).

Wage inequality may also play a role, but we do not consider this in our work.

After the analysis of the relationship between the educational level of emigrants and each of the previous characteristics of the host countries (language, physical distance, and economic situation), we introduce a novel contribution to the literature by exploring different combinations of host country characteristics, since emigrants may make their location choice considering, for example, all those features simultaneously. Additionally, it is arguable that some of the characteristics of the destination countries may be more important for emigrants than others, so a separate analysis for each would not be sufficient to study the location choice of emigrants. In this work, we propose a methodology that allows us to examine which of the characteristics of the destination country (or which among them) is (are) the most relevant, without imposing any restriction on the location choice process. Results suggest that, with low URs, English-and Spanish-speaking countries are the most likely to become the host countries for more-educated individuals, regardless of their location. However, it is also observed that the greater the proximity, the less important is the economic situation; whereas, the greater the distance, the more central is the labor economic situation of the country for highly educated individuals.

The remainder of the paper is organized as follows: Section 2 presents our empirical strategy, Section 3 describes the data, results are discussed in Section 4, and Section 5 concludes the paper.

The main purpose of this research is to determine whether a relationship exists between the level of education of Spanish emigrants and certain characteristics of the host country (characteristics considered both separately and jointly). This analysis is necessary because, as explained above, the incentives, costs, and potential gains that emigrants can obtain in the destination country may differ depending on their level of education. We focus on three different characteristics: the language of the destination country, the physical distance from the emigrants’ home country, and the economic situation of the host country, measured by its UR, which have been extensively considered in the literature as crucial factors in the emigration choice. In most cases, prior work has analyzed those characteristics separately, using aggregate data. In our case, we suppose that emigrants may decide to settle in a country by considering just one of those characteristics or several of them. Since we have no information on what motivates emigrants to settle in one particular country rather than in another, we are not able to analyze how the location choice process takes place. Nevertheless, we know the outcome of that decision process, that is, the place of residence of emigrants. This gives us information, for example, on whether emigrants with a low level of education tend to locate in countries with certain specific characteristics, whereas those with a high level of education choose other countries with different characteristics. Thus, we can study whether emigrants choose their destination country depending on their level of education. To address this issue, we use a model for nominal outcomes, specifically a multinomial logit model (MNL), in which we estimate a separate binary logit for each pair of outcome categories (Nerlove and Press 1973). This methodology has also been considered in other papers to analyze migration choices (Huber and Nowotny 2013).

Initially, we examine the relationship between each of the characteristics of the host country and the level of education of the emigrants. Within this framework, we explore whether emigrants with high level of education tend to live in countries with a language other than their native language, or not, relative to those with low levels of education. We also analyze whether emigrants with high levels of education settle in distant countries, that is, far from their home countries, and whether they choose countries with a better economic situation or not. We then consider all the possible combinations of the characteristics of the destination countries and, again, we examine their relationship with the level of education of the emigrants. The results of the joint analysis are compared with the initial separate analysis, to study whether the latter is useful. Note that we do not impose any order in the location choice process, neither in the separate analysis nor in the joint analysis, but with the methodology proposed here, we can observe the characteristics of the destination country that are more (or less) likely to be relevant for emigrants in the choice of a destination country, depending on their level of education. Formally, we estimate the following equation: (1) lnϕm/b=lnPr(y=m/x)Pr(y=b/x)=xβm/b for m=1…J \ln {\phi _{m/b}} = \ln {{\Pr \left( {y = m/{\bf{x}}} \right)} \over {\Pr \left( {y = b/{\bf{x}}} \right)}} = {\bf{x}}\beta_{m/b} \;{\rm{for}}\;m = 1 \ldots J

In Eq. (1), b is the base category and m varies from “1” to J, with J being the total number of outcome categories (we revisit this issue later). The vector x includes not only the educational variables (defined as “secondary education completed”, “some college education or more”, and “primary or less than primary education completed”, which is the omitted measure) but also a range of other socioeconomic characteristics of Spanish emigrants. These characteristics may be considered to have an impact on the location decision for reasons independent of their level of education and include age, gender (whether the emigrant is a man), and employment status (whether the emigrant is employed).

We have repeated the analysis without those controls, and our results do not vary.

Regarding the outcome categories in the separate analysis, when the dependent variable is the official or the most commonly spoken language of the host country, we consider three categories: English, Spanish, and Others. As explained earlier, it would be expected that more-educated individuals, that is, those with greater human capital (Barro and Lee 2001), would be more easily capable of living in non-Spanish-speaking countries. Nonetheless, because of the opportunity costs to acquire language proficiency, it would be expected that these more-educated individuals choose to live in countries with language in which they have already acquired proficiency, with the most-likely countries being Spanish- and English-speaking countries. As Chiswick and Miller (2014) explain, Spanish is the first language in their country of origin and English is normally the second language in Spanish schools, in addition to being one of the most common languages in business. The outcome categories in the case of physical distance are Europe, the Americas, and Others. For the UR, the categories are as follows: low unemployment (UR <4%), medium unemployment (UR between 4% and 9.9%), and high unemployment (UR ≥10%).

It has been observed that all these outcome categories are indistinguishable on testing whether any of the independent variables significantly affects the odds of m versus alternative n categories (Anderson 1984). We test the following:

H₀ : β_1,m|n = … = β_i,m|n = … = β_I,m|n = 0, where β_i is the coefficient associated with the explanatory variable x_i. We cannot reject the hypothesis that those outcome categories are indistinguishable when using logistic regression likelihood-ratio test (LR) and Wald tests.

We recognize that the use of a model for nominal outcomes in all cases can generate some concerns, since some of the destination country characteristics can be considered as ordered variables. The language of host countries is clearly a nominal variable because all the categories herein are assumed to be unordered. Thus, in principle, the use of a MNL for this analysis would be appropriate. In contrast with the language outcome category, the distance and the UR can be considered as partly ordered – or even ordered – which would imply using an ordinal model in our empirical analysis (McKelvey and Zavoina 1975). However, the parallel regression assumption implicit in the ordinal model is not satisfied in any of these cases. We calculate the approximate likelihood-ratio test of proportionality of odds across outcome categories and detect that the parallel regression assumption can be rejected at the 5% level of significance. In addition, it is possible to argue that these variables are not considered as ordered by emigrants. Because of vacation periods, the schedule, and perhaps even climate, among others, emigrants may see one part of the world as being different from the other. Of course, this can be correlated with physical distance (close countries usually have more similar climates and even customs), but emigrants can observe it as unordered choice. In these cases, the outcomes can be treated as nominal (Marcén and Molina 2012). In the case of the joint analysis, this problem does not exist. The combinations of all these variables (language, distance, and UR) cannot be considered ordered outcomes. Then, it is arguable that the models for nominal outcomes, in particular, the MNL model, are suitable.

It is worth noting that we test the property of independence of irrelevant alternatives. This property is convenient in MNL models, since if alternatives are not truly independent, then the estimates will be inconsistent. Our results indicate that the MNL model is appropriate, using the Hausman test (Hausman and McFadden 1984) and the Small–Hsiao test (Small and Hsiao 1985).

With respect to the outcome categories in the joint analysis, we include all possible combinations without imposing any order on the location process decision. We repeat the analysis by incorporating, as the dependent variable, a combination of the location of the destination country and the official (or the most common) language of that country. Within this framework, we are interested in examining whether emigrants with higher levels of education are more likely to choose a destination country with a specific language, depending on its location. For example, it is possible to surmise that emigrants with low levels of education are more likely to settle in less-distant countries (low travel costs), regardless of the language, since acquiring language proficiency is less important to their low-skilled employment, whereas highly educated emigrants who decide to migrate to a distant country (high travel costs) may prefer those countries with the same language as the home country (or countries where they do not have to spend a lot of time to acquire language proficiency) in order to reduce the migration costs. Similarly, we run the entire analysis defining the dependent variable as a combination of the location of the destination country and its economic situation. As before, in this case, we can explore whether emigrants with high levels of education tend to settle in destination countries with a better economic situation, if they are distant, whereas the economic situation is less important for them when the destination is close to their home country. Because all the characteristics considered here appear to be important, it is logical to propose a combination of all of them in the dependent variable. With this approach, we are able to study whether, depending on the level of education of emigrants, they decide to settle in a country considering just one of those characteristics or several of them.

We analyze the dynamics among the outcome categories by utilizing odds ratios, which is an intuitive method of interpreting the estimates (Greene 2008; Long and Freese 2006). Holding other variables constant, the change factor in the odds of outcome category m versus outcome category n when x_i increased by δ is represented as follows: (2) ϕm/n(x,xi+δ)ϕm/n(x,xi)=eβi,m/nδ {{{\phi _{m/n}}\left( {{\bf{x}},{x_i} + \delta } \right)} \over {{\phi _{m/n}}\left( {{\bf{x}},{x_i}} \right)}} = {e^{{\beta _{i,m/n}}\delta }}

For a unit change in x_i, δ = 1, the odds of m versus n are expected to change by a factor of exp(β_i,m|n), holding all other variables constant. For a standard deviation change in x_i, δ = s_xi, the odds of m versus n are expected to change by a factor of exp(β_i,m|n × s_xi). To simplify the odds analysis, odds-ratio plots are also presented (Long and Freese 2006).

To implement our analysis, we use data from the Integrated Public Use Microdata Series–International of the 2000–2007 Censuses (Minnesota Population Center 2014).

This period was chosen since it is possible to get the greatest number of destination countries of Spanish emigrants for those years.

Table 1 shows the summary statistics of the relevant variables by destination country. Our individuals are 45.34 years old on average, with those living in Argentina being older than those in other groups (52 years) and those in Ireland being the youngest (30 years). Almost 50% of emigrants are men, although there are differences, with more men than women living in the countries of South and Central America. Most of the emigrants are married and employed. With respect to the education level of Spanish emigrants, Table 1 presents considerable variations, with less-educated individuals living in Argentina, Portugal, and the Philippines, whereas more highly educated ones are located in Colombia, Peru, and El Salvador. The differences in education may point to a potentially important role of education in the location choice of Spanish emigrants. With respect to the number of observations, we recognize that the large number of observations from the France Census can generate some concerns.

All the estimates are obtained using weights from the Census. Results without weights do not change.

Columns 7–9 of Table 1 present the summary of the characteristics of the destination countries. Column 7 includes the official or the most common language in each host country, according to data from the US Central Intelligence Agency (CIA). Twelve countries have Spanish as the official language, three have English, and the remaining six countries have other languages (Portuguese, Greek, German, French, and Filipino). As mentioned earlier, we consider three categories (English, Spanish, and Others). In the literature, it is possible to find several classifications of languages when many home and destination countries are analyzed (Chiswick and Miller 2014). However, in this work, we only consider one home country and relatively few host countries, with the number of languages being limited. For this reason, we favor the use of that simple classification.

With respect to the UR, the data come from the World Data Bank. Column 8 only includes information on the UR for the year in which the Census was collected in each country. We do not detect a clear regional pattern, with the highest rates being for Argentina and South Africa and the lowest for Mexico, Cuba, and Austria. The analysis has been repeated using the average UR from 1990 to 2000, to overcome differences in the years in which the information is taken. Results are maintained. Regarding the physical distance between the home country and the host country, we separate the countries by continent. There are five countries from Europe, 14 from the Americas, and just two from other parts of the world (Column 9). Although not included in this document, we have also used the physical distance measured in kilometers from the capital of the country of origin to the capital of the destination country, as in Belot and Hatton (2012). Results do not vary.

This paper examines the relationship between the education level of Spanish emigrants and their destination country. Since Spanish emigrants were born under the same laws, economic conditions, and institutions, the differences in their location choice can be due to dissimilarities in their level of education. To study this issue, our work uses microdata from the Censuses of 21 countries, for the period from 2000 to 2007. Destination countries are classified using three characteristics: the official (or the most common) language, the physical distance between the country of origin and the destination, and the level of unemployment. In this way, we take into consideration the migration decisions of individuals based on language, migration costs, and job search.

Our results suggest that Spanish emigrants with secondary and tertiary levels of education have greater preferences for English-speaking countries, more-distant countries, and countries with low URs, when each of these factors is analyzed separately. Our findings are quite robust to the inclusion of several sociodemographic characteristics and to the use of different subsamples.

By combining these characteristics, since Spanish emigrants may consider more than one characteristic of the potential host country when deciding the location, we draw interesting conclusions. There are still similarities in the location choice between those with secondary and tertiary levels of education. English-speaking countries, regardless of the location (Ireland, the US, and the UK), and Spanish-speaking countries with low URs are the most-likely residence choices for more educated emigrants. The highly educated are more likely to choose to live in distant countries (with the exception of the UK and Ireland), but if it is a Spanish-speaking country, the economic situation is a significant factor. However, when those individuals decide to move to a European country, they appear to pay less attention to the economic situation, since the least-likely option for them is a European country with a low UR (excluding the UK and Ireland).

To sum up, our results reveal that the level of education plays a role in the location choice of those who decide to migrate, but that the differences appear to be more significant between those with a primary level of education and those with a secondary or higher level of education. Our findings denote the importance of considering several of the characteristics of the destination country, to determine how potential emigrants choose a country to live in. The conclusions provided by this work can also be useful for potential destination countries, in order to implement policies to attract specific groups of emigrants.