Chapter 3 illustrates that foreign-born individuals whose mother tongue is different from the language of the test tend to have lower literacy and numeracy proficiency (when these are assessed in the language of their country of residence) and poorer labour market outcomes than individuals whose mother tongue matches the language spoken in the country. However, language penalties in information processing skills and labour market outcomes vary considerably, both across countries and within countries across different migrant groups. This chapter illustrates that the depth of the language penalty in skills and labour market outcomes is related to the degree of proximity between the mother tongue spoken by migrants and the language spoken in the country of destination. Individuals whose mother tongue is very different from the language spoken in their country of residence have very low proficiency relative to the native born if they arrived in the host country after the age of 12, and the negative impact persists irrespective of length of stay. Furthermore, these individuals are less likely to have access to gainful employment, irrespective of their age, gender or educational level.
Skills on the Move
Chapter 3. Language matters: language disadvantage and the outcomes of foreign-born adults in PIAAC
Abstract
The statistical data for Israel are supplied by and under the responsibility of the relevant Israeli authorities. The use of such data by the OECD is without prejudice to the status of the Golan Heights, East Jerusalem and Israeli settlements in the West Bank under the terms of international law.
The role of language in shaping the integration of migrants
Previous empirical investigations of some of the factors that explain differences in information processing skills and labour market outcomes between migrants and native populations identify language as a crucial element (Bonfanti and Xenogiani, 2014[1]; Isphording, 2014[2]);. On average individuals whose mother tongue is different from the language spoken in the host country have been shown to have lower levels of literacy and numeracy than individuals whose mother tongue matches the language spoken in the country. Moreover, language is important in explaining differences in labour market participation and wage levels (Isphording, 2014[2]; OECD/EU, 2014[3]).
One possible explanation for the observed differences in information processing skills (such as literacy and numeracy) between native language speakers and individuals whose mother tongue is not the same as the language of the assessment in which these skills are measured is that measurements of these skills capture both skills proficiency and language fluency. To the extent that migrants whose mother tongue is different from the language of the host-country are less fluent in the language of the host country than natives, they will tend to perform less well in standardised tests that require language proficiency given similar levels of underlying skills.
Language fluency has been considered in the literature as a key driver of immigrants’ integration, because it facilitates individuals’ access to job opportunities, job retention, and career progression. Individuals who are fluent in the host-country language are also more likely to participate in the social life of their communities, to be able to access public services and contribute to local activities (Dustmann and Van Soest, 2001[4]; Dustmann and Fabbri, 2003[5]; Bleakley and Chin, 2004[6]; Bleakley and Chin, 2010[7]). The labour market returns to language fluency are linked to the ability to access and use information on job opportunities, ability to perform during the hiring process, and higher productivity while on the job.
Several factors influence the process that leads to the acquisition of language fluency. In a seminal work, Chiswick and Miller (1995[8]) developed a theoretical framework that classifies the determinants of language skill acquisition into factors that affect the level of exposure migrants have to the language of the host country and factors that shape the ability and the efficiency migrants have of becoming fluent in a new language.
The level of exposure migrants have to the host-country language can be considered to be a function of the time spent in the host-country, the number of interactions that on average occur per unit of time, and the efficacy of such interactions. First and foremost, the level of exposure to the host-country language is associated with the number of years they spent in the host country: other things being equal, migrants who lived in their host country for longer will have been more exposed to the host country language. The relationship between the number of years spent in a country and language fluency does not, however, need to be linear. For example, the marginal returns to time spent in the country may be decreasing, such that each additional year may be associated with a smaller improvement in language fluency. Alternatively, the marginal returns to time spent in the country may be increasing, if a certain level of language proficiency is a precondition for individuals to be able to acquire new language skills and the more fluent an individual is, the faster the pace of additional improvements will be.
Fluency may be the result not only of the quantity of individuals’ exposure to the foreign language, but also of the quality of exposure and the ease with which migrants acquire a new language. Irrespective of overall length spent in the host country, individuals who arrived as young children may be able to acquire the host country language with much greater ease (Newport, 2002[9]). In fact, supporters of the Critical Period Hypothesis argue that the age of 12 marks an important threshold after which the efficiency with which individuals acquire language skills in a foreign language decreases markedly. Individuals who have high levels of education and who have high levels of literacy in their native language may also be better able to understand what is required to become proficient in a new language, how to seek and how to access support to do so and as a result may be better placed to be able to gain fluency. For example, Dustmann (1994[10]) and Isphording and Otten (2011[11]) illustrate that non-native German immigrants in Germany who had good writing abilities in the mother tongue acquired greater fluency in Germany than non-native German speakers who had poor writing abilities in their mother tongue.
Economic incentives have been shown to play an important role in motivating migrants to gain fluency in their host country language. For example, investments in language acquisition are positively associated with the expected duration of stay in the host country (Dustmann, 1999[12]; Isphording and Otten, 2014[13]). Migrants who expect or seek employment are likely to have a greater incentive to acquire language proficiency, as they will be required a higher degree of communication and interaction with others. For instance, in a study on female migrants in Germany, Dustmann (1994[10]) find that those that had not worked before had lower German-speaking fluency, irrespective of their level of education. Language requirements can differ markedly across occupations that are of equal social status or which command similar incomes.
Quality differences in levels of exposure not only depend on the characteristics of individuals, but also on the opportunities they have to interact with native speakers and the depth of such interactions. Migrants who live in neighbourhoods which are predominantly occupied by native speakers have greater and higher quality exposure to the host country language per unit of time spent in the host country when compared to migrants who live in neighbourhoods which are densely populated by non-native speakers. Evidence from Australia (Chiswick and Miller, 1995[8]) and the United Kingdom (Dustmann and Fabbri, 2003[5]) suggests that there is a negative relationship between levels of ethnic minority density in a given location and the language skills of migrant residents. Other studies have analysed the language skills of intermarried immigrants in Australia (Meng and Gregory, 2005[14]) and immigrant spouses in the U.S, suggesting a positive effect of depth of exposure to the host country language on language fluency.
The make-up of non-native language speakers in PIAAC participating countries
PIAAC reveals that, although on average around 12% of adults in PIAAC participating countries are not native speakers, countries differ greatly in the language composition of their adult populations (Figure 3.1). Not native speakers in PIAAC are defined as those individuals who reported to have spoken at birth (and still understand) a language that is different from the language of the test. For example, in Singapore over 71% of adults surveyed in PIAAC were not native speakers, because many sat the test in English but at birth they spoke Malaysian, Tamil, or Chinese or other languages. Apart from Singapore, non-native speakers represent over one in five of the adult population in Canada (22%) and Israel (22%).
While there is an association between the probability that migrants, defined in PIAAC as participants who were not born in the country in which they sat the PIAAC test and being a non-native speaker, the correspondence is far from perfect: on average across PIAAC participating countries, around 59% of the migrant population are not native speakers and around 5% of the native population is as well not native speakers. This means that, in PIAAC-participating countries, 70% of not native speakers are also migrants. Figure 3.1 illustrates that countries differ in the composition of their resident adult populations: in some countries there are more individuals who are not native language speakers than there are migrants, while in others there are more migrants than not native language speakers. In Lithuania a higher percentage of the adult resident population is not a native language speaker than is foreign-born. These are countries with established language minorities. By contrast, in New Zealand, Australia, Ireland, Estonia, Spain, the United Kingdom and France a higher percentage of the adult resident population is foreign-born than is not a native language speaker. These are predominantly countries where migrant communities come from countries where the same language of the country of destination is spoken.
While Figure 3.1 indicates that there is no exact match between language status and migration status and that countries differ greatly in the extent to which they are home to not native speakers, it does not shed light on the extent to which migrants in PIAAC participating countries vary in the composition and mix of languages that are spoken. Figure 3.2 shows the different linguistic groups that co-exist in selected PIAAC participating countries. Some of the countries with the largest overall percentages of not native speakers, such as Singapore and Israel have a low level of language diversity, since they are home to a small group of well-defined language groups. In Singapore, for example, three large language groups coexist: individuals whose mother tongue is Chinese (representing 76% of the non-native speaker population), individuals whose mother tongue is Malay (representing 14% of the non-native speaker population), and finally individuals whose mother tongue is Tamil (4% of the non-native speaker population). In a second group of countries most language minority residents belong to one large language minority group, with the presence of a few small groups. This is the case of the United States, where Spanish is the language spoken at birth by the majority of the non-native speaker population (representing almost 60% of this group). Chinese is the second most spoken language representing around 6% of the adult resident population, followed by Vietnamese (0.31%), Tagalog (0.27%), Russian (0.25%), German (0.23%) and other languages. Similarly, in Estonia, Russian is the most prevalent language group, with around 58% of the non-native speaker population speaking Russian as their mother tongue. A third group of countries comprises countries where a large variety of language groups coexists: countries such as Canada and Italy are countries that differ greatly in their migration regimes and history as countries of destination for migrants but both are now home to a large variety of language minority populations.
Figure 3.1. Percentage of migrants and non-native speakers in PIAAC participating countries
Note: Non-native speakers are defined as those participants who reported to have spoken at birth a language that is different from the language of the test. Migrants are defined as those participants whose country of birth is different that the country at which they are doing the test. Estimates for the Russian Federation are missing due to the lack of language variables.
Source: (OECD, 2015[15]) Survey of Adult Skills (PIAAC) (2012, 2015), Table 3.A.1, www.oecd.org/skills/piaac/publicdataandanalysis
Figure 3.2. Percentage of non-native speakers, by first language spoken at home and understood in selected PIAAC participating countries
Note: Native speaker refers to whether the first or second language learned as a child is the same as the language of assessment, and not whether the language has official status.
* Romanian; Moldavian; Moldovan.
Source: (OECD, 2015[15]) Survey of Adult Skills (PIAAC) (2012, 2015), Table 3.A.2, www.oecd.org/skills/piaac/publicdataandanalysis
Standard characterisations of the impact migration has on social diversity of host countries are based on statistics indicating the percentage of foreign-born individuals who live in the country or indices indicating birthplace or ethnic diversity through concentration/diversity indices (Easterly and Levine, 1997[16]; Collier, 2001[17]; Alesina, Harnoss and Rapoport, 2015[18]). Figure 3.3 illustrates the level of language diversity in PIAAC participating countries using information reported by individuals who participated in PIAAC on their native language. The figure illustrates two linguistic diversity indices: the within diversity index indicates the diversity of languages spoken by migrant communities, the between diversity index represents the overall share of population whose mother tongue is a language that is not the language in which they sat the PIAAC assessment. The two indices range between 1 and 0 with 1 representing higher diversity and 0 the case in which all individuals speak the same language.
Figure 3.3. Language diversity in PIAAC participating countries
Note: Estimates for Australia and Germany are missing due to the lack of language variables.
Source: (OECD, 2015[15]) Survey of Adult Skills (PIAAC) (2012, 2015), Table 3.A.3, www.oecd.org/skills/piaac/publicdataandanalysis
Figure 3.3 indicates that countries differ greatly across the two components of the language diversity index. Chile is a country where few individuals do not speak Spanish both in the overall population and within the migrant population: the between linguistic diversity index is as low as 0.03 in the whole population and 0.18 when considering only foreign-born adults (within linguistic diversity index). At the other side of the spectrum lie the English speaking community of Canada and Israel, where diversity is high on both dimensions: in Israel, between linguistic diversity at the population level is as high as 0.66 and within the foreign-born population it stands at 0.75. In English speaking Canada, the between level of diversity is 0.54 and within migrant communities it is as high as 0.92.
Language-related disadvantage
PIAAC reveals that the mother tongue language is an important determinant of differences in literacy proficiency. Figure 3.4 illustrates the average performance in literacy in PIAAC participating countries of natives, migrants whose mother tongue is the same as the language in which they sat the PIAAC test and migrants whose mother tongue is different from the language in which they sat the PIAAC test. The average difference between foreign-born and native-born individuals in PIAAC participating countries is 22 points. However, while the difference in the PIAAC scores of migrants who are native speakers and of non-immigrant native speakers is 10 points, this difference is as large as 27 score points between natives and migrants whose mother tongue is different from the language in which the PIAAC test was conducted.
However, Figures 3.4 reveals large differences across countries. In Lithuania, Estonia and the Czech Republic there are no significant differences in the literacy proficiency of migrants whose mother tongue is the same as the language of the PIAAC assessment and those whose mother tongue is different. In a second group of countries the migrant gap is largely explained by the fact that migrants speak a language that is different from the language in which the PIAAC assessment was conducted. For example, in Australia, Austria, Greece, Finland, the Netherlands and Singapore over 70% of the migrant gap in literacy scores can be explained by the language penalty, controlling for the influence of age, gender and the level of education attained. Similarly, Figure 3.5 shows that in Australia, Austria, Singapore and Finland over 75% of the migrant gap in numeracy scores can be explained by the fact that migrants often speak a language that is different from the language in which the PIAAC test was conducted.
In no country do migrants who are native or non-native speakers have higher literacy proficiency than natives. Israel is the only country in the PIAAC sample in which migrants who are non-native speakers have better results than migrants who are native speakers.
PIAAC reveals that being a non-native speaker is associated with lower literacy proficiency. However, individuals whose mother tongue is different from the language of the host country tend to have higher returns to skills than native language speakers. As shown in Figure 3.6, the returns to literacy skills for non-native speakers are generally similar or higher than for natives, controlling for factors such as educational attainment and years of experience. On average, and increase in 25 points in literacy proficiency is associated with a wage premium of 5.4% for non-native speakers, compared to 4% for natives or migrants whose mother tongue is the same as the host country language. Results are similar when considering the returns to numeracy skills in Figure 3.7. An increase in 25 points in numeracy proficiency is related to a 4.6% increase in wages for native speakers, and over 5.1% for non-native speakers.
Several explanations can lie behind the observed differences in returns. One possibility is that, because of language difficulties, given similar PIAAC test results among non-native speakers and native speakers, non-native speakers may have better unobserved characteristics and skills than native speakers (such as being proficient in another language). The positive wage premium that is associated with literacy and numeracy skills among non-native speakers could also be due to differences in the sectors in which non-native speakers are employed.
Figure 3.4. Gap in literacy performance between natives and migrants, in PIAAC participating countries
Note: Statistically significant differences are marked in a darker tone. Proficiency in literacy ranges between 0 and 500 score points. Adjusted differences are based on a regression model and take account of differences associated with all or some of the following variables: age, gender, education, and language background. Estimates based on a sample size less than 30 are not shown. Estimates for the Russian Federation are missing due to the lack of language variables.
Countries are ranked in descending order of the unadjusted migrant gap (natives minus migrants).
Source: (OECD, 2015[15]) Survey of Adult Skills (PIAAC) (2012, 2015), www.oecd.org/skills/piaac/publicdataandanalysis
Figure 3.5. Gap in numeracy performance between natives and migrants, in PIAAC participating countries
Note: Statistically significant differences are marked in a darker tone. Proficiency in numeracy ranges between 0 and 500 score points. Adjusted differences are based on a regression model and take account of differences associated with all or some of the following variables: age, gender, education and language background. Estimates based on a sample size less than 30 are not shown (Japan, Poland and Turkey). Estimates for the Russian Federation are missing due to the lack of language variables.
Countries are ranked in descending order of the unadjusted migrant gap (natives minus migrants).
Source: (OECD, 2015[15]) Survey of Adult Skills (PIAAC) (2012, 2015), www.oecd.org/skills/piaac/publicdataandanalysis
Figure 3.6. Returns to literacy skill for individuals, by language background
Note: Hourly wages, including bonuses, in PPP-adjusted USD (2012). The regressions are estimated with log wages as the dependent variable separately for language native and non-native workers and includes controls for years of education, years of experience and experience squared, part-time work and gender. The wage distribution was trimmed to eliminate the 1st and 99th percentiles. The estimated coefficients have been multiplied by 47 which corresponds to the standard deviation in literacy proficiency. Statistically significant coefficients are marked in a darker tone. Estimates based on a sample size less than 30 are not shown.
Countries are ranked in descending order of the percentage change in wages of the non-native speakers.
Source: (OECD, 2015[15]) Survey of Adult Skills (PIAAC) (2012, 2015), Table 3.A.4, www.oecd.org/skills/piaac/publicdataandanalysis
Figure 3.7. Returns to numeracy literacy skills for individuals, by language group
Note: Hourly wages, including bonuses, in PPP-adjusted USD (2012). The regressions are estimated with log wages as the dependent variable separately for language native and non-native workers and includes controls for years of education, years of experience and experience squared, part-time work and gender. The wage distribution was trimmed to eliminate the 1st and 99th percentiles. The estimated coefficients have been multiplied by 52 which corresponds to the standard deviation in numeracy proficiency. Statistically significant coefficients are marked in a darker tone. Estimates based on a sample size less than 30 are not shown.
Countries are ranked in descending order of the percentage change in wages of the non-native speakers.
Source: (OECD, 2015[15]) Survey of Adult Skills (PIAAC) (2012, 2015), Table 3.A.4., www.oecd.org/skills/piaac/publicdataandanalysis
Language distance and migrant’s outcomes
The analyses presented in previous sections clearly distinguish migrants into those whose native language is the same as the language in which the PIAAC test was conducted and those whose native language is different to conclude that in several countries speaking a language that is different from the language of the assessment is associated with lower literacy and numeracy scores. However, a dichotomous differentiation between same/different languages is inevitably simplistic and does not consider the rainbow of variability and degrees of similarity that exist between languages. For example, the language barrier that migrants from Spanish speaking countries face when settling in Italy is not the same that Spanish speaking migrants face when they settle in Finland. This section develops a more detailed analysis to capture the extent to which some of the between-country variation in the language penalty gap that is observed in PIAAC is related to language composition of resident migrant populations. Furthermore, this section attempts to quantify the association between the degree of linguistic proximity between individuals’ mother tongue and the language in which they sat the PIAAC assessment and their literacy, numeracy and wage levels.
This section builds on previous work on the issue based on data from IALS (Isphording, 2014[2]) and expands the analysis to include literacy, numeracy and wage levels but also considerably extends the generalisability of findings because of the wider spectrum of languages and countries covered in PIAAC compared to IALS. This section attempts to establish if the relative difficulty in learning a distal language explains differences in skill levels between migrants, especially among those who recently settled in the country, or arrived beyond the age of 12, a critical age for language proficiency acquisition.
To analyse the differences in literacy and numeracy proficiency between native and non-native speakers, and study the extent to which language dissimilarity explains part of the variability in the performance gap and labour market outcomes of individuals who are non-native speakers, this chapter uses a measure of linguistic proximity. The Automatic Similarity Judgement Program (ASJP), developed by the German Max Planck Institute for Evolutionary Anthropology, is based on the comparison of the pronunciation of words that have the same meaning in pairs of languages. Languages can differ along a number of dimensions: vocabulary, grammar, pronunciation, scripture and phonetic inventories. The overall distance between any two pairs of languages reflects the degree of dissimilarity across all key language dimensions and reflects the ease/difficulty with which individuals speaking one language can acquire proficiency and mastery in the other language. Box 3.1 includes an explanation of how this measure is computed as explained in Bakker et al., (2009[19]).
Box 3.1. Estimation of the Language Distance index: Levenshtein Distance and the ASJP programme
The Levenshtein distance is a metric developed to identify the difference between two sequences. When comparing words, the Levenshtein distance characterises the minimum number of single-character edits (insertions, deletions or substitutions) that one is required to perform in order to change one word into the other (Levenshtein, 1966[17]). This chapter is based on previous work exploiting the Levenshtein distance to compute the level of dissimilarity across combinations of languages (Bakker et al., 2009[15]). The Max Planck’s ASJP programme developed a composite indicator based on the automatic comparison of the pronunciation of 40 words that have the same meaning from 4 664 languages. The indicator is built using the following procedure. First, each pair of words i with the same meaning is judged according to their similarity in pronunciation, by counting the number of insertions, deletions or substitutions of consonants and vowels that are necessary to transfer the phonetic transcription of one word (in language x) to the other correspondent word in language y, obtaining a measure of the distance between language x and y for the pair or words i,. For example, the English word person – expressed phonetically as pers3n – needs 2 insertions, deletions or substitutions to be transformed into the same word in Spanish, persona.
To aid interpretations, the table below displays some examples of the Levenshtein distance between words of different languages with the same meaning. This first value that we estimate, for each pair of words, is then normalised by the potential maximum distance between both words, obtaining. The average the normalised distances estimated for the 40 words in the list is then computed obtaining the normalised language distance between languages x and y, . This estimate is normalised again by dividing it by the global distance, which is the average distance between any word in the list in language α with any word in the list for language β.
Finally, to obtain our definitive measure of language distance, we divide our previous value of normalised language distance by the global distance to obtain the normalized normalised and divided Levenshtein distance.
Table 3.1. Examples of measurement of language distance between words
Source: (Brown et al., 2008[20])
In order to identify if the variability in the outcomes of migrants is associated with how similar/dissimilar their mother tongue is to the language of their host country, a linguistic proximity indicator was computed for all pairs of languages present in the PIAAC respondent’s sample. The Language Distance index is then used as a control in models that estimate gaps in literacy and numeracy.
A shortcoming of the diversity indices represented in Figure 3.3 is that while they capture quantitative differences in the size of the different language groups, they do not account for the degree to which languages differ from each other and, even more crucially, for the degree to which language minorities differ from the main language spoken in a country. In order to account for these qualitative differences, Figure 3.8 displays not only the level of language diversity that exists within migrant populations, but also the extent to which, on average, the languages that migrant communities speak are very similar or very different from the language in which they sat the PIAAC test which, for the large majority of countries, corresponds to the official language spoken either in the country as a whole or in the local community in which the respondent lives.
Figure 3.8 suggests that the two dimensions are correlated: countries with migrant populations that are non-homogeneous tend to be countries in which, on average, the distance between the languages spoken by migrants and the official language spoken in the country is largest and, conversely, countries with homogeneous migrant populations tend to be countries in which the average language distance is smallest. At the two extremes are Sweden and Chile. Sweden is home to a large number of migrants from several communities, most of which speak a language that is very different from Swedish. By contrast, most migrants in Chile speak the same language and this language is Spanish, therefore the language distance indicator in Chile is very low. The United States and English speaking parts of Canada, Slovenia and the Czech Republic have similar levels of language distance, but they differ importantly in the degree of diversity within language groups that are observed in the country. The level of diversity of languages spoken in Finland and the Slovak Republic is roughly the same, but the average language distance between Finnish and the languages spoken by migrants in Finland is considerably greater than the average language distance observed in the Slovak Republic.
Within countries there are important differences in the language distance that different groups of migrants face. For example, among migrants who live in Italy the language distance that Romanian migrants face is considerably lower than that faced by Albanians. The Language Distance index between Romanian and Italian is in fact 57 while it is 93 between Albanian and Italian.
The level of diversity of languages spoken by migrants and the degree to which on average such languages differ from the official language spoken in the country is an important consideration when assessing the potential language training needs of migrant communities. The greater the distance between the languages spoken by minorities and the official language spoken in the country is, the more intense and long term language training needs are likely to be and the more difficult obtaining language fluency will be for migrant communities. The greater the diversity of languages spoken by migrants, the more difficult it may be to find trainers who will be able to cater to a large variety of needs but, in the absence of large communities of migrants who speak the same language, the greater the incentive for migrants will be to learn the official language in the country, because opportunities for communication within the migrant community will be lower.
Figure 3.8. Language distance in PIAAC participating countries
Note: Estimates for Australia, Germany and the Russian Federation are missing due to the lack of language variables.
Countries are ranked in descending order of the average language distance.
Source: (OECD, 2015[15])Survey of Adult Skills (PIAAC) (2012, 2015), Table 3.A.3., www.oecd.org/skills/piaac/publicdataandanalysis
Despite being widely used in previous empirical work (Bakker et al., 2009[19]; Isphording and Otten, 2014[13]), a key limitation of the ASJP is that the resulting indicator only captures differences between languages in pronunciation. However, as previously mentioned, languages differ along a range of dimensions, such as grammatical structure, alphabet or the extension of its vocabulary. This is particularly relevant because the ASJP measures distance between languages only in their spoken form, while the PIAAC measures are based on a written text. However, the ASJP measure is a good approximation of broader similarity in other characteristics between two languages, and it is a strong predictor of family relations between languages. Compared to other measures of linguistic dissimilarity, the ASJP measure is easily and transparently computed for any pair of languages and thus, readily available for analyses. For this reason, it is the most widely used measure of linguistic distance in empirical work.
Linguistic gaps in literacy, numeracy and labour market outcomes
To identify the association between language distance and literacy scores, a series of models were estimated and results are reported in Tables 3.2 and 3.3. For each outcome of interest, literacy score, numeracy score, employment status and wages, analyses were conducted on the pooled PIAAC sample, weighting each country for population size and weighting individual respondents to derive estimates representative at the population level. All models control for country of destination as well as for the level of economic development in the country of origin of foreign-born adults through an indicator of GDP per capita. Although this means that analyses account for confounders related to differences in host countries’ migration selection processes (such as use of points-based systems that select migrants by ability) and economic incentives for migrating, because country of destination effects may also account for between-country differences in the quality of support and integration systems for migrants, estimates may be over-controlling for country policy choices.
For each outcome, four models were developed. In the first two models the entire PIAAC sample was considered to identify the additional hurdle immigrants face while in the last two models the foreign-born population was the focus of analyses to identify the extent to which language explains differences within the immigrant population. In the first model, each outcome of interest was considered to be a function of whether the respondent was a migrant, the linguistic distance between the mother tongue language of the individual and the language in which the PIAAC test was conducted, accounting for socio-economic and demographic differences such as age, gender, individuals’ own educational attainment and parental educational attainment. In the second model the role of language distance is not considered to affect outcomes homogeneously and differences in the relationship between language distance and outcomes across genders and across individuals with poorly and highly educated parents are examined.
Following the theoretical framework set by Chiswick and Miller (1995[8]), the third model focuses on the migrant subsample and controls for the number of years since arrival into the host country, arguably the most important variable that affects the exposure of migrants to the host country language and well as age at arrival. Length of stay is included by deriving a categorical indicator discriminating between native-born individuals, individuals who lived in the country for more than 5 years, and individuals who have been in the country for 5 years or less. Age at arrival is considered through an indicator of whether immigrants reported having arrived in the country in which they sat the PIAAC test at age 12 or older, to test for the validity of the Critical Period Hypothesis (Newport, 2002[9]). Finally, the fourth identifies specificities in the extent to which language distance interacts with exposure and arrival during/after the critical period for language acquisition.
Results reported in Table 3.2 indicate that the greater the linguistic dissimilarity between the mother tongue of an individual and the language in which the individual sat the PIAAC test, the lower his or her proficiency in literacy and numeracy will be. Results presented in Models 1 and 2 in Table 3.2 indicates that, among PIAAC participating countries, migrants whose mother tongue is the same as the language in which the PIAAC test, and when they match natives on background characteristics (including the level of economic development of the country from which they migrated) they perform on a par in literacy and numeracy with native born individuals. However, when comparing foreign-born individuals whose mother tongue matches the language in which they sat the PIAAC test and foreign-born individuals whose mother tongue is different, the performance gap increases by around 7 points in literacy and 6 points in numeracy for each additional 50 points on the linguistic distance scale. This means, for example, that in Italy, Albanians can be expected to suffer an additional penalty of over 12 score points in literacy due to the fact that their language is very different from Italian (93.40) while Romanians can be expected to suffer an additional penalty of around 7.5 score points, because the common roots between Italian and Romanian make the two languages rather similar (56.77). Interestingly, the penalty associated with language distance is very similar in numeracy.
The association between language distance and literacy and numeracy scores can be useful not only to interpret variation in scores across foreign-born individuals within the same country of destination, but also to better understand the role language composition plays in shaping between country differences in the literacy and numeracy gap between natives and foreign-born individuals. Figure 3.9 illustrates the additional average gap in literacy that can be expected in PIAAC participating countries given the average level of linguistic distance among migrant populations in the country, after controlling for individual characteristics.
Table 3.2. Literacy and numeracy proficiency as a function of language distance
Note: Estimates for Australia, Germany and the Russian Federation are missing due to the lack of language variables.
Source: (OECD, 2015[15]) Survey of Adult Skills (PIAAC) (2012, 2015), Table 3.A.3., www.oecd.org/skills/piaac/publicdataandanalysis
Figure 3.9. Additional migrant gap in literacy due to the average linguistic distance among foreign-born populations, by country
Note: Estimates for Australia, Germany and the Russian Federation are missing due to the lack of language variables.
Note: Countries are ranked in descending order of the score point difference.
Source: (OECD, 2015[15]) Survey of Adult Skills (PIAAC) (2012, 2015), Table 3.A.3., www.oecd.org/skills/piaac/publicdataandanalysis
When comparing individuals who have the same gender, age, own education and parental education, and come from countries with similar level, a gap of 10 points in the linguistic distance measure corresponds to 1.44 score points in literacy and 1.11 in numeracy. In order to aid the interpretation of the quantitative relevance of estimated effects in explaining both between country and within country variations, Figure 3.6 illustrates, for each country, the average level of language distance observed among migrant communities, as well as the interquartile range in the linguistic distance measure. The comparatively high level of language distance observed, on average, among foreign-born individuals resident in Sweden or Norway suggest that, in these countries, migrant gaps can be expected to be very large in international comparisons. However, differences in the language spoken by different migrant groups explain very little of the variations observed within the country across different migrant communities. On the other hand, in countries such as Spain, the average language distance is small, in comparative terms and therefore observed gaps in these countries in literacy and numeracy between native born individuals and foreign-born individuals are not attributable to language factors. However, the comparatively large interquartile range measure observed in these countries suggests that when evaluating variations in the migrant gap across individuals within the country, language may play a key role.
Table 3.1 supports the critical period hypothesis: individuals who arrived in the country of destination after the age of 12 have lower literacy and numeracy scores than individuals who arrived prior to age 12. On average, the late arrival penalty corresponds to almost 17 score points in literacy and 14 points in numeracy. Length of residency appears to be positively associated with proficiency in PIAAC. Other things being equal, individuals who resided in the country for 5 years or more prior to siting the PIAAC test score, on average, 11 points in literacy and 10 points in numeracy above those who had been resident in the country for less than 5 years. Interestingly, although the negative association between language distance and performance in literacy and numeracy does not depend on whether individuals resided in the country for over 5 years, the negative association between language distance and both literacy and numeracy is stronger among individuals who arrived after the age of 12. A difference of 10 points in the language distance index is associated with a difference of an additional 1.8 score points in the PIAAC literacy assessment and an additional 1.3 points in the numeracy assessment, effectively indicating that the language distance disadvantage is twice as large among late arrivals.
Previous sections of this chapter revealed that countries differ considerably in the language make-up of their resident populations. Analyses presented in this section suggest that individuals whose mother tongue is very different from the language in which they sat the PIAAC test have, other things being equal, lower scores in literacy and numeracy. Table 3.3 identifies the relationship between the mother tongue of individuals and how different this is from the language spoken in the country in which they reside and their labour market outcomes, most notably, their probability of being in work and their wage level when employed.
Table 3.3 suggests that, other things being equal, individuals whose language is very different from the language in which they sat the PIAAC test are less likely to be employed, although the association is quantitatively small. For example, other things being similar, the difference in the probability that an Arabic speaking migrant in Italy (the maximum observed language distance observed in Italy) will be employed compared to a Romanian speaking migrant in Italy (the minimum observed language observed in Italy) is 2.7 percentage points. Results confirm that individuals who arrived after the age of 12 are more likely to be employed than individuals who arrived in the country before the age of 12. This result may be due to selection mechanisms: it is possible that when older individuals migrate or families with older children migrate, they do so because of employment prospects or because individuals who migrated to a country as older children or as adults tend to create fewer non-labour market bonds with their local communities and are more likely to resettle in their country of origin or to look for an alternative destination when they are out of work or reach retirement. Table 3.3 indicates that among individuals who are employed and receive a salary, wage levels are not associated with the distance between the mother tongue individuals speak and the language spoken in their country of residence. Unfortunately, a key limitation of analyses examining labour market outcomes is that PIAAC did not collect information on the language used in the labour market by the respondent and therefore the language distance index is calculated using the language in which the PIAAC test is taken as the reference. However, this may not reflect the language spoken by the respondent at work and how sought after proficiency in particular languages may be in local labour markets (for example English).
Table 3.3. Employment and wages as a function of language distance
Note: Estimates for Australia, Germany and the Russian Federation are missing due to the lack of language variables.
Source: (OECD, 2015[15]) Survey of Adult Skills (PIAAC) (2012, 2015), Table 3.A.3., www.oecd.org/skills/piaac/publicdataandanalysis
Conclusions and Implications
This chapter indicates that the literacy and numeracy proficiency of non-native language speakers is, other things being equal, lower than that of native language speakers. As such, language proficiency can be a major hurdle for the social and economic integration of migrants into the labour market and social life of their receiving communities. The acquisition of literacy proficiency in the host country language is importantly related to the mother tongue of the migrant and the age at which migration took place. In particular, this chapter suggests that individuals with a linguistically distant background face a distinctively higher challenge to reach a sufficient level of literacy and numeracy proficiency in the host country language than individuals whose language background is more homogeneous, in particular when they migrate after the age of 12.
Results presented in the chapter suggest that differences in the linguistic make-up of migrant populations explain both between country differences in the migrant gaps in literacy and numeracy scores, as well as the within-country variability in literacy and numeracy across migrants. These results suggest that language training is crucial if migrant communities are to be able to be fully integrated in the labour markets and social lives of their communities but also that the time and intensity of language training provided to should be tailored to the specific language group migrants belong to. More intense and longer training should be devoted to individuals coming from linguistically distant groups and training should account for the specific communalities and differences across languages to be maximally effective.
The finding that the skills of non-native speaking migrants tend to be lower in particular when they migrate after the age of 12 and when their mother-tongue is very different from the language spoken in their host communities suggests that strong language support should be given in the context of pre-school and primary school to the extent possible, so that children are maximally supported in a crucial period for language acquisition but also that intense training should be given to those who arrive when they have missed such opportunities for language development. Finding new, creative ways to help individuals who arrive as teenagers or adults gain valuable language skills, should become a priority for educators and education scientists so that current barriers are eliminated or lowered. Even though the costs of providing adequate language support may be large, analyses presented in this report suggest that the cost of inaction is likely to be considerably larger than investments in adequate and effective training.
Traditional modes of instruction could be complemented, for example, by the use of technology, which has proven to be effective for supporting non-native language acquisition in some contexts. While technology cannot replace real classroom instruction, it can be used to complement and supplement the work of trained teachers and professional working with non-native language learners. Opportunities for language learning can be ubiquitous when learners use mobile technologies to access information and communicate with other learners or educators. Technology can change traditional teacher-centred instructional settings, which are ill suited to promote language acquisition, and open new possibilities for collaboration, social interaction and access to multiple resources to enhance non-native language learning (Eamer, 2013[21]).
Three examples of how technology can promote language acquisition are: “Digital communities of practice” where non-native speakers can engage with native speakers through online discussions. Non-native speakers can as or even more participative than native speakers, and gain a legitimate status through academic socialisation. Such communities can promote learners’ motivation by enabling social collaboration. (Kim, 2010[22]).; “Digital storytelling” where non-native language learners use compilations of photo, video, audio and text to produce a meaningful output in the language they need to learn (Rowinsky-Geurts, 2013[23]). Students may find this approach cognitively challenging (e.g., having difficulties with vocabulary and verb conjugation), but rewarding in terms of complex thinking and using complex strategies to complete in-depth artefacts; and “Computer-assisted language learning” (CALL) which uses computers to monitor students’ progress and provide targeted feedback (Presson, Davy and MacWhinney, 2013[24]). CALL materials can be useful to aim the learning of specific vocabulary, grammatical forms, or pronunciation skills.
References
[18] Alesina, A., J. Harnoss and H. Rapoport (2015), “Birthplace diversity and economic prosperity”, No. 18699, National Bureau of Economic Research, Cambridge, MA, http://dx.doi.org/10.3386/w18699.
[26] Alesina, A. and E. La Ferrara (2002), “Who trusts others?”, Journal of Public Economics, Vol. 85, pp. 207-34.
[19] Bakker, D. et al. (2009), “Adding typology to lexicostatistics: A combined approach to language classification”, Linguistic Typology, Vol. 13/1, pp. 169-181.
[7] Bleakley, H. and A. Chin (2010), “Age at arrival, English proficiency, and social assimilation among US immigrants”, American Economic Journal. Applied Economics, Vol. 2/1, p. 165.
[6] Bleakley, H. and A. Chin (2004), “Language skills and earnings: Evidence from childhood immigrants”, Review of Economics and Statistics, Vol. 86/2, pp. 481-496.
[1] Bonfanti, S. and T. Xenogiani (2014), “Migrants' skills: Use, mismatch and labour market outcomes – A first exploration of the International Survey of Adult Skills (PIAAC)”, in Matching Economic Migration with Labour Market Needs, OECD Publishing, Paris.
[20] Brown, C. et al. (2008), “Automated classification of the world’s languages: a description of the method and preliminary results”, STUF-Language Typology and Universals Sprachtypologie und Universalienforschung, Vol. 61/4, pp. 285-308.
[8] Chiswick, B. and P. Miller (1995), “The endogeneity between language and earnings: International analyses”, Journal of Labor Economics, Vol. 13/2, pp. 391-409.
[17] Collier, P. (2001), “Ethnic diversity: An economic analysis”, Economic Policy, Vol. 16/32, pp. 127-166, https://www.researchgate.net/publication/235737439_Ethnic_Diversity_an_Economic_Analysis (accessed on 02 October 2018).
[12] Dustmann, C. (1999), “Temporary migration, human capital, and language fluency of migrants”, The Scandinavian Journal of Economics, Vol. 101/2, pp. 297-314.
[10] Dustmann, C. (1994), “Speaking fluency, writing fluency and earnings of migrants”, Journal of Population Economics, Vol. 7/2, pp. 133-156.
[5] Dustmann, C. and F. Fabbri (2003), “Language proficiency and labour market performance of immigrants in the UK”, The Economic Journal, Vol. 113/489, pp. 695-717.
[4] Dustmann, C. and A. Van Soest (2001), “Language fluency and earnings: Estimation with misclassified language indicators”, Review of Economics and Statistics, Vol. 83/4, pp. 663-674.
[21] Eamer, A. (2013), “Computer-assisted language learning in the new frontier: Exploring the i4”, in Schwieter, J. (ed.), Innovative research and practices in second language Acquisition and Bilingualism, Benjamins, Amsterdam and Philadelphia.
[16] Easterly, W. and R. Levine (1997), “Africa's growth tragedy: Policies and ethnic divisions”, The Quarterly Journal of Economics, Vol. 112/1, http://dx.doi.org/10.1162/003355300555466.
[2] Isphording, I. (2014), “Disadvantages of linguistic origin—Evidence from immigrant literacy scores”, Economics Letters, Vol. 123/2, pp. 236-239.
[13] Isphording, I. and S. Otten (2014), “Linguistic barriers in the destination language acquisition of immigrants”, Journal of Economic Behavior & Organization, Vol. 105, pp. 30-50.
[11] Isphording, I. and S. Otten (2011), “Linguistic Distance and the language fluency of immigrants”, Ruhr Economic Paper, No. 274.
[22] Kim, H. (2010), “Promoting communities of practice among non-native speakers of English in online discussions”, . Computer Assisted Language Learning, Vol. 24/4, pp. 353-370.
[25] Levenshtein, V. (1966), “Binary codes capable of correcting deletions, insertions, and reversals”, Soviet Physics Doklady , Vol. 10/8, pp. 707-710.
[14] Meng, X. and R. Gregory (2005), “Intermarriage and the economic assimilation of immigrants”, Journal of Labor Economics, Vol. 23/1, pp. 135-174.
[9] Newport, E. (2002), “Critical periods in language development”, in Nadel, L. (ed.), Encyclopedia of Cognitive Science, Macmillan Publishers Ltd./Nature Publishing Group, London.
[15] OECD (2015), Survey of Adult Skills (PIAAC) (database), http://www.oecd.org/skills/piaac/publicdataandanalysis.
[3] OECD/EU (2014), Matching Economic Migration with Labour Market Needs, OECD Publishing, Paris, http://dx.doi.org/10.1787/9789264216501-en.
[24] Presson, N., C. Davy and B. MacWhinney (2013), “Experimentalized CALL for adult second language learners”, in Schwieter, J. (ed.), Innovative Research and Practices in Second Language Acquisition and Bilingualism, Benjamins, Amsterdam/Philadelphia.
[23] Rowinsky-Geurts, M. (2013), “Digital stories in L2 classes: High-impact practices and affective learning.”, in Schwieter, J. (ed.), Studies and Global Perspectives of Second Language Teaching and Learning, Benjamins, Amsterdam/Philadelphia.
Annex 3.A. Data tables
Annex Table 3.A.1. Percentage of adults, by immigrant and language background
Note: Native speaker refers to whether the first or second language learned as a child is the same as the language of assessment, and not whether the language has official status. Non-native speaker refers to whether the first or second language learned as a child is not the same as the language of assessment. Thus in some cases, foreign language might refer to minority languages in which the assessment was not administered. Estimates are missing for the Russian Federation due to the lack of language variables.
Source: Survey of Adult Skills (PIAAC) (2012, 2015), www.oecd.org/skills/piaac/publicdataandanalysis
Annex Table 3.A.2. Percentage of non-native speakers, by first language spoken at home and understood.
Note: Non-native speaker refers to whether the first or second language learned as a child is not the same as the language of assessment. Thus in some cases, foreign language might refer to minority languages in which the assessment was not administered. Estimates are missing for Australia, Germany and the Russian Federation due to the lack of language variables. Estimates based on small sample size are not shown (Chile, Japan, Korea and Northern Ireland (UK)).
Source: Survey of Adult Skills (PIAAC) (2012, 2015), www.oecd.org/skills/piaac/publicdataandanalysis
Annex Table 3.A.3. Diversity of languages within and between migrants and distribution of the language distance among migrants
Note: Estimates for Australia and Germany are missing due to the lack of language variables.
Source: Survey of Adult Skills (PIAAC) (2012, 2015), www.oecd.org/skills/piaac/publicdataandanalysis
Annex Table 3.A.4. Percentage change in wages associated with an increase of one standard deviation in proficiency for native and non-native speakers
Note: Hourly wages, including bonuses, in PPP-adjusted USD (2012). The regressions are estimated with log wages as the dependent variable separately for language native and non-native workers and includes controls for years of education, years of experience and experience squared, part-time work and gender. The wage distribution was trimmed to eliminate the 1st and 99th percentiles. The estimated coefficients have been multiplied by 47 for literacy and 52 for numeracy which correspond to the standard deviation in the proficiency. Estimates based on a sample size less than 30 are not shown (Chile and Japan). Estimates for the Russian Federation are missing due to the lack of language variables.
Source: Survey of Adult Skills (PIAAC) (2012, 2015), www.oecd.org/skills/piaac/publicdataandanalysis
Annex Table 3.A.5. Language distance diversity in PIAAC participating countries
Minimum and maximum language distance between the language of the test and the language spoken at birth of the foreign-language migrants
Note: Only language couples with more than 30 observations in the country are taken into account to extract the minimum and the maximum distance language.
1. Note by Turkey: The information in this document with reference to “Cyprus” relates to the southern part of the Island. There is no single authority representing both Turkish and Greek Cypriot people on the Island. Turkey recognises the Turkish Republic of Northern Cyprus (TRNC). Until a lasting and equitable solution is found within the context of the United Nations, Turkey shall preserve its position concerning the “Cyprus issue”.
2. Note by all the European Union Member States of the OECD and the European Union: The Republic of Cyprus is recognised by all members of the United Nations with the exception of Turkey. The information in this document relates to the area under the effective control of the Government of the Republic of Cyprus.
Source: Survey of Adult Skills (PIAAC) (2012, 2015), www.oecd.org/skills/piaac/publicdataandanalysis