CAN THE LOW UNEMPLOYMENT RATE OF SWEDISH-SPEAKERS IN FINLAND BE ATTRIBUTED TO LANGUAGE-GROUP AND INDUSTRIAL STRUCTURE?

Transcription

1 CAN THE LOW UNEMPLOYMENT RATE OF SWEDISH-SPEAKERS IN FINLAND BE ATTRIBUTED TO LANGUAGE-GROUP AND INDUSTRIAL STRUCTURE? * Jan Saarela and Fjalar Finnäs Abstract. This paper attempts to explain why the unemployment rate of the Swedishspeaking minority in Finland is substantially lower than that of the Finnish-speaking majority. Specifically, we are concerned with the role of language-group and industrial structure. The study mainly utilise cross section data from the end of 1990, 1995 and 1998, of all labour force participants in bilingual municipalities. The analysis results of logistic regression models suggest that the language-group difference in unemployment rates is larger in areas with a higher proportion of Swedish-speakers, but that it only to a limited extent can be explained by industrial structure, age, gender and education. Our conclusion is that the reasons behind the unemployment gap can hardly be found by analysing existing register data. Keywords: Unemployment, Native language, Industrial structure, Ethnic enclaves JEL-codes: J15, J64, J71 * Financial support from Rektor för Åbo Akademi is gratefully acknowledged (Saarela). Comments from seminar participants at EEA 2001 and anonymous persons have been helpful. A previous version of this paper was entitled Ethnicity, Industrial Structure and Unemployment in Finland.

2 2 Can the low unemployment rate of Swedish-speakers in Finland 1. Introduction Belonging to an ethnic or linguistic minority does not inevitably mean that a person is disadvantaged in the labour market. There are groups, for example American-born Chinese, Japanese and Jews in the United States (Brenner and Kiefer, 1981; Sowell, 1981; Chiswick, 1983a; 1983b), as well as Welsh-speakers in Wales (Drinkwater and O Leary, 1997), which have an overall better labour market position than people belonging to the majority. The reasons for this have not been fully understood, since the between-group differences cannot be totally attributed to socio-economic and demographic factors. In Finland, Swedish-speakers constitute the largest minority group, barely six per cent of the total population. 1 They are guaranteed equal constitutional rights as the Finnishspeakers, and are also at a clear advantage in the labour market. The national unemployment rate of Swedish-speakers was 2.6 per cent at the end of 1990, in comparison with 5.7 per cent for Finnish-speakers. During the mid-1990s Finland was hit by a deep economic recession, which resulted in a dramatic increase in unemployment rates. The language-group difference although remained; the national unemployment rate at the end of 1995 was 11.4 per cent for Swedish-speakers, in comparison with 19.3 for Finnishspeakers. Also after the recession, at the end of 1998, there was a large disparity; the unemployment rate of Swedish-speakers was 7.9 per cent, in comparison with 14.9 per cent for Finnish-speakers. Practically all Swedish-speakers reside very concentrated on the southern and western coastlines of the country. These regions are those experiencing the lowest unemployment rates, which suggests that the between-group difference in unemployment rates may be due to regional factors. Also in this area (the shaded parts on the map in Figure 1), however, there are noteworthy disparities. For males, the unemployment rate of Swedishspeakers was 2.8 per cent at the end of 1990, whereas it was 3.7 per cent for Finnishspeakers. At the end of 1995 it was 11.7 per cent for Swedish-speakers, and as high as 18.3 per cent for Finnish-speakers. Also at the end of 1998, there was a large difference; the unemployment rate of Swedish-speakers was 7.5 per cent, as compared with 11.7 per cent for Finnish-speakers. For females, the pattern does not seem to be equally strong. At the end of 1990, the unemployment rate of Swedish-speaking females was in fact somewhat higher than that of Finnish-speakers; 2.4 per cent, in comparison with 1.9 per cent. At later cross section dates, however, the similar pattern as for males can be observed, although the between-group difference not being equally large. The unemployment rate of Swedish-speaking females was 11.8 per cent at the end of 1995, as compared with 14.1 per cent for Finnish-speaking females. At the end of 1998, it was 9.0 per cent for Swedish-speaking females, in comparison with 9.9 per cent for Finnish-speaking females. 1 Less than two per cent of the population has a native language other than Finnish or Swedish.

3 Can the low unemployment rate of Swedish-speakers in Finland 3 Vasa Per cent Swedish-speakers Åland Islands Turku Helsinki Figure 1. The geographical concentration of Swedish-speakers in Finland (1990) Saarela and Finnäs (2002a) were the first to explore why Swedish-speakers in Finland have lower unemployment rates than Finnish-speakers. They show that the betweengroup difference in unemployment rates cannot be attributed to age, education, gender or municipality of residence, because it decreases only marginally when accounting for the impact of these factors. The present understanding of the reasons behind the lower unemployment rates of Swedish-speakers is thus fairly limited. This paper attempts to shed some further light onto this issue. Specifically, we are interested in the contribution of structural factors, i.e. whether the language-group structure and the industrial structure of individuals place of residence may help to explain the unemployment gap between Swedish-speakers and Finnish-speakers.

4 4 Can the low unemployment rate of Swedish-speakers in Finland Half of the Swedish-speaking population reside in municipalities where they form the local majority (see Figure 1). Swedish-speakers are thus a minority at the national level, but to some extent a majority at the local level. There are a number of studies showing that the geographical concentration of an ethnic group may affect its performance in the labour market (see e.g. Borjas, 1995; Clark and Drinkwater, 2002), which plausibly could be the case also here. The specific industrial structure of a minority group may be another important determinant of the group s overall performance in the labour market (cf. Brenner and Kiefer, 1981; Shackett and Trapani, 1987). It is plausible that this could explain some of the unemployment gap between Swedish-speakers and Finnish-speakers, because they are to some extent employed in different industries. In order for the reader to better understand the outline of the paper, the next section discusses the most essential potential factors that may contribute to the relatively low unemployment rate of Swedish-speakers in Finland. 2. Theoretical considerations The present paper does not compare immigrants and natives, as many previous studies of ethnic minorities do (see Chiswick, 1977; 1978; Borjas, 1985; 1987; 1992; 1993; 1994). Swedish-speakers constitute a native group in Finland, which has been less mobile with regard to within-country migration than the Finnish-speakers. Many Finnish-speakers are in fact first or second generation of migrants into the Swedish-speakers main settlement area. The latter implies that the language structure of the area has experienced a fairly dramatic change during the 20th century. During several centuries, up to 1809, Finland was an integrated and equal part of the realm of Sweden. Swedish was then, and also after that, the dominant language of government, business and culture. Many Swedish-speakers were, in comparison with Finnishspeakers, well educated, had a relatively good socio-economic position, and owned some piece of land. They have therefore, traditionally, worked in different industries than Finnish-speakers. The language-group difference in socio-economic position is nowadays much less prevalent. Some differences in industrial distribution although still remain. Table 1 shows that the most notable feature in this context is the large proportion of Swedish-speakers employed in the agricultural sector. Agriculture although amounts to a fairly small part of total employment. The proportion of people employed in this sector has experienced a downward sloping trend during the whole 1990s. It is therefore not reasonable to expect that agricultural employment alone could be the main reason behind the lower unemployment rate of Swedishspeakers.

5 Can the low unemployment rate of Swedish-speakers in Finland 5 Table 1. Proportion of people employed in each industry by gender and native language at the end of 1990 Males Females Swedishspeakers Finnishspeakers Swedishspeakers Finnishspeakers Industry Agriculture Manufacturing Construction Trade, hotels and restaurants Transport and communications Financial intermediation Public and other services Industry unknown n 59, ,100 53, ,464 The description refers to the bilingual part of Finland, except for Åland that is excluded. Industry classification is according to SIC 1995: Agriculture, hunting, forestry and fishing (code 01-05), Manufacturing (code 10-41), Construction (code 45), Trade, hotels and restaurants (code 50-55), Transport and communications (code 60-64), Financial intermediation, insurance and business activities (code 65-74), Public and other services (75-97), Industry unknown (code 99). Source: The cross section data from Statistics Finland, which are utilised in the analysis. The geographical concentration of Swedish-speakers in Finland has, fairly evidently, favoured the vitality of the group. The most important institutional elements of the Swedish-speaking community are political and educational, as well as cultural. These include political pressure groups, a complete school system, media, theatres, a diocese, and a number of different co-operative organisations (see Saarela and Finnäs, 2002b). 2 These elements are commonly regarded as important for the social networks of Swedishspeakers. It is plausible that they favour job search, and thus reduce unemployment propensity (cf. Montgomery, 1992). There are no data available, however, that would provide us with explicit information about such issues. Neither are there any existing data that would provide information about language proficiency of the two groups. It is still fairly obvious that Swedish-speakers to a higher extent than Finnish-speakers are bilingual, i.e. that they speak both Swedish and Finnish fluently (Finnäs, 2000; Saarela and Finnäs, 2002a). Considering that Swedish-speakers form a majority in certain local areas, and that language proficiency requirements (ability to speak Swedish, or both Swedish and Finnish) most likely are higher in areas with a higher proportion of Swedish-speakers, there are strong reasons to believe that the language-group difference in unemployment propensity may increase with the proportion of Swedish-speakers in that area. 2 O Leary and Finnäs (2002) provide a discussion about similarities between Protestants in the Republic of Ireland and Swedish-speakers in Finland, with focus on the metropolitan regions.

6 6 Can the low unemployment rate of Swedish-speakers in Finland There are a number of theoretical arguments as to why this may be the case. Potential customers and employers in an ethnic enclave are more likely to be co-ethnics, and tastes for discrimination against the other group consequently more prevalent (Becker, 1971; Holzer and Ihlandfeldt, 1998). Shared language and culture may improve group-specific opportunities (Aldrich et al., 1985; Lazear, 1999). Language skills may also, besides the direct effect on economic outcomes (cf. Grenier, 1987; Hayfron, 2001; Dustmann and van Soest, 2001; Chiswick and Miller, 2002; Shields and Wheatley Price, 2002), benefit from the proximity of co-ethnic employers (McManus, 1990; Carliner, 1995). In this paper, we cannot adjudicate between the above explanations other than by analysing the potential impact of language-group structure and industrial structure. 3. Data and methodology We will utilise cross section data, consisting of the total population in Finland at the end of the years 1990, 1995 and 1998 respectively. For each year we have a multidimensional matrix that includes all individuals and information about their native language (a resident in Finland may have only one native language), gender, age, education, municipality of residence, labour market status and, for those employed, industry. These data have been compiled by Statistics Finland, on the basis of the Central Population Register. For each of the cross section dates, we restrict the data to Swedish-speakers and Finnish-speakers living in the bilingual part of Finland (municipalities which have Swedishspeakers; the shaded parts on the map in Figure 1). The Åland Islands are excluded, since they form a monolingual Swedish, self-governed, county with specific legislative practices. Saarela and Finnäs (2002a) have shown that there are no language-group differences in labour force participation. We therefore study only labour force participants (employed and unemployed). We also restrict the data to people aged years. Our aim is to study whether differences in unemployment propensity between Swedish-speakers and Finnish-speakers can be attributed to the language-group structure of the place of residence, on the one hand, and to the industrial structure of the place of residence, on the other hand. Language-group structure refers to the proportion of Swedish-speakers in each municipality, where the municipalities have been compiled into five different categories: less than 15 per cent Swedish-speakers, 15 to 34 per cent, 35 to 64 per cent, 65 to 84 per cent, and at least 85 per cent. We have attempted a number of different categorisations of this variable, including a continuous version of it. The categorisation chosen seemed to be the most appropriate one. Industrial structure refers to the proportion of people employed in each of the 46 municipalities in the data, at the end of This cross section date is used, since the un-

7 Can the low unemployment rate of Swedish-speakers in Finland 7 employment rate at that time was very low (3.2 per cent at the national level). Utilising the industrial distribution at any of the other two cross section dates would not, however, affect the results. In order to account for differences in the sector of work between genders and between language groups, we have calculated the industrial distribution separately for male Swedish-speakers, male Finnish-speakers, female Swedish-speakers and female Finnishspeakers. These relative proportions of people employed in each of the eight different industry categories are not, as such, utilised in the statistical estimations. Instead, we assign them into quartiles. In the results to be reported, we have not utilised all of the industry variables, but only those that appeared to be most important. Table 2. Distribution of some of the explanatory variables (proportions) in the cross section data by gender and native language Males Females Swedishspeakers Finnishspeakers Swedishspeakers Finnishspeakers Age, years Education, Basic Lower vocational Upper vocational Undergraduate Graduate Language-group structure, <15% Swedish n 180, , , ,187 The description refers to labour force participants aged years in the bilingual area, and is for pooled cross section dates. The distribution of the above variables is very similar between cross section dates. Descriptive statistics for cross section date, municipality of residence and industrial-structure variables are not shown. The distributions for the latter are almost similar to those outlined in Table 1. Some descriptive statistics of the utilised data are provided in Table 2. Language-group differences in demographic composition are reflected by the fact that the proportion of

8 8 Can the low unemployment rate of Swedish-speakers in Finland older people is higher among Swedish-speakers than among Finnish-speakers. In correspondence with previous arguments, we can see that Swedish-speakers to a great extent live in areas where they are in majority. This pattern is even more apparent for Finnishspeakers. In fact, of all Finnish-speakers in the bilingual area, as much as 41 per cent live in Helsinki (where the Swedish-speaking proportion of the total population is about eight per cent). Only 14 per cent of all Swedish-speakers live in Helsinki. There are also some differences in education levels. Finnish-speakers seem to have a somewhat higher education. This is although an artefact of regional differences in education and the regional distribution of each language group. Finnish-speakers are, as noted above, heavily concentrated to the metropolitan area, where education levels generally are higher than in more rural areas. Within-municipality differences in education (which we are able to take into account in the statistical analysis) are although in favour of the Swedish-speakers. Descriptive statistics for the variables representing industrial structure are not outlined in the table, since the distribution is almost similar to that shown in Table 1. As a supportive dataset, we will utilise an extract from the Finnish Longitudinal Census Data File, which contains linked individual data from the censuses of 1970, 1975, 1980, 1985, 1990 and This file contains data on all individuals who were residents in Finland at the time of one or more of the censuses. The file is maintained by Statistics Finland. There is information about each individual s native language, gender, age, education, area of residence (in order to guarantee anonymity of the data, the 46 municipalities had to be grouped into 15 larger regions on the basis of geographical position and urbanisation, upon completion of data transfer from Statistics Finland), family situation, labour market status and, for those employed, industry. The longitudinal nature of this dataset enables us to perform an individual-level analysis of the impact of previous industry of work (in 1990) on unemployment propensity (in 1995), unlike the case with the cross section data that utilise a structural industry variable. This dataset is restricted to those who have been employed in We can thus study if there, for those with work experience, are similar language-group differences in unemployment propensity as for all labour force participants, and whether previous industry of work may explain some of these potential differences. Practical limitations of the data imply that we must restrict the analysis to males years of age and females years of age. 3 3 In the utilised extract of the longitudinal file there are some problems with defining labour market status for those not employed, specifically males in compulsory military service and females nursing children at home. We avoid this classification problem by the age restriction.

9 Can the low unemployment rate of Swedish-speakers in Finland 9 Table 3. Distribution of some of the explanatory variables (proportions) in the longitudinal data by gender and native language Males Females Swedishspeakers Finnishspeakers Swedishspeakers Finnishspeakers Age, years n.a. n.a Education, Basic Lower vocational, lower level Lower vocational, upper level Upper vocational Undergraduate Graduate Language-group structure, <15% Swedish Industry in 1990, Agriculture Manufacturing + Construction Trade, hotels, restaurants Transport, communications Financial intermediation Public and other services Family situation, Married with children Married without children Partner, with children (consensual union) Partner, without children (consensual union) Sole supporter Single Living with senior parent n 46, ,368 34, ,563 The description refers to those employed at the end of 1990 in the bilingual area. In the data, the unemployment rate at the end of 1995, for those employed in 1990, is 9.7 per cent for Swedish-speaking males, 15.0 per cent for Finnish-speaking males, 9.8 per cent for Swedish-speaking females, and 11.9 per cent for Finnish-speaking females. The distribution for language-group structure does not coincide with that in the cross section data, because the variable has been constructed on the basis of individuals area, not municipality, of residence. Some descriptive statistics of the longitudinal data set can be seen in Table 3. As expected, the distribution over age, education and industry is fairly similar to that depicted

10 10 Can the low unemployment rate of Swedish-speakers in Finland by the cross section data. The variable representing language-group structure had to be constructed on the basis of individuals area, not municipality, of residence. It is therefore somewhat more imprecise than that used in the cross section data. This explains why the distribution differs somewhat between the two data sets. In these longitudinal data, we also have information about a third type of structural variable: each individual s family situation. Finnäs (1997) has argued that marital stability of Swedish-speakers is higher than of Finnish-speakers, and that this may be considered an indication of a higher level of social integration. The table also shows that the proportion of people living in traditional families (married with children) is substantially higher among Swedish-speakers than among Finnish-speakers. If social integration in terms of social networks affects labour market performance (as argued by Montgomery, 1992), between-group differences in family situation may potentially explain some of the unemployment gap. As an additional feature, we wish to see whether this may be the case. In order to quantify the effects of the explanatory variables on unemployment propensity, logistic regression models will be estimated. Since there are gender-differences in industrial distribution, separate models are estimated for males and females. 4. Results The estimation results are provided in Table 4 and Table 5. They are reported in a number of steps, in order to outline the impact of each set of variables on the language-group difference in unemployment propensity. The first model includes the individual-level variables native language, age, education and cross section date. The subsequent one adds the language-group structure of the municipality, whereas the following accounts for the interaction between native language and language-group structure. Model 4 incorporates industrial-structure variables, instead of those related to language-group structure. The final model includes both sets of structural variables. For males, we use the industrial variables representing agriculture, manufacturing and construction, whereas for females manufacturing and public services. We have experimented with a number of different sets of industrial variables. The chosen ones are those improving the fit of the model mostly. As can be seen, the language-group difference in odds for being unemployed, when accounting for the impact of age, education and cross section date, are substantial higher for males than for females; 42 per cent against 13 per cent. 4 This gender disparity, however, can to a large extent be attributed to the fact that the overall unemployment propensity for females is higher in municipalities with a higher proportion of Swedish-speakers 4 1-exp(-0.549)=0.42 and 1-exp(-0.140)=0.13.

11 Can the low unemployment rate of Swedish-speakers in Finland 11 (Model 2). This explains why language-group differences in aggregate unemployment rates are smaller for females than for males (see the introduction). Table 4. Logistic regression results for unemployment, males in the cross section data (n=1,066,414) Model 1 Model 2 Model 3 Model 4 Model 5 Native language, Finnish Swedish (0.010) (0.012) (0.018) (0.012) (0.025) % Swedish-speakers, < (0.011) (0.012) (0.015) (0.012) (0.014) (0.025) (0.020) (0.030) (0.037) (0.023) (0.053) (0.060) Interaction 0 0 Swe % Swe (0.033) (0.038) Swe % Swe (0.029) (0.036) Swe % Swe (0.041) (0.046) Swe. 85- % Swe (0.060) (0.063) % in Agriculture, Q Q (0.021) (0.023) Q (0.021) (0.029) Q (0.026) (0.037) % in Manufacturing, Q Q (0.011) (0.012) Q (0.013) (0.016) Q (0.015) (0.021) % in Construction, Q Q (0.017) (0.021) Q (0.018) (0.023) Q (0.025) (0.028) Constant (0.013) (0.013) (0.013) (0.021) (0.025) Log likelihood -322, , , , , Cox & Snell R Square Nagelkerke R Square The numbers outlined refer to unexponentiated coefficients. Standard errors are in parenthesis. The variables age, education and cross section date are also included in each model, but the estimated effects are not displayed here. The reference categories used for these variables are years, lower vocational, lower level education and the end of According to Model 3, the language-group difference in unemployment propensity seems to increase with the proportion of Swedish-speakers in the municipality, suggesting that the geographical concentration of Swedish-speakers has a favourable effect with regard to their labour market position, in comparison with that of Finnish-speakers. For

12 12 Can the low unemployment rate of Swedish-speakers in Finland females, the impact of language-group structure is somewhat less emphasised than for males. Table 5. Logistic regression results for unemployment, females in the cross section data (n=1,081,712) Model 1 Model 2 Model 3 Model 4 Model 5 Native language, Finnish Swedish (0.010) (0.013) (0.020) (0.012) (0.022) % Swedish-speakers, < (0.012) (0.013) (0.015) (0.013) (0.015) (0.019) (0.020) (0.029) (0.030) (0.022) (0.045) (0.050) Interaction 0 0 Swe % Swe (0.037) (0.038) Swe % Swe (0.032) (0.034) Swe % Swe (0.042) (0.046) Swe. 85- % Swe (0.054) (0.055) % in Manufacturing, Q Q (0.009) (0.010) Q (0.012) (0.015) Q (0.017) (0.020) % in Services, Q Q (0.022) (0.026) Q (0.023) (0.027) Q (0.022) (0.027) Constant (0.016) (0.016) (0.016) (0.027) (0.031) Log likelihood -281, , , , , Cox & Snell R Square Nagelkerke R Square The numbers outlined refer to unexponentiated coefficients. Standard errors are in parenthesis. The variables age, education and cross section date are also included in each model, but the estimated effects are not displayed here. The reference categories used for these variables are years, lower vocational, lower level education and the end of Industrial structure seem to be important for overall unemployment propensity, and the model fit is also clearly improved (compare Model 4 with Model 1). It cannot, however, explain the difference between language groups. For males, including the industry variables reduce the between-group difference in unemployment propensity somewhat, whereas the difference even is emphasised for females. Finally, we can in Model 5 see that, when accounting for both the impact of languagegroup structure and industrial structure, there remains a substantial difference in unem-

13 Can the low unemployment rate of Swedish-speakers in Finland 13 ployment propensity between the two groups. In municipalities with a low concentration of Swedish-speakers (less than 15 per cent), the difference in odds is 28 per cent for males. It increases with geographical concentration and is in municipalities with at least 85 per cent Swedish-speakers as much as 57 per cent. For females, the pattern is not equally clear. In the most Finnish-dominated areas, the language-group difference in odds for unemployment is 32 per cent, but it does not increase to more than 46 per cent in municipalities with at least 85 per cent Swedish-speakers. 5 It may also be pointed out that we have analysed whether the depicted language-group difference in unemployment propensity differ between cross section dates. Since Swedish-speakers have a more favourable labour market position than Finnish-speakers, it may be argued that they also are less sensitive to changes in the business cycle. We did not, however, find any clear indications that this would be the case. There are although some, but not very convincing, evidence in favour of a time trend. For males, the impact of language-group structure on the language-group difference in unemployment propensity seems to have become more emphasised over time. For females, similar effects can be found only with respect to municipalities with at least 85 per cent Swedish-speakers. In order to study if the similar pattern is the case when restricting the analysis to those previously employed (five years earlier), and to study the contribution of industry in terms of an individual-level variable, we have analysed the longitudinal data in a similar manner as the cross section data. The estimation results can be seen in Table 6 for males and in Table 7 for females. Despite that the estimation results from the two data sets cannot be considered as directly comparable, it is obvious that, for both males and females, the previous conclusions are confirmed. The language-group difference in unemployment propensity is higher in regions with a higher proportion of Swedish-speakers, but the pattern is not perfectly monotonous. Gender-differences in this respect are now smaller, which is a consequence of restricting the data to previously employed (the impact of genderdifferences in underlying factors, such as for instance labour force participation, should be reduced). Previous industry is important in terms of explaining overall unemployment propensity, but to a quite limited extent in terms of explaining the language-group difference in unemployment propensity. The similar arguments are the case for the family situation variable. 5 Replacing the categorical language-group structure variable in Model 5 with a continuous one results in the estimate for the interaction term Swedish-speaker % Swedish-speakers for males. This means that the language-group difference in odds for unemployment is 12.5 per cent higher (1-exp[ ]=0.125) if comparing a municipality with, say, 5 per cent Swedish-speakers with a municipality with, say, 90 per cent Swedish-speakers. For females, the corresponding estimate is

14 14 Can the low unemployment rate of Swedish-speakers in Finland Table 6. Logistic regression results for unemployment, males in the longitudinal data (n=256,460) Model 1 Model 2 Model 3 Native language, Finnish Swedish (0.028) (0.028) (0.029) % Swedish-speakers, < (0.030) (0.030) (0.031) (0.020) (0.020) (0.021) (0.057) (0.057) (0.058) (0.106) (0.107) (0.109) Interaction Swe % Swe (0.072) (0.073) (0.074) Swe % Swe (0.043) (0.043) (0.044) Swe % Swe (0.079) (0.079) (0.081) Swe. 85- % Swe (0.119) (0.120) (0.122) Previous industry, Agriculture (0.049) (0.050) Manufacturing + Construction (0.016) (0.017) Trade, hotels, restaurants (0.019) (0.020) Transport, communications (0.024) (0.024) Financial intermediation (0.022) (0.023) Public and other services 0 0 Family situation, Married with children 0 Married without children (0.020) Partner, with children (0.027) Partner, without children (0.024) Sole supporter (0.038) Single (0.016) Living with senior parent (0.033) Constant (0.017) (0.021) (0.024) Log likelihood -98, , , Cox & Snell R Square Nagelkerke R Square The numbers outlined refer to unexponentiated coefficients. Standard errors are in parenthesis. The variables age and education are also included in each model, but the estimated effects are not displayed here. The reference categories used for these variables are years and lower vocational education. It may be noted that the results indicate that people having been employed in public services are least likely to be unemployed five years later, in comparison with those having experience from other sectors. People who are married and have children are also, in comparison with those in other family situations, less likely to become unemployed. The estimation results although point out that, when having controlled for the impact of age, education, previous industry and family situation, the language-group difference in the odds for being unemployed is 14 per cent higher for Swedish-speaking males than Fin-

15 Can the low unemployment rate of Swedish-speakers in Finland 15 nish-speaking males, in regions with at most 15 per cent Swedish-speakers. In regions with at least 85 per cent Swedish-speakers, the difference in odds is 57 per cent. For females, the corresponding differences in odds are 17 per cent and 48 per cent. Table 7. Logistic regression results for unemployment, females in the longitudinal data (n=218,016) Model 1 Model 2 Model 3 Native language, Finnish Swedish (0.034) (0.034) (0.034) % Swedish-speakers, < (0.037) (0.038) (0.038) (0.022) (0.022) (0.022) (0.062) (0.063) (0.063) (0.096) (0.097) (0.098) Interaction Swe % Swe (0.091) (0.092) (0.093) Swe % Swe (0.050) (0.050) (0.050) Swe % Swe (0.088) (0.089) (0.090) Swe. 85- % Swe (0.113) (0.114) (0.115) Previous industry, Agriculture (0.055) (0.055) Manufacturing + Construction (0.020) (0.020) Trade, hotels, restaurants (0.018) (0.019) Transport, communications (0.033) (0.033) Financial intermediation (0.022) (0.022) Public and other services 0 0 Family situation, Married with children 0 Married without children (0.022) Partner, with children (0.037) Partner, without children (0.031) Sole supporter (0.023) Single (0.020) Living with senior parent (0.138) Constant (0.020) (0.023) (0.025) Log likelihood -74, , , Cox & Snell R Square Nagelkerke R Square The numbers outlined refer to unexponentiated coefficients. Standard errors are in parenthesis. The variables age and education are also included in each model, but the estimated effects are not displayed here. The reference categories used for these variables are years and lower vocational education.

16 16 Can the low unemployment rate of Swedish-speakers in Finland 5. Conclusions It has previously been shown that the difference in unemployment rates between Swedish-speakers and Finnish-speakers in Finland cannot be explained by individual-level factors such as age, gender, education and place of residence. In this paper we have investigated whether structural determinants, specifically industrial structure and language-group structure, have any effect on the unemployment gap. Industry turns out to be an important determinant of overall unemployment propensity, but it cannot explain the language-group difference in unemployment rates. In line with much of the previous research on ethnic enclaves, our results suggest that a population group benefits in the labour market from constituting the local majority. The language-group difference in unemployment propensity is higher in municipalities with a higher proportion of Swedish-speakers. Specifically noteworthy is the difference between municipalities with a very low proportion of Swedish-speakers and those with a very high proportion of Swedish-speakers. Our results thus correspond with traditional views on majority-minority relationships. A major caveat in the present case is that there is an overall language-group difference, which is independent of the proportion of Swedish-speakers in the municipality. There consequently remains a difference in unemployment propensity, even when having accounted for language-group structure. We have also tried to include individuals family situation into the estimations. This factor turned out to be important for overall unemployment propensity, but it could not explain the language-group difference. The data having been used for the study of the unemployment gap between Swedishspeakers and Finnish-speakers in Finland, and possible reasons behind it, are not ideal in all respects. Based on our analyses and findings, we are convinced that there is a difference between the language-groups, which cannot be fully eliminated using any existing data. The reason for this is that we believe that the explanations should be sought among latent factors, such as language proficiency and social networks. Empirical studies based on even more detailed and designed register data may certainly improve the analysis to some extent. However, in order to fully understand the mechanisms, one would have to collect other types of data.

17 Can the low unemployment rate of Swedish-speakers in Finland 17 References Aldrich, H., Cater, J., Jones, T., McEvoy, D. & Velleman, P. (1985). Ethnic Residential Concentration and the Protected Market Hypothesis. Social Forces, 63, Becker, G. (1971). The Economics of Discrimination. University of Chicago Press, Chicago. Borjas, G. (1985). Assimilation Changes in Cohort Quality and the Earnings of Immigrants. Journal of Labor Economics, 3, Borjas, G. (1987). Self-selection and the Earnings of Immigrants. American Economic Review, 77, Borjas, G. (1992). Ethnic Capital and Intergenerational Mobility. Quarterly Journal of Economics, 107, Borjas, G. (1993). The Intergenerational Mobility of Immigrants. Journal of Labor Economics, 11, Borjas, G. (1994). Immigrant Skills and Ethnic Spillovers. Journal of Population Economics, 7, Borjas, G. (1995). Ethnicity, Neighborhoods and Human-capital Externalities. American Economic Review, 85, Brenner, R. and Kiefer, N. M. (1981). The Economics of the Diaspora: Discrimination and Occupational Structure. Economic Development and Cultural Change, 29, Carliner, G. (1995). The Language Ability of US Immigrants: Assimilation and Cohort Effects. NBER Working Paper, No National Bureau of Economic Research, Cambridge. Chiswick, B. R. (1977). Sons of Immigrants: Are They at an Earnings Disadvantage? American Economic Review, 67, Chiswick, B. R. (1978). The Effect of Americanization on the Earnings of Foreign-born Men. Journal of Political Economy, 86, Chiswick, B. R. (1983a). An Analysis of the Earnings and Employment of Asian- American Men. Journal of Labor Economics, 1, Chiswick, B. R. (1983b). The Earnings and Human Capital of American Jews. Journal of Human Resources, 18, Chiswick, B. R. & Miller, P. W. (2002). Immigrant Earnings: Language Skills, Linguistic Concentrations and the Business Cycle. Journal of Population Economics, 15, Clark, K. & Drinkwater, S. (2002). Enclaves, Neighbourhood Effects and Employment Outcomes: Ethnic Minorities in England and Wales. Journal of Population Economics, 15, Drinkwater, S. J. & O Leary, N. C. (1997). Unemployment in Wales: Does Language Matter? Regional Studies, 31, Dustmann, C. & van Soest, A. (2001). Language Fluency and Earnings: Estimation with Misclassified Language Indicators. Review of Economics and Statistics, 83,

18 18 Can the low unemployment rate of Swedish-speakers in Finland Finnäs, F. (1997). Social Integration, Heterogeneity, and Divorce: The Case of the Swedish-speaking Population in Finland. Acta Sociologica, 40, Finnäs, F. (2000). Tvåspråkiga familjer i statistikens ljus. Forskningsrapport, No. 37. Institutet för Finlandssvensk Samhällsforskning, Vasa. Grenier, G. (1987). Earnings by Language Group in Quebec in 1980 and Emigration from Quebec between 1976 and Canadian Journal of Economics, 20, Hayfron, J. E. (2001). Language Training, Language Proficiency and Earnings of Immigrants in Norway. Applied Economics, 33, Holzer, H. J. & Ihlandfeldt, K. R. (1998). Customer Discrimination and Employment Outcomes for Minority Workers. Quarterly Journal of Economics, 113, Lazear, E. (1999). Culture and Language. Journal of Political Economy, 107, S95-S126. McManus, W. S. (1990). Labor Market Effects of Enclaves: Hispanic Men in the United States. Journal of Human Resources, 25, Montgomery, J. D. (1992). Job Search and Network Composition: Implications of the Strength-of-Weak-Ties Hypothesis. American Sociological Review, 57, O Leary, R. & Finnäs, F. (2002). Education, Social Integration and Minority-Majority Group Intermarriage. Sociology, 36, Saarela, J. & Finnäs, F. (2002a). Unemployment and Native Language: The Finnish Case. Journal of Socio-Economics, forthcoming. Saarela, J. & Finnäs, F. (2002b). Ethnicity and Unemployment in Finland. Ethnic Studies Review, 25, Shackett, J. R. & Trapani, J. M. (1987). Earnings Differentials and Market Structure. Journal of Human Resources, 22, Shields, M. A. & Wheatley Price, S. (2002). The English Language Fluency and Occupational Success of Ethnic Minority Immigrant Men Living in English Metropolitan Areas. Journal of Population Economics, 15, Sowell, T. (1981). Markets and Minorities. Basil Blackwell, Oxford.