Migrants’ subjective well-being in Europe: does relative income matter?

ABSTRACT

This paper contributes to the growing field of inquiry that investigates migrants’ subjective well-being by analysing the role of income, relative to two reference groups: natives and other migrants. Using data collected by the European Social Survey from 2002 to 2018, we constructed two measures of economic distance to compare each migrant’s economic situation with that of natives and other migrants with similar characteristics. Our results indicate that when the disadvantage between the migrant and the reference groups becomes smaller, eventually becoming an advantage, the migrant’s life satisfaction increases. Such relationship is stronger when migrants’ income is examined relative to natives than when compared with migrants’. This suggests that upward comparison is more important than downward comparison for migrants’ subjective well-being. We also show that the relationship between relative income and subjective well-being is stronger for second-generation migrants and for those with more formal education. Finally, we show that subjective measures both at the individual (feelings about one’s own economic condition) and societal (feelings about the national socio-economic-institutional condition) levels moderate the relationship between relative income and subjective well-being.

KEYWORDS:

• Subjective well-being
• relative income
• migrants
• subjective measures

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 Austria, Belgium, Bulgaria, Croatia, Cyprus, Czech Republic, Denmark, Estonia, Finland, France, Germany, Greece, Hungary, Iceland, Ireland, Israel, Italy, Lithuania, Luxembourg, Netherlands, Norway, Poland, Portugal, United Kingdom, Russia, Serbia, Slovakia, Slovenia, Spain, Sweden, Switzerland, Ukraine, Turkey.

2 We consider migrants both first generation (born abroad) and second generation (born in country with one or both parents born abroad). Natives are those born in country with both parents born in country.

3 After McBride (2001), we used ten-years age classes (except the first and last open ones): < = 30 years old, >30 & < = 40 years old, >40 & < = 50 years old, >50 and < = 60 years old, >60 years old.

4 We coded education as a dummy: low education (lower secondary education – EISCED2 as the highest level) and medium-high education (high secondary education – EISCED 3 and higher). We used the same threshold to split the sample by level of education (section 4.3).

5 After Arpino and de Valk (2018) and Senik (2014), we used area of origin instead of country for two reasons. First, the small sample size of some nationalities in some receiving countries can lead to imprecise estimations. Second, for round 1 only the area of origin was available. A different coding of area of origin has been used to run a robustness check (see Appendix).

6 For the first three rounds of ESS, income was measured via a variable with twelve modalities with which the respondents indicate their total net income. To harmonise these data with income codified in deciles of income in subsequent rounds of ESS, we collapsed the first two modalities into one and the last two categories into one, obtaining ten modalities also for the rounds ESS1-3. The distribution of this new variable is similar to the distribution of income level in ESS4-9, further supporting our choice of variables. We conducted a robustness check to test our strategy (results reported in the appendix).

7 Squared age was included to account for a possible non-linear effect. We did not apply any upper age threshold, because we chose to consider migrants of all ages, and few elder migrants were surveyed. In addition, applying a threshold (e.g. 65 years) did not affect estimates (results available upon request).

8 We could not add variables describing type of occupation (ISCO classification, type of organisation the individual works, or has worked for) because these variables had too many missing values. This causes a reduction of the sample, especially in the models on subgroups. We ran some models using dummies for type of organisation (this variable included fewer missing values), but none of them were significant. Results are available upon request.

9 Following ONU classification, we included Mexico in Central America (https://data.un.org/en/iso/mx.html). We also included the rest of North America (USA and Canada) and Oceania in the same group, because they have similar economic conditions. In addition, low numbers of immigrants from Oceania could introduce bias into model estimates.

10 To avoid problems of collinearity, we used second generation as reference. This allowed us to exploit at the same time variation in SWB by second-generation-immigrant status (reference) and, for first generation immigrants, by area of origin. A similar approach has been used in Holland and de Valk (2013). Differences by generation are analysed in depth in section 4.2.

11 We did not add destination regions, because both the prevalence of missing values (>50%) and small sample size of migrants for each region and year could lead to biased estimates.

12 Cluster standard errors were used to correct for possible violations of independence between individuals in the same country, i.e. factors that did not vary across individuals in the same cluster but varied across clusters.

13 We added three dummies (low income: from the 1st to the 4th decile; medium income: from the 5th to the 7th decile; high income: from the 8th to the 10th decile) instead of all deciles for two reasons. First, some deciles’ variables were not significant. Second, collinearity with main explanatory variables may cause erroneous subgroup estimates.

14 To avoid drastic reduction of the sample, we used only those variables for which the number of missing values was reasonable, following Bălţătescu (2005) and partially Hendriks and Burger (2019).

15 We ran a PCA with varimax rotation to the set of original variables, obtaining a first component with an eigenvalue of 2.938, explaining around 60% of the initial variance.

16 Country-fixed variables (not reported) are all statistically significant but have different signs. In all four models, living in Switzerland, Denmark, Finland, Iceland, Luxemburg, Netherlands, Norway or Sweden is positively related to migrants’ SWB, while the relationship is negative for all other countries. This result is consistent with Kogan et al. (2018) who found that migrants were likely to be more satisfied in countries that offer more welcoming social settings.

17 In the online appendix, we have reported separate estimations by gender (Table A2). A Wald chi-square test shows that difference by gender for income relative to natives is not significant (Χ²(1) = 0.18, p = 0.667), while the difference for the relative income with migrants is significant at 5% level (Χ²(1) = 4.10, p = 0.043).

18 A Wald chi-square test confirmed that the differences between the coefficients of the variable ‘relative income with natives’ of the first and the second generation are significant at 1% in the total sample (models 1 and 2), at 0.1% in the women’s sample (models 3 and 4) and not significant in the men’s sample (models 5 and 6). The same test conducted for the variable ‘relative income with migrants’ shows that the differences between the coefficients of the first and the second generation are statistically significant at 1% for the total sample (models 7 and 8) and not significant for the two genders.

19 We used the same threshold here that was used to segment the reference groups.

20 A Wald chi-square test confirmed that the differences between the coefficients of the variable ‘relative income with natives’ of the ‘low educated’ and ‘medium-high educated’ groups are significant at 1% in the total sample (models 1 and 2) and at the 5% level in the women-only (models 3 and 4) and men-only samples (models 5 and 6). The same test conducted for the variable ‘relative income with migrants’ indicated that the differences between the coefficients of the low and medium-high educated migrants were statistically significant at 5% for the total sample (models 7 and 8) and not significant for the two genders.

Additional information

Notes on contributors

Manuela Stranges

Manuela Stranges is Assistant Professor of Demography at the University of Calabria (Italy).

Daniele Vignoli

Daniele Vignoli is Full Professor of Demography at the University of Florence (Italy).

Alessandra Venturini

Alessandra Venturini is Full Professor of Political Economy at the University of Turin (Italy).