In this section, we present our empirical model which extends that of Cawley et al. (). Using a similar methodology we examine the experiences of immigrants to the UK in the years 2004 and 2006, a time when the UK labor market was relatively strong and the UK experienced a large inflow of migrants.
4.1 Description of the data
The British Household Panel Survey (BHPS) is a nationally representative annual survey in the UK that began in 1991 with roughly 5,000 households that include over 9,000 adults. Several sub-samples have been added and removed from the survey over time: the European Community Household Panel (ECHP) from 1997 to 2001, the Scotland and Wales Extension from 1999 onward, and the Northern Ireland Household Panel Survey (NIHPS) from 2001 onward. In addition to these extensions, new members enter the survey when they join an original survey household by marriage, birth or cohabitation.
The BHPS allows us to identify immigrant status, obesity, and labor market outcomes. We use two questions to identify and categorize immigrants by country of origin and relocation duration: “Where were you born?” and “In what year did you first come to this country to live (even if you have spent time abroad)?” Both of these questions are only asked once; the first time an individual is interviewed if they report that they have not always lived at the same address and that their country of birth is not the UK. Given our limited sample, we form parsimonious groupings by country of origin as detailed in Table 1.
The BHPS collected information on height and weight only in two waves of the data, in 2004 and 2006. Weight and height are self reported and then used to calculate the BMI and indicator variables for the clinical weight classifications of overweight and obesity. In keeping with Cawley et al. (), we conduct our analyses on four different indicators of labor market outcomes. First, we examine whether immigrants face a wage penalty or premium for being obese or overweight by looking at labor income reported for the month prior to the interview. Monthly labor income is a derived variable (fimnl) that includes income from multiple sources including overtime and self-employment. Next we look at whether or not individuals are employed (jbft), which includes both full-time (at least 30 hours per week) or part-time work. We create an indicator of work limitations from several variables in the BHPS that change during the sample period. These include direct measures of work limitations due to health or mental health conditions, and more specific questions about difficulty performing work5. Finally, we follow [Balia and Jones (2008]) in deriving social classifications from the Registrar General’s social classification which is based on three-digit SIC codes (jbrgsc). We use this classification because it has the fewest missing observations, though the sample size for this regression is still slightly smaller than those of the other outcome measures. The social classifications are: professional and managerial, skilled including those in the military, and semi and unskilled. A dichotomous indicator of a professional or managerial occupation is our dependent variable for the regressions on white collar work.
Our sample is restricted to respondents who report BMI data in at least one of the two waves resulting in a sample of 21,083 person-year observations from 14, 408 individuals. Of these individuals, 583 are immigrants. We observe 254 of these immigrants for both waves resulting in 837 person-year observations for immigrants with BMI information. Immigrants comprise 4.4 percent of the female sample and 3.4 percent of the male sample. This sample omits 5 observations for pregnant immigrants and 204 observations for pregnant natives because of the obvious association between pregnancy and reported weight. In addition, seven observations were deleted because of unreasonably low BMIs resulting from low reported weight. Complete summary statistics by gender and immigration status pooled for person-waves are presented in Table 2.
Notable in the unadjusted summary statistics is that immigrants, both men and women, have a higher average wage and higher proportions in managerial/professional jobs than UK natives. This reflects a positive selection of immigrants to the UK consistent with the findings of Dustmann et al. () who analyzed data from the British Labour Force Survey from 1981 – 2005. The proportion of immigrants who are employed is similar to that of natives, although immigrant women are substantially more likely to report work limitations than native women. Also consistent with Dustmann, Glitz and Vogel, the majority of immigrants in our sample are from the West which includes other OECD countries. The average duration spent in the UK among immigrants in our sample is high (31.57 years and 33.99 years for women and men respectively). This long duration reflects the sample construction of the BHPS which only adds individuals to the sample if they join a household of an original sample member from 1991 or are part of one of the sub-samples which are not likely to include many immigrants. That said, we do observe a range of duration from 1 – 85 years with 10% of our immigrant observations residing in the UK for less than 8 years and 5% for less than 3 years.
Female immigrants have lower BMIs on average and a lower probability of being obese than native born women, and these differences are statistically significant at the one percent level. No other weight differences are statistically significant across immigrant status for either men or women.
The demographic variables reported in Table 2 suggest that immigrants to the UK are comparable in age and family structure compared with natives although immigrants are more likely to be married. We classify education as “college or professional” which includes teaching and nursing degrees, “vocational” which includes high school and apprenticeships, and “no education” which includes those still in school. Immigrants are overrepresented in both the highest and lowest education categories. Following Cawley et al. (), we include variables on current smoking status and whether or not the individual reports a problem with drinking in order to account for myopic time preferences6. Not surprisingly, immigrants are more likely to report being racial or ethnic minorities. We report these race categories in the summary statistics, but include region of origin rather than race in the regressions to better capture the effects of immigration7. Wheatley [Price (2001]) notes that country of origin variables may also capture variations in the quality of schooling, the transferability of skills between national labor markets, average English language ability, and systematic differences in unobserved ability, determined before migration by the prevailing characteristics of the origin country. Unfortunately, our data does not include information on English language proficiency.
4.2 Healthy immigrant hypothesis
To examine the presence of a healthy immigrant effect among immigrants to the UK, we estimate health differentials for immigrants controlling for individual characteristics as represented in the following equation:
We look at two measures of health (Hit): BMI and an indicator of obesity (BMI ≥ 30). IMM is a dichotomous variable equal to one if the respondent is an immigrant, and DUR is the number of years the respondent had been in the UK at the time of the survey (equal to zero if they are native born). Y2006 is a year fixed effect. The vector X contains the demographic variables described above plus a quadratic age term.
The results in Table 3 generally support the healthy immigrant hypothesis. The coefficient on immigrant status is negative for both BMI and the likelihood of obesity for women and men, although the effects for men are smaller and do not reach conventional levels of significance. Controlling for individual characteristics such as age and education, female immigrants have a BMI that is slightly more than 2 points lower than that of comparable natives. The lower BMI levels among female immigrants translate into obesity rates that are 10.5 percentage points lower than that of native-born women. The positive significant coefficient on the duration variable shows evidence of an assimilation effect. However, the magnitude of this coefficient is relatively small, 0.037, suggesting that while the BMIs of immigrant women begin to approach those of natives, it would take nearly 60 years (2.154/.037) for immigrant women to gain the weight associated with the higher BMI of non-immigrant women. This slow rate of assimilation with respect to weight is also reflected in a very small and insignificant coefficient on duration in the obesity regression.
There are differences in weight associated with immigrant region of origin. The groupings are more parsimonious than those presented in Table 1. The omitted category is immigrants from former British colonies and Europe. The results for immigrants from India and Pakistan are the most strongly significant and in opposite directions for men and women. Immigrant women from India and Pakistan have a BMI that is 1.08 points lower (−2.154 + 1.074) than native women in the UK8. In contrast, men from India and Pakistan are have a BMI that is 2.027 BMI points lower (−.458 + −1.569) than comparable natives. Other significant immigrant effects by region include results for Africa and South America9. We see evidence contrary to the healthy immigrant hypothesis in that male immigrants from these regions experience higher BMI and a significant 3.5 percentage point higher (−0.075 + .11) probability of obesity. Weight measures for immigrants from East/Asia are not significantly difference from the omitted category of immigrants from Europe and the former colonies.
The associations between BMI and obesity and other covariates are in line with previous studies. For example, BMI and obesity increase with age at a diminishing rate for both women and men. Marriage and children are associated with higher BMI for men, but not women. Though existing work consistently shows that BMI tends to rise after marriage, OLS models that combine the selection of thinner individuals into marriage and the impact of marriage on weight demonstrate much weaker correlations (Averett et al. ). Lower levels of education (college professional is the omitted category) are associated with higher weight while smoking and drinking are strongly associated with lower weight for both men and women.
4.3 Labor market outcomes: immigrants only
We begin our analysis of immigration, obesity and labor market outcomes by estimating models similar to those reported by Cawley et al. () on the immigrant-only sample. Our results, like those of Cawley et al. (), can only be viewed as associations since we are unable to instrument for BMI and small sample and cell sizes make individual fixed-effects models infeasible. Our empirical model takes the following form:
L represents one of four labor market outcomes previously described. We classify individuals as overweight (OW) and obese (OB) with recommended weight and underweight combined as the omitted category. We estimate log wage equations using OLS and logit models for employment, work limitations and white collar work. Estimates of equation (2) are presented in Table 4 for each of our four labor market outcomes. Marginal effects are reported for logit models of employment, work limitations and white collar work.
Despite our smaller sample size, we find more significance in the effects of overweight and obesity on labor market outcomes among an immigrant-only sample in the UK than Cawley et al. () found among immigrants to the US. However, beyond the different institutional settings between the UK and the US, our sample includes a large proportion of immigrants from developed countries including the US, Canada and Australia while their sample consists of immigrants from developing countries.
Overall, the results demonstrate a negative association between overweight and obesity and labor market outcomes for immigrants to the UK. The wage penalties range from −12.7 percent for overweight women to −17.5 percent for obese men, though the large standard errors preclude any of these estimates reaching conventional levels of significance. These models also control for length of time in the U.K. Duration has a small positive effect only on the earnings of women indicating that wages for women gradually increase with duration but at a rate of 1 percent per year.
We find a sizeable and strongly significant association between overweight and obesity and the report of health-related work limitations for immigrant women. Female immigrants who are overweight or obese are 12.3 percentage point and 18.4 percentage points more likely to report such limitations than healthy weight immigrants. For men, we find a significant difference of 11.6 percentage points for obese men only.
Finally, we find large negative associations between overweight and obesity and being in a white collar job. The associations are very similar for men and women at approximately 11 percentage points for those overweight and 17 percentage points for those obese. We do not find any significant associations between weight and being employed.
4.4 Labor market outcomes: immigrants and natives
Since our sample includes both immigrants and natives, we can extend Cawley et al. () to directly compare the immigrant and native populations with respect to the effect of weight on labor market outcomes. We do this by estimating the following equation which augments equation (2) by including an indicator of immigration status (IMM) and interaction effects between immigration and weight classifications and a control for duration in the UK for immigrants (this variable is 0 for natives):
The key coefficients of interest are ϕ and λ, the coefficients on the interactions of immigrant and weight status. Table 5 reports the results from this analysis.
The strongest finding of these regressions is a wage penalty for overweight and obese immigrant men of 25 percent and 27 percent respectively. This wage penalty for immigrant men is in contrast to the wage premium of 6–7 percent that we find for overweight and obese native born men. The finding of a wage premium for overweight and obese native men is consistent with findings from other studies that have used OLS (see Brunello and d’[Hombres 2007]; [Morris 2006], ). We find negative coefficients of 15.8 and 19 percent for overweight and obese immigrant women, but while sizeable they are not significant at conventional levels.
Turning to the probability of employment, we find that obese immigrant women experience a nearly 9 percentage point greater probability of working. This finding is marginally significant. This stands in contrast with a strongly significant finding that obese native women are associated with a nearly 5 percentage point lower probability of working. However, a joint F-test on the coefficients of obese and the interaction between obesity and immigrant does not reach conventional levels of significance (p = .405). We find a positive association between being overweight and working for native born women and men. One potential explanation for this finding is, as noted earlier, BMI may not be the best measure of adiposity since someone who is very muscular may be classified as overweight on the basis of BMI.
Work limitations are more prevalent among the obese for both native men and women. Notably, overweight immigrant women are 11.9 percentage points more likely to have work limitations all else equal, and this finding is strongly significant. Associations for overweight immigrant men and for obese immigrants of both genders are smaller and not statistically significant.
The last labor market outcome in Table 5 is whether or not the respondent is in a professional or managerial job, which in keeping with Cawley et al. (), we refer to as “white collar.” We find significant associations between overweight and obesity and white collar work for men, but not for women. Native men are less likely to be in professional or managerial jobs if they are overweight (2.3 percentage points) or obese (4.8 percentage points). As we saw in the means, immigrants overall are more likely to be in white collar occupations. Overweight and obese male immigrants, however, are less likely to be in these jobs (−10.5 and −12.1 percentage points respectively) although the association is statistically significant only for overweight immigrant men.
All of the models include indicators for the immigrant’s region of origin with the omitted category consisting of other countries in Europe, the US, Canada and Australia. As noted, a majority of the immigrants in our sample come from this omitted category, and our broad groupings were dictated by our limited sample size. Nonetheless, we find consistent negative associations for female immigrants from India and Pakistan across all of the labor market outcomes. The wage penalty for women from this region is estimated to be over 34 percent, second in magnitude only to the nearly 70 percent penalty associated with having no education10. Women from India/Pakistan are 13.2 percentage points less likely to be working, 14.7 percentage points more likely to report work limitations, and nearly 25 percentage points less likely to be in white collar jobs. We do not find a significant wage penalty for men from India/Pakistan but we do find a strong negative association with employment and a positive but not statistically significant association with white collar work11.
Briefly looking at the other covariates in the model, we find associations that are consistent with economic theory. Higher educational attainment is strongly associated with better labor market outcomes. Age is positively associated with wages, employment and white collar work, but also positively associated with the probability of reporting work limitations. We find a marriage wage penalty for women and premium for men though marriage is positively associated with working and negatively associated with work limitations for both. Children are negatively associated with wages, working and white collar work for women, but not for men. Finally, we find a strong negative association between smoking and labor market outcomes and to a lesser extent between drinking and labor market outcomes.