# 6 Descriptive analysis

## Archived Content

Information identified as archived is provided for reference, research or recordkeeping purposes. It is not subject to the Government of Canada Web Standards and has not been altered or updated since it was archived. Please "contact us" to request a format other than those available.

We begin by estimating the individual component of immigrant earnings in (4). To obtain Ycjit we simply demean logYcjit within each cjt (cohort×arrival age×year) cell by regressing logYcjit on a constant. Later, additional explanatory variables will be added to this regression to determine their effect on earnings inequality and earnings instability. In what follows, computations are performed for each c and j separately so subscripts c and j are dropped to simplify the notation.

Table 1 shows the results of variance decomposition for all immigrants in each cohort and for different arrival age groups within each cohort. The between and within variances do not sum up to the total variance because the panels are unbalanced. For all cohorts, the between component is larger than the within component, although the between-within difference differs from cohort to cohort.

The first notable result sheds some light on the issue of whether earnings inequality among recent immigrants is higher than among those who arrived in Canada in the past. The between variance component computed for the first four periods after arrival (t=4) is 46% higher for the 1998-to- 2000 cohort and 28% higher for the 1995-to-1997 cohort than for the 1980-to-1982 cohort. For t=7 the comparison between 1998 to 2000 and 1980 to 1982 is not available; however, the between variance computed for the 1995-to-1997 period is 27% higher than for the 1980-to-1982 period. Finally, for t=10 the between variance is about 9% higher for the 1992-to-1994 cohort than for the 1980-to-1982 cohort. This is considerably lower than the 16% difference between these cohorts computed for t=7 and 23% difference computed for t=4. Judging by these results, more recent immigrant cohorts experienced much higher earnings inequality in the first several years after arrival than did previous cohorts; however, in the longer run the cross-cohort differences may not be as pronounced, and all immigrant cohorts eventually reach comparable levels of earnings inequality.

The within variance component appears to follow more pro-cyclical paths. For instance, those who arrived during the 1989-to-1991 period and entered the labour market in the midst of the 1990-to-1993 recession, have the highest four-year within variance. Not surprisingly, however, computed for seven- and 10-year periods, the within variances for this cohort are almost the same as the within variances of the two previous cohorts. Generally, those who entered the labour market in the mid-1980s have substantially smaller σ2within,t=4 than those who entered the labour market later. As may be expected, the cross-cohort differences are smaller for t=7 and t=10, although even in these cases the within variance computed for the 1980-to-1982 cohort is considerably lower than for any other cohort.

Breaking these trends down by arrival age groups, we find that, while for the earlier cohorts the between variance is considerably higher for older immigrants than for younger ones, there is little cross-age difference for the recent cohorts. The cross-age equalization appears to be mostly due to the rising between variance among younger immigrants. The σ2between,t=4 is 47% higher for the 1980-to-1982 cohort than for the 1998-to-2000 cohort (0.627 compared with 0.427) in the 25-to- 29 category, 68% higher in the 30-to-34 category (0.692 compared with 0.413) and only 16% higher in the 45-to-49 age category (0.635 compared with 0.549). Therefore, it appears that cross-cohort differences in between variance are mostly driven by the rising inequality among immigrants who arrived at younger ages. The between variances computed for t=7 and t=10 seem to follow similar patters.

Table 2 shows the effect of education, language ability and origin on earnings inequality and earnings instability. As mentioned above, this is achieved by adding each of these explanatory variables into the first-stage regression and then re-computing variance decomposition. Although controlling for education, language and origin has a clear impact on inequality, it has very little effect on instability. This is not surprising, considering the differences in the sources for inequality and instability. Education, language and cultural background are skill-related characteristics that are absorbed into the persistent component of earnings variability and have a long-term effect.

Controlling for the birthplace brings about the largest reduction in σ2between,t=4 for all arrival cohorts. The relative effect of education and language ability, on the other hand, is different for different cohorts. For most cohorts, the effects of foreign education and language ability are similar for t=4; the impact of foreign education is somewhat weaker for the 1980-to-1982 and 1986-to-1988 cohorts but stronger for the 1995-to-1997 cohort. On the other hand, the effect of a foreign education seems to grow when we consider σ2between,t=10 .

For instance, for the 1989-to- 1991 and 1992-to-1994 cohorts, σ2between,t=4 is about the same for both categories, while σ2between,t=10 is smaller for foreign education.

These results provide an interesting insight into the role of foreign education in the economic progress of immigrants. Although, shortly after arrival, a foreign education may have less impact on labour market prospects of immigrants than more easily recognizable skills, such as the ability to speak English or French, in the longer run, immigrants to Canada with higher educational attainment have a greater ability to adjust to the demands of the Canadian labour market.

It is also worth pointing out that even after controlling for all three factors, a large part of immigrant earnings inequality remains unexplained. For t=4, controlling for all three variables reduces the between variance by 14% to 26 %, depending on a cohort; for t=10, the reduction is from 20% to 35%. Furthermore, the combined effects of language, education and birthplace appear to be stronger for earlier cohorts. For the 1980-to-1982 cohort, for instance, controlling for all three variables reduces σ2between,t=4 by 26% (35% for t=10) compared with 17% for the 1992-to-1994 cohort (24% for t=10) and 15 % for the 1998-to-2000 cohort.

Differences in education, the ability to speak one of the official languages or ethnic background can be broadly viewed as differences in cohorts' human capital, so the impact of these variables should be absorbed in the between variance component. The within variance component, on the other hand, measures the 'unexplained' earnings variation, which is not skill related. It may be related, among other things, to local labour market fluctuations or seasonal oscillations in the demand for goods and services. Although the within variance may be affected indirectly by the changes in the cohort skill composition, controlling for education, language and ethnic background should not have any direct effect on the within variance. Indeed, Table 2 shows that the between variance component absorbs virtually all the effect of controlling for extra variables in the first-stage regressions. This result holds both for t=4 and t=10.

In Table 3 the results are broken down by arrival-age groups. The birthplace has the strongest impact on σ2between,t=4 for all cohorts and arrival-age groups. The relative effects of education and language, however, vary considerably with cohorts, age groups and the number of years for which variance is computed. For t=10 , the impact of foreign education is stronger than the impact of language for immigrants in all age categories for the 1989-to-1991 and 1992-to-1994 cohorts. It is also stronger for immigrants in the 30-to-34 arrival-age groups for all cohorts. Generally, it seems that compared with language ability, the longer-term effect of foreign education is somewhat stronger for more recent cohorts in all age groups.

In sum, the descriptive results seem to indicate the following: earnings inequality accounts for a larger portion of the immigrant earnings dispersion than earnings instability; earnings inequality is higher for more recent cohorts than for those who arrived in the early 1980s; earnings instability is pro-cyclical—immigrants who arrived just before or during the recession in the early 1990s have experienced higher levels of earnings instability than earlier cohorts; the region of birth has the strongest impact on earnings inequality, while the impacts of foreign education and the ability to speak an official language vary from cohort to cohort and across arrival age groups; although controlling for education, language ability and origin reduces earnings inequality, it has very little effect on earnings instability; and, even after controlling for education, language and birthplace, a large portion of immigrant earnings inequality remains unexplained.

In the next section, we will examine cohorts' earnings inequality and earnings instability dynamics using a more flexible econometrics model.