The effect of education and cognitive skills on earnings

View the most recent version.

Results without cognitive skill variables

In this section, we use our sample of paid workers to estimate earnings regressions with and without controlling for individual skills. The dependent variable is the log of weekly earnings. As a first step, we estimate a specification that includes a quadratic in experience, the education dummy variables specified earlier, a dummy for immigrant status, a quadratic in years since entering Canada for immigrants, and a dummy for first language other than English or French. This specification is similar to immigrant – Canadian born earnings equations estimated with cross-sectional data that have been reported in previous studies.

The first column in Tables 5.1 (males) and 5.2 (females) presents these results. They reflect commonly observed patterns. In particular, male returns to experience are approximately 7% per year just after leaving school but decline to zero by 29 years later. As is typically found to be the case, female returns to experience are lower, 5.4% per year early in the career and also declining to zero by 25-30 years later. There are also substantial returns to education that are on the order of those found in earlier studies, with women experiencing much higher returns to schooling than men. Male immigrants receive weekly earnings that are over 50% less than earnings of Canadian-born workers with the same level of total experience and education. For female immigrants the magnitude of this negative entry effect is somewhat lower, but the gap is still a substantial one – approximately 44 percent. Immigrant earnings then rise at rates of approximately 2.5% (males) and 2.8% (females) more per year compared to similar Canadian born workers in the years just after the immigrant enters Canada. As indicated by the negative coefficients on the years-since-migration (YSM) squared variables, this rate of catch-up to the Canadian-born diminishes over time. If male and female immigrant earnings actually follow these "years since migration" profiles, then their earnings would equal those of a comparable Canadian-born worker after approximately 28 years in Canada. This, however, is a big "if". As Borjas (1985) points out, if immigrants arriving in different years (i.e., in different cohorts) face different entry earnings and/or years since migration earnings profiles, then a cross-sectional years since migration profile will represent a combination of actual profiles and the effects of shifts across cohorts. Thus, the cross-sectional profile is not necessarily the relevant earnings assimilation profile for any set of immigrants. With only a single cross-section of IALSS data, there is no way to address this problem. The immigrant dummy variable and years since migration profile summarize a combination of cohort effects and assimilation profiles rather than a profile that bears behavioural interpretation. Since our focus is on effects of cognitive skills rather than cohort patterns, this is not a central concern. It is only important that we control for the combination of cohort and assimilation effects, not that we can separately identify them.¹

The specification in column 1 imposes equal returns to education and experience for immigrants and the Canadian born but allows immigrants to have separate entry earnings and an earnings progression with years since arrival. However, the latter YSM effects can be difficult to interpret even in the absence of the cohort effect complication just described. For individuals arriving in Canada after they have completed their education, YSM corresponds to experience in the Canadian labour market. For individuals completing their education in Canada, YSM will equal years of experience in the Canadian labour market plus the number of years between arrival and entry into the labour market. Since the latter years may include time when the immigrant is quite young, their impact on earnings is likely quite different from that of labour market experience. For that reason, we implement an adjusted specification (reported in column 2) that allows the immigrant entry effects and Canadian experience effects to differ between immigrants who arrive after completing their education and immigrants who obtain some or all of their education in Canada. Differences between these two groups of immigrants in the coefficients on Canadian experience variables could represent some combination of differential returns to experience and differential cohort effects.

The adjusted specification in column 2 includes both the separate immigrant experience variables described above and three dummy variables corresponding to immigrants whose source country was either: 1) the United States or United Kingdom; 2) any continental European country; or 3) Asia. We include these variables because previous studies have placed a great deal of emphasis on region of origin effects in explaining immigrant earnings patterns (e.g., Baker and Benjamin, 1994). In interpreting the estimates reported in column 2 note that the various experience coefficients are reported so that they can be read directly rather than as comparisons to, say, the Canadian experience variables.

Although the results for males and females share many common features, there are also some noteworthy gender differences. We therefore discuss the male results (Table 5.1) and female results (Table 5.2) separately. The estimated coefficients relating to the Canadian experience of male immigrants who obtained some of their education in Canada and the overall male experience coefficient (which corresponds mainly to the experience effects for the Canadian born) are very similar in size, and tests of the hypothesis that they are equal to each other cannot be rejected at conventional significance levels. In contrast, male immigrants without Canadian education receive significantly greater returns to Canadian work experience. The intercept coefficients for the two groups of immigrants are both negative and significantly different from zero. Nonetheless, the implication from the coefficients is that immigrants who complete their education abroad have earnings that are almost 65% lower than comparable Canadian-born workers, whereas those with some Canadian education receive earnings that are about 16% lower than otherwise comparable native-born Canadians. These estimates apply to the base category, those whose first language spoken was French or English and who are not from the United States, the United Kingdom, Europe, or Asia. For those whose first language was other than English or French, average weekly earnings are another 3% to 5% lower, although this effect is not precisely estimated and is not significantly different from zero. Finally, the source region coefficients suggest that immigrants from continental Europe have earnings that are over 20% higher than those of other immigrants. Immigrants from the United States/United Kingdom also receive higher earnings than the base group, although these estimated effects are smaller than for Europe (about 12%) and are not precisely estimated. Immigrants from Asia have earnings that are slightly lower than the base group, although this effect is also not statistically significant.

Table 5.1 reports the results for females. As was the case for men, the negative entry effect experienced by foreign-educated immigrants is much larger than that for immigrants with Canadian education. However, both entry effects are smaller (in absolute value) than the corresponding estimates for male immigrants, and the coefficient associated with Canadian-educated female immigrants is not significantly different from zero. Another feature common to both male and female immigrants is the result that foreign-educated immigrants experience substantially higher returns to Canadian experience than do native-born Canadians. However, in contrast to the results for men, female immigrants with Canadian education also obtain higher returns to Canadian experience than the Canadian born, albeit lower than those obtained by their foreign-educated counterparts. Finally, although the estimated consequences of having a first language other than English or French are similar to those for men, the earnings differences associated with alternative source countries are very different. In particular, after controlling for other factors, female immigrants from the United States/United Kingdom and Europe do not experience higher earnings than those from other source regions, in contrast to the situation for male immigrants.

The adjusted basic specification is still, potentially, too restrictive. In particular, it restricts the returns to foreign experience (in terms of earnings in Canada) to be the same as returns to Canadian experience for the native born. The specification in column 3 of Table 5.2 permits a separate return to foreign experience. This is important because Friedberg (2000) finds, using Israeli data, that negative immigrant entry earnings effects can be completely explained by a lower return to foreign experience than native experience. For immigrants from some countries, she found that foreign experience was worth zero in the Israeli labour market. These results are replicated for Canada by Alboim et al (2003) and Ferrer, Green and Riddell (2006). Green and Worswick (2002) study this further and show that this is a recent phenomenon for Canada since immigrant cohorts in the early 1980s earned returns on foreign experience that were similar to the returns the Canadian born earned for Canadian experience. Similar to results in those papers, when we introduce foreign experience variables in column 3, the immigrant entry effect coefficients are no longer significantly different from zero; indeed, three of the four estimated coefficients are positive. At the same time, for both men and women the returns to Canadian experience for the two immigrant groups are not significantly different from those for the Canadian born. Finally, note that introducing the foreign experience effect does not change the returns to education, language impacts, and country of origin effects.

Among both male and female immigrants the return to foreign experience itself is essentially zero. It is this low rate of return on foreign experience that is the source of the negative immigrant entry effects in columns 1 and 2 of the table. Comparing immigrant earnings to those of Canadian-born workers with the same total number of years of experience shows that immigrant earnings are significantly lower. This occurs because the immigrants are obtaining zero returns to some of those years of experience. Once we control for foreign experience, we are effectively comparing immigrants to Canadian-born workers with the same number of years of Canadian experience, and it turns out that immigrant and Canadian-born workers have earnings that are much more similar when compared on that basis. This does not negate the fact that immigrants have lower earnings. However, it does help us understand that a major source of those lower earnings is an inability to transfer human capital acquired in a foreign labour market to Canada. It is worth noting, as well, that foreign experience does not suffer from the same interpretation difficulties as Canadian experience for immigrants. That is, there is no cohort dimension to the number of years an immigrant worked before arriving. Immigrants arriving in recent cohorts and cohorts from decades ago could all have the same distribution of foreign experience before arriving. The same is not true of Canadian experience: those arriving in earlier cohorts necessarily have more. This means that we can give the coefficient on foreign experience a standard human capital acquisition interpretation much as we have given to Canadian experience.²

Column 4 of Table 5.2 contains our preferred specification which we reach by first allowing a complete set of interactions among all immigrant, experience and education variables and then eliminating sets of interactions where testing indicates it is appropriate. Thus, for example, we allowed for different returns to education for immigrants who obtained some education in Canada. For both males and females we could not reject the restriction that the differences between these returns and those for the Canadian born were zero at any conventional significance level. However, we do find that returns to education are significantly lower for male and female immigrants without Canadian education. We also allowed for the possibility that each type of experience (whether foreign or Canadian-acquired) might interact with each type of education. We do find evidence of significant interactions of Canadian experience with education for the Canadian born and immigrants who obtained some education after arrival. These interaction coefficients are negative for both men and women, and increase (in absolute value) with educational attainment. Thus among the Canadian born and immigrants with Canadian education the positive impact of experience on earnings diminishes as educational attainment rises. However, there is no evidence of similar interactions between experience and education for immigrants educated before arrival.

To aid in interpretation of the results in column 4 of Table 5.2, we present in Table 5.3 fitted average earnings for a set of specific cases characterized by differing levels of education and experience. To generate the entries in this table, we formed fitted average log earnings values for a base case person – a Canadian-born worker who has not graduated from high school and has no Canadian experience. We also formed average log earnings for Canadian-born and immigrant workers with differing levels of Canadian and foreign experience and education. For the immigrants, we formed the fitted averages such that they are relevant for an individual who completed schooling outside Canada, who is not from the United States, the United Kingdom, Europe or Asia, and whose first language is English or French. The various fitted earnings are differenced relative to those of the base case Canadian-born individual.

An examination of the table entries corresponding to Canadian-born and foreign-educated immigrants who have not graduated from high school (the 1^st and 2^nd rows in the first and fourth columns, respectively) indicates that low educated immigrants earn considerably more than similarly educated Canadian-born workers when they first enter the Canadian labour market. By moving along the first and second rows, we can see the effects of increasing Canadian experience for low educated workers. The larger increase as we move along the first row rather than the second arises because foreign-educated immigrants receive a lower return to Canadian experience than do native-born Canadians.

Thus, the immigrant advantage right out of school is whittled away by the fact that the Canadian born get a higher return to Canadian experience than do immigrants.³

Moving across rows 1, 4 and 7 shows the effects of greater Canadian experience for Canadian-born workers with different levels of education. Earnings increase with experience for all three education categories, but the magnitude of the increase is smaller among well educated workers than among the poorly educated.

It is also instructive to move down each column to see the impacts of increased educational attainment for a given level of Canadian experience. Comparing rows 1, 4 and 7 to rows 2, 5 and 8 within each column illustrates the higher estimated returns to education for Canadian-born compared to foreign-educated immigrant workers.

Finally, a noteworthy feature of the immigrant – Canadian-born differences is that the patterns are very similar for men and women.

Results with cognitive skill variables

In Tables 5.4 and 5.5, we use the preferred specification from Tables 5.1 and 5.2 but include the average skill score. A comparison of the first column in Tables 5.4 (males) and 5.5 (females), where we simply add the skill variable without any interactions, and the fourth column in Tables 5.1 and 5.2 respectively reveals the direct impact of cognitive skills and their indirect impacts on other returns. The returns to skills are substantial, with a 100-point increase in cognitive skills raising earnings by almost 30 percent.⁴ The impact of skills on earnings is remarkably similar for men and women. As in Green and Riddell (2003), there is little, if any, change in the experience effects or experience interactions when we control for skills. However, estimated returns to education for Canadian-born and Canadian-educated immigrants decline to a significant extent, indicating that an important component of conventional estimates of the return to schooling arises from the impact of education on skills and the value placed on skills in the labour market.⁵ With the inclusion of controls for cognitive skills, estimated returns to foreign-educated immigrants decline even more than was the case for the Canadian born. Indeed, for men returns to education fall by about 50%; for women the decline is even greater, and, after controlling for skills, the remaining returns to education are no longer significantly different from zero. Thus, cognitive skills constitute a significant amount of what foreign education seems to deliver – at least in terms of the skills that are valued by Canadian employers.

As discussed earlier, a major question of interest is whether returns to skills are lower for immigrants. To investigate this issue we report in column 2 estimates based on a specification that allows the returns to skills to differ between immigrants and native-born Canadians, but does not allow interactions between skills and human capital inputs like education and experience. As discussed previously, in the absence of such interaction effects differences in the coefficient on the average skill measure between immigrants and the Canadian born can be interpreted as a clear measure of discrimination. The estimates in column 2 provide no evidence of discrimination in the sense of immigrants receiving a lower return to cognitive skills. Indeed, male immigrants receive a rate of return that is about 50% greater than that experienced by Canadian-born men (earnings gains for immigrants of 37% associated with a 100 point increase in cognitive skills, versus 24% for native-born Canadians) while female immigrants receive returns equal to those of native-born women – earnings gains of approximately 28%.

In column 3 we report a more general specification that allows returns to skills to differ across the three groups, again estimated without interactions with skills. Among males, foreign-educated immigrants receive the largest returns to skills, followed by immigrants with Canadian education, and native-born Canadians receive the lowest (although still substantial) returns. The differences in returns between Canadian-born and foreign-educated immigrants are statistically significant, but the pair-wise differences between the other two groups (native-born Canadians versus Canadian-educated immigrants, Canadian-educated versus foreign-educated immigrants) are not statistically significant at conventional levels. A test of the hypothesis that all three coefficients equal each other cannot be rejected at the 10% level. Among females, foreign-educated immigrants also experience the greatest returns, followed by native-born Canadians, with Canadian-educated immigrants receiving the lowest returns. The three coefficients are also not significantly different from each other at the 10% level. However, in a pair-wise comparison of Canadian and foreign-educated immigrants, we can reject equality of earnings impacts at the 10% level.

Once we allow the impact of cognitive skills on earnings to differ across the three groups, returns to education decline markedly for foreign-educated immigrants, and remain much smaller than for the Canadian born. Indeed, for both male and female foreign-educated immigrants the coefficients on educational attainment are no longer significantly different from zero. Immigrants who finished their education prior to arrival in Canada receive substantially greater returns to skills than do native-born Canadians but lower returns to formal education after we control for skills. Within our analytical framework, the implication is that education acquired abroad produces cognitive skills such as literacy, numeracy and problem solving skills (since the educational attainment coefficients change substantially with the introduction of the skill variable) but does not produce other skills that are valued in the Canadian labour market (since the educational attainment coefficients are not significantly different from zero once we control for skills).

In summary, when we allow the impacts of skills on earnings to differ across the three groups, we find no evidence that the returns to skills for native-born Canadians exceed those for Canadian-educated or foreign-educated immigrants. Indeed, among foreign-educated immigrant men the earnings gains associated with additional skills are significantly greater than those for native-born Canadians, while those for Canadian-educated immigrants are not significantly different from those for natives. Among women the skills-related earnings gains received by the Canadian born are not significantly different from those received by Canadian-educated or foreign-educated immigrants. Thus these estimates provide no evidence of discrimination in the sense of employers paying immigrants less for the same skills as Canadian-born workers. It is worth emphasizing that this result refers to what we call "usable" skills. Immigrants may have higher cognitive skill scores if tested in their native language and one could argue that those skills are being undervalued. But immigrants are receiving returns to skills as measured in English or French that are no worse than those obtained by Canadian-born workers.

The last specification in column 4 is the result of a specification search involving interactions of average skill scores with education and experience. The results indicate some interaction of skills with experience for Canadian-born men and women and for male (but not female) immigrants. The two immigrant groups are pooled in this case because the (foreign) experience-skills interaction effects of the two groups were not significantly different from each other. These interaction effects are positive, suggesting that immigrant men and Canadian-born men and women with higher skill levels receive greater returns to work experience. The interaction effects are moderate in size; for example, an extra 25 skill points raises the returns to experience by about 2 percentage points. For men the interaction is somewhat larger among immigrants than native-born Canadians, but this is not the case for women as there is no evidence of an interaction between experience (Canadian or foreign) and skills for immigrant women.

Perhaps the most interesting implication from the fourth column is the strong evidence of a positive interaction between education and skills for female immigrants educated outside Canada. The magnitude of this effect is largest at the university level, followed by the high school graduate level. Including these interaction terms results in the coefficient on the average skill score of foreign-educated female immigrants falling to zero. This implies that women with less than a high school education receive zero returns to additional skills, while women with at least a high school education experience earnings gains from a 100-point increase in skills in the order of 35% (in the case of non-university post-secondary graduates) to 71% (for university graduates). However, returns to skills do not vary with educational attainment for Canadian-born women and Canadian-educated immigrant women. The interaction terms between education and skills were also not statistically significant among men. The reasons for these differences between foreign-educated immigrants and the other two groups in the way skills and education interact in influencing earnings – and the differences in the nature of these interactions between men and women – warrant further analysis.

Finally, the dummy variable corresponding to men whose skill scores were imputed indicates that inability (or unwillingness) to complete the main test is positively associated with earnings. These men, many of whom are immigrants, earn approximately 10% more than one would anticipate given their imputed skill scores and other characteristics. This positive coefficient is consistent with results in other studies indicating the importance of immigrant enclaves in allowing immigrants to do better than expected when they do not acquire the host country language (Edin et. al. (2003)). However, the result does not hold for females, for whom there is no difference in earnings associated with not completing the main tasks.

To aid in the interpretation of the results in the fourth column of Tables 5.4 and 5.5, we repeat the exercise of forming fitted average log earnings for various types of workers but, in this case, holding the average skill score constant at 283 (the sample weighted average value for men and women). The results are contained in Table 5.6. Comparing these results to those in Table 5.3, one sees patterns similar to those when skills are not held constant.

One interesting question arising out of these estimates is the relative importance of lower immigrant skill levels in explaining immigrant/Canadian-born earnings differentials. To investigate this, we constructed a series of fitted average earnings differentials, all based on the fourth column in Tables 5.4 and 5.5. We first construct an estimate of average log earnings for immigrants and the Canadian born separately using the estimated coefficients in conjunction with the appropriate average values for the regressors.⁶ Those estimates imply an overall average immigrant earnings disadvantage of 11 log points over the Canadian born among high school educated men and an immigrant advantage of 1 log point among high school educated women. The corresponding estimates among those with university education imply an immigrant disadvantage of 22 log points for men and 19 log points for women. We next repeated this exercise but gave immigrants the same return to foreign experience as the Canadian born receive for their Canadian experience. The result is a shift from the immigrant disadvantage of 11 log points to an advantage of 46 log points for high school educated men – a net change of 57 percentage points in the earnings gap – and a shift from the immigrant advantage of 1 log point to an immigrant advantage of 13 log points for high school educated women – a net change of 12 percentage points. Among the university educated the immigrant disadvantage changes from 22 log points to an immigrant advantage of 28 log points for men – a net change of 50 percentage points in the earnings gap, and from an immigrant disadvantage of 19 log points to a disadvantage of 8 log points for women, a net change of 11 percentage points. These estimates fit with results in earlier papers, described above, indicating that lower returns to foreign experience play an important role in understanding immigrant/Canadian-born earnings differentials, especially for men. The importance of low returns to foreign experience is much more important for men than for women, but for each gender is similar across education groups.

In our next counterfactual, we set the returns to foreign experience back to their original values but gave immigrants the average skill scores observed for Canadian-born workers with the same level of education. For high school educated men, this reduces the immigrant disadvantage from the 11 log points mentioned above to an advantage of 5 log points, a net change of 16 percentage points, and among high school educated women it increases the immigrant advantage from 1 log point to 14 log points, a net change of 13 percentage points. For the university educated, it reduces the immigrant disadvantage from 22 log points to 11 log points for males and from 19 log points to 1 log point for females, net changes of 11 and 18 percentage points respectively. Again, the changes in the earnings differential are similar across the two education groups. Low skills thus appear to be an important factor for understanding male immigrant earnings differentials, though not nearly as important as low returns to foreign experience. However, for females low skills are a somewhat more important factor in explaining immigrant – Canadian-born earnings differences than are low returns to foreign experience.

---

Notes

1. See Green and Riddell (2007) for an analysis of cohort and aging effects among native-born Canadians using the 1994 IALS data and the 2003 IALSS data, and Willms and Murray (2007) for an analysis of skill loss and gain over time. In future work we plan to examine cohort effects using the 1998 OILS data and the Ontario observations in the 2003 IALSS data.
2. However, Green and Worswick (2002) point out that Canadian-born earnings can also be organized in a cohort format and that doing so provides insights into the cross-cohort patterns in immigrant cohorts. In particular, they find that approximately 60% of the cross-cohort decline in immigrant earnings in the 1980s can be attributed to general declines across cohorts of new entrants of all kinds into the Canadian labour market.
3. Note that all of these statements are based on interpreting coefficients on Canadian experience as reflecting true returns to experience rather than cohort effects.
4. However, these estimated returns to skills are lower than those obtained in our previous research for Canadian-born workers with the IALS data (Green and Riddell, 2003) and for immigrants with the OILS data (Ferrer, Green and Riddell, 2006). These differences warrant further investigation.
5. For example, the earnings gain associated with a university education (relative to less than high school) falls by 18 percentage points for men and 15 percentage points for women.
6. We constructed fitted average earnings separately for the two immigrant groups; the estimates for immigrants as a whole are weighted averages of the fitted earnings for immigrants with and without Canadian education.