Notes

^1.

The ACS asks the same questions as the decennial census every year to a representative sub-sample of the population.

^2.

To reduce data processing time, a 10% random sample of the domestic-born sample is used in all years in both countries, with the exception of the 2005 ACS data for the U.S.

^3.

This exclusion is to make the domestic-born more comparable between Canada and the U.S. Aboriginal peoples tend to have lower earnings than the remaining domestic-born population, and the population shares of Aboriginal peoples are different in the two countries. About 5.4% of Canada's population reported Aboriginal ancestries in the 2006 Census, while about 0.8% of the U.S. population reported their ancestry as American Indian or Alaska Native in the 2005-2007 ACS.

^4.

The question on the 1991, 2001, and 2006 Canadian censuses was "In what year did this person first become a landed immigrant?" Additional instructions were included for providing the first time that individuals obtained landed-immigrant status. In 1981, the question was "In what year did you first immigrate to Canada?" In the U.S., the question was "When did this person come to the United States to stay?" in 1980 and 1990, and "When did this person come to live in the United States?" on the 2000 U.S. census and the 2005 ACS, with additional instructions to list the latest year the person came to live/stay or the year the person came to stay permanently.

^5.

Several studies in the U.S. have shown that a large proportion of immigrants to the U.S. have lived there before obtaining legal immigration status, that many have arrived and left the U.S. multiple times, and that the year of arrival reported in the census does not systematically correspond to either the first or the last time a person entered the U.S. (see for example Lubotsky (2007), Ellis and Wright (1998), Redstone and Massey (2004), and Massey and Malone (2002)).

^6.

Estimates from the combined Longitudinal Administrative Databank (LAD) and the Longitudinal Immigration Database (IMDB) show that about 21% of all immigrants who landed in 1997 had worked or lived in Canada prior to their landing. The corresponding estimate is 25% for the 1998 landing cohort, 20% for the 1999 cohort, and 22% for the 2000 cohort.

^7.

Lubotsky (2007) compares the difference in log annual earnings between immigrants and U.S.-born individuals of all education levels when the immigrants' year of arrival is measured in different ways. Using a single longitudinal data set, he defines year of arrival by (1) answers to a census-type question, (2) the first year the immigrants' earnings are observed in the longitudinal Social Security data, and (3) the earlier of the two. He finds that when measuring year of arrival by a census-type question, the disadvantage of new immigrants is overestimated by 3.5 to 4.7 percentage points for the 1975-79 arrival cohort, but underestimated by 4.0 to 7.1 percentage points for the 1985-89 arrival cohort.

^8.

The sample of recent immigrants in the U.S. includes individuals who arrived in the following years: 1975 to 1980, 1985 to 1990, 1995 to 2000, and 2000 to 2005. The sample of recent immigrants in Canada includes the following cohorts: 1976 to 1980, 1986 to 1990, 1996 to 2000, and 2001 to 2005. It is typical in the Canadian immigration literature to exclude from analysis immigrants who arrived in the census (or other survey) year and immigrants who arrived in the year before, ensuring that all individuals in the sample were in the country for the entire reference year (the calendar year preceding the census year). This is particularly important when the outcome of interest is annual earnings, but perhaps less relevant in case of weekly or hourly earnings. Charts 1, 2, 3 and 4 in the Appendix show distributions of weekly wages for immigrants who arrived in Canada one, two, and three years prior to the census, respectively. The distributions look fairly similar in the Canadian data, with an obvious improvement in weekly wages for workers who have been in the country longer. Therefore immigrants who arrived one year before the census are retained in the sample in order to keep the sample of recent immigrants as comparable to that of the U.S. sample as possible.

^9.

There is a major change in how education is measured in the 2006 Census of population of Canada, compared to earlier censuses. This study uses the derived variable for the highest degree, certificate or diploma, available consistently until the 2001 Census and matches it with the corresponding variable on the 2006 Census. As a robustness check for the comparability of the four education groups between the 2006 Census and earlier censuses, a synthetic cohort of domestic-born men was formed between 2001 and 2006, and their distribution was calculated across the individual education categories in the derived education variables for each of the two censuses and the broader education categories created for this study. The cohort is 30 to 39 years old in the 2001 Census and 35 to 44 in the 2006 Census. Results are presented in Text table 2. In the U.S. there is a change in how education is measured between the 1980 and 1990 censuses. The 1980 U.S. census records the highest grade of school or year of college attended or completed by the respondent without specifying whether a credential was obtained or not. The 1990 and later censuses, and the ACS, measure the highest level of educational attainment among those who hold high school diplomas, and the highest school grade completed for those without a high school diploma. Once again a synthetic cohort of domestic-born men was formed between 1980 and 1990 and their distribution calculated across the individual education categories in the census education variables and the broader education categories created. The cohort is 30 to 39 years old in the 1980 Census and 40 to 49 in the 1990 Census. The summary of this exercise is presented in Text Table 3.

^10.

The exception is the U.S. 1980 data, where the highest grade or post-secondary level attended, rather than completed, is known.

^11.

For example, in Canada the high school graduate category includes individuals with completed trades certificates or diplomas, as well as those who have some but not completed post-secondary education at any level, including university. In the U.S. the high school graduate category includes any post-secondary training below a (two-year) college associate degree, whether completed or not, but excludes individuals who have started but did not complete a college degree (associate degree or higher).

^12.

The question on the U.S. census about the language spoken at home is aimed at identifying individuals who speak a non-English language at home, regardless of whether they also speak English or not. A similar question was constructed from the more general language questions in the Canadian census.

^13.

The measure of English language ability in the U.S. (English or French in Canada) is based on different census questions in Canada and the U.S. In Canada, census respondents were asked whether they can speak English or French well enough to conduct a conversation. In the U.S., respondents reported how well they spoke English. English speakers are defined (in the U.S.) as those who reported speaking English well, those who reported speaking English very well, or those who speak only English at home. Non-English speakers are defined as those who reported speaking English not well or not at all.

^14.

The log point gap, or difference in mean log wages between two groups, can be interpreted as a percentage difference for small gaps. The log point gap and percentage difference are virtually the same for log point gaps smaller than 0.1. For larger gaps the approximation is less accurate. For example, the gap of -0.25 log points in mean log wages between university-educated new immigrant and Canadian-born men in 1980 represents a difference of 22.1%, while the gap of -0.67 log points in 2005 represents a difference of 48.8%.

^15.

Borjas and Friedberg (2007) have pointed out that the ACS data appear to overstate the difference in earnings of immigrant and domestic-born workers relative to census data. Paid workers were found to have lower average weekly wages in the 2000 ACS data than in the 2000 5% census sample, and this is more so for new immigrants than for the domestic-born. If this pattern is in some way related to differences in how or when the census and ACS data are collected, then the gap in wages between university-educated immigrants and university-educated domestic-born workers is overestimated for 2005.

^16.

Note that, when 25-to-30-year-old domestic-born workers (university graduates) are the comparison group (rather than 25-to-54-year-olds), the gap in wages between new immigrants and the domestic-born in Canada is much smaller in any given year, and the change in this gap across time is also smaller in magnitude. The difference in relative wages of immigrants between Canada and the U.S. in 2005, for example, falls by one-third, from 30 log points when 25-to-54-year-old domestic-born workers are the reference group to 20 log points when 25-to-30-year- old domestic-born workers are used as a comparison group.

^17.

Note that the levels of wages reported are not adjusted for purchasing power parity (ppp) and thus should not be directly compared across the two countries. However, in the years covered by this study, the ppp varied between 0.79 and 0.85 U.S. dollars per Canadian dollar, so that university-educated immigrants were not just earning less than the Canadian-born; they were also earning less than new immigrants in the U.S., at least from 1990 onwards.

^18.

Note that the educational attainment of individuals classified as high school graduates is somewhat different in Canada and the U.S., as described in Section II, "Data and definitions." As a result, some of the difference in the measured university wage premium between the two countries may be due to differences in how education is measured in each country.

^19.

Among new immigrants in the U.S., the premium leveled off or decreased slightly between 2000 and 2005.

^20.

Note that Charts 5, 6, 7 and 8 plot mean wages, while Charts 9, 10, 11 and 12 plot differences in mean log wages.

^21.

This study does not make any attempts to correct for the self-selection of women into the labour force.

^22.

The unadjusted wage gap between university-educated new immigrant and domestic-born workers is between 0.01 and 0.05 log points higher when only residents of Census Metropolitan Areas (metropolitan areas in the U.S.) are considered. To the extent that immigrants are not distributed across host countries proportionally to the domestic population, their settlement choice is likely endogenous. This study is ultimately concerned about studying the country-level differences in relative wages of immigrants in Canada and in the U.S.

^23.

Among men, the unadjusted wage gaps show that the disadvantage in wages of university-educated new immigrants from non-traditional source regions is more than two times as large as that for immigrants from traditional countries (North America, Oceania, Northwest Europe) in Canada. In the U.S., immigrant men from traditional source countries earned (on average) more than the U.S.-born in three of the four years in the study, while the disadvantage was restricted to immigrants from non-traditional source countries.

^24.

To create the synthetic cohorts, labour market outcomes of immigrants aged 25 to 44 who have lived in the host country for no more than five years are compared to the outcomes of domestic-born individuals aged 25 to 44, and ten years later the outcomes of immigrants aged 35 to 54 who have lived in the country for 11 to 15 years are compared to those of domestic-born individuals aged 35 to 54. The age bracket is reduced from 25-54 to 25-44 so as to exclude individuals close to retirement age. Any differences in the labour market status and outcomes of individuals aged 55 to 64 between immigrants and the domestic-born could confound the trends this study is interested in highlighting. There are several difficulties in 'following' a synthetic cohort of individuals across cross-sectional data sets. This study focuses only on individuals with a university education. Although by age 25 most individuals who are likely to pursue a university education will have completed their first degree, some may still have been missed. The sample focuses on individuals who worked at some point in the reference year and for whom paid employment was the major source of income. It is possible that some individuals might move between paid employment and self-employment between the two cross-sections and that individuals not included in the sample of workers in the first cross-section would be part of the sample in the second cross-section. This may be particularly true of immigrant women, whose participation rates are rather low in the first five years after arrival.

^25.

Note also that at 11 to 15 years since migration, a cohort covers five years of arrival in both Canada and the U.S., unlike the recent immigrant definition, where a cohort in the U.S. includes individuals who arrived in the census year (hence covering more than five full years).

^26.

Although these individuals arrive as temporary workers, they would still be captured in the U.S. sample of immigrants in this study.