3 Data and measures

View the most recent version.

This study draws data from Statistics Canada's 2002 Ethnic Diversity Survey (EDS). The EDS collects a national representative sample of over 42,000 non-Aboriginal Canadian residents aged 15 years or over. The survey was designed to provide information to better understand how Canadians of different ethnic backgrounds interpret and report their ethnicity and how people's backgrounds affect their participation in the social, economic and cultural life in Canada. For these purposes, the EDS covers a wide range of topics that include ethnic ancestry, ethnic identity, place of birth, visible minority status, generation status, knowledge of languages, family background, social networks, life satisfaction and socioeconomic activities. The survey over- samples visible/ethnic minority groups and thus obtains relatively large samples to allow comparisons between these minority groups and more-established, large ethnic communities in various characteristics (Statistics Canada 2003).

This study focuses on a sample of 7,400 working-age (25 to 64 years) immigrants and the second generation of immigrants (persons born to immigrant parents, thereafter referred to as the Canadian born) who resided in the eight largest metropolitan areas—Toronto, Montréal, Vancouver, Ottawa–Hull, Calgary, Edmonton, Winnipeg and Hamilton. The majority of Canada's immigrants—about 77% in 2001—and visible minorities—about 88% in 2001—are located in these large urban areas. This sample does not include those who suffered long-term illness, were on maternity/parental leave or were retired. It also excludes the third-plus generation population—those with both parents born in Canada—since this consists primarily of British, French and other western European ethnic origins (over 95%). Moreover, due to longer periods of settlement in Canada than the first or second generations and with greater rates of inter-group marriage, the third-plus generation is also more ambivalent about its ethnic identity, which is directly linked to our measure of ethnic concentration at the workplace (as will be discussed below). The study sample does not include people with British or French origins—the two charter groups in Canada—since the focus of this study is on minority groups. This study also does not include Aboriginals, since persons declaring an Aboriginal origin or identity were excluded in EDS's sample selection (Statistics Canada 2003).

The above-selected sample is used to compare the characteristics among those who did not work, those whose co-workers were mostly of the same ethnic origin, and workers with few or no co- ethnic co-workers. Within this sample, about 5,600 worked in the year prior to the survey date, either as paid employees or self-employed, had co-workers at the workplace, earned positive income and had employment income as their main income source. This sub-sample (worker sample) is used to examine the association between ethnic concentration at the workplace on the one hand, and earnings and self-perceived life satisfaction on the other.

The focal variable of interest—co-ethnic concentration at the workplace—is based on the survey question "As far as you know, how many people that you worked with had the same ethnic ancestries as you?" This question was only asked among respondents who had reported at least one ethnic ancestry, other than "Canadian," and had rated it either important or very important to them. In the worker sample, about 1% did not report at least one ethnic ancestry other than "Canadian" and a further 32% did not rate their ethnic ancestry as important to them. Thus, only 67% of our sample directly answered the survey question. However, we do not simply discard those people who did not directly answer the survey question on co-ethnic concentration at the workplace: they provide a very meaningful reference, based on the comparison that can be made between those whose ethnic ancestry was not important to them and those who rated their ethnicity as important but with different levels of workplace concentration.²

Co-ethnic concentration at the workplace is originally coded as a categorical variable with six groups: 1) all co-workers had the same ethnic origins as the respondent, 2) most of them did, 3) about half of them did, 4) a few of them did, 5) none of them did, and 6) ethnic ancestry was not important to the respondent. Since relatively few people were in the first category and there were no large differences between the first and second categories in the outcomes, the first two are combined in the analysis. The 4th and 5th categories are also combined, as little difference exists between them in the outcomes. In the analysis, workers in an ethnically homogenous workplace refer to individuals who share the same ethnic origin with all or most of their co-workers. This workplace concentration variable reflects the ethnic composition in the individuals' immediate work environment.

Multinomial logistic regression models are used to examine characteristics that distinguish those who mostly work with co-ethnics from other workers and from those who are not employed. The models are constructed separately for immigrants and for the Canadian born. The models include two sets of explanatory variables. The first set contains basic sociodemographic variables, including sex (women=1, men=0), age (in single years), education, family structure, place of residence, self-reported official language ability, visible minority/ethnic groups and the years since immigration for immigrants. Education is coded as three dummy variables: university degree, some postsecondary and high school graduation, with the common reference being less than high school. Family structure is represented by two variables: current marital status (married =1, else =0), and the number of children aged 14 years or under in the household, ranging from 0 to 4. The place of residence is coded as seven dummies for Toronto, Montréal, Ottawa–Hull, Calgary, Edmonton, Winnipeg and Hamilton, with Vancouver as the common reference.

The language variable captures both the proficiency of the official language and the ability in a minority language. It has five categories for immigrants: 1) individuals who have English/French as their mother tongue and still speak it; 2) individuals who have a minority language as mother tongue and still speak it, but they spoke mostly English/French with parents and siblings by age 15; 3) individuals who have a minority language as mother tongue and still speak it, and they did not speak English/French with parents and siblings by age 15, but speak English/French with friends; 4) individuals who have a minority language as mother tongue and still speak it, and they did not speak English/French with parents and siblings by age 15, and they do not speak English/French with friends, but they can speak English/French; and, 5) individuals who do not speak English/French. For the Canadian born, the last category does not apply.

Since different visible minority/ethnic groups may vary in the extent of co-ethnic concentration at the workplace, it is important to include detailed visible minority/ethnic categories in the model. The following 13 visible minority/ethnic groups are identified, each with a minimum sample size of about 50 persons among immigrants and the Canadian born. They include five visible minority groups: Chinese, South Asians, Blacks, Filipinos, and other visible minorities (Arab/West Asians, Latin Americans, Koreans, Japanese, Southeast Asians, visible minorities not included elsewhere and multiple visible minorities).³ There are also four ethnic groups with European background: Germans, Italians, Portuguese, and other European minority groups.⁴ See Table 1 for the sample size for each identified group.⁵

While the demographic variables show whether co-ethnic concentration at the workplace is more prevalent among those with less marketable human capital, the second set of variables is used to capture the role of individuals' preference for a co-ethnic work setting. This set of two variables reflects individuals' ethnic belongings and networks. One variable is whether the respondents reported that at least half of their friends were from the same ethnic group by age 15; the second variable is whether the respondents reported a strong sense of belonging to their ethnic or cultural group.

To examine the association of co-ethnic concentration at the workplace with earnings and self- perceived life satisfaction, regression models are constructed for employed men and women separately, and for immigrants and the Canadian born separately. One dependent variable is log annual income. A direct measure of employment income (wages and self-employment income) would have been preferable, but the EDS did not collect such information. To mitigate the potential bias, the analysis is restricted to those whose main income source is from employment. The other dependent variable is self-perceived life satisfaction,⁶ originally an ordinal variable with value ranging from 1 (not satisfied with life at all) to 5 (very satisfied). Since very few people reported 'not satisfied at all' (1.2%) or 'somewhat not satisfied' (1.5%), a dummy variable is created to contrast 'low levels of life satisfaction' (1, 2 and 3) and 'high levels of life satisfaction' (4 and 5). A dichotomous logit regression is used for this dependent variable.⁷ There are no other measures of psychological well-beings in the EDS.

For both outcomes, regression analyses include the aforementioned sociodemographic variables, but the age variable is replaced by potential years of work experience. Potential years of work experience are defined as 'age minus years of education minus 6.' The squared term of this variable is also included in the models. Moreover, the models include an additional set of explanatory variables related to individuals' job attributes, including self-employment status (self-employed=1, paid workers=0), weeks worked in the previous 12 months, hours usually worked in a week in the previous 12 months, occupation (six categories: management, natural and applied sciences, other professionals, sales and service, trades and transportation, and others—mostly related to occupations unique to primary industry, processing, manufacturing and utilities), and industry (five categories—goods-producing industries, trade and transportation, business services, public services and personal services).

With the above control variables, three Ordinary Least Squares regression models are built sequentially for earnings. Model 1 includes workplace concentration and basic demographic variables—visible minority/ethnic groups, potential years of work experience and family structure. This model provides the average earnings gaps across levels of co-ethnic concentration when differences in basic demographic characteristics are accounted for. Model 2 adds in education, language proficiency, immigrant status, and ethnic networks and belonging. This model will show how much the estimated average gaps in Model 1 can be accounted for by differences in characteristics that are associated with working in ethnically homogenous settings. Model 3 adds in work attributes.

For the second outcome variable—life satisfaction—two logistic models are built sequentially. Model 1 includes individual sociodemographic characteristics, job attributes and metropolitan areas of residence. Model 2 adds in log annual earnings to Model 1.

For both outcomes, models are constructed separately for immigrants and the Canadian born and also for men and women. To examine whether the associations of workplace concentration with outcomes differ across visible minority/ethnic groups, separate models are constructed for each large group—with at least 300 observations when combining immigrants and Canadian born as well as men and women. Because of small sample size at the group level, immigrants and Canadian born, as well as men and women, are combined in these models with controls for sex and immigrant status.

The EDS is a probabilistic survey and a survey weight is assigned to each respondent to represent the target population at the national level. This weight is used in all descriptive results. In regression models, the survey weight is standardized by dividing it with the average weight in the study sample. This standardized weight has the advantage of maintaining the same distributions as those of non-standardized weights, but avoiding an overestimation of the critical level in testing the significance of regression coefficients (Statistics Canada 2003).⁸

 

² It is possible that those who are highly motivated and with higher-than-average abilities are also the ones who do not rate their ethnic ancestry as important or who have a lower propensity to work in an ethnically homogeneous environment. In Section 5, we discuss the implications of such unobserved heterogeneity.

³ In this study, visible minorities are defined by Canada's Employment Equity Act as "persons, other than Aboriginal peoples, who are non-Caucasian in race or non-white in colour." The regulations that accompany the Act identify the following visible minority groups: Chinese, South Asians, Blacks, Arab/West Asians, Filipinos, Southeast Asians, Latin Americans, Japanese, Koreans, and others (Renaud and Costa 1999).

⁴ Two other European groups—Dutch and Polish—also meet the specified minimum sample size criteria. But there are hardly any workers who share the same ethnic ancestry with most of their co-workers in these two groups. They are combined into 'other European minority groups.'

⁵ In the 2001 Census, the population share of the selected seven single groups in the eight largest metropolitan areas is 6.6% for Chinese, 5.6% for South Asians, 3.8% for Blacks, 1.9% for Filipinos, 1.7% for Germans, 4.1% for Italians and 1.4% for Portuguese.

⁶ The survey question is "All things considered, how satisfied are you with your life as a whole these days?"

⁷ Two alternative modelling approaches have been tested. One approach uses ordered logit models for the original 5-category ordinal variable. However, the score test fails to support the proportional odds assumption. The other approach is to fit a dichotomous logit model by contrasting 'very satisfied' (5) with all other categories (1 to 4). The results are similar to those reported in the paper.

⁸ This is an issue only with certain procedures in some statistical software (e.g., Proc logistic in SAS).