Krishna
Pendakur and Ravi Pendakur
Introduction
A workshop
cosponsored by Metropolis and Strategic Research and Analysis
(Canadian Heritage) was held on 1-2 February 2002, in Ottawa,
with participation by economists, sociologists and geographers
from the various Canadian Metropolis Centers. The goal of
the workshop was to reach some consensus on what we actually
know about discrimination against ethnic minorities in Canadian
labour markets, and to flesh out priorities for future research
goals and instruments. The purpose of the workshop was to
try to ascertain what we know and what we would like to know
about the place of ethnic minorities in Canada's labour markets.
What Do
We Know?
Over the
past decade there has been a steady flow of research devoted
to examining the degree to which ethnic minorities are subject
to labour market discrimination in Canada (see for example,
Akbari, 1992; Howland and Sakellariou, 1993; Stelcner and
Kyriazis, 1995; Christofides and Swidinsky, 1994; Baker and
Benjamin, 1997; Hum and Simpson, 1998; Pendakur and Pendakur,
1998; Lian and Matthews, 1998). While these authors have generally
concluded that immigrant groups often face significant and
substantial labour market disadvantage, there is debate over
the degree to which minorities born in Canada are subject
to similar disadvantage. There is also debate over which ethnic
groups are subject to labour market disadvantage, and whether
that disadvantage is context-specific. This debate is frustrated
somewhat by the use of a variety of empirical approaches,
data sets and time periods.
Broad
consensus has emerged supporting the view that immigrants,
especially visible minority immigrants, have poorer labour
market outcomes than Canadian-born workers. This view is complicated
by the fact that the labour market disadvantage faced by visible
minority immigrants seems much stronger for men than for women.
Many papers have suggested that Canadian-born visible minority
workers earn less than (or, not equivalently, have lower wages
than) Canadian-born white workers. Again, the earnings disadvantage
is larger for men than for women, and is larger for some particular
visible minority ethnic groups than others, notably Blacks.
Recent work by Pendakur and Pendakur (2002) shows that the
labour market disadvantage faced by Canadian-born visible
minorities is not a new phenomenon, and has persisted for
at least 25 years. They also show considerable variation in
earnings disparities for various ethnic groups, and considerable
variation in these disparities across Canadian cities.
Both sociologists
and economists have focussed their research on the differential
labour market attainment of various ethnic minorities. Economists
have typically focussed narrowly on earnings or wages, while
sociologists have brushed more broadly exploring occupational
differences as well. Geographers have looked into occupational
segregation as it relates to residential segregation. Not
surprisingly, the findings from these quarters are roughly
consistent with the findings for earnings. We know that occupational
attainment (as measured by Pineo-Porter or other scales) is
worse for immigrants, and that it is somewhat correlated with
residential segregation. We know this to be especially true
for visible minority ethnic groups. However, less is known
about the occupational attainment of Canadian-born visible
minorities.
What we
know is roughly that some ethnic groups in some cities do
poorly in Canada's labour markets, but we know hardly anything
about why this is so. Pendakur and Pendakur (2002, IMR) suggest
that poor labour market outcomes for members of particular
ethnic groups is connected with ethnically-based labour enclaves
and to racist exclusion from majority labour markets, but
their research is suggestive rather than conclusive. Hiebert
(1998) suggests that occupational segregation is correlated
with residential segregation, but does not explore much why
this is so.
Much of
the discussion in the workshop focussed on what we do not
know. Although we know a lot about 'the way things are', we
know much less about why they are that way. The discussion
can be broken down into two parts. The first relates to defining
the object of interest by exploring concepts and differentiating
between such issues as discrimination and inequality. The
second part relates to defining and exploring the venues in
which discrimination and inequalities can emerge.
Conceptual
Issues
Discrimination
or Inequality?
We may
define discrimination as unequal treatment and inequality
as unequal outcomes. Typically we assume that discrimination
results in inequality, but this may not be so. The two inequities
do not imply each other. One may have unequal treatment without
unequal outcomes, as in a society with two equally well-funded
but antagonistic populations. One may have unequal outcomes
without unequal treatment, as in a completely non-racist society
where one group has a large amount of inherited wealth and
the other does not.
Not all
discrimination is bad, nor is all inequality. We are discriminating
when we choose some people as friends and others not, but
this sort of unequal treatment is typically not of concern.
In contrast, we are typically worried about discrimination
against groups of people on the basis of ascribed characteristics,
especially when it results in inequality. However, not all
inequality correlated with ascribed characteristics can be
assumed to be bad. While some inequality may be due to discrimination,
some inequality may be a product of choice, such as choosing
not to pursue further education or a high paying job. The
latter may also be a product of constraints such as being
unable to pay for higher education. Deciphering the rationales
for such choices is thus important in determining the nature
of income differences. It is also an important theoretical
distinction.
Academics
described as welfarists care about inequality independent
of cause and are thus interested in examining inequality that
may not be linked to discrimination. Academics focussing upon
ethnic or racial discrimination, argue that the discrimination
is special because it is caused by socially created barriers
which are insurmountable, based on colour. Thus differentiating
between ascribed and attained characteristics is key in determining
the nature of inequality. This leads to an interesting caveat
in that race and ethnic identity while overlapping are fundamental
different. Race while admittedly socially constructed is,
at least in the short term, immutable. Simply declaring yourself
to be white, black or oriental is generally insufficient if
people do not consider you to be a part of the group. Race
is thus an ascribed characteristic. Ethnicity, under certain
constraints can be attained or ascribed in that, it is possible
for some people to take on the characteristics of another
ethnic group and 'pass' for the majority, thereby making their
ethnic affiliation an 'attained' characteristics.
What dimensions
of inequality?
The group
argued back and forth for hours about which are the most important
types of inequality across ethnic groups, and ended up concluding
(inconclusively) that many types of inequality matter. Importantly,
most of the types of inequality that matter have not been
studied. In this category fall inequality of family support,
inequality of education attainment and quality, inequality
of social and job networks, and inequality of goals and desires,
all of which feed into the fairly well understood and measured
inequality in labour market outcomes such as occupation, earnings
and wages. On top of these inequalities lie the inequalities
which help or hinder in turning money into well-being, such
as inequality in credit market access, inequality in home-buying
markets, and other inequalities.
Model
Building
It seemed
clear to the group of researchers gathered that the dominant
model used by economists to study labour market discrimination
is a nonstarter. The competitive model of perfect information,
homogeneous buyers and sellers forces us to look for segregation
as the result of discriminatory preferences rather than inequality
in outcomes. Good models of economic discrimination must depart
from the competitive model in some significant way. Several
avenues are currently under investigation by economic theorists,
including embedding discriminatory preferences in a model
of imperfect competition where some jobs are simply better
than others, and even though lots of workers might
be able to fill these goods jobs, only some workers get to.
In the other social sciences, models of social cohesion
may allow us to understand the inequalities we see in labour
markets. These models push us to look at segmented labour
markets as they relate to segmented interactions in other
domains, such as social and job networks and links of trust
and friendship.
Measurement
Issues
The group
was largely empirical researchers, and as such, focussed for
some time on questions of empirical strategy. Part of the
challenge in measuring differences in outcomes is related
to the concepts we wish to measure. The opportunities are
broadly related to standard of living which can be a product
of job quality, work quantity and earnings. The inputs are
related to age, education, family type and possibly ethnic
origin or colour.
Family
Characteristics
Including
family characteristics in a model is also subject to debate.
It is generally recognized that marital status and whether
or not a couple has children effects income, however including
such characteristics in a model of income discrimination is
debatable since these characteristics should not theoretically
have an impact on the type of work. If decisions about the
type of work are family decisions, than it is important to
try to take these decisions into account, particularly if
the decision type varies by minority status. However, while
basic information about family type is available, the complex
decision-making information about job choice and family preference
is not. Thus at best, researchers are able to control for
only the coarsest of household decisions such as the decision
to get married, or have children.
The issue
of family decisions is an area that we know very little about
in part because the data does not exist. It would be worth
exploring this area in future surveys. There has been some
qualitative work on the way in which family decisions impact
on work and education choices, however this is not the case
on the quantitative side. It is thus difficult to determine
if the impact of family decisions are widespread.
There
is work that suggests that for immigrants, labour force decisions
can be family decisions. Wives from immigrant households,
for example, may decide to enter the labour force immediately
at a relatively low level in order to give husbands a chance
to get Canadian credentials and labour force experience. With
the added time, it is possible that husbands can attain better
jobs than if they enter the labour force immediately. Families
may also make decisions about the ideal level of schooling.
Indeed there is evidence that this varies markedly across
ethnic groups with some striving for high levels of schooling
and others, on average finishing with much lower levels of
schooling.
There
is an argument to be made that immigrants migrate for the
benefit of their children. They are willing to pay the costs
of migration in terms of loss of friend and family contacts,
work contacts and other benefits found in the home country
so that their children will have a better standard of living.
If this is the case, cohort effects matter. The labour force
performance of minorities born in Canada as compared to immigrants
and even those who immigrated while still in school as compared
to those who came in adulthood are markedly different. Minorities
born in Canada and those who came at a relatively young age
have much better labour force prospects than immigrants who
arrive later in life.
There
are also class effects. The socioeconomic status of parents
(both minority and majority) can be expected to affect the
socioeconomic status of children. Unfortunately, the characteristics
of parents in relation to children are rarely asked on major
surveys. There are some datasets that include information
on the education of parents (the upcoming ethnic diversity
survey) for example, however these are the exception rather
than the rule.
Education
While
there is agreement that there is a relationship between income
and education, the way in which education is included in the
model varies from one study to another. Many researchers use
years of education and years of education squared as controls
in income models. And, while it is true that income generally
goes up by years of schooling, it is also true that the payback
for years of schooling is not linear. Rather, it is the credential
which should be linked to income. Thus including separate
controls for different credentials will generally provide
more effective controls than simply including years of schooling.
This is particularly important for national comparisons in
Canada where the number of years of highschool vary from one
province to another.
While
an obvious control in determining income differences, education
can be measured many different ways depending on the goal
of measurement and the variable detail available. The most
common choices available are years of schooling, highest level
achieved and major field of study. Some datasets also include
place of schooling and parents education. Additional information
which could prove useful, includes whether the schooling was
at a public or private school or university.
Much of
the choice as to which education characteristics are measured
is at least in part a product of availability. Years of schooling
is often used as a control for schooling because it is an
interval level variable and thus has a minimal impact on the
degrees of freedom in a regression model. However, years of
schooling does not capture the real relationship with income
and education because the degrees or certificates associated
with education are not evenly distributed across the years
of schooling spectrum. Thus for example, a Bachelors degree
likely derives a greater income boost than the 4 years of
additional schooling after high school would indicate. Including
individual variables for different levels of schooling is
thus probably a better indicator of the payoffs to schooling
because such a strategy allows each level of schooling to
vary independently of the others.
One largely
unexplored area is that of the major field of study for degree
holders. If minorities are distributed across disciplines
in the same way as majority members, than it is probably not
important to include the field of study as a criteria. However
if this is not the case and if there are earnings differentials
associated with the discipline than it is important to take
such differences into account. Including the field of study
in the measurement strategy would have the advantage of controlling
for differences in earnings across different disciplines.
Of particular
relevance to immigration is the issue of foreign credentials.
With the exception of the Survey of Labour and Income Dynamics,
most datasets do not include information on where academic
credentials are obtained. In order to determine where an immigrant
obtained his or her schooling, it is thus necessary to impute
the place in which credentials were obtained by using a combination
of years of schooling, degree, age, age at immigration and
year of immigration. Such a strategy means that in a dichotomous
variable defining whether credentials are foreign or not,
a portion of the people will be misclassified. Including a
third classification for people who are difficult to classify
into either of the 2 camps resolves the problem to a degree,
but the strategy is still open to misclassification. Thus
the best strategy is to explicitly include information on
where the highest level of schooling was obtained on the dataset.
Thus far
the discussion has implicitly assumed that higher levels of
schooling lead to higher levels of income and that respondents
are in general agreement that this is the case. Given this
logic, than discrimination is defined as earnings differentials
that are apparent after schooling has been taken into consideration.
Thus, poorly schooled minorities will only experience a negative
earnings differential if their earnings are lower than poorly
schooled majority workers. In the same way, a highly educated
minority worker could be defined as being subject to discrimination
if he or she is paid substantially less than majority workers
with the same high level of schooling, even if the minority
worker is well paid in comparison to the average level of
income for society. However, it is also true that minority
groups, even those born in Canada, have very different higher
education participation rates. These differences could be
due to (1) parental pressure - minority group members having
very different preferences for schooling, (2) perceived barriers
- people feeling that they cannot get any further up the education
ladder and therefore it is not worth it to try, or (3) a form
of ethnic comparative advantage. If education is not chosen
'rationally' than we need another model for schooling. In
these instances it may be necessary to explore the impact
of education streaming or family choices in education.
Job Characteristics
Equal
access to different industry sectors- is a prime goal of the
employment equity program. However measuring and comparing
job quality is difficult. It is not easy to compare the quality
of different jobs across different firms, let alone across
different sectors of the economy. Further while sociologists
have devised scales to measure job prestige, they are hampered
by the same problems of comparability across sectors and firms.
The fallback
position has been to use occupation as a control, but not
pay too much attention to it. Controlling for occupation over
time however has two major drawbacks. The first is that occupational
distributions change over time and controlling for these structural
shifts is problematic. The second, which is more specific
to the Canadian case is that the occupational categorizations
have changed over time, making it impossible to create a comparable
occupational legend which is stable.
There
are two possible scenarios to this issue. The first scenario
questions whether work related variables should even be included
in a model designed to measure income differences. For, if
limited access to good jobs in good industries is one of the
forms taken by labour force discrimination (i.e.: minorities
cannot get into good jobs because they are blocked from doing
so), then including job characteristics in the model would
effectively mask the differences caused by this form of discrimination.
The second
scenario argues that an occupational classification which
is stable over time is unnecessary if occupation is being
used simply as a control and not as a variable intended for
detailed explanation. Given that the researcher only wishes
to control for the effects of occupation, including a large
number of (not necessarily the same) job characteristics in
each period is sufficient to capture the differentiation of
interest. While the first strategy is perhaps more consistent
and therefore will reveal more comparable results, the second
has the advantage of including a stronger set of controls.
However it should be realized that the question answered in
each is different. The first strategy assumes that occupational
segregation is at least in part responsible for the degree
of income difference. It thus asks if minorities are paid
less even though they have similar personal attributes. The
second compares minority and majority workers in the same
job and asks if minorities are paid less for the same work.
In the
same way, including other work related variables such as the
number of hours, weeks worked, full-time or part-time status
and industry, while adding a layer of complexity, may mask
the very differences we are trying to measure if there is
unequal access to any of those job related criteria. Thus
if ethnic origin is one of the criteria used to define access
to full-time jobs, including full-time status in the model
would effectively mask the differences in income which are
a result of this differentiation. However there is a case
for inclusion of these variables if it can be argued that
some job decisions are made by choice and these choices differ
by ethnic origin. For example, if it can be argued that British
origin women prefer to work part-time while women from other
ethnic groups prefer to work full-time, than omitting full-time
status from the model will create a false picture of income
differences by comparing a full-time minority workers to part-time
majority workers. While this is probably less of an issue
today, such differences may be important when looking at earnings
gaps over time.
Ethnic
Identity
Social
scientists argue that identity is multidimensional and therefore
unidimensional means of categorizing identity are inappropriate.
At the same time enlarging the scope of what constitutes ethnic
identity can be complex and amorphous. Ethnic identity weaves
together a broad spectrum of concepts, including ancestry,
religion, language and socialization to name a few. In the
same way, there are myriad of ways in which people may define
membership in a particular ethnic group. In general terms,
however, ethnicity refers to a group's distinctiveness and
is therefore a measure of culture which includes within it
a bundle of membership criteria such as: (1) Self-categorization
/ identification both as a means of identifying group members
and non-members; (2) Shared descent: a notion of history that
suggests commonality; (3) Specific cultural traits such as
custom or language; and (4) A social organization for interaction
both within the group and with people outside the group.
Many of
these aspects of identity are relatively easy to capture using
markers such as ethnic origin, language, immigrant status,
etc. However, we do not often ask whether people treat the
respondent as a minority. Further, in a settler society with
a broadly heterogeneous population, people can have multiple
affiliations, identifying with a number of different ethnicities
or groups.
Ethnic
identity is also fluid - it changes over time and place. Analyzing
the impact of ethnic origin on earnings is therefore becoming
increasingly complex and methodologically difficult. In part
this is because regression and other techniques require that
categorical variables be included in the form of dummy variables
in which each individual is categorized only once in any dummy
variable set. This is relatively easy for people claiming
only a single ethnic origin, however an increasing number
of people claim more than one origin meaning that they potentially
fit into more than one dummy variable in the set (for example
people claiming both British and Italian could conceivable
be in both the dummy variable for British and the variable
for Italian). Such a coding would effectively give that individual
the bonus (or penalty) for each group, but would not give
an indication of the unique impact of being both British and
Italian.
One common
solution is to categorize people into one of a set of origins
and allocate people claiming more than one origin into only
one of the origins claimed. In such a categorization, the
person claiming British and Italian origins would be allocated
into only the British or only the Italian ethnic dummy. However,
discounting the impact of an origin seems both methodologically
and theoretically incorrect. Another solution is to create
a dummy variable set that includes a series of multiple origins
which can be uniquely categorized (for example a category
for 'British and Southern European Origins'. This method has
the advantage of allowing people with that combination of
origins to be categorized by a unique effect, but it has the
disadvantage of creating a large number of possible origin
combinations.
Data issues
The discussion
above hints at a variety of research programs which would
surely be fruitful in aiding our understanding of how ethnic
minorities fit into Canadian labour markets and Canadian society.
These new research programs led us to a wish list for future
research instruments and data sets.
Many micro-data
sources exist for the study of discrimination and inequality
as they relate to ethnic minorities in Canada. Research has
focussed almost exclusively on individual-level data to the
exclusion of family-level data. Family-level data are more
appropriate if decisions are actually made at the level of
the family, as they certainly are in some communities.
The first
hope expressed by the group was that currently available panel
datasets be much more aggressively explored and used. In panel
datasets, we see two important advantages over repeated cross-sections:
(1) definitions are held constant over time, which is especially
important for occupational and industry definitions, and for
ethnic origin definitions themselves; and (2) individuals
are follow over time, which allows us a glimpse at dynamic
phenomena such as promotions and life-cycle choices.
The LIDS,
IMDB, SLID, LAD and LSI are all fairly large panels which
include information on ethnic origin and immigration status
as well as other interesting dimensions of labour market and
other choices and outcomes. Although these data tend to exhibit
'small number' problems when the object of interest is the
behaviour or outcomes of a small group-such as Canadian-born
visible minorities who make up a mere 1% of the working age
population-it is important to push these data to their limit.
None of
the currently available public-use data are seriously focussed
on issues of work organisation. The structure of work, including
who bosses whom, who promotes whom, who gets promoted, who
gets favourable dismissal and who gets fired, who gets rehired,
and who gets a golden handshake, is essentially unstudied
by social scientists interested in ethnic minorities in Canada.
This could be remedied with new instruments which focus on
work organisation.
Two possibilities
for new instruments to explore these areas are focus groups
and firm-level micro data. Management and business scholars
have used focus groups extensively for decades, and economists
have used them almost not at all. Focus groups may be relatively
inexpensive route to learning about work organisation and
model building in general. Similarly, experimental work (eg,
Henry and Ginsberg 1989) might be used to learn more about
the processes that create and sustain differential treatment.
On more
familiar terrain, firm-level micro data would be an excellent
supplement to the many individual-level micro data sets currently
available. Firm level data would enable researchers to 'put
the workplace together' and see how workplaces vary across
minority and majority bosses and workers. In addition, the
study of ethnically based labour market enclaves would be
greatly enriched by knowledge of the ethnic makeup of firms.
For example, are firms ethnically segregated, and does this
relate to residential segregation and to ethnically-based
differentials in labour market outcomes?
Summary
Linking
economic disparity and labour market segmentation to identity
requires an understanding of how gender, language and ethnicity
affect the set of social and labour market processes that
generate success in labour markets.
The goal
of the workshop was to map a path on how to illuminate the
degree to which ethnic origin and identity effect labour market
aspirations, performance and outcomes drawing from a multidisciplinary
team of researchers. This constitutes an extensive research
agenda, which in our view could more easily be attained through
a multidisciplinary framework which brings together researchers
from a variety of theoretical and academic backgrounds. However
such research teams are neither common nor easy to develop.
It is our hope that the costs of developing such interdisciplinary
research agendas do not outweigh the
goal of explaining the intersections of labour market differentials
and ethnicity.
