Job-EducationMismatch and Its Impact on the Earnings of Immigrants: Evidence from Recent Arrivals to Canada

Using themost recent Longitudinal Survey of Immigrants to Canada, this papermeasures the incidence of job-educationmismatch, particularly over-education, examines its determinants, and estimates its impact on the earnings of immigrants. Job-education mismatch is measured using the realized match method, and the corresponding earnings impact is estimated using an overrequired-under education technique. Determinants of over-education are examined using a bivariate probit model to account for selectivity into employment. Panel data estimation methods are used to estimate earnings equations and the analysis is strati�ed by gender. Results show that recent immigrants to Canada have a persistent high incidence and intensity of over-education with a substantial negative impact on their earnings. In particular, two-thirds of recent immigrants to Canada are over-educated with a wage loss of 8%, while an under-educated immigrant loses around 2% on average. Results also show that pro�ciency in English or French and post-immigration investment in education and training signi�cantly reduce the likelihood of being over-educated. e �ndings of this study could bene�t policies directed to help immigrants integrate in the labour market.


Introduction
e contribution of immigrants to a host country's welfare largely depends on the degree to which their foreign education translates into useable quali�cations and skills in the host labour market. A common �nding is of the imperfect transferability of human capital across countries (see, e.g., [1][2][3]). New immigrants typically face barriers when searching for a job which, in principle, matches their quali�cations. Such barriers either prevent or delay integration into the host labour market. In particular, immigrants who �nd a job on arrival in the host country usually work in a job which requires a level of education which is less than they actually possess. is form of job-education mismatch is known in the literature as over-education. If a worker is employed in a job requiring more years of schooling than the worker actually has, then the worker is under-educated. Most of the job-education mismatch literature focuses on studying over-education in view of its high incidence and signi�cant adverse effects.
Several studies have examined the occupational outcomes of immigrants relative to their source-country educational attainment (see, e.g., [2,[4][5][6]). e general �nding is that there is a substantial earnings disadvantage and a high incidence of over-education among immigrants. For example, Li et al. [5] �nd that, during 1993-2001, 52% of recent immigrants to Canada were over-educated compared to 28% of their Canadian-born counterparts; also recent immigrants are twice as likely to remain over-quali�ed compared to native Canadians. Using census data, Galarneau and Morissette [6] �nd that 28% of recent immigrant males and 40% of corresponding females, each with a university degree, were in jobs with lower educational requirements compared, respectively, to 10% and 12% for Canadianborn workers. Wald and Fang [7] �nd that half of the immigrants into Canada who arrived between 1989 and 1997 were over-educated. ere is similar evidence from Europe: Nielsen [8] �nds that the incidence of over-education among Danish immigrants with foreign-acquired education was 47% compared with 33% among native Danes. For additional international evidence, see: Lindley and Lenton [9] for the UK, Green et al. [2] and Kler [10] for Australia and Galarneau and Morissette [6] for Canada. e adverse effects of over-education have been established in, for example, Galarneau and Morissette [6], Tsang and Levin [11], Tsang [12], Tsang et al. [13], Conference board of Canada [14]. For example, the Conference Board of Canada (2001) estimates that the Canadian economy loses up to 5 billion Canadian dollars per annum due to over-education. In addition, several theories have been put forward to explain the existence of over-education based on signaling from incomplete information [15]; spatial mobility [16]; differential over-quali�cation of secondary earners [17] and the human capital hypothesis. e most recent empirical literature on the incidence and impact of over-education is based on the human capital hypothesis. Immigrants are forced, initially on entering host-country labour market, to accept jobs with lower educational requirement to compensate for their lack of host-country-speci�c human capital. ese include job experience, contacts in the host labour market, language skills, and social capital (e.g., from relatives and friends). Other reasons for over-education include poor quality of education, nonrecognition of foreign experience and credentials, market entry in a period of recession, and competition from domestic workers. e pro�le of personal characteristics of the immigrant also matters for being overeducated. e probability of being over-educated is expected to be high for immigrants who immigrate in old age or who are member of a visible minority. Mother tongue and country of origin are key reasons for an increased degree of over-education for recent and established immigrants to Canada [6]. Licensing requirements and regulations required by professional associations also play a role.
e main objective of this paper is to measure the incidence of job-education mismatches, particularly overeducation, to examine their main determinants and to estimate their corresponding impact on the earnings of recent immigrants to Canada. e extant literature in this area is mostly based on cross-sectional data or surveys. In this paper, longitudinal data are used from the most recent Longitudinal Survey of Immigrants to Canada (LSIC). Such data permit control of unobserved individual heterogeneity as well as the evolution of over-education, in particular, over time, leading to a more comprehensive analysis than hitherto. e paper proceeds as follows. Section 2 presents the empirical methodology followed, in Section 3, by a description of the data. e results are presented in Section 4 and discussed in Section 5. Finally, the conclusions are summarized in Section 6.

Measuring the Job-Education Mismatch. A job-education mismatch for the th employed immigrant in occupation
; ; is measured as which is de�ned as − . (1) In which is the actual years of schooling completed by employed immigrant in job , is the required years of schooling for job . In job , worker is over-educated if > , under-educated if < , and just-educated if . In addition to the employed immigrants, there are unemployed immigrants who are seeking employment and . Required schooling for each occupation is here de�ned as the modal number of years of schooling of the workers in occupation . e arithmetic mean number of years could have been used yet the results are robust between the two [18].

Determinants of Over-Education.
Being over-educated is observable only for those who are employed. Sampling only from the over-educated will generally lead to sample selection bias which can be avoided by allowing for the probability of selection into employment. is is achieved by applying a bivariate probit model with sample selection within which the probability of being over-educated, conditional on being employed, is estimated. e resulting estimates are consistent and asymptotically efficient [19,20].
e �rst probit equation, (2) below, estimates the probability of being employed for each of immigrants, using explanatory factors listed in the appendix: In which Pr * is an vector representing probabilities of being employed, is a matrix of explanatory factors of order is a vector of coefficients, and is a vector of residual errors of order . e explanatory variables in the matrix include gender, age, marital status, immigration class, level of education, region of origin, �eld of study, language pro�ciency, visible minority status, whether foreign work experience is recognized in the Canadian labour market. In addition, to account for the effect of any postimmigration investments in Canadian human capital, a set of dummy variables that indicate whether the immigrant has a Canadian work experience or has taken any education or training aer arrival are used.
e probability of being employed ( * is latent, but what is observed is a dichotomous variable indicating whether the individual is employed or not, such that e second probit equation, (4) below, estimates the probability of being over-educated for each of employed immigrants as follows: In which Pr * is an vector representing probabilities of being over-educated, is a matrix of explanatory factors of order is a vector of coefficients, and is an vector of residual errors. e matrix includes most of the explanatory variables included in the �rst probit equation in ISRN Economics 3 addition to the number of education credentials and whether the job is part time or full time. * 2 is also latent, but what is observed is a dichotomous variable ( 2 ) for whether the individual is over-educated or not, such that 2 = 1 if * 1 > 0, 2 = 0 otherwise. (5) e error term in each of the probit equation is speci�ed as follows: e error terms 1 and 2 consist of a part that is speci�c to that equation and a second part that is common to both equations. is means that the error terms in the two probit equations may be correlated due to the existence of unobserved factors that jointly affect the probability of employment and over-education. e correlation between the two error terms is estimated and tested for its statistical signi�cance. If 0, the standard probit method applied only to the over-education equation produces biased results. However, the biprobit model with sample selection gives consistent and asymptotically efficient estimates [19,20].
e objective is to estimate the probability of an immigrant being over-educated conditional on being employed, which is given by e joint estimation of (2) and (4), the testing of the correlation of corresponding estimates of 1 and 2 , and the estimation of the probability of over-education conditional on being employed are all undertaken using the biprobit model in STATA 11.

e Earnings Impact of Job-Education
Mismatches. e impact of job-education mismatch on earnings is estimated using the over-required-under education (ORU) model [18]. It has been shown that the ORU model is superior to the conventional Mincerian earnings function which uses total schooling as an explanatory variable. e standard earnings function may give misleading results, because the return to surplus schooling-beyond what is required by the job-is likely to be lower than the return to required schooling [21]. e ORU model may be presented as follows: e subscripts , , and index, respectively, individual, occupation, and time period. denotes required years of schooling by the job, denotes years of over-education, and denotes years of under-education.
is a vector of control variables: age, gender, �eld of study, marital status, immigration class, work experience, language pro�ciency, provincial dummies, and visible minority status. is the standard time varying residual term. It should be noted that and are mutually exclusive, and for each individual, one of them or both must be zero. e ORU model is reduced to the standard Mincerian earning equation if 2 = 3 = | 4 |. However, if this does not hold, the ORU model will yield a better �t and the return to required education will be greater than the return to total education [21].
Under the ORU model, over-educated and under-educated workers are compared to coworkers, that is, workers with the same required schooling who are just-educated. Accordingly, the coefficients in the ORU model are interpreted as follows: 2 : return to an additional year of required schooling, 3 : return to an additional year of surplus schooling relative to coworkers, 4 : wage loss of an additional year of de�cit schooling relative to coworkers, 3 − 2 : return to an additional year of surplus schooling relative to workers, and with the same total year of schooling which is adequately used.
Several �ndings concerning the earnings impact of jobeducation mismatch are documented in the literature (see: [18,[22][23][24][25]). First, the return to over-education is positive ( 3 > 0) but smaller than required education ( 3 < 2 ), while the return to under-education is negative ( 4 < 0). Second, over-educated workers earn less than workers with the same educational attainment in jobs which require that level of schooling ( 3 − 2 ) < 0. However, under-educated workers earn more than workers with the same educational level working in jobs requiring the level of education that they have, and earn less than coworkers possessing the required level of education. As a baseline model, (9) is estimated by a pooled crosssectional method, using ordinary least square (OLS), with standard errors corrected for clustering at the individual level. Pooled OLS produces consistent estimators, if the error term in the ORU model is uncorrelated with the explanatory variables in the model. However, if unobserved individual characteristics are crucial for determining earnings, then the error term will be correlated with the included explanatory variables, and hence it is better to use panel data estimation methods to yield consistent estimates. To bene�t from the longitudinal structure of the LSIC in accounting for unobserved individual heterogeneity, (9) may be estimated using panel data methods, using both of the between effects and random effects models. e between effect is obtained when OLS is performed on the average over time for each individual in (9). e �xed effects model is also estimated but is not reported, because the key variables of interest were dropped in the �xed effects estimation. Accordingly, (9) can be rewritten as Here, the error term , from (9) becomes + , where represents time invariant individual-speci�c effects and is the standard residual term. In the random effects model, is assumed to be uncorrelated with other covariates in the model [19]. STATA 11 soware package is used to conduct the multivariate analyses, and all the estimation as well as the calculations of (2) and (4) are weighted using the LSIC sampling weights.

Data and Sample Characteristics
e main data source for this study is the LSIC. e survey was conducted by Statistics Canada and Citizenship and Immigration Canada using a sample from the 164,200 immigrants who immigrated to Canada between October 2000 and September 2001. e survey consists of three waves of interviews of the same cohort of immigrants. In the �rst wave, 12,000 immigrants aged 15 years and above were interviewed between April 2001 and May 2002, six months aer becoming permanent residents in Canada. In the second wave, 9,300 of the same immigrants were interviewed in 2003, two years aer landing. In 2005, about 7,700 of the same immigrants were reinterviewed, four years aer their arrival. e LSIC contains comprehensive information on all standard labour market and sociodemographic variables. For more information about the LSIC see Statistics Canada [26].
To achieve the objectives of the current study, data from the three waves of the LSIC survey are used. Data from the con�dential 2001 Canadian census of population is used to derive the required years of schooling based on 508 occupations which, in each occupation, is determined by the modal number of years of schooling of workers in that particular occupation.
e multivariate analyses include a number of economic and sociodemographic variables commonly used in the literature. Age is represented in continuous form. Gender is captured by two dummy variables. Marital status is represented by two dummy variables: married, single, or separated. An individual's educational attainment is captured by the total number of years of schooling. Work experience in the Canadian labour market is captured by the number of weeks that an immigrant has worked in Canada. Foreign work experience has two speci�cations. In the �rst speci�cation, a dummy variable is used to indicate whether the immigrant had full-time foreign-work experience before immigration. In the second speci�cation, potential work experience (age minus years of schooling minus 6) is used. Immigration class is captured by four categories: family class, skilled workers, business class (reference group), and refugees. Provincial or regional �xed effects are represented in �ve categories: Ontario, Quebec, British Colombia (reference group), Atlantic provinces (New Brunswick, Prince Edward Island, Nova Scotia and Newfoundland and Labrador), and Western provinces (Alberta, Saskatchewan and Manitoba). Immigrants' �elds of study appear as seven categories: educational; �ne arts, humanities and social sciences; engineering; health professions; commerce (reference group); agriculture and mathematics; and no specialization. Language pro�ciency is captured by an indicator variable: English/French is the mother tongue and English/French is not the mother tongue (reference group). Experience recognition is captured by an indicator variable with foreign experience not recognized in the Canadian labour market as the reference group. Working status comes as full time (1), part time (0). e analysis also includes a dichotomous variable indicating whether an individual belongs to a visible minority (1) group as de�ned by Statistics Canada or not (0). Table 1 reports the summary statistics of the variables included in the analysis. Half of the immigrants are males, 80% are married, 60% have at least a bachelor degree, and 61.3% are skilled workers. For the �eld of study, 17.3% have an engineering background, and 19% are in commerce or management-related �elds. 9.5% of the immigrants reported English or French as their mother tongue. 48% are residing in Ontario. e average age of individuals in the sample is 35 years.

e Incidence and Intensity of the Job-Education
Mismatch. e longitudinal feature of the LSIC facilitates the study of how the incidence and intensity of over-education has evolved over time. e incidence of the job-education mismatch at each wave of the LSIC is reported in Table 2. e results indicate a high incidence of over-education among recent immigrants to Canada since 76.3% of the immigrant males and 71.8% of the females is over-educated in the �rst wave of the LSIC (6 months aer becoming permanent residents). ese �gures did not improve much aer four years from arrival, when 70.4% of the males and 64.6% of the females are over-educated. e results also indicate that 15.7% of the immigrant males and 16.6% of the females are under-educated aer 6 months from arrival in Canada. As a benchmark for comparison purpose, the incidence of over-education among Canadian-born is estimated as 43.85% using data from the 2001 Canadian census.
e persistent high incidence of over-education among recent immigrants is in line with the �ndings of several previous studies which examined earlier cohorts of immigrants to Canada (e.g., [4,5,7]). e wide range of variables in the LSIC enables measuring the incidence of job-education mismatches among different subgroups of immigrants. Accordingly, the sample is strati-�ed by gender, marital status, age, and level of education. Table 3 shows considerable differences in the incidence of over-education among different subgroups of immigrants. Based on the LSIC data, the incidence of over-education is higher among recent immigrant males than among females, and is also higher among married people than singles. e incidence of over-education decreases with age. is is consistent with the job search behavior of young workers and the fact that young workers lack enough job experience or training, relative to older workers.
Not surprisingly, the incidence of over-education increases with educational attainment. e results show that immigrants with a high level of education face substantial di�culties in transferring their quali�cations to the Canadian labour market than the less educated immigrants. For  Source: author's compilation using data from LSIC. All statistics are population weighted using the LSIC sampling weights.
example, 90% of recent immigrants with a bachelor degree, and 94% with a master degree are over-educated. Immigrants with a degree in regulated occupations such as dentistry, medicine, veterinary medicine, optometry, and law have the highest incidence of over-education at 96%. ese �ndings are consistent across the three waves of the LSIC and are in line with the �ndings of Li et al. [5], Wald and Fang [7]. A complete understanding of the phenomenon of overeducation and under-education requires studying not only its incidence but also its intensity (how many years of surplus or de�cit schooling), and how each evolves over time. Studying how the job-education mismatch intensity evolves over time helps in understanding the dynamics of the assimilation process, where a reduction in the intensity of the job-education mismatch could be an indicator of job match improvement and assimilation. e general �nding from this analysis is that the incidence and intensity of over-education decreases with the length of stay in Canada (see Table 4).
e distribution of job-education mismatch intensity and its evolution by the length of stay in Canada are shown in Figures 1, 2, and 3. e distribution is clearly not symmetric, but it is negatively skewed. us the le tail is longer, with a few negative values (under-educated), the mass of the distribution being concentrated on the right (over-educated). However, the skewness of the distribution has decreased according to the time spent in Canada.

Determinants of Over-Education.
Results from the bivariate probit model for the determinants of employment and over-education are presented in Table 5.
Results show that recent immigrants who are members of a visible minority are less likely to be employed and more likely to be over-educated. is is consistent with the �ndings  Table 4.  Table 4.  of several studies which reveal that immigrants may be subject to discrimination in the Canadian labour market (e.g., [4, 7�). Previous studies show that language pro�ciency and region of origin are important determinants of employment and over-education. Consistent with these �ndings, results of the biprobit model show that immigrants with excellent or well-established English or French pro�ciency are more likely to be employed and less likely to be over-educated relative to immigrants with poor language ability. Results also show that post-immigration education and training increase the probability of employment and reduce the probability of being over-educated. is is known in the assimilation literature as the time since immigration effect. As immigrants stay more in the host country and start to accumulate hostcountry-speci�c human capital, their labour market outcome is expected to improve. Having relatives in Canada increases the chances of being employed, showing the importance of social capital in helping immigrants to integrate. As recognition of the importance of social capital, the immigration point system in Canada gives 5 points for having relatives in Canada. Also the likelihood of employment increases with age where older workers are expected to have more accumulated human capital to offer in the labour market.
Married immigrants are less likely to be employed and overeducated, relative to single immigrants. e recognition of foreign work experience in the Canadian labour market decreases the probability of being over-educated. Part-time job holders are more likely to be over-educated. As the number of credentials increases, the probability of overeducation increases, because it will be longer and more costly to recognize them. ese results are in line with the �ndings of Li et al. [5] using the Canadian Survey of Labour and Income Dynamics.
Note that the correlation coefficient between the two error terms is positive and statistically signi�cant, suggesting that sample selection is a signi�cant problem, and that there are unobserved factors that jointly affect the likelihood of employment and over-education. is means that the standard probit model applied only to the over-education equation produces biased results. However, as indicated earlier in the paper, the biprobit model with sample selection yields consistent and asymptotically efficient estimates.

e Earnings Impact of the Job-Education Mismatch.
Results from estimating the ORU earnings equation are reported in Table 6 for the whole population. In addition, the results for males and females are reported in Table 7.
Consistent with previous �ndings, the return to overeducation is positive, but smaller than the returns to required education for both males and females, while the return to under-education is negative. ese �ndings are robust to changing the model speci�cation and estimation method. Given that the results of the three estimation methods pooled OLS, random effects and between effects are very similar, results obtained from the pooled OLS model have been  used in what follows. e return to over-education is 36.8% of the returns to required education for females, while for males, the return to over-education is 13.2% of the returns to required education. Over-educated males (females) earn 8.7% (7.6%) less than workers with the same education attainment in jobs which require that level of schooling. is means that for females there is no wage loss to an additional year of de�cit schooling relative to coworkers. However under-educated males lost 3.25% for an additional year of de�cit schooling relative to coworkers. Although the return to under-education is not statistically signi�cant for females, the effect of under-education has almost the same magnitude as for over-education. e effects of the other explanatory variables are consistent with a priori expectations. Earnings increase for immigrants with English or French as a mother tongue, and with having full-time foreign experience recognized in the Canadian labour market. Working in a parttime job naturally reduces earnings. An immigrant's earnings also increase with age and having longer Canadian work experience.

Discussion
e high incidence of over-education among recent immigrants to Canada could be explained by several reasons that are widely discussed in the literature [25]. Immigrants are forced, at least during the early period of entering the host country's labour market, to accept jobs with less educational requirement due to lack of host country-speci�c human capital (e.g., job experience, contacts in labour market, and language skills). Based on data from the LSIC, aer four years from arrival to Canada, lack of Canadian work experience was mentioned to be the main difficulty (49.8%), followed by lack of contacts in the job market (37.1%), nonrecognition of foreign experience (37%), and foreign quali�cations (35.4%). About one-third of job seekers who experienced difficulties stated language barriers as a problem. Entering the labour market during periods of recession could be a possible reason for the high incidence of over-education [27]. It is evident that the LSIC participants have arrived to Canada during a period of recession. As a result, these immigrants are le with no option other than survival jobs. About one-third of the job seekers who experienced difficulties reported lack of employment opportunities as a problem. Costly accreditation and licensing requirements by professional associations in many regulated occupations also constitute an entry barrier in many occupations. Poor sourcecountry schooling quality is another reason for the high incidence of over-education among recent immigrants [27]. Recent data show that there has been a shi from countries with a high quality educational system to countries with a low quality educational system. According to the 2006 Canadian Census of Population, 58.3% of recent immigrants came from Asia (including the middle-east) compared to 12.1% in 1977, while those who came from Europe was 16.1% compared with 61.6% in 1977. Picot and Hou [28] �nd that Canadian employers have no reliable information about the real occupational skills and education quality of graduates from Asian institutions. Another challenge is that recent immigrants have less command of official languages. For example, only 9.5% of the immigrants in the LSIC report English or French as their mother tongue. Another reason is that there could be some discrimination in the labour market toward visible minorities. About 15% of the immigrants covered by the LSIC reported discrimination as the main difficulty in �nding a suitable job. is is supported by several studies which argued that immigrants are subject to discrimination in the Canadian labour market (see: [4,7,28]) For example, Oreopoulos [29] conducted a study using thousands of resumes sent in response to online job postings for occupations in Toronto. e author �nds considerable employer discrimination against applicants with ethnic names on the resume in terms of lower callback rates and interview requests, compared to those with Englishsounding names.

Conclusion
is paper attempted to measure the incidence of jobeducation mismatch, particularly over-education, examine its determinants, and estimate its impact on the earnings of recent immigrants to Canada using panel data. Results show that two-thirds of recent immigrants to Canada are over-educated with a wage penalty of about 8%. Results also show that acquisition of host-country-speci�c human capital, particularly, language pro�ciency and education and training signi�cantly reduces the likelihood of being over-educated. ough the time span covered by this study is short, allowing only for an assessment of initial assimilation, yet, the �ndings of this study could bene�t policies directed to help immigrants integrate in the labour market.