Grades and labour market earnings in Canada: new evidence from the 2018 National Graduates Survey

ABSTRACT

This study uses the 2018 National Graduates Survey (NGS) to examine how postsecondary education grades are related to the labour market earnings of workers, and the potential moderating effects of work experience during school, work placements related to the field of study, and training acquired since graduation. Using simple regression models, we find non-causal evidence that the overall grade point average is positively related to earnings. This suggests that higher grades may translate into higher labour market earnings independently of other employment and training effects, indicating that the strength of the grade point average signal to employers is not weakened.

KEYWORDS:

• Education
• Canada
• income
• grades
• work experience

Acknowledgements

We are grateful to two anonymous referees who provided valuable comments that greatly improved this research. The analysis presented in this paper was conducted at the Lethbridge Research Data Centre (RDC) which is part of the Canadian Research Data Centre Network (CRDCN). The services and activities provided by the Lethbridge RDC are made possible by the financial or in-kind support of the SSHRC, the Canadian Institute of Health Research (CIHR), the Canadian Foundation for Innovation (CFI), Statistics Canada and the University of Lethbridge. The views expressed in this paper do not necessarily represent those of any of these organisations.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Declaration of disclosure statement

The authors report that there are no competing interests to declare. This is a heavily revised version of the first author’s MA thesis completed under the supervision of the second author.

Notes

1 Basically, good grades help to open doors for students entering the labour market. Once employed for a period of time, job performance then better “reveals” the true productivity of the individual. We thus expect grades to be of less importance the longer an individual has been employed or, according to the signaling/screening hypothesis, once a better signal is available (such as actual work performance).

2 Statistics Canada only allows weighted data to be used and these were all rounded to the nearest 1000. The observations not retained are, on average, similar to the retained observations, but a marginally different in some respects. Namely, those omitted are less likely to work or have a work placement during their programs, less likely to have additional education since graduation, are older, more likely to be visible minorities and born outside of Canada and are less likely to have business as their field of study.

3 Given the theoretical possibility that it is the level of education that may be endogenous (e.g., college versus university), previous attempts were also made to find an instrument for this variable using the Public Use Microdata File (PUMF). However, endogeneity and weak instrument tests often failed to identify valid instruments. In the cases where valid instruments were found, the results were both quantitatively and qualitatively like those presented below for the coefficients of interest. Similarly, different data limitations and codings of missing/unstated values were also imposed on the data set and again the results were robust to these different samples. Bootstrapped standard errors were also calculated using the weights provided by Statistics Canada with the results robust to this change. Below, the grade-earnings relationship is also estimated separately by subsample. We disaggregate regressions based on model 2 by level of study, work experience during enrollment, work placement during study, gender, other education or training since graduation, and field of study. Results based on these disaggregated regressions are presented in Tables 3 through 7 and Table A1 in the Appendix.

4 Work placements are defined in the NGS as being part of the graduate’s program and include co-ops, internships, practicums, clinical placements, field experience and community service learning. The definition excludes any work placements or experiences that were not part of the program. In other words, work placements are different than working while attending school in these data. Both variables are used individually in the analysis that follows.

5 Coefficient estimates presented here are only approximations of the marginal effects. Thus, because the grade dummies are binary variables and the dependent variable is in log form, the more accurate estimate can be obtained by using the formula: marginal effect = 100·[e ^{β^} - 1], where e is the mathematical constant 2.71828127.

6 Field of study was interacted with work placement to see if there were differential effects of having a work placement by field of study. Estimates of these interactions were almost all statistically zero and did not change the main results presented here.

7 This could be the result of at least two factors: (1) that individuals who pursue additional education were out of the labour force while studying or had fewer hours while training, reducing their experience and thus having a negative impact on their incomes; or (2) that they sought additional training to improve their lower-than-average salaries and were still lagging at the time of the 2018 interview. In a previous version of this paper, we limited the sample to include only those whose level of study at the end of their programs was the same as the highest level of education at the time of graduation and at the time of the interview to isolate the effects of on-the-job training rather than additional formal education. This did not markedly change the value or significance of the coefficient estimate.

8 There is a difference in coefficient estimates between college and bachelor’s levels of previous education (p = 0.0001), but no difference between bachelor’s and master’s (p = 0.0630) and master’s and doctorate (p = 0.7854).

9 A test for coefficient differences between one and two disabilities shows a significant difference (p = 0.0001), but there is no statistical difference between two and three or more disabilities (p = 0.1630).

10 In the previous version of this paper using the PUMF, some 92.9% of respondents were living in the region where they obtained their certification.

11 Given that may differ by field of study (Achen and Courant 2009) we disaggregate results by field of study (Table A1) and find that the coefficient on the A-range grade is positive and statistically significant at at least the 5% level for only respondents in six of the 12 categories: visual and performing arts, social and behavioural sciences and law; business, management, and public administration; mathematics, computer and information science; architecture, engineering and related technologies; and personal, protective and transportation services. The coefficient on the B-range grade is insignificantly different from zero in all but four cases: visual and performing arts; social and behavioural sciences and law; mathematics, computer and information sciences; and architecture, engineering and related technologies. Tests for differences in coefficient estimates on grade in the A-range and B-range yielded significant differences (at least the 5% level) only in three cases: business, management, and public administration; mathematics, computer and information sciences; and architecture, engineering and related technologies. These estimated relationships between grades and incomes by field of study are comparable to the pattern of results in both Finnie et al. (2016) and Britton et al. (2022).

12 Here we include only the three largest levels of education (as measured by the number of graduates) plus doctoral graduates. Results for the two university certificate or diploma programs are available upon request.

13 In the previous version of this paper using the PUMF, less than one percent of doctoral and master’s students had grades below the B-range.

14 When using coop placement instead of work placement, the coefficient for college graduates decreased to 0.032 but was statistically insignificant but increased slightly to 0.061 and was a significant at 0.1% for bachelor’s degree holders. Coefficient estimates for both master’s and doctoral graduates were not statistically significant in either case. These results are not shown in Table 3 but are available upon request.

15 Substituting a coop placement variable in place of the work placement variable results in little change to the coefficient estimates on both grade variables and in both cases (i.e., working and not working). However, the coefficient estimate on coop for those working almost quadruples from a statistically insignificant 0.013–0.050 and significant at the 0.1% level. The estimate on coop for those not working remained significant at 1% and did not change markedly in value.

16 Separating the sample into those who took coop programs (a subset of work placement) and those who did not, did not markedly change these results. The main difference was that the coefficient on the B-range for those who participated in work placement changes from 0.062 and significant at 1% to an insignificant 0.026. This does suggest that grades are more important income determinants in the absence of a coop placement, but tests for the equality of A-range and B-range grades between the coop and no coop groups failed to yield any statistical differences.

17 In the weighted raw data, the average male earnings premium for college graduates is 19.3% compared to premiums of 13.7%, 5.3%, and 1.3%, respectively, for bachelor’s, master’s and doctoral graduates.