Results

View the most recent version.

Archived Content

Information identified as archived is provided for reference, research or recordkeeping purposes. It is not subject to the Government of Canada Web Standards and has not been altered or updated since it was archived. Please "contact us" to request a format other than those available.

Archived

This page has been archived on the Web.

Descriptive

Table 2 presents the means of math and science scores at age 15 by first university program taken as well as by gender.  The first thing to note is that, on average, males have significantly higher math scores than females, 589 vs. 569 (significantly different at .01 level) while, in terms of science scores, males have a higher average score but, in this case, it is not statistically different from females.  This is not entirely surprising as earlier work has found similar results (Bussière et al. 2007).

Table 2 Mean of mathematics and science PISA scores by first program type in university, by gender

In terms of first university program type, significant gender differences in mathematics PISA scores are observed for the social sciences and business programs. In contrast, men who enter STEM, health or other programs had higher average mathematics scores at age 15, but the difference with average female scores was not statistically significant. 

For both males and females, however, average mathematics scores are highest for those who enter STEM programs; the same is also true when examining average science scores. Thus, for both males and females, higher mathematics/science scores at age 15 translate into a greater chance of a first university program being in STEM fields. However, the absence of a statistically significant gender difference across male and female averages suggests that male and female mathematics/science scores appear to operate quite similarly in affecting whether or not the first university program is in a STEM field.

The lack of significant gender differences for STEM programs might be a result of examining average levels of mathematical/science ability, which may not capture enough of the distributional differences between men and women.  Table 3 presents the proportion of youth entering each type of university program by different levels of mathematical/science ability and gender.  High mathematical/science ability is defined as the 4th proficiency level or higher, while lower ability is defined as the 3rd proficiency level or lower.

Table 3 First university program choice of YITS-PISA respondents who attended university, by category of PISA scores and gender

Table 3 shows that the most common first university programs for females, regardless of mathematical or science ability, are the social sciences, followed by STEM programs, business, health, and finally all other programs.  For example, around 50% of women, regardless of their mathematical/science ability, choose the social sciences.  Interestingly, women with lower mathematical and science skills, while also more likely to choose social science programs, however, choose STEM programs in much lower proportions than their female counterparts who are strong in mathematics or science (about 15% vs. 23%). Thus, among young women there appears to be a relatively strong link between mathematical/science ability at age 15 and university fields of study that require substantive science and mathematics backgrounds (the STEM fields) compared with those that do not.

Conversely, among young men, the most common first university programs, regardless of mathematical or science ability, are in the STEM fields, with the exception of males of lower science ability who have a larger proportion who go into the social sciences.  Similar to women, men also go into the social sciences in similar proportions (slightly more than 30%), regardless of mathematical/science ability. However, like their female counterparts of lower mathematical and science ability, men of lower mathematical and science ability go into STEM programs much less frequently than men of higher mathematical and science ability. The difference is about 7 percentage points (45.7-38.5) in the case of mathematics, and even greater at 12 percentage points (42.5-30.1) for science ability. Thus, as with women, men also appear to have a relatively strong link between mathematical/science ability at age 15 and choosing STEM fields; these within gender differences across ability do not extend to other disciplines in the same magnitude.

These descriptive results suggest that, as in Table 1B, young women are most likely to choose the social sciences as their first university program, and this does not change at high or low levels of mathematical or science ability. In other words, the most popular first university program for young women, regardless of mathematical/science ability, is in the social sciences. In contrast, for young men, the most popular programs are in the STEM fields, except for males of lower science ability, who are most likely to choose social science programs.  Finally, there is some indication that youth, regardless of gender, with strong mathematical/science skills are more likely to choose STEM programs than their counterparts with lower skills.  The gap is somewhat greater for science than for mathematics. These relationships will be examined further in the following section, using multivariate techniques that take other important covariates into account.

Multinomial Logistic Regression

Table 4 presents the estimates from three different specifications of multinomial logistic regression models on both the mathematics and science subsamples. The first model is the bivariate case, which only includes mathematical/science ability by gender. The second model includes the mathematics/science by gender variable, as well as a wide range of covariates (discussed earlier) known to be either related to mathematical/science ability and/or program choice. The only two covariates that are not included in Model 2 are mathematics/science marks and self-assessed mathematical ability. These remaining two variables are considered in Model 3.  Table 5 presents the results from all control variables from Model 3 (the full model).  In Tables 4 and 5, the results have been transformed from multinomial logits to average marginal effects for ease of interpretation.  They can be interpreted as the effect of a one-unit change in any given explanatory variable on the probability of choosing each of the university programs.  Four categories of mathematical/science ability by gender are specified: female-high PISA, female-low PISA, male-high PISA, and male-low PISA. In all analyses, young women with high PISA mathematics/science scores are taken as the reference category.

Table 4 Average marginal effects of mathematical/science ability at age 15 by gender, based on three different multinomial logit models predicting university program choice

The effect of mathematical/science ability and gender on program choice

The bivariate results at the top of Table 4 replicate the proportions given in Table 3; thus, the focus here will be on how the coefficients change across models given the addition of control variables. Two relationships from the bivariate results, and how they change with the inclusion of control variables, will be assessed: (1) Does the gender difference remain in choosing STEM, and social science programs? (2) Are males and females of high mathematical/science ability, as measured through PISA scores, more likely to go on to STEM programs than their lower ability counterparts, even after controlling for factors such as mathematics/science marks and self-assessed mathematical ability?¹

First, it was observed that males, regardless of mathematical ability, have a higher probability of entering STEM programs than females. For instance, in the bivariate case, males of high mathematical ability have a 22.4% higher probability of going into a STEM field than females of equally high mathematical ability (the 0.224 effect indicated in Table 4). Also, even males of lower mathematical ability have a 15.3% higher probability of entering a STEM program than females who have high mathematical ability.  This male advantage in entering STEM programs holds in Model 2, and even in Model 3, once mathematics marks and self-assessed mathematical ability are added. In the full model, males of high mathematical ability have a 17.8% higher probability of choosing a STEM program in university than females of equally high ability, while males of lower mathematical ability have a 19.4% greater probability.²

The story for science, gender and choosing STEM is more or less the same, except that males with lower science ability do not have a higher probability of entering STEM programs than women of high science ability. For instance, in model 3, males of high science ability have an 11.3% higher probability of entering STEM programs than females of an equally high level, while the value for males of low science ability is only just significant at the 10% level.  Part of the discrepancy between males and females of high science ability choosing STEM is explained through the covariates because in the bivariate case, males of high ability had an 18.7% higher probability of entering STEM, which was reduced about 7 percentage points once science marks and self-assessed mathematical ability were included.³

Females, conversely, regardless of mathematical or science ability, are more likely to choose social science programs. For instance, according to Model 3, males of high mathematical ability have a 12.6% lower probability of entering social science programs than females of equally high mathematical ability. For males of lower mathematical ability the probability (15.9%) of entering social science programs is even lower when compared to females of high mathematical ability.⁴ A similar effect is observed in the science results.

Thus, it appears that mathematical and science ability have little effect on explaining gender differences in program choice, at least when considering STEM and social science programs. Granted, females of high mathematical ability have a higher probability of entering STEM programs than their counterparts with lower mathematical and science ability, but their higher mathematical ability is not sufficient to erase the difference in STEM attendance from males with even lower mathematical ability.  Likely, something other than ability (as measured through the PISA assessments) is driving the difference in the decision of males and females to choose one of these two major programs in university. It is also something other than mathematics/science marks, self-assessed mathematical ability, and numerous other factors related to student achievement and educational interests at age 15.⁵

The second issue to be examined involves the question of whether males and females of high mathematical/science ability (as measured through PISA scores) are more likely to go into STEM programs than their lower ability counterparts (same gender), after controlling for factors such as mathematics/science marks and self-assessed mathematical ability. For women, results from Table 4 show that marks and self-assessed mathematical ability may have more of an impact on program choice than ability as measured via PISA scores. For instance, in the bivariate case, women with lower mathematics PISA scores had an 8% lower probability of entering a STEM program than females with high mathematics PISA scores. In Model 2, this difference, while still significant, was reduced to 7.3% and, by Model 3, the difference in choosing STEM based on mathematics PISA scores was completely removed for women. Thus for women, mathematical ability, as measured through PISA scores, does little to explain differences in STEM attendance in university, once mathematics marks and self-assessed mathematical ability are taken into account.⁶  A very similar finding was discovered for women with respect to science PISA scores and choosing STEM programs: science marks, as well as one’s perceptions of his or her abilities in mathematics, explained away the difference between young women of higher and lower science ability.

For men, there is no real difference in the probability of entering a STEM program based on mathematics PISA scores. In the bivariate model, the difference between these two groups is only significant at the 10% level.  In the other two models, there is no difference.  In the science results, in the bivariate case, males with high science PISA scores have a 12% (p < .01) higher probability of entering a STEM program than their lower ability counterparts. However, this difference is completely removed in model 2 once most of the control variables are included.⁷

The effect of marks and self-assessed mathematical ability on program choice

Table 5 presents the marginal effects for the covariates used in Model 3, the full model. Given that models are not estimated separately by gender, these marginal effects cannot reveal anything about gender differences in program attendance. This would require additional analyses with males and females separated. Nonetheless, it is informative to present the results for mathematics and science marks and self-assessed mathematical ability, given the importance that these two measures have with regard to gender, PISA scores, and university program choice.

Table 5 Average marginal effects of the controls used in Model 3 (full model) from Table 4: Based on multinomial logit models predicting university program choice

In terms of marks, the highest marks (average marks of 90% to 100%) are associated with a 6.7% and a 9.4% greater probability (than average marks of 80% to 89%) of entering a STEM field, for mathematics and science marks respectively. Conversely, individuals who had average marks below 80% are significantly less likely to enter STEM fields than individuals who had average marks of 80% to 89%.   Youth who had average mathematics marks of 90% to 100% were however less likely to enter the social sciences than youth who had average marks of 80% to 89%. Marks had less impact on the other fields, except that individuals with the lowest mathematics marks were significantly less likely to go into health programs than individuals with average mathematics marks of 80% to 89%.  Science marks had a similar, albeit, somewhat weaker effect on program choice.

Self-rated mathematical ability (rated on a scale from poor to excellent), not surprisingly, had a very similar effect on program choice as did mathematics marks; namely that youth who rated themselves as having higher ability in mathematics were also more likely to choose STEM programs and less likely to choose social-science programs. With each increase in the scale, the probability of choosing a STEM program increased by 9.7%, while the probability of choosing a social science program decreased by 10%.  Self-rated mathematical ability had a similar effect in the models using the science subsample.

---

Notes

1. The results related to STEM and social sciences are focused on, since these are the two most common first university programs of choice.
2. In some additional analyses, which stepped in mathematics marks and self-assessed mathematical ability individually into Model 3, self-assessed ability had a larger impact on reducing the difference between males and females of higher mathematics ability choosing STEM programs. These results are available upon request.
3. In some additional analyses, self-assessed mathematical ability was more influential than science marks in reducing the male–female difference in STEM attendance. These results are available upon request.
4. Also, while not shown explicitly in Table 4, females of low mathematical ability have an even greater probability of entering social science programs than males, regardless of their mathematical ability. This is known because of the positive marginal effect associated with females of lower mathematical ability (0.012).
5. Finnie and Childs (2010) also found that mathematics and science marks did little to explain the gender gap in STEM attendance.
6. This paper was mainly focused on the intersection between PISA scores and gender and so interaction terms between marks and gender, and self-assessed mathematical ability and gender were not introduced into the models. These relationships are interesting but are left for future work.
7. It is unclear which of the covariates presented in Model 2 explained away the difference between males with high versus lower PISA science scores; further analysis would be needed to isolate each of the individual factors.