Sex Differences in School Performance During High School: Puzzling Patterns and Possible Explanations

AbstractWhile past explanations for sex differences in school performance are heavily geared toward explaining boys' small advantage on standardized math tests, we broaden the focus to include educational outcomes for which girls typically do well—standardized reading tests and math and English grades. Among adolescents in the National Education Longitudinal Study, girls enjoy better English and math grades and higher reading test scores than boys in eighth grade and these advantages all increase during high school. In contrast, boys earn slightly higher math test scores in eighth grade and this gap also increases slightly by the end of high school. This set of patterns leaves us with a puzzle—why do girls and boys excel at different components of schooling? We explore models that assess the degree to which these different trajectories can be explained by sex differences in classroom behavior and out-of-class activities. We conclude that a major reason for sex differences in grades is boys' poorer classroom behavior while sex differences in time spent outside of school should supplement previous explanations for test scores gaps. ACKNOWLEDGMENTSWe thank Brian Powell and Maureen Tobin for helpful comments.NOTESNotes1 Data from the recently collected Early Childhood Longitudinal Study of 1998 through 1999 suggest that sex differences in math skills may be evident as young as first grade (CitationDenton and West 2002). This new information suggests that sex differences in math skills may emerge earlier than previously thought.2 For a critique of this position see CitationSommers (2000).3 While it is possible that girls earn better grades and reading test scores than boys despite poorer treatment from teachers, girls' typically superior performance on most educational outcomes suggests that alternative explanations—other than discrimination against girls in the classroom—merit attention.4 In support of this position, math grades and math test scores correlate only .40 and English grades and reading test scores correlate only .37 in NELS:88.5 For example, "'playing with' equipment such as seesaws or cameras seemed to be an especially powerful kind of learning activity, with many students stating that they learned something from using real equipment that help them on the … items" (CitationHamilton 1998:189).6 Because we require valid test score data at two points in time and valid teacher evaluations, our sample is potentially less generalizable than the complete NELS sample. We caution that the students persisting in our sample are typically more advantaged socioeconomically than those who did not persist in the sample. In supplementary analysis we found that sex differences on test scores and grades for the whole sample were similar to those among our subsample and that differences across outside-of-school activities and in-class citizenship were also unchanged. This pattern is consistent with the view that our results would generalize to the whole sample if it were possible to use all cases.7 We also estimated models including measures of effort and disruptiveness from eighth grade classes. The results were similar to those presented.8 While we focused on activities students were involved in outside of class, NELS also has measures of students' perceptions of how useful math will be in their future and whether math will be useful in their job. Boys are more likely than girls to perceive math as useful in their future but equally likely to perceive math as useful in their job. Including the variable measuring how useful math will be in the future in our regression models reduces the difference in gains in math test scores slightly.9 Previous studies suggest that students have exhibited a slight tendency to overestimate their own grades versus the more objective transcript reports. It is not clear whether this is the case in NELS or whether this tendency varies by sex in systematic ways. It is likely that there is greater measurement error in self-reported grades and that this reduces the likelihood of finding statistically significant differences, therefore biasing our models in a conservative way.10 Most models predicting educational outcomes control for other common predictors such as race, family structure, and number of siblings. While there is no a priori reason to believe that these variables would vary by sex of student, we did find that boys enjoy slightly higher socioecomic status (SES) and fewer siblings in the NELS data. Controlling for these factors in multivariate models, however, did not alter the pattern of results presented in our tables. We also considered whether SES position affects girls and boys differently. We found little evidence that the patterns in our tables vary in important ways by socioeconomic status.11 CitationAllison (1990) suggests that under some conditions there may be advantages to predicting change scores (Y2 - Y1) rather than our models (Y2 = Y1 + X). We also estimated the Y2 - Y1 model to see if our pattern of results was sensitive to this alternative specification and the general patterns persisted.12 In supplemental analyses we explored how the patterns in Tables 2–5 are conditioned by several student characteristics. Similar to CitationCatsambis (1994), we noted that female Hispanics are especially vulnerable to losing ground in math grades and test scores.In addition, some past work suggests that sex differences in math test scores are restricted to the top end of the performance distribution (CitationHedges and Nowell 1995), and so we tested whether the processes we uncovered were different among the highest performing students. Our results provide some indication that among high school students with high test scores (one standard deviation above the mean), boys gain the most, while among students with low test scores (one standard deviation below the mean), girls gain the most. For grades, the patterns were different for the high-achieving group. While low-performing boys gained less in grades than low-performing girls, this gap was even greater among high-performing boys/girls.13 We recognize that these associations may not reflect causal relationships but rather associations between our activities indicators and some omitted variable. To address this issue we attempted to estimate random effects models that would help us control for unobserved differences between boys and girls, but these were difficult to estimate for technical reasons. Ideally we would employ similar measures of outside activities at each wave of data but NELS sometimes changed the wording of questions across waves, complicating our efforts to estimate true change in an activity.14 It may be that we could explain the entire math test score difference if we had additional indicators of after-school activities. We explored some of these possibilities. For example, we found that boys more often participate in school and community sports, but that these activities were not consistently related to gains in math test scores (but see CitationBroh (2002) for slightly different results based on varsity sports only). We also found that girls spend more time talking on the phone with friends than boys but that this activity is negatively related to gains in most educational outcomes and appears to play no role in mediating sex differences.