Bryan J. Pesta, Peter J. Poznanski (2008)
The relationship between IQ scores and elementary cognitive task (ECT) performance is well established, with variance on each largely reflecting the general factor of intelligence, or g. Also ubiquitous are Black–White mean differences on IQ and measures of academic success, like grade point average (GPA). Given C. Spearman’s (Spearman, C. (1927). The Abilities of Man. New York: Macmillan) hypothesis that group differences vary directly with a test’s g loading, we explored whether ECT performance could mediate Black–White IQ and GPA differences. Undergraduates (139 White and 40 Black) completed the Wonderlic Personnel Test, followed by inspection time and choice reaction time ECTs. Despite restriction of range, ECT performance completely mediated Black–White differences on IQ (d=.45). Group differences on GPA (d=.73), however, were larger and ECT performance did not mediate them. We discuss findings in light of Spearman’s hypothesis.
A Black–White group difference on intelligence test scores has persisted in the literature for over 90 years. Currently, the group IQ mean for Blacks (85) remains about one-standard deviation below the group IQ mean for Whites (100; see, e.g., Neisser et al., 1996; Lynn, 2006; Rushton & Jensen, 2006). […]
The literature also shows that Blacks, on average, are less likely than Whites to attend college, and those that do have lower standardized test scores, grade point averages (GPAs), and higher dropout rates (Dreary, Strand, Smith, & Fernandes, 2007; Roth et al., 2001; U.S. Department of Education, 2001). Academic achievement and IQ, however, are strongly related. Gottfredson (2005b) summarized data showing a median correlation of .60 between standardized tests of school performance and IQ test scores (.80 when aggregating the different academic tests into a single composite; see also Gottfredson, 2004). Clearly, a large percentage of the variance in academic performance is shared by variance in IQ. The direction of causality, however, and whether third variables explain the relationship, remain empirical questions.
One prominent explanation for Black–White differences is Spearman’s hypothesis (Spearman, 1927; see also Jensen, 1985), which posits that race differences on IQ test scores reflect race differences on the general factor of intelligence, or g. Evidence for Spearman’s hypothesis comes from findings that a test’s g loading strongly correlates with the magnitude of the Black–White difference the test produces (see, e.g., Hartmann, Kruuse, & Nyborg, 2007; Jensen, 1998; Lynn & Owen, 1994; Nyborg & Jensen, 2000; te Nijenhuis & van der Flier, 2003). If the race difference on IQ tests is a g difference, then other valid measures of g (i.e., beyond traditional paper and pencil IQ tests) should mediate the relationship between race and IQ. Consistent with this prediction, our goal is to explore whether performance on basic measures of information processing, which themselves are highly g-loaded, can mediate race differences on IQ.
Information processing ability is inferred by subject performance on so-called elementary cognitive tasks (ECTs). Examples of commonly used ECTs are those that measure processing speed, or reaction time (RT), and those that measure speed of information intake, or inspection time (IT; see Jensen, 1998, for an overview of various ECTs). The literature shows that basic information processing ability, as measured by ECTs, correlates about .50 (after correcting for attenuation) with g, as measured by traditional IQ tests (for meta-analytic reviews, see Grudnik & Kranzler, 2001; Kranzler & Jensen, 1989 see also Jensen, 1998). To our knowledge, however, studies exploring the relationship between race and ECT performance have used RT but not IT (for reviews, see Jensen, 1998; Rushton & Jensen, 2005). And, although the relationship between grades, race and IQ is clear (see, e.g., Dreary et al., 2007; Gottfredson, 2004; 2005b), how ECT performance might mediate these relationships is unknown.
We administered the 12-minute version of the WPT in classrooms which varied in size between 11 and 38 students. Students later completed the ECTs in a computer laboratory. Instructions for the ECTs were presented onscreen. They included an overview of each task, and sample trials illustrating how participants should respond. For the IT task, we emphasized that speed of response was not important. Subjects were told to take as much time as they needed, focusing only on responding accurately. For the RT task, we told subjects to respond as fast as possible while maintaining high accuracy.
On each trial of the IT task, a fixation cross (+) appeared centered on the screen for 1000 ms. The fixation cross was then blanked for 100 ms. Thereafter, the IT stimulus (i.e., two vertical lines of differing length, connected by a smaller horizontal line) appeared. On a random half of trials, the left line of the IT stimulus was longer than the right. For the other half of trials, the opposite was true. The IT stimulus remained onscreen for a varying amount of time, as described below. After the appropriate IT duration had passed, a lightening bolt mask (where both vertical lines were of equal length, see, e.g., Luciano et al., 2004) appeared for 300 ms. The screen then went blank, and subjects responded by pressing “z” if they thought the left line was longer than the right, or by pressing “m” if they thought the right line was longer than the left.
The critical manipulation in the IT task involves varying the duration of the IT stimulus from trial to trial. All durations, however, had to be multiples of the computer monitor’s refresh rate (one frame every 13.33 ms, given a refresh rate of 75 Hz). To achieve this, we used an adaptive staircase method which varied the IT duration across trials, as outlined by Luciano et al. (2004). All subjects first completed three practice trials, and then started the experimental trials with the IT duration set at 133 ms. Four correct answers in a row at any IT duration resulted in a “reversal,” which caused the IT duration for the next trial to decrease (by four frames for the first two reversals; two frames for the next two reversals, and one frame for every reversal thereafter).
The IT duration also increased (by four, two or one frame, depending on the current reversal count) whenever a wrong answer was made. Reversals for wrong answers, though, occurred only if the previous IT duration was longer than the IT duration that forced the mistake. Hence, a long string of wrong answers would result in a constant increase of the IT duration. These would not count as reversals. Wrong-answer reversals only occurred when subjects made mistakes just after the IT duration had been decreased (due to four correct responses in a row on trials using a longer IT duration). The program ran for 96 trials or 15 reversals, whichever came first.
For the RT task, three letters (always two Capital “Ss” and one Capital “A”) appeared centered on the screen, and remained until the subject responded. The subject’s task was to rapidly indicate in which of the three positions the letter “A” appeared. For example, if the display showed: S A S, the subject would indicate that the letter “A” appeared in position two. Subjects used the number keypad on the keyboard, and responded with their right hand. The first six trials for each subject were practice, and not included in data analyses. Subjects then completed 60 experimental trials (where the target letter, “A,” appeared randomly, 20 times in each of the three possible positions).
We used p<.05 as the level of significance for all analyses. Table 1 shows mean WPT scores and GPAs by race. The WPT has a reported population mean of 22, with a population standard deviation of 7.0 (Wonderlic & Associates, 2002). From Table 1, the range of IQ scores for our participants seems somewhat restricted, as the sample standard deviation of 5.8 is only 83% of the population standard deviation of 7.0. The partial restriction of range might explain why the Black–White IQ difference here (d=.45) was smaller than the typical one-standard deviation effect (d=1.0) reported in the literature. A larger race difference emerged, however, for GPA where Black students averaged .73 standard deviations lower than the White student mean (this value is similar to those reported by Gottfredson, 2005a,b,c see Table 18.3, p. 536). In sum, significant race differences appeared for both IQ scores and GPAs.
Table 2 shows group means and standard deviations for the IT and RT tasks. We calculated each person’s overall IT as his/her average IT duration across all reversals except the first. The column labeled “IT variability” is a measure of intra-individual variability. It represents the mean of the standard deviation of IT durations for subjects by race across all reversals. This measure is not typically reported in the literature. We included it here because it showed significant race differences, and also correlated with various other measures in the study (see Table 3). For the RT task, we calculated median RTs and standard deviations for each subject across the 60 experimental trials, after first excluding error trials and any single trial with an RT greater than 1000 ms. Error and excluded trials combined averaged only 2.72 occurrences (SD=5.02, out of 60 possible) per subject. Note that all group differences in Table 2 are significant.
Table 3 shows simple correlations for the demographic variables, IQ, GPA, and the ECT scores. Small but significant effects appeared for race and its correlation with every other variable in the table. The last variable in Table 3 is a factor score derived from a principal components factor analysis conducted on the four ECT-task variables listed in Table 2. We conducted it to avoid multicollinearity in the regression analyses, as all ECT scores were intercorrelated. The analysis revealed a single factor accounting for 55% percent of the variance in the ECT scores. Each of the four ECT-task scores loaded significantly on the factor (loadings ranged from .58 for RT to .84 for IT).
Table 4 shows regression analyses which test whether ECT factor scores mediated the race difference on IQ (Hypothesis 1). For each regression equation in the table, both age and gender were entered as control variables. Baron and Kenny (1986) identified three criteria that must be met for a variable to achieve the status of a mediator. First, the independent variable (race) must correlate with the mediator variable (ECT), as shown in Eq. (1). Second, the independent variable (race) must correlate with the dependent variable (IQ), as shown in Eq. (2). Third, entering both the mediator and the independent variable should result in the latter now being non-significant. From Eq. (3), the standardized beta weight for race, controlling for ECT-task performance, is no longer significant (B=−.124, p=.09). Complete mediation, in theory, should result in the beta weight dropping to zero, which did not happen here. Nonetheless, ECT-task performance qualifies as a statistical mediator of race and IQ, as controlling for it reduced the relationship between race and IQ from a significant weight of −.254 in Eq. (2), to a non-significant rate of −.124 in Eq. (3) (a 49% decrease). This mediation occurred even after controlling for race differences on both age and gender.
Table 5 reports tests of whether the ECT factor scores mediated the relationship between race and GPA (Hypothesis 2). Again, all equations included age and gender as control variables. In Eq. (1), the independent variable (race) correlated with the mediator (ECT). In Eq. (2), the independent variable correlated with the dependent variable (GPA). However, Eq. (3) shows that race still significantly predicted GPA, even after controlling for ECT-task performance (B=−.285). In fact, (1) controlling for ECT-task performance in Eq. (3) only reduced the race/GPA correlation by .03 (10%), relative to Eq. (2), and (2) ECT-task performance itself was not significant in Eq. (3) (B=−.093; p=.23). Hence, contrary to Hypothesis 2, ECT-task performance failed to mediate the relationship between race and GPA reported here.
Summarizing key results: (1) Small (relative to the literature) race differences existed on IQ test scores. (2) Large (relative to the IQ difference here) race differences existed on GPA. (3) Race differences also existed on all ECT measures, including the intra-individual variability of IT and RT. (4) A factor score derived from the ECT measures statistically mediated the relationship between race and IQ. (5) The ECT factor scores failed to even partially mediate the relationship between race and GPA. In fact, ECT scores added little in the way of incremental variance explained when race, age and gender were also in the equation predicting grades. The data therefore support our first but not second hypothesis.
Given Spearman’s hypothesis, it is perhaps not surprising that cognitive indicators of g (i.e., our ECTs) would mediate the relationship between race and psychometric measures of g (i.e., the WPT). The variation in both ECT performance and IQ scores seems largely driven by individual and group differences in the general factor. On the other hand, more than g contributes to a person’s GPA, including variables like motivation, conscientiousness, family environment, work status, etc. (though recent research also points to a genetic cause for within-group differences in academic achievement, see, e.g., Luo, Thompson, & Detterman, 2003; Wainwright, Wright, Geffen, Luciano, & Martin, 2005; Wainwright et al., 2006). Considering just the present data set, however, basic measures of information processing do little to explain the Black-White difference on GPA.
Although consistent with Spearman’s hypothesis, our data offer no insights as to possible causes for race differences on the ECTs. Whether these differences might arise from differences in environment, nutritional levels, genes or some other factor is an issue in need of further study. Further limits to the present study include: (1) a relatively small sample size for Black students, though statistical power did not seem to be an issue given the pattern of consistent, significant results was found. (2) A restricted range of participants, as we ran only college students. (3) Use of WPT scores as a proxy for g (i.e., we did not derive g factorially). Future studies with multiple measures of g might show an even clearer picture of the mediation effects reported here. (4) For unknown reasons, both age and gender differed by race. Although we included each as control variables in the mediated regressions, a more random sampling of race in a future study would offer stronger evidence that neither played a role in the data patterns reported here.
We believe the present data illustrate the potential of explaining and understanding group differences on IQ test scores by appeal to group differences on basic information processing tasks. In the present study, controlling for the latter eliminated the former. In addition, ECTs are so simple, that many possible explanations for observed differences (e.g., motivational levels) are reduced if not diminished. More research is needed, both on the criterion-related validity of ECTs, and on the underlying theoretical reasons for Black–White differences on these measures.