The Causes of Group Differences in Intelligence Studied Using the Method of Correlated Vectors and Psychometric Meta-Analysis

The Causes of Group Differences in Intelligence Studied Using the Method of Correlated Vectors and Psychometric Meta-Analysis

Daniel Metzen, Master thesis, 2012.

Abstract

The huge IQ gap between non-Western immigrants and ethnic Dutch has emerged as one of the primary explanations for the large differences in school and work achievement between these groups. Is there a genetic component in the IQ gap between immigrants and ethnic Dutch? Meta-analyses have shown that the group differences on IQ subtests correlate almost perfectly with the cognitive complexity of these subtests; moreover, the cognitive complexity correlates perfectly with heritability and strongly with physical characteristics of the brain. If no other causes for IQ differences show a strong correlation with g loadings, this would point to a strong genetic component in IQ differences between immigrants and ethnic Dutch. In the present study, we first seek support for the hypothesis that only variables under genetic influence show a strong positive relationship with general intelligence. These are group differences, heritability, and physical characteristics of the brain. Second, we test whether differences in IQ due to variables not under genetic influence, namely biological-environmental factors, aging, and autism show a negligible to weak correlation with general intelligence. Support for both hypotheses would suggest that group differences are primarily driven by genetic factors and only to a minor extent by non-genetic factors. Therefore, group differences between non-Western immigrants and ethnic Dutch should be regarded as stable over time.

Concerning the first analysis, we first conducted a full-fledged meta-analysis on reaction time differences between Whites and higher-IQ groups, and Whites and lower-IQ groups, and we conducted several bare-bones meta-analyses and analyses of individual studies on differences in IQ profile between groups of different ethnicity. Second, we explored subgroups on school type, and religion. Third, we carried out a meta-analysis on the question whether g-loadedness of reaction time measures predicted the heritability of these measures. Fourth, we carried out a meta-analysis on the link between g loadings and brain volume. Concerning the second analysis, we first conducted several bare-bones meta-analyses and analyses of individual studies on biological-environmental variables. Second, we conducted bare-bones meta-analyses on the psychological phenomena autism and aging.

The hypothesis was strongly supported: heritabilities and most group differences showed moderate to strong positive correlations with g, but the correlation of brain volume with g was quite modest. All other phenomena showed no strong positive correlation with g.

It is concluded that these findings are strongly in line with a substantial genetic component in group differences in intelligence. This suggest that the large group differences in school achievement and work achievement are stable and that I/O psychologists should find ways to deal with them instead of ways of trying to change them.

Study 1: Group Differences

Study 1a: Difference Between Whites and Higher-IQ and Lower-IQ Groups in Reaction Time

In almost all studies that were included in this meta-analysis the reaction time measure of Jensen (1993) were used. The reaction time measure of Jensen (1983) consists of an apparatus with a home button in the middle and several reaction buttons located in a half circle around the home button. This apparatus is connected to a computer screen on which stimuli is presented. The participant has to rest his hand on the home button until a stimulus is presented. When the stimulus is presented the participant removes his hand from the home button and presses the correct reaction button. There are three different tasks: A) Simple reaction time (SRT): In this task, the participant has to press a designated reaction button. There is no uncertainty which button will light up, but there is uncertainty with regard to when the button will light up. This task measures pure reaction time. B) Choice reaction time (CRT): In this task the participant has to press a not designated reaction button. Hence, there is uncertainty with regard to which button will light up and when the button will light up. This task is slightly more complex than the SRT. The time needed to choose the correct reaction button is an additional indication of the speed of cognitive processing. C) Discrimination reaction time (Oddman RT): In this task three different reaction buttons light up. The button that is most far away from the other two buttons is the correct reaction button. In addition to the uncertainty concerning which reaction buttons light up and when they light up, the participant has also to identify the correct reaction button. Since the Oddman RT involves the most cognitive processing of the three tasks, Jensen (1983) suggests that the Oddman RT should show the highest correlation with IQ from the three tasks.

Four different reaction time measures are recorded. The RT (reaction time) consists of the time interval between the onset of the stimulus and the removal of the hand from the home button. The MT (movement time) consists of the time interval between the removal of the hand from the home button and the activation of the reaction button. SDRT (Standard deviation of reaction time) measures the intraindividual variability in RTs. Every participant has a mean reaction time with a standard deviation. The SDRT is the average of this SD of all participants; SDMT (Standard deviation of movement time) is the analogue measure for movement time. So, for all measures a low score is a good score and a high score is a bad score.

In the study of Vernon and Jensen (1984) different reaction time measures were used. In this study the measures DIGIT (Speed of processing), SD2 & SA2 (Speed of retrieval of information from long-term memory), DT2 and DT3 (Efficiency of short term memory storage and processing), and RT (Simple and choice reaction time or decision making) were used. For more information, we refer to the original article of Vernon and Jensen (1984).

The purpose of this study is to determine whether the correlation between the magnitude of g loadings and difference scores on a set of reaction time measures between Whites and lower-IQ groups, and Whites and higher-IQ groups is strongly positive in sign. We will test this by conducting two full-fledged psychometric meta-analyses on studies that reported reaction time measures for Whites, Blacks, and Asians.

Results

To test whether differences in reaction time between higher-IQ groups and lower-IQ groups are related to g we conducted meta-analyses on the correlations d x g between Asians and Whites, and Whites and Blacks. The meta-analysis on the correlation d x g between Asians and Whites included eight data points. Figure 6 depicts a scatterplot for the obtained correlations d x g and the harmonic mean. The outcomes of the meta-analysis are reported in Table 7. It shows the number of studies (K), the number of participants (N), the bare-bones meta-analytic correlation d x g (r), the standard deviation of this correlation (SDr), the meta-analytic correlation d x g corrected for four statistical artifacts (rho-4), the SD of this correlation (SDrho-4), correlation d x g corrected for all five statistical artifacts (rho-5), the percentage of variance explained by all artefactual errors (%VE), and the 80% confidence interval (80%CI). The meta-analysis yields a correlation d x g with a value of -.64 with 1% of explained variance only. This percentage of variance explained is so low, we tested the origin of comparisons as a moderator. Most of the values d x g derived from reaction time measures reported in the same study. However, we did also compute vectors of reaction time measures derived from different studies. When we exclude the latter comparisons, the rho-5 becomes .48. The percentage of explained variance is still low (3%).

The meta-analysis on the correlation d x g for Black-White differences included six data points. Figure 7 depicts a scatterplot between the obtained correlations d x g and the harmonic means. The outcomes of the meta-analysis are reported in Table 8. It shows the number of studies (K), the number of participants (N), the bare-bones meta-analytic correlation d x g (r), the standard deviation of this correlation (SDr), the meta-analytic correlation d x g corrected for four statistical artifacts (rho-4), the SD of this correlation (SDrho-4), correlation d x g corrected for all five statistical artifacts (rho-5), the percentage of variance explained by all artefectual errors (%VE), and the 80% confidence interval (80%CI). The meta-analysis yields a correlation d x g with a value of .15 with 1% of explained variance only. This percentage of variance explained is so low, we tested the origin of comparisons as a moderator. Most of the values d x g derived from reaction time measures reported in the same study. However, we did also compute vectors of reaction time measures derived from different studies. When we exclude the latter comparisons, the rho-5 becomes .37. However, there is virtually no increase in explained variance.

Conclusion

The present study was designed to test whether differences in reaction time measures between Whites and lower-IQ groups, and Whites and higher-IQ groups are related to general intelligence. We intended to broaden the datasets by combining reaction time scores derived from different studies. However, since reaction time scores presumably are vunerable to even slight changes in the measurement apparatus, we decided to base our conclusions on the results of original studies. The results of the analysis on data derived from original studies offer moderate support for our hypothesis. The meta-analytical correlation d x g for reaction time differences between Asians and Whites had a moderate value of value of rho-5 = .48.

Study 1b: Differences Between Germans and Immigrants

To test whether there is a strong positive correlation between the magnitude of g loadings and the difference on IQ subtests scores between migrant and German children in Germany, an analysis was performed on the data from one study that reported IQ scores of subtests from migrant children in Germany.

Results

The results of the study on the correlation between g loadings and the score differences between German and migrant children (d) are shown in Table 9. The Table gives data derived from one study, with participants numbering 218. It also lists the reference for the study, the cognitive ability test used, the correlation between g loadings and d, and the sample size. The correlation is substantially positive.

Conclusion

The goal of this study was to test whether differences in IQ profile between migrant and German children in Germany are strongly related to general intelligence, or g. We obtained a substantially high correlation d x g of .68. Findings from this single study are in line with previous meta-analyses on group differences. We therefore conclude that our analysis, although limited by a modest sample size, offers further support for the Hypothesis that differences in IQ between the Western population and non-Western immigrants are related to g, and, in consequence, are stable over time. However, additional studies to confirm this result are required.

Study 1c: Differences between European Jews and Oriental Jews, and European Jews and non-Jewish Whites

The purpose of this study is to determine whether the correlation between the magnitude of g loadings and difference scores on IQ subtest scores between European Jews and Oriental Jews, and between European Jews and non-Jewish Whites is strongly positive in sign. We will test this by performing an exploratory psychometric meta-analysis on studies that reported IQ-subtests scores from Jews.

Results

The results of the studies on the correlation between g loadings and the score differences between non-Jewish Whites and European Jews (d) are shown in Table 10. The Table shows data derived from four studies, with participants numbering a total of 302. It also lists the reference for the study, the cognitive ability test used, the correlation between g loadings and d, and the sample size. The correlations are positive in sign and substantial in magnitude. Table 11 presents the results of the bare-bones meta-analysis of the four data points. The preschool children tested by Levinson (1959) and Levinson (1960) had a mean age of about 5.5 years. The comparison group, that is the lowest age range from the WISC, however, was two years older. Therefore, we conducted an additional analysis without these two data points. Table 11 shows the number of correlation coefficients (K), total sample size (N), the true correlation (rho), and its standard deviation (Sdr). The last column presents the percentage of variance explained by sampling errors (%VE). The analysis of all four data points yields an estimated correlation (rho) of .52, with 11.50% of the variance in the observed correlations explained by sampling errors. When excluding the aforementioned studies from the analysis, the estimated correlation (rho) rises to .80 with 82% of variance explained by sampling errors.

The results of the studies on the correlation between g loadings and the score differences between European Jews and Oriental Jews (d) are shown in Table 12. The Table reports data derived from one study, with participants numbering a total of 870. It also lists the reference for the study, the cognitive ability test used, the correlation between g loadings and d, and the sample size. The correlations are positive in sign and substantial in magnitude. Table 13 presents the results of the bare-bones meta-analysis of the four data points. It shows the number of correlation coefficients (K), total sample size (N), the true correlation (rho) and their standard deviation (SDr). The last column presents the percentage of variance explained by sampling errors (%VE). The analysis of four data points yields an estimated correlation (rho) of .70, with 40.81% of the variance in the observed correlations explained by sampling errors.

Analysis of Verbal and Performance component of the Wechsler scale. We obtained substantially high correlations d x g and might therefore conclude that differences between European Jews and non-Jewish Whites, and differences between European Jews and Oriental Jews are indeed related to differences in general intelligence, or g. Since European Jews were also found to have better scores on verbal tests than on performance tests we also want to include an analysis with regard to scores on verbal and performance IQ tests included in the test battery. Since all studies report scores of Wechsler tests, we decided to compare mean d scores on tests of the Verbal IQ scale with mean d scores of tests of the Performance IQ scale. Table 14 presents the results of the bare-bones meta-analysis of two data points for differences between European Jews and non-Jewish Whites on the Verbal scale and the Performance scale. It shows the number of d scores (K), total sample size (N), the true effect size (dt) and their standard deviation (SDd). The last column presents the percentage of variance explained by sampling errors (%VE). The analysis of two data points yields an estimated effect size (dt) on the Verbal scale of 1.42 with 100% of the variance in the observed effect sizes explained by sampling errors. On the Performance scale, we found an estimated effect size of .26 with 100% of the variance in the observed effect size explained by sampling errors. These results are visualized in Diagram 1.

Table 15 presents the results of the bare-bones meta-analysis of two data points for differences between European Jews and Oriental Jews on the Verbal scale and the Performance scale. It shows the number of d scores (K), total sample size (N), the true effect size (dt) and its standard deviation (SDd). The last column presents the percentage of variance explained by sampling errors (%VE). The analysis of two data points yields an estimated effect size (dt) on the Verbal scale of .71 with 100% of the variance in the observed effect sizes explained by sampling errors. On the Performance scale, we found an estimated effect size of .55 with 100% of the variance in the observed effect size explained by sampling errors. The results are visualized in Diagram 2.

Conclusion

The goal of this study was to explore whether differences in IQ profile between European Jews and non-Jewish Whites, and European Jews and Oriental Jews have a strong correlation with general intelligence, or g. We obtained a meta-analytic correlation of .80 for differences between European Jews and non-Jewish Whites, and a meta-analytic correlation of .70 for differences between European Jews and Oriental Jews. As these findings are based on only a limited amount of studies with a rather small total N it is not possible to draw strong conclusions. Still, this study offers support for the Hypothesis that IQ group differences between European Jews and non-Jewish Whites and European Jews and Oriental Jews are related to general intelligence, and, in consequence, should be stable over time. A further analysis of effect sizes on differences between Verbal scale subtests and Performance scale subtests, revealed a meta-analytic effect size of 1.42 for differences between European Jews and non-Jewish Whites on the Verbal scale and a meta-analytic effect size of .27 for differences between European Jews and non-Jewish Whites on subtests of the Performance scale. Since the effect size for the Verbal scale is more than five times larger than the effect size of the Performance scale, we can conclude that differences in IQ profile between European Jews and non-Jewish Whites are substantially stronger for Verbal tests than for Performance tests. Differences in effect size between the Verbal and the Performance scale between European Jews and Oriental Jews were negligible.

Study 1d: Differences Between Jews and Arabs

The purpose of this study is to determine whether the correlation between the magnitude of g loadings and difference scores on IQ subtest scores between Jewish and Arab groups in Israel is strongly positive in sign. We will test this by performing an exploratory psychometric meta-analysis on studies that reported IQ scores of at least seven subtests from Jews and Arabs residing in Israel.

Results

The results of the studies on the correlation between g loadings and the score differences between Jews and Arabs (d) are shown in Table 16. The Table gives data derived from three studies, with participants numbering a total of 1443. It also lists the reference for the study, the cognitive ability test used, the correlation between g loadings and d, and the sample size. The correlations are small and negative. Table 17 presents the results of the bare-bones meta-analysis of the five data points. It shows the number of correlation coefficients (K), total sample size (N), the true correlation (rho) and their standard deviation (SDr). The last column presents the percentage of variance explained by sampling errors (%VE). The analysis of five data points yields an estimated correlation (rho) of -.25, with 216.96% of the variance in the observed correlations explained by sampling errors.

The analysis of data points yielded estimated effect sizes, with a percentage of variance explained by sampling errors larger than 100. This phenomenon is called “second-order sampling error”, and results from the sampling of studies in a meta-analysis. Percentages of variance explained greater than 100 are not uncommon when only a limited number of studies are included in an analysis. The proper conclusion is that all the variance is explained by statistical artifacts (see Hunter & Schmidt, 2004, pp. 399-401, for an extensive discussion).

Analysis of verbal and performance component of the Wechsler scale. The bare-bones meta-analysis of the correlation d x g between Jews and Arabs yielded a correlation that points to a small negative relationship of IQ differences between Jews and Arabs, and general intelligence. We also conducted an analysis of difference scores on the Verbal and the Performance scale. We averaged d scores of subtests for the Verbal and the Performance scale in every study that uses a Wechsler test and conducted a bare-bones meta-analysis of effect sizes with the average d score of each study as data points. Table 18 presents the results of the bare-bones meta-analysis of 11 data points for the Verbal and the Performance scale. It shows the number of d scores (K), total sample size (N), the true effect size (dt) and their standard deviation (SDd). The last column presents the percentage of variance explained by sampling errors (%VE). The analysis of five data points yields an estimated effect size (d) of .42 for differences on the Verbal scale, with 195.45% of the variance in the observed effect sizes explained by sampling errors. For differences on the Performance scale we received an estimated effect size of .61, with 96.49% of the variance in the observed effect sizes explained by sampling errors.

Conclusion

In the study on differences between the IQ profile of Jews and Arabs, we expected a strong correlation d x g. The results of this analysis clearly show no support for this hypothesis. The correlation d x g is small and negative in sign, which points to a rather minor role of general intelligence as an explaining factor for differences in IQ between Jews and Arabs. A further analysis on subtest scores on the Verbal and the Performance scale of Wechsler tests yielded comparable effect sizes for both scales. In conclusion, this study does not support the Hypothesis that group differences in IQ between Jews and Arabs residing in Israel can be explained by differences in general intelligence.

Study 2: Subgroup Differences

Study 2a: Explorative Comparison of School Types

To test whether there is a strong positive correlation or a negligible correlation between the magnitude of g loadings and the differences on IQ subtest scores of children of different school types, an exploratory psychometric meta-analysis was performed on a number of studies that reported IQ scores of at least seven subtests from children of different school types.

Results

The results of the studies on the correlation between g loadings and the score differences between school types (d) are shown in Table 19. The Table reports data derived from three IQ battery manuals and one study, with participants numbering a total of 9849. It also lists the reference for the study, the cognitive ability test used, the correlation between g loadings and d, and the sample size. The correlations range from substantially negative to extremely positive. Table 20 presents the results of the bare-bones meta-analysis of 32 data points. It shows the number of correlation coefficients (K), total sample size (N), the true correlation (rho) and their standard deviation (SDr). The last column presents the percentage of variance explained by sampling errors (%VE). The analysis of all data points yields an estimated correlation (rho) of .20, with 3.01% of the variance in the observed correlations explained by sampling errors. However, it is clear that the comparison between Mavo 2 and Lbo 2 in the study by Evers and Lucassen (1983) is an extreme outlier. Taking the reduced sample of 31 studies, the value of r = -.42 is more than three SD below the average sample-sized weighted correlation of .29. Taking out this one extreme outlier increased the percentage of variance to a value of 5.85.

Conclusion

An earlier meta-analysis by te Nijenhuis et al. (2007) showed a correlation d x g of +1 for differences in IQ profile between gifted individuals and standardized groups and a correlation d x g of .74 for differences in IQ profile between mentally retarded individuals and standardized groups. We expected that differences between school types would result in smaller IQ differences, which presumably are not substantially related to general intelligence. The results of our analysis show a positive meta-analytic correlation d x g of rho = .29, but only 5.85% variance explained by sampling errors. Based on these results we can conclude that the relationship between differences in IQ profile of different school types and general intelligence is modest, but not as strong as in the meta-analyses on gifted and mentally retarded; differences might be more pronounced on lower levels of the intelligence hierarchy.

Study 2b: Explorative Comparison of Religious Groups

The analysis was performed on the data from a study that reported IQ scores of seven subtests from different religious groups.

Results

The results of the studies on the correlation between g loadings and the score differences between different religious groups in the Netherlands (d) are shown in Table 21. The Table reports data derived from one study, with participants numbering a total of 1913. It also lists the reference for the study, the cognitive ability test used, the correlation between g loadings and d, and the sample size. The correlations show no clear pattern with regard to magnitude and sign. Table 22 presents the results of the bare-bones meta-analysis of the six data points. It shows the number of correlation coefficients (K), total sample size (N), the true correlation (rho) and their standard deviation (SDr). The last column presents the percentage of variance explained by sampling errors (%VE). The analysis of both data points yields an estimated correlation (rho) of -.21, with 1.31% of the variance in the observed correlations explained by sampling errors.

Conclusion

The study on the relationship between IQ profiles of different religious groups and general intelligence was of exploratory nature only. Based on a sample size of 1913 we obtained a meta-analytic correlation d x g of rho = -.21 with 1.31% variance explained by sampling errors. We therefore conclude that differences between religious groups in the Netherlands are not related to general intelligence.

Study 3: Heritability

h² Reaction Time Measures

To test whether the herditability coefficients of reaction time measures are related to general intelligence, or g, we conducted a bare-bones meta-analysis on two studies. To compute the correlation h² x g we had to obtain g loadings of RT measures. Since the correlation matrix of RT scores was not an option, we used the correlation of RT measures with IQ batteries like the ASVAB and the WISC-R as an estimate for the g loadedness of the RT measures.

Results

The results of the studies on the correlation between g loadings and h² of RT measures are shown in Table 23. The Table gives data derived from two studies, with participants numbering a total of 389. It also lists the reference for the study, the reaction time measures used, the correlation between h² and g loadings of reaction time measures, and the sample size. Both correlations are substantially positive. Table 24 presents the results of the bare-bones meta-analysis of the two data points. It shows the number of correlation coefficients (K), total sample size (N), the true correlation (rho) and their standard deviation (SDr). The last column presents the percentage of variance explained by sampling error (%VE). The analysis of data points yields an estimated correlation (rho) of .51, with 65.22% of the variance in the observed correlations explained by sampling error.

Conclusion

Since scores of measures of cognitive processing were found to be substantially heritable and show a positive correlation with g, we explored whether the heritability pattern in reaction time measures is similar to the heritability pattern found in measures of IQ test batteries. Te Nijenhuis and Jongeneel-Grimen (2007) carried out a meta-analysis based on a sample size of 2,590 and found a correlation of +1 between the vector of g loadings and the vector of heritability coefficients of IQ batteries. Based on the assumption A) that reaction time measures have a small to substantial correlation with general intelligence and B) that reaction time measures are substantially heritable, in the present study we expected a substantial positive correlation between the vector of g loadings and the vector of heritability coefficients of reaction time measures. Based on a bare-bones meta-analysis of two data points we found a rho of .51 with 65.22% of the observed variance explained by sampling error. Although the correlation we found is not as positive as the correlation between heritability coefficients and g loadings of IQ batteries in te Nijenhuis and Jongeneel-Grimen (2007), we can conclude that the present, relatively small, dataset confirms our hypothesis that general intelligence, as it is reflected in measures of reaction time, is substantially heritable.

Study 4: Physical characteristics of the brain

Brain Volume

The purpose of this study is to determine whether the correlation between the magnitude of g loadings and the correlation between brain volume and IQ subtest scores is strongly positive in sign. We will test this by performing an exploratory psychometric meta-analysis on all studies that report the correlation of brain volume with IQ subtest scores.

Results

The results of the studies on the correlation between r (brain volume x subtest scores) (r1) and g loadings are shown in Table 25. The Table gives data derived from two studies, with participants numbering a total of 246. It also lists the reference for the study, the cognitive ability test used, the correlation r1, and the sample size. Table 26 presents the results of the bare-bones meta-analysis of the four data points. It shows the number of correlation coefficients (K), total sample size (N), the true correlation (rho) and their standard deviation (SDr). The last column presents the percentage of variance explained by sampling errors (%VE). The analysis of both data points yields an estimated correlation (rho) of .07, with 10.15% of the variance in the observed correlations explained by sampling errors. However, it is clear that the study by Flashman, Andreasen, Flaum, and Swayze (1998) is an extreme outlier: taking the reduced sample of three studies, the value of r1 = -.41 is more than four SD below the average sample-sized weighted correlation of .35. Taking out this one extreme outlier increased the percentage of variance to a value of 34.21.

Conclusion

Similar to other physical characteristics of the brain, we expected that brain volume shows a strongly positive correlation with g. After excluding one outlier from the analysis we derived at a true correlation between brain volume and g of .35. Compared to other characteristics of the brain this correlation is rather small. However, one needs to consider that the sample size of the present study was relatively small.

Study 5: Biological-environmental factors

Although Spitz (1987) suggested that biological-environmental variables should mimic the pattern found in genetic variables, previous research did not find a strong positive correlation d x g for biological-environmental variables. Therefore, we want to explore whether changes in IQ profile due to biological-environmental variables are strongly related or unrelated to differences in general intelligence. We expect increased or decreased IQ subtest scores due to biological-environmental influences to have a correlation close to zero with the subtests’ g loadings. If there is a correlation close to zero, we will also explore whether differences can be found on broad or narrow cognitive abilities measured by Wechsler tests.

Study 5a: Iodine supplementation/deficiency

The objectives of this analysis are twofold. First, we explore the correlation d x g between the magnitude of g loadings and difference scores on IQ tests of children with iodine deficiency that were supplemented with iodine and children with iodine deficiency that received a placebo. Second, we explore the correlation d x g between children deficient in iodine and a control group that is not deficient in iodine.

Results

The results of the study on the correlation between g loadings and the score differences between an iodine deficient group that was supplemented with iodine and an iodine deficient group that received a placebo (d) are shown in Table 27. The Table gives data derived from one study, with participants numbering a total of 72. It also lists the reference for the study, the cognitive ability test used, the correlation between g loadings and d, and the sample size. The correlation d x g is substantially negative in sign. The results of the study on the correlation between g loadings and the score differences between iodine deficient groups and control groups are reported in Table 28. The Table gives data derived from one study, with participants numbering a total of 196. It also lists the reference for the study, the cognitive ability test used, the correlation between g loadings and d, and the sample size. Correlations d x g range from substantially negative to substantially positive. Table 29 presents the results of the bare-bones meta-analysis of six data points. It shows the number of correlation coefficients (K), total sample size (N), the true correlation (rho) and their standard deviation (SDr). The last column presents the percentage of variance explained by sampling errors (%VE). The analysis of all data points yields an estimated correlation (rho) of .01, with 51.09% of the variance in the observed correlations explained by sampling error.

Conclusion

The goal of this study was to tests whether differences in IQ profile due to iodine deficiency are related to general intelligence. One study on the effects of iodine supplementation of iodine deficient children on change in IQ profile yielded a correlation d x g of -.54. Furthermore, a bare-bones meta-analysis of a study on differences in IQ profile between iodine deficient groups and non-iodine deficient control groups yielded a rho of .01 with 51% variance explained by sampling error. Taken together, the results of this study confirm our expectations as they indicate that the effect of Iodine supplementation/deficiency is not strongly related to g.

Study 5b: Prenatal Cocaine Exposure

In the present study, we explore the correlation d x g between the magnitude of g loadings and difference scores on IQ battery subtest between children who were exposed to cocaine prenatally and a control/standardized group. We conducted an exploratory psychometric meta-analysis on a number of studies that reported IQ scores of at least seven subtests from subjects exposed to cocaine prenatally. If we observe a negligible correlation, we will explore whether observed differences between IQ are related to broad dimensions of cognitive abilities.

Results

The results of the studies on the correlation between g loadings and the score differences between children exposed to cocaine prenatally and control groups (d) are shown in Table 30. The Table presents data derived from two studies, with participants numbering a total of 215. It also lists the reference for the study, the cognitive ability test used, the correlation between g loadings and d, and the sample size. The correlations are opposite in sign with nearly the same mild magnitude. Table 31 presents the results of the bare-bones meta-analysis of the two data points. It shows the number of correlation coefficients (K), total sample size (N), the true correlation (rho) and its standard deviation (SDr). The last column presents the percentage of variance explained by sampling errors (%VE). The analysis of both data points yields an estimated correlation (rho) of -.23, with 16.98% of the variance in the observed correlations explained by sampling errors.

Analysis of broad cognitive abilities. Since we found a rho of -.23 in a bare-bones meta-analysis based on two data points that are nearly completely the opposite in magnitude and sign, we also explored whether differences in IQ profile between children exposed to cocaine prenatally and control groups not exposed to cocaine prenatally are due to differences on broad or narrow abilities. The study of Singer et al. (2004) reports only subtests of Crystallized Intelligence and Broad Visual Perception. Therefore, a meta-analysis of effect sizes could not be conducted. Instead, we computed a weighted average of all d scores of broad abilities of both studies. This analysis yielded mean d scores shown in Table 32. These results are also visualized in Diagram 3.

Conclusion

The goal of this analysis was to explore the correlation of differences in IQ profile between children prenatally exposed to cocaine and control groups. Based on a bare-bones meta-analysis of two data points with a sample size of 215 we obtained a rho of -.23 with 17% of observed variance explained by sampling error. Clearly, this result does not indicate a strong relationship between general intelligence and IQ impairment due to prenatal cocaine exposure. A further analysis of differences on broad cognitive abilities showed that effect sizes for subtests of Fluid Intelligence are much larger than effect sizes of Crystallized Intelligence, Broad Visual Perception, and General Memory and Learning. We therefore conclude that prenatal cocaine exposure has a differential effect on broad cognitive abilities, with Fluid Intelligence as the cognitive ability on which differences are most strong. The studies included in this meta-analysis both have shortcomings. One study only reported six IQ subtests. We stated earlier to include only studies with at least seven subtests. We deviated from this rule, because there was only a small amount of studies available. However, the correlation in this study could be different if seven subtests were reported. Participants in the other study were also more likely to be exposed to other drugs than cocaine, for instance, marijuana and alcohol. Therefore, the effects on IQ subtests scores are not solely due to the use of cocaine.

Study 5c: Fetal Alcohol Syndrome

To explore the correlation d x g between the magnitude of g loadings and IQ subtest scores of individuals who suffered from fetal alcohol syndrome, an analysis was performed on the data from a study on subjects who suffered from FAS. If we observe a negligible correlation, we will explore whether observed differences between IQ are related to broad dimensions of cognitive abilities. Furthermore, we test whether there is a correlation d x g close to zero between the magnitude of g loadings and IQ subtest scores of individuals with different degrees of severity of FAE/FAS. If we observe a negligible correlation, we will explore whether observed differences between IQ are related to broad dimensions of cognitive abilities.

Results

The results of the study on the correlation between g loadings and score differences between FAE/FAS and a control group are presented in Table 33. The Table gives data derived from one study, with participants numbering a total of 110. It also lists the reference for the study, the cognitive ability test used, the correlation between g loadings and d, and the sample size. The correlation is positive and small in magnitude. The results of the study on the correlation between g loadings and score differences between different degrees of FAE/FAS are presented in Table 34.

The Table gives data derived from one study, with participants numbering a total of 125. It also lists the reference for the study, the cognitive ability test used, the correlations between g loadings and d, and the sample size. The correlations are small and positive as well as small and negative in sign. Table 35 presents the results of the bare-bones meta-analysis of the three data points. It shows the number of correlation coefficients (K), total sample size (N), the true correlation (rho) and their standard deviation (SDr). The last column presents the percentage of variance explained by sampling errors (%VE). The analysis of all data points yields an estimated correlation (rho) of .12, with 83.04% of the variance in the observed correlations explained by sampling errors.

Analysis of broad cognitive abilities. The results of both studies revealed a small positive correlation d x g. Such a result does not indicate a strongly positive relationship between differences in IQ profile caused by fetal alcohol syndrome and general intelligence. To explore differences on broad abilities we computed average d scores for all subtests with regard to the broad ability they measure. Table 36 shows the mean d scores and sample sizes of the study of differences between FAE/FAS groups and a control group for the broad abilities Fluid intelligence, Crystallized Intelligence, Broad Visual Perception, and General Memory and Learning. These results are also shown in Diagram 5. Table 37 shows the mean d scores and sample sizes of the study of differences between degrees of severity of FAS for the broad abilities Fluid intelligence, Crystallized Intelligence, Broad Visual Perception, and General Memory and Learning. These differences are also depictured in Diagram 6.

Conclusion

The goal of this study was to explore the correlation of differences in IQ profile between individuals that suffer from fetal alcohol syndrome and control groups. Results showed a correlation d x g of .16 for differences between an FAE/FAS group and a control group, and a meta-analytic correlation d x g of .12 for differences between degrees of severity of FAE/FAS. Based on a relatively small sample size we can conclude that differences in IQ profile between an FAE/FAS group and a control group as well as differences between degrees of severity of FAE/FAS are nearly completely unrelated to general intelligence, or g. A further analysis of broad cognitive abilities revealed that differences between an FAE/FAS group and a control group are stronger for tests of Fluid Intelligence, Crystallized Intelligence, and General Memory and Learning, and less strong for tests of Broad Visual Perception. Concerning the difference in IQ profile between different degrees of severity of FAE/FAS we found that differences on test of Crystallized Intelligence and General Memory and Learning are stronger than differences on tests of Fluid Intelligence and Broad Visual Perception. We therefore conclude that a) IQ impairment due to fetal alcohol syndrome does affect the broad cognitive abilities Fluid Intelligence, Crystallized Intelligence, and General Memory and Learning more strongly than Broad Visual Perception. b) IQ impairment due to a more severe degree of FAE/FAS does affect the cognitive abilities Crystallized Intelligence and General Memory and Learning more strongly than Broad Visual Perception and Fluid Intelligence. This study has several shortcomings. First, a higher degree of severity of FAE/FAS does not necessarily correspond to a lower IQ score. The Full scale IQs of FAE, FAS 1stº, and FAS 2ndº were nearly the same. Only FAS 3rdº had a lower Full scale IQ score than the other degrees. Since we could only make a comparison between FAE, FAS 1stº, FAS 2ndº, and FAS 3rdº, the implications of the results of the correlation d x g between different degrees of severity are rather limited. We should also consider that a diagnosis with FAE/FAS does not lead to a pronounced IQ profile at all. The FAE group had a Full Scale IQ range of 46 to 117 with a mean of 77 and the FAS 1stº group had a Full scale IQ score range from 44 to 132 with a mean of 79. Therefore, these groups comprised individuals that can be considered severely mentally retarded on the one hand and individuals that can be considered gifted on the other hand. Although we can expect that there are individuals that would have had a rather low IQ score without the effect of FAE/FAS and also individuals that would have been extremely gifted without the effects of FAE/FAS, this is an enormous IQ range for a diagnosis.

Study 5d: Air Pollution

To explore the correlation between differences in IQ profile due to exposure with polluted air and general intelligence, or g, we computed the correlation d x g between children from a highly polluted and a lowly polluted city. If we observe a negligible correlation d x g, we will explore whether differences in intelligence lie on broad cognitive abilities.

Results

The result of the study on the correlation between g loadings and the score differences between children exposed to high levels of air pollution and a control group (d) are shown in Table 38. The Table gives data derived from one study, with participants numbering a total of 55. It also lists the reference for the study, the cognitive ability test used, the correlation between g loadings and d, and the sample size. The correlation is small and negative in sign.

Analysis of broad cognitive abilities. We did only obtain one study on the effects of air pollution on IQ profile, so our result is based on a very small sample size and cannot be considered representational for the whole population of children exposed to air pollution. Nevertheless, we computed the correlation d x g, which resulted in a value of -.17. Since this correlation does not suggest a strong positive relationship between general intelligence and effects of air pollution on IQ subtest scores, we computed mean d scores for each broad cognitive ability. The results of this analysis are shown in Table 39. Overall, mean d scores are not very high. Still positive d scores were obtained for all broad cognitive abilities. The highest mean d score was observed for tests of Fluid Intelligence. Slightly lower difference score were obtained for tests of Crystallized Intelligence, Broad Visual Perception, and General Memory and Learning. These results are also shown in Diagram 7.

Conclusion

The study on the relationship between differences in IQ profile caused by air pollution and general intelligence is different to the previous studies on the effect of prenatal cocaine exposure, and fetal alcohol syndrome on general intelligence, respectively, because in the air pollution study, children were not exposed to a toxin prenatally. Still, we did not obtain results that suggest a strong positive relationship between general intelligence and IQ impairment due to the biological-environmental variable air pollution. Therefore, this study does suggest that the assumption of Spitz (1987), that biological-environmental variables mimic the pattern of genetic variables, is wrong. A further analysis of broad cognitive abilities indicated that the overall effect of air pollution on IQ impairment is rather weak. Nevertheless, the strongest impairment of IQ was found on tests of Fluid Intelligence and a slightly less strong impairment was found on tests of Crystallized Intelligence, Broad Visual Perception, and General Memory and Learning. Since our results are based on only one study we should not overestimate the theoretical implications. Still, in the context of the results of studies on other biological-environmental variables as iodine deficiency/supplementation, prenatal cocaine exposure, and fetal alcohol syndrome, we can conclude that the results on the pollution and children from an area with low levels of air pollution on tests of Fluid Intelligence, Crystallized Intelligence, Broad Visual Perception, and General Memory and Learning impairment of IQ by air pollution fit into the pattern of a rather minor role of general intelligence as an explaining factor of IQ differences between average groups and groups exposed to biological-environmental variables that have a detrimental effect on IQ.

Study 5e: Traumatic Brain Injury

To explore the correlation d x g between the magnitude of g loadings and IQ subtest scores of individuals who suffered from TBI, an exploratory psychometric meta-analysis was performed on a number of studies that reported IQ scores from TBI subjects. If we observe a negligible correlation d x g, we will explore whether differences in intelligence lie on broad cognitive abilities.

Results

The results of the studies on the correlation between g loadings and the score differences between a TBI group and control/standardized groups (d) are shown in Table 40. The Table reports data derived from nine studies, with participants numbering a total of 629. It also lists the reference for the study, the cognitive ability test used, the correlation between g loadings and d, and the sample size. The correlations show no clear pattern with regard to magnitude or sign. Table 41 presents the results of the bare-bones meta-analysis of the 14 data points. It shows the number of correlation coefficients (K), total sample size (N), the true correlation (rho) and their standard deviation (SDr). The last column presents the percentage of variance explained by sampling errors (%VE). The analysis of 14 data points yields an estimated correlation (rho) of -.07, with 35.43% of the variance in the observed correlations explained by sampling error. Sample sizes were highly comparable, which most likely led to much lower %VE.

Analysis of broad cognitive abilities. The bare-bones meta-analysis of the correlation d x g between TBI groups and control groups did not yield a correlation that points to a substantial positive relationship of IQ impairment due to TBI and general intelligence. Therefore, we also conducted an analysis of difference scores on broad cognitive abilities. We averaged d scores of subtests for every broad ability in every study that used a Wechsler test and conducted a bare-bones meta-analysis of effect sizes with the average broad ability d score of each study as data points. Table 42 presents the results of the bare-bones meta-analyses of 11 data points for Fluid Intelligence, Crystallized Intelligence, Broad Visual Perception, and General Memory and Learning. It shows (from left to right): the number of d scores (K), total sample size (N), the true effect size (dt) and their standard deviation (SDd). The last column presents the percentage of variance explained by sampling errors (%VE). The analysis of 11 data points of differences on tests of Fluid Intelligence yields an estimated effect size (dt) of .53, with 79.73% of the variance in the observed effect sizes explained by sampling errors. The analysis of 11 data points of differences on tests of Crystallized Intelligence yields an estimated effect size (dt) of .49, with 132.90% of the variance in the observed effect sizes explained by sampling errors. The analysis of 11 data points of differences on tests of Broad Visual Perception yields an estimated effect size (dt) of .52, with 77.34% of the variance in the observed effect sizes explained by sampling errors. Finally, the analysis of 11 data points of differences on tests of General Memory and Learning yields an estimated effect size (dt) of .65, with 105.54% of the variance in the observed effect sizes explained by sampling errors. These results are also shown in Diagram 8.

The analysis of data points of Crystallized Intelligence and General Memory and Learning yield estimated effect sizes, with a percentage of variance explained by sampling errors larger than 100. This phenomenon is called “second-order sampling error”, and results from the sampling of studies in a meta-analysis. Percentages of variance explained greater than 100% are not uncommon when only a limited number of studies are included in an analysis. The proper conclusion is that all the variance is explained by statistical artifacts (see Hunter & Schmidt, 2004, pp. 399-401, for an extensive discussion).

Conclusion

The study on the relationship between differences in IQ profile caused by TBI reveals further evidence against the assumption of Spitz (1987), that environmental-biological variables mimic the pattern of genetic variables. Our analysis resulted in a meta-analytic correlation d x g for differences between TBI groups and control groups of -.07. Therefore, general intelligence is nearly completely unrelated to differences in IQ profile between TBI and control groups. A further analysis of differences on broad cognitive abilities revealed higher difference scores on test of General Memory and Learning and roughly the same magnitude of difference scores on tests of Fluid Intelligence, Crystallized Intelligence, and Broad Visual Perception. Although TBI does clearly lead to an impairment of all cognitive abilities, the impairment of General Memory and Learning is most severe. These results present further evidence against a major role of general intelligence in IQ impairment due to biological-environmental variables.

Study 5f: Malnutrition

To explore the correlation between the magnitude of g loadings and IQ scores of malnourished individuals, an exploratory meta-analysis was performed on all studies that reported IQ scores from malnourished individuals. If we obtain a negligible correlation d x g, we will explore whether differences lie on broad cognitive abilities.

Results

The result of the study on the correlation between g loadings and the score differences between a malnourished group and a control group (d) are shown in Table 43. The Table gives data derived from one study, with participants numbering a total of 51. It also lists the reference for the study, the cognitive ability test used, the correlation between g loadings and d, and the sample size. The correlation is small and negative in sign.

Analysis of broad and narrow abilities. The correlation d x g between a malnourished group and a control group does not point to an either pronounced positive or negative relationship between the IQ profile of a malnourished group and general intelligence. Therefore, we did also compute mean d scores for all broad abilities. Table 44 shows the results of this computation.

Overall, d scores are not very high. However, Fluid Intelligence and Broad Visual Perception seem slightly more affected by malnutrition than Crystallized Intelligence and General Memory and Learning. These results are also shown in Diagram 9.

Conclusion

The study on the relationship between differences in IQ profile caused by malnutrition reveals further evidence against the assumption of Spitz (1987), that environmental-biological variables mimic the pattern of genetic variables. Our analysis resulted in a correlation d x g for IQ differences between a malnourished group and a control group of -.15. Therefore, general intelligence is nearly completely unrelated to differences in IQ profile between a malnourished and a control group. A further analysis of differences on broad cognitive abilities revealed higher difference scores on test of Fluid Intelligence and Broad Visual Perception and roughly the same lower magnitude of difference scores on tests of Crystallized Intelligence, and General Memory and Learning. Overall, the magnitude of difference scores is rather low. These results present further evidence against a major role of general intelligence in IQ impairment due to biological-environmental variables.

General Conclusion

The goal inherent to all studies in this section was to test whether differences in IQ due to biological-environmental variables show no strong correlation with general intelligence. Table 45 gives an overview of all correlations between g loadings and differences on biological-environmental variables we obtained in our studies. With one exception, mean correlations are very close to zero. Based on the results of seven biological-environmental variables we can conclude that Spitz assumption is wrong, and biological-environmental variables do not mimic the pattern of genetic variables. Group differences due to biological-environmental variables are rather unrelated to general intelligence, or g. The effects of biological-environmental variables should therefore be studied at the level of broad abilities.

Study 6: Aging and Autism

Study 6a: Aging

To explore the correlation between the magnitude of g loadings and the decline on IQ subtest scores of individuals due to aging, an exploratory analysis was performed on data of a longitudinal study (Schaie and Willis, 1993) and the standardization study of the Spanish WAIS-III (Juan-Espinosa, 2002). If we observe a strong correlation d x g this would indicate that general intelligence declines over the course of a lifetime.

Results

The results of the longitudinal study on the correlation between g loadings and the score differences between different age groups (d) are shown in Table 47. The Table reports data derived from one study, with participants numbering a total of 2,677. It also lists the reference for the study, the cognitive ability test used, the correlation between g loadings and d (r1), the correlation between g loadings and d when subtests that measure crystallized intelligence are left out (r2), and the sample size. In general, correlations are substantially positive. Table 48 presents the results of the analysis of the standardization study of the Spanish WAIS-III, with participants numbering a total of 850. It also lists the reference for the study, the cognitive ability test used, the correlation between g loadings and d (r1), the correlation between g loadings and d when subtests that measure crystallized intelligence are left out (r2), and the sample size. The correlations are highly positive. Table 49 reports the results of the bare-bones meta-analysis of 23 data points from both studies. It shows the number of correlation coefficients (K), total sample size (N), the true correlation (rho) and their standard deviation (SDr). The last column presents the percentage of variance explained by sampling errors (%VE). The table reports meta-analytical results for correlations d x g including all subtests and for correlations only including subtests that do not belong to the Crystallized Intelligence domain. The analysis of all data points yields an estimated correlation (rho) of .45 for all subtest, with 28% of observed variance explained by sampling errors, and a correlation of rho = .59 when subtests of crystallized intelligence are left out, with 15% of observed variance explained by sampling error.

Type of study as a moderator. Since the two studies included in the analyses were very different in nature, we did explore whether the meta-analytic correlation rho differs between studies. Table 50 reports the results of this analysis. It shows the number of correlation coefficients (K), total sample size (N), the true correlation (rho) and their standard deviation (SDr). The last column presents the percentage of variance explained by sampling errors (%VE). The true correlation rho when subtests of Crystallized Intelligence are included and when they are excluded differed extremely between the longitudinal study and the standardization study of the WAIS-III. When subtests measuring Crystallized Intelligence are included in the analysis the true correlation rho for the WAIS-III is only .37 compared to a correlation of .48 in the longitudinal study. This result changes dramatically when subtest measuring Crystallized Intelligence are left out of the comparison. Then the true correlation rho of the standardization study of the WAIS-III rises to .76. The true correlation rho for the longitudinal study enhances to .53, only. Clearly, the effects of aging on general intelligence are very different in the longitudinal study and in the standardization study of the WAIS-III. The effect of the exclusion of subtests of Crystallized Intelligence on the true correlation rho is shown in Diagram 12.

Age range as a moderator. A close inspection of the data from the longitudinal study reveals that when both groups in the comparison are above age 60 the correlations are much stronger. So, we tested the age range as a moderator. Table 51 reports the results of this analysis. It shows the number of correlation coefficients (K), total sample size (N), the true correlation (rho) and their standard deviation (SDr). The last column presents the percentage of variance explained by sampling errors (%VE). When subtests that measure Crystallized Intelligence are included the true correlation rho for comparisons between groups younger than 60 is .24. For groups older than 60 this correlation rises to .55. When subtests that measure Crystallized Intelligence are excluded from the analysis the true correlation rho for comparisons between groups younger than 60 is .34. This correlation also shows a dramatic increase to .52 when computed for comparisons between groups older than 60. The effect of age range on the true correlation rho is shown in Diagram 11.

Age cut-off point as a moderator. A close inspection of the data of the longitudinal study revealed that the correlations d x g become higher when the age of the second group increased. To test whether the age cut-off point moderates the true correlation rho, eight cut-off points were selected. An age cut-off point is always one year above the age of the age groups reported in the study; except for the oldest age group, of course. The results of this analysis are shown in Table 52 and Diagram 12. Table 52 shows the number of correlation coefficients (K), total sample size (N), the true correlation (rho) and their standard deviation (SDr). The last column presents the percentage of variance explained by sampling errors (%VE). Clearly, the true correlation rho increases when the age cut-off point increases. This is true for comparisons between age groups when tests of Crystallized Intelligence are included, as well as when they are excluded.

Conclusion

The goal of this study was to explore the correlation between the declines in IQ scores due to aging and general intelligence. Since score decline in old age on tests of Crystallized Intelligence is a lot lower than score decline on tests of Fluid Intelligence, we expected that the correlation d x g is not strongly positive. Based on two studies with a total sample size of 3,527 we obtained a meta-analytic correlation d x g of rho = .45 between groups younger than 68 and groups older than 68, when subtests that measure Crystallized Intelligence are included. When these subtests are removed from the analysis the true correlation rho increases to .59. These results indicate a modest positive relationship between decline in IQ scores due to aging and general intelligence, which becomes stronger when subtests of Crystallized Intelligence are excluded. A further analysis of moderator variables in the longitudinal study showed additional support for this conclusion. First, correlations d x g between groups older than 60 were substantially larger than correlations between groups younger than 60. Second, correlations d x g increase when the cut-off point for age comparisons increases. In conclusion, the correlation between IQ decline due to aging and general intelligence is not strongly positive. This is in line with our expectations. However, the correlation d x g becomes substantially high when test of Crystallized Intelligence are excluded.

Study 6b: Autism

To test whether there is a strong positive correlation or a negligible correlation between the magnitude of g loadings and difference scores on IQ subtests between an autistic and a comparison group, an exploratory psychometric meta-analysis was performed on a number of studies that reported IQ scores of at least seven subtests from subjects with autism. If we do not find a strongly positive correlation, we will also explore whether differences lie on broad or narrow cognitive abilities. Autistic children tend to show less deviation on tests of perceptual motor organization skills and rote memory skills, but more deviation on tests of verbal abstraction and verbal comprehension (Lincoln, et al., 1988). In almost all studies we analyzed cognitive abilities were tested with a Wechsler test. In Wechsler tests Perceptual motor organization is administered with the tests: Picture Completion, Picture Arrangement, Block Design, and Object Assembly. Rote memory skills are measured with the test Digit Span. Verbal abstraction and verbal comprehension are measured with the tests Vocabulary, Similarities, Information, Arithmetic and Comprehension. If the above statement is true, we would expect less deviation on the subtests Picture Completion, Picture Arrangement, Block Design, Object Assembly, and Digit Span and more deviation on the subtests Vocabulary, Similarities, Information, Arithmetic, and Comprehension.

Results

The results of the studies on the correlation between g loadings and the score differences between autistic individuals and non-autistic individuals (d) are shown in Table 53. The Table gives data derived from 12 studies, with participants numbering a total of 249. It also lists the reference for the study, the cognitive ability test used, the correlation between g loadings and d, and the sample size. The correlations range from substantially positive to substantially negative. Table 54 presents the results of the bare-bones meta-analysis of 12 data points. It shows the number of correlation coefficients (K), total sample size (N), the true correlation (rho) and their standard deviation (SDr). The last column presents the percentage of variance explained by artifactual errors (%VE). The analysis of all data points yields an estimated correlation (rho) of .00, with 42.88% of the variance in the observed correlations explained by sampling errors. However, it is clear that the studies of Allen, Lincoln, and Kaufman (1991b), and Bartak, Rutter, and Cox (1975) are extreme outliers: taking the reduced sample of 10 studies, the values of r = -.67 and .52 are more than three SD below the average sample-sized weighted correlation of .04. Taking out these extreme outliers increased the percentage of variance to a value of 58.94%.

Analysis of broad cognitive abilities. The bare-bones meta-analysis of the correlation d x g between autistic groups and control groups did yield a meta-analytic correlation of rho = .04. This implies that differences in IQ profile between autistic groups and control groups are completely unrelated to general intelligence. To test whether differences in IQ profile lie on lower levels of the intelligence hierarchy, we also conducted an analysis of difference scores on subtest level. We conducted a bare-bones meta-analysis of d scores for each Wechsler subtest reported in the studies. In all studies that used a Wechsler test the same subtests were reported. Table 55 presents the results of the bare-bones meta-analyses of 10 data points for 11 different Wechsler tests. It shows the number of d scores (K), total sample size (N), the true effect size (dt) and their standard deviation (SDd). The last column presents the percentage of variance explained by sampling errors (%VE). According to previous research (Lincoln, et al., 1988 ), we should find higher differences on the tests Vocabulary, Similarities, Information, Arithmetic, and Comprehension, and smaller differences on the tests Picture Completion, Picture Arrangement, Block Design, Object Assembly, and Digit Span. In general, these expectations are strongly confirmed. The results are visualized in Diagram 13.

Conclusion

The goal of this study was to explore whether differences in cognitive profile between autistic groups and control groups are related to general intelligence. Our analysis revealed a meta-analytic correlation d x g of rho = .04. Therefore, we have to conclude that IQ differences between autistic groups and control groups are completely unrelated to general intelligence. A further analysis of differences on subtest level revealed that there are only minor differences between autistic and control groups on the tests Block Design and Object Assembly, which measure perceptual motor organization, and major differences on tests that measure verbal abstraction and verbal comprehension. We therefore conclude that differences in IQ profile between autistic and control groups are not related to general intelligence, but are due to differences in in verbal comprehension and verbal abstraction.

General Conclusion and Discussion

The huge IQ gap between non-Western immigrants and ethnic Dutch has emerged as one of the primary explanations for the large differences in school and work achievement between these groups (te Nijenhuis, et al., 2004). Is there a genetic component in the IQ gap between these immigrants and ethnic Dutch? Previous meta-analyses have shown that the group differences on IQ subtests correlate almost perfectly with the cognitive complexity of these subtests; moreover, the cognitive complexity correlates perfectly with heritability and moderately to strongly with physical characteristics of the brain.

The purpose of the present study was to test a fundamental hypothesis: Only variables under genetic influence are strongly and positively related to general intelligence, or g, and variables not under genetic influence are not. We conducted analyses on a manifold of variables to test this hypothesis (see Table 66). The hypothesis was strongly supported: variables under genetic influence, namely heritabilities and most group differences, showed moderate to strong, positive correlations with g; the meta-analytical correlation of brain volume with g was weak to modest. All other phenomena showed no strong positive correlation with g.

It is concluded that group differences are moderately to strongly positively related to general intelligence, heritabilities are moderately to strongly positively related to general intelligence, and a physical characteristic of the brain, namely brain volume, is weakly to moderately positively related to general intelligence. Biological-environmental factors show weakly negative to weakly positive correlations. These findings in combination suggest a strong genetic component in group differences with regard to general intelligence, and that biological-environmental variables presumably do not affect these differences. Therefore, group differences in general intelligence should been seen as rather stable over time. Previous research on Spearman’s hypothesis showed that IQ differences between Whites and Blacks and IQ differences between non-Western immigrants and ethnic Dutch (te Nijenhuis & Dragt, 2010; te Nijenhuis & Repko, 2011; te Nijenhuis & Willigers, 2011) are primarily caused by differences in general intelligence. Concerning the role of IQ differences in debate on the integration of non-Western immigrants into the Dutch society, we can conclude, that first, there are large differences in IQ between the second immigrant generation and ethnic Dutch (te Nijenhuis, et al., 2004). Second, we showed that group differences in IQ are related to general intelligence, which is in line with the results of previous research (te Nijenhuis & Dragt, 2010; te Nijenhuis & Repko, 2011; te Nijenhuis & Willigers, 2011). Third, general intelligence is largely heritable (te Nijenhuis & Jongeneel-Grimen, 2007), and in the present study, we also showed that general intelligence as it is reflected in reaction time is heritable. Finally, we showed that environmental variables do only weakly affect general intelligence. In sum, differences in IQ between non-Western immigrants and ethnic Dutch are primarily related to general intelligence, which is under strong genetic influence, but only weakly affected by biological-environmental factors. Therefore, further IQ gains of the second generation of immigrants, should only constitute non-g related gains in IQ, which should leave the gap in school and work achievement unaffected. In consequence, I/O psychologists should find ways to deal with differences in general intelligence instead of ways of trying to change them.

An analysis of differences in reaction time between Whites and lower-IQ groups, and Whites and higher-IQ groups revealed that differences in reaction time between these groups are only weakly to moderately related to g. This finding could be due to different reasons. First, reaction time is only an indirect measure of general intelligence, and previously established moderate to substantial positive correlations between general intelligence and reaction time do not necessarily imply that also group differences in reaction time are moderately to strongly related to general intelligence. In addition, the reliability of the reaction time measures showed serious flaws, which works against finding strong results. An analysis of a study on differences in IQ profile between German and immigrant children showed a moderate to strong positive correlation d x g. Similarly, analyses of studies on differences in cognitive profile between Jews and non-Jewish Whites, and European Jews and Oriental Jews indicated a strong positive correlation d x g, too. However, an analysis on differences in IQ between Jews and Arabs residing in Israel showed a negative correlation d x g. In conclusion, we find modest support for the hypothesis that group differences in IQ are strongly related to g on reaction time measures, strong support that three other comparisons between ethnic groups are strongly on the g factor, and the comparison involving Arabs in Israel shows the first recorded exploratory meta-analytic finding of a negative link between group differences. It should also be noted that immigrants represent a sub-population of an ethnic groups, in the sense, that migrants from a certain country, do not necessarily represent a random sample from this country’s population. It could even be assumed that lower cognitive abilities of migrants compared to the average population of the home country, were a reason for low achievement in the home country, which led migrants to leave the home country. Although this is only speculation, the point is that migrants’ cognitive abilities might be structurally lower than the cognitive abilities of the average home-country population. Therefore, it should be emphasized that the comparison between populations on country level are not necessarily comparable to comparisons between whole country populations and migrant groups. Still, the comparison of country level differences between Jews and non-Jewish Whites, the comparison of within country differences between Jews of European ancestry and Jews of Oriental ancestry, and the comparison of German and migrant children lead to similar results: (Ethnic) group differences in IQ are strongly and positively related to general intelligence. Only the group differences between Jews and Arabs residing in Israel could not be explained with differences in general intelligence.

The previous group comparisons concerned ethnic groups, and additionally we explored differences between subgroups within an ethnic group. The analyses on subgroups that differ with regard to religious belief and the school type the subgroups attend, respectively, did not yield strong positive correlations, as is generally the case in comparisons between ethnic groups. In particular, differences in IQ between religious groups, namely Catholics, Protestants, and atheists, showed a correlation d x g of close to zero. Differences in IQ between school types showed a small correlation d x g. Therefore, subgroup differences do not seem to have a particularly strong relationship with g. These findings can be construed as providing support for the hypothesis that when comparing samples only group differences and generally not subgroup differences are strongly and positively related to g. However, it should be mentioned that te Nijenhuis et al. (2009) reported a rho = +1 for differences between gifted persons and average persons. Clearly, more exploratory meta-analyses are required to see which subgroup differences act like ethnic group differences.

In previous studies, genetic variables were found to have a strong positive relationship with g. Therefore, we expected that also the heritability coefficients of reaction time measures would show a strong correlation with g loadings. A bare-bones meta-analysis on two studies revealed a correlation h² x g of .51. This finding provides modest support for our hypothesis.

A range of physical characteristics of the brain was found to have a substantial correlation with g. In the present study, we explored the relationship between brain volume and g. Results indicate a modest correlation d x g. Spitz (1987) hypothesized that biological-environmental variables mimic the pattern of genetic variables. Previous studies on this topic, however, did not indicate a pronounced relationship between the biological-environmental cocaine-, lead-, and smoke-exposure. Nonetheless, it was still possible to find support for Spitz’ hypothesis in studies of other biological-environmental variables. Therefore, we explored the correlation d x g of the variables iodine supplementation/ deficiency, prenatal cocaine exposure, fetal alcohol syndrome, air pollution, traumatic brain injury, and malnutrition. The picture that emerged from all this studies is straightforward: Differences in IQ caused by these biological-environmental variables are virtually unrelated to general intelligence.

A further exploration of the psychological phenomena aging and autism revealed that IQ decline in aging has a substantial relationship with general intelligence, but the meta-analytical correlation is clearly not +1. The analyses showed that an analysis of gains on broad abilities is more likely to be helpful for understanding this phenomenon than the g factor. The IQ profile of autistic groups clearly does not correlate with general intelligence.

Limitations of the Studies

In general, many studies were based on a relatively low N (see Table 66). Clearly, more studies are needed to confirm our results. It should also be taken into consideration that the method of correlated vectors is a unidimensional approach to the nature-nurture debate in intelligence. The effect of other possible factors that might influence intelligence like SES cannot be determined by this method. The least we can conclude from the results of our analysis is that if only variables under genetic influence show a strong correlation with g, and non-genetic variables do not, the effect of non-genetic variables on g will be rather limited.

The meta-analysis on group differences in reaction time had several limitations. First, the tasks used in Jensen (1993), Jensen and Whang (1994), Ja-Song and Lynn (1992), Lynn and Holmshaw (1990), Lynn and Shegisa (1991), and Lynn, Chan, and Eysenck (1991) yielded two types of reaction time, namely reaction time (RT) and movement time (MT). The former measures the time interval between onset of stimulus and release of the home button and the latter measures the time interval between release of the home button and activation of the response button. A close inspection of reaction time differences between groups indicated that when one group scores better on the first measure it has as strong tendency to score worse on the second measure, and vice versa. Since the same pattern emerged in the reaction time and movement time measures of all three tasks, it is likely that the observed differences in reaction time are due to a response tendency, which makes the tasks less suitable as a measure of reaction time to be analyzed by the method of correlated vectors. A possible solution to this problem would be to add reaction and movement time for each group and to correlate differences between groups on these sum measures with appropriate g loadings. Although the resulting d vectors would have only six values instead of 12, confounds due to response tendency on reaction time measures would be removed. Second, we could not base the computation of g loadings on the correlation matrix of subtests scores. As a substitute procedure, we used the reaction time subtest’s correlation with the SPM to estimate the subtests’ loadings on g. Although, from a theoretical perspective, this procedure should yield a good estimate of g loadings, some of the subtests showed negative correlations with the SPM. Due to a relatively low sample size, this result is not surprising. Still, a negative g loading does not exist, and using a negative g loading in the computation of the correlation d x g, would have made the comparison useless. We decided to reverse the sign of the correlation and use the resulting positive correlation as the best estimate of the g loading. We acknowledge that this procedure is not optimal; still, it posed the best possibility to conduct the analysis under these circumstances. Better estimates of g loadings in future analyses on differences in reaction time are desirable.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s