Jan Smit, Master thesis, 2011.

Abstract

The central question addressed in this study is whether some potential biological and environmental factors might cause true differences in general mental ability (g) between groups, or just “hollow” score differences between groups.

Four bare-bones psychometric meta-analyses (MAs) were performed to test these premises. We predicted strong positive correlations between vectors of lowered IQ scores following lead exposure, prenatal cocaine exposure, and prenatal smoke exposure on the one hand; and vectors of g loadings on the other hand. We also predicted a strong positive correlation between the vector of increased IQ scores following breastfeeding on the one hand and vectors of g loadings on the other hand.

In line with our hypotheses, the meta-analyses showed correlations of .79 on breastfeeding (total N=7847) and .91 on prenatal cocaine exposure (total N=391). However, the variance explained by artifactual errors is too low to draw strong conclusions about the link between cocaine exposure and g, and about the link between breastfeeding and g. Contrary to our hypotheses, the meta-analyses showed correlations of -.19 on lead exposure (total N = 702) and -.19 on prenatal smoke exposure (total N = 443). We were not able to come up with strong theoretical explanations for these values of the correlation between lead exposure and g.

In sum, two of the four meta-analyses showed mixed support for the theory: high correlations between g loadings and effects, but little variance explained in the data points in the meta-analysis. The other two meta-analyses showed no support at all for the theory: an absence of correlations between g loadings and effects and virtually no variance in the data points explained. The amount of support for the theory in these four meta-analyses is therefore modest.

General and specific rules for inclusion

Following previous studies on the MCV (te Nijenhuis & Dragt, 2010; te Nijenhuis & Franssen, 2010; te Nijenhuis, de Pater, van Bloois, & Geutjes, 2009), we used four inclusion criteria which had to be met: 1) the cognitive batteries had to have a minimum of three composite scores or a minimum of seven subtests of the intelligence test (this to obtain a reliable estimate of the true correlation); 2) the IQ test had to be well-validated; 3) the mean composite scores for the high lead-level groups, cocaine-exposed groups, and smoke-exposed groups had to be lower than the mean scores of the comparison group (the control group or the standardization sample of the IQ test). For the breastfed group, the composite scores had to be higher than the mean scores of the comparison group; 4) only studies published in English were used.

Results

Study 1: Lead exposure

Table 2 shows the results on the correlations between a) score differences between lead exposed and average groups (d) and b) g loadings. Results are derived from 11 studies, with a total of 702 participants. It shows also the study reference, the intelligence test, the observed correlation between g and d, sample size, and mean age (and age range).

The results of the meta-analysis of 11 studies are shown in Table 3. The outcomes of the first analysis are reported in Table 1. K represents the study number, N represents the accumulated harmonic sample sizes, r represents the mean bare-bones correlation, SD_{rho}represents the standard deviation after sample size corrections, %VE represents the percentage of explained variance, and 80%CI represents the confidence interval, which gives us the expected correlation for 80% of the cases.

Our analysis gives us a bare-bones correlation of .17, and the explained variance due to artifactual errors is 4.5%.

Conclusion

Although we predicted a high correlation between lead exposure and g, the meta-analysis shows no strong link between lead level and the g factor; the 11 studies show a large variation in the value of r. It should be mentioned that there were also some studies with very little lead-level differences between the two groups in that study. However, when we left out these studies in a moderator analysis the resulting correlation did not increase in value. At this point we can find no theoretical explanations for the weak and variable correlation between lead level and the g factor.

Study 2: Breastfeeding

Table 4 shows the results on the correlations between a) score differences between breastfed and average groups (d) and b) g loadings. Results are derived from five studies, with a total of 7847 participants. It shows also the study reference, the intelligence test, the observed correlation between g and d, sample size, and mean age (and age range).

The results of the meta-analysis of five studies are shown in Table 5. K represents the study number, N represents the accumulated harmonic sample sizes, r represents the mean bare-bones correlation, SD_{rho} represents the standard deviation after sample size corrections, %VE represents the percentage of explained variance, and 80%CI represents the confidence interval. This interval gives us the expected correlation for 80% of the cases.

Our analysis gives us a bare-bones correlation of .79, and the explained variance due to artifactual errors is 20.3%.

Conclusion

When we look at the observed correlation, we can conclude from the exploratory meta-analysis that the effects of breastfeeding are quite strongly linked to the g vector. Yet, only 20.3% of the variance can be explained by artifactual errors, so most of the variance cannot be explained by artifactual errors. No strong conclusions can therefore be drawn about the link between prenatal breastfeeding and g.

In an outlier analysis, we left out the study from Kramer et al. (2008), because the comparison group also received breastfeeding, albeit to eliminate degree. However, this did not change the percentage variance explained. It may be there are additional moderators in this dataset that we are unable to find. However, there is a strong link between breastfeeding and g.

Study 3: Prenatal cocaine exposure

Table 6 shows the results on the correlations between a) score differences between cocaine-exposed and average groups (d) and b) g loadings. Results are derived from four studies, with a total of 391 participants. It shows also the study reference, the intelligence test, the observed correlation between g and d, sample size, and mean age (and age range).

The results of the meta-analysis of four studies are shown in Table 7. K represents the study number, N represents the accumulated harmonic sample sizes, r represents the mean bare-bones correlation, SD_{rho} represents the standard deviation after sample size corrections, %VE represents the percentage of explained variance, and 80%CI represents the confidence interval. This interval gives us the expected correlation for 80% of the cases.

Our analysis gives us a bare-bones correlation of .91, and the explained variance due to artifactual errors is 3.9%.

Conclusion

The meta-analysis shows that the effects of prenatal cocaine exposure are quite strongly linked to the g vector. Yet, only 3.9% of the variance can be explained by artifactual errors (%VE), which is extremely low: the large majority of the variance cannot be explained by artifactual errors. No strong conclusions can therefore be made about the link between prenatal cocaine exposure and g.

In an outlier analysis we left out the study from Asanbe and Lockert (2006) because there were some subjects in the cocaine-exposed group that also were exposed to other drugs, although. cocaine was the main drug used in this group, and it was unclear how many of the subjects used other drugs. However, the percentage of variance explained remained low.

Study 4: Prenatal smoke exposure

Table 8 shows the results on the correlations between a) score differences between lead exposed and average groups (d) and b) g loadings. Results are derived from three studies, with a total of 443 participants. It shows also the study reference, the intelligence test, the observed correlation between g and d, sample size, and mean age (and age range).

The results of the meta-analysis of three studies are shown in Table 9. K represents the study number, N represents the accumulated harmonic sample sizes, r represents the mean bare-bones correlation, SD_{rho} represents the standard deviation after sample size corrections, %VE represents the percentage of explained variance, and 80%CI represents the confidence interval. This interval gives us the expected correlation for 80% of the cases.

Our analysis gives us a bare-bones correlation of -.17, and the explained variance due to artifactual errors is 1.2%.

Conclusion

Although we predicted a high correlation between prenatal smoke exposure and g, the meta-analysis shows no strong link between prenatal smoke exposure and the g factor. No outlier analysis was possible, due to the small amount of data points. The data are completely non-supportive of the theory.

Discussion

The central question addressed in this study is whether some potential biological and environmental factors might cause true differences in general mental ability (g) between groups, or just “hollow” score differences between groups. Cognitive test scores are the number one predictor of school and workplace achievements, and it is mainly the g loadings of these cognitive tests that accounts for this effect.

Results from previous meta-analyses using the MCV lead us to the following theory: if the correlation between the g vector and a second vector is close to +1.00, variation in scores on the variable is caused by biological factors; If the correlation is close to -1.00, the variation in scores is caused by non-biological factors; If the correlation is close to 0.00, the variation is caused by both biological and non-biological factors (te Nijenhuis et al., 2009; Nijenhuis & Dragt, 2010). In this study we focused on differences in general mental ability between groups, where there was a focus on children and the effect from four different variables on their general mental ability: lead exposure, breastfeeding, prenatal cocaine exposure, and prenatal smoke exposure. We hypothesized a link between these variables and g loadings.

In line with our hypotheses, the meta-analyses showed correlations of .79 on breastfeeding (total N=7847) and .91 on prenatal cocaine exposure (total N=391). However, the variance explained by artifactual errors is too low to draw strong conclusions about the link between cocaine exposure and g, and about the link between breastfeeding and g.

Contrary to our hypotheses, the meta-analyses showed correlations of -.19 on lead exposure (total N = 702) and -.19 on prenatal smoke exposure (total N = 443). We were not able to come up with strong theoretical explanations for these values of the correlation between lead exposure and g. As regards to the low correlation between smoke exposure and g, two of the three studies used the McCarthy scales to measure the children’s intelligence. When we looked at However, the g loadings of the McCarthy scales show a quite strong variation between different age groups, suggesting that the g loadings are quite unreliable for young children, which could explain the unexpected findings.

Limitations

Some limitations for the present studies have to be addressed. First, the method of correlated vectors (MCV) is not without limitations. With this method there can be some strong positive vector correlations between two groups when these groups also differ on factors that are not correlated with g (Dolan and Lubke, 2001). Also spurious correlations can be found between vectors because subtest’s g loadings are sensitive of other subtests in a cognitive battery.

There are also some limitations in the included studies for the meta-analyses. In the studies for lead exposure, there were differences in lead measurement methods. Some studies used teeth for lead measures while other studies used blood lead measures. Although there is no evidence implying an inconsistency between those methods, it is possible that there is a difference in reliability between the methods.

When we look at the studies for breastfeeding, we observe a large difference between studies concerning the duration of breastfeeding. In some studies there was a large difference between the study participants regarding duration of breastfeeding. When a large group of participants in a study did not receive long-term breastfeeding, their test data could influence the correlation between breastfeeding and g. This could also be the case for cocaine exposed children, where we also found a difference between the duration of cocaine usage during pregnancy between studies and within studies. Also, the level of cocaine intake during pregnancy could vary between participants, which could alter the outcome of the meta-analyses. It is highly possible that longer cocaine abuse will have a greater impact on the child’s cognitive development.