Skin color and IQ in the GSS

Share on FacebookShare on Google+Email this to someoneTweet about this on Twitter

A question from Jason Malloy prompted a quick search of the GSS for data on the cause of the Black-White IQ gap. In 1982, the GSS characterized the skin color of Black participants on a 5-point scale (1:very dark brown to 5:very light brown). The very dark/light categories consist of only 50 and 14 individuals, respectively, and so in the following analysis I merged them with the dark/light brown categories, to give three COLOR levels: dark, medium, and light. In the web application, use COLOR(r:1-2;3;4-5) instead of COLOR. The WORDSUM variable is a 10 question vocabulary test, which I’m treating as a proxy for IQ. It is correlated with educational attainment (~.4), and also correlates (~.4-.5) with tests of reasoning and basic knowledge that were given in some years. These other tests are not available for 1982. In the all-subject all-year GSS data set, WORDSUM varies by SEX, and in 1982 COLOR also varies by SEX. Thus, SEX is controlled for in each analysis. WORDSUM is lower in the youngest and oldest age groups, so an AGE(25-65) filter was used.

Table 1. Mean WORDSUM score by COLOR and SEX with ANOVA

Main Statistics



Cells contain:
-Std Devs
-N of cases
1: Dark 4.15
2: Medium 5.39
3: Light 6.04

color indicates T-statistic, and thus p-value

Color coding:<-2.0<-1.0<0.0>0.0>1.0>2.0T
Mean in each cell:Smaller than averageLarger than average

Analysis of Variance
SSQ Eta_sq df MSQ F P
Main effects 89.443 .061 3 29.814 6.956 .0002
COLOR 89.099 .061 2 44.549 10.394 .0000
SEX .691 .000 1 .691 .161 .6884
Interaction 2.569 .002 2 1.285 .300 .7412
Residual 1,375.884 .937 321 4.286
Total 1,467.896 1.000 326

We can quantify the effect size of each skin color class using Cohen’s d statistic, which measures the mean difference in standard deviation units. In the 1982 dataset, the overall d for the Black-White gap on WORDSUM is -0.63 (among males d=-0.51, among d=-0.74). For comparison, the 1982 male-female gap among Whites is d=-.12, favoring females.

Table 2. Effect size (d) of COLOR on WORDSUM using “light” as a control group

Color Male Female Total
Dark -0.99 -0.53 -0.68
Medium -0.31 -0.20 -0.22
Light 0.00 0.00 0.00

We can also use Whites as the control group.

Table 3. Effect size (d) of COLOR on WORDSUM using Whites as a control group

Color Male Female Total
Dark -0.99 -1.10 -1.04
Medium -0.35 -0.69 -0.54
Light -0.07 -0.46 -0.33

Thus, there are substantial (moderate to large effect size) differences in WORDSUM scores between the darkest and lightest Blacks in 1982.

As reported by Rushton and Jensen (2005), Shuey (1966) reviewed 18 studies which used skin color as a measure of racial admixture to compare with IQ. Of those 18, 16 found a significant effect of the kind found here, but the overall correlation with IQ was low (r=.1). In this data, the COLOR WORDSUM correlation is r=.31 among males and r=.18 among females, with an overall correlation of r=.23. Off the top of my head, I’m not certain what the expected correlation would be between IQ and skin color among Blacks for a given measure of “between-group heritability” (BGH) as described by Jensen (1998). I’ll leave it as an exercise for our mathematically skilled commentators to derive a formula for this relationship and to evaluate the signficance of this finding in explaining the cause of the Black-White IQ gap.

Labels: , , , ,


  1. Be very careful. The GSS survey is very addictive.

  2. “I’m not certain what the expected correlation would be between IQ and skin color among Blacks for a given measure of “between-group heritability” (BGH) as described by Jensen (1998). “ 
    Light skin vs dark skin IQ gap is probably more a result of successful high IQ black men taking light skin women as wives rather than due to white admixture (the actual effect of which is reduced as a result of assortive mating).

  3. 1: Nat Genet. 2004 Nov;36(11 Suppl):S54-60. 
    Implications of correlations between skin color and genetic ancestry for 
    biomedical research. 
    Parra EJ, Kittles RA, Shriver MD. 
    Department of Anthropology, University of Toronto at Mississauga, Mississauga, 
    Ontario L5L 1C6, Canada. 
    Skin pigmentation is a central element of most discussions on ‘race’ and 
    genetics. Research on the genetic basis of population variation in this 
    phenotype, which is important in mediating both social experiences and 
    environmental exposures, is sparse. We studied the relationship between 
    pigmentation and ancestry in five populations of mixed ancestry with a wide 
    range of pigmentation and ancestral proportions (African Americans from 
    Washington, DC; African Caribbeans living in England; Puerto Ricans from New 
    York; Mexicans from Guerrero; and Hispanics from San Luis Valley). The strength 
    of the relationship between skin color and ancestry was quite variable, with the 
    correlations ranging in intensity from moderately strong (Puerto Rico, rho = 
    0.633) to weak (Mexico, rho = 0.212). These results demonstrate the utility of 
    ancestry-informative genetic markers and admixture methods and emphasize the 
    need to be cautious when using pigmentation as a proxy of ancestry or when 
    extrapolating the results from one admixed population to another. 
    PMID: 15508005 [PubMed - indexed for MEDLINE]

  4. Table 1 Relationship of melanin content and individual ancestry 
    Samplea Spearman’s rho (95% c.i.)b 
    Relationship of melanin index and ancestry 
    African Americans (W-AF) 0.440 (0.330 to 0.538) 
    African Caribbeans (W-AF) 0.375 (0.239 to 0.496) 
    Mexicans (I-AM) 0.212 (0.057 to 0.357) 
    Puerto Ricans (W-AF) 0.633 (0.457 to 0.761) 
    Relationship of skin reflectance and ancestry 
    Hispanics (I-AM) -0.259 (-0.141 to -0.369) 
    aAncestral proportion axes are indicated by the abbreviations W-AF (West African) and I-AM (Indigenous American). 
    bNinety-five percent confidence intervals for the correlation estimates are shown in parentheses.

  5. I saw the evidence. Now can we need to know why.

  6. Two major hypotheses: 
    1. white admixture causes (some fraction of) skin color and IQ differences each thru hereditary mechanisms 
    2. skin color differences leads to different social/cultural environments which causes IQ differences 
    Hypothesis #1 predicts that direct measurement of individual ancestry thru genotyping would produce stronger correlations than those seen here; and that if MALD were used, individual loci would be found that contribute to the association. I’m not sure what Hypothesis #2 predicts. 
    One additional data point to note: the skin color effect is stronger in the North and West than in the South, and that for all groups/races WORDSUM scores are higher in the North and West than in the South.