Blanchard, Ray. Birth order and sibling sex ratio in homosexual versus heterosexual males and females. Annual Review of Sex Research, 8:27-67.
Costa, Paul T., Jr., and McCrae, Robert R. 1992. NEO PI-R Professional Manual. Odessa, FL: Psychological Assessment Resources.
Ernst, C‚cile, and Jules Angst. 1983. Birth Order: Its Influence on Personality. Berlin and New York: Springer-Verlag.
Harris, Judith Rich. 1995. Where is the child's environment? A group socialization theory of development. Psychological Review, 102:458-89.
Harris, Judith Rich. 1998. The Nurture Assumption: Why Children Turn Out the Way They Do. New York: Free Press.
Loehlin, John C. 1997. A test of J. R. Harris's theory of peer influences on personality. Journal of Personality and Social Psychology, 72:1197-1201.
Modell, John. 1996. Family niche and intellectual bent. Review of Born to Rebel, by Frank J. Sulloway. Science, 275:624.
Rosenthal, Robert. 1987. Judgment Studies: Design, Analysis, and Meta-analysis. Cambridge: Cambridge University Press.
Salmon, Catherine A. 1998. The evocative nature of kin terminology in political rhetoric. Politics and the Life Sciences, 17:51-57.
Salmon, Catherine A. In press. On the impact of sex and birth order on contact with kin. Human Nature.
Salmon, Catherine A., and Daly, Martin. 1998. Birth order and familial sentiment: Middleborns are different. Human Behavior and Evolution, 19:299-312.
Sulloway, Frank J. 1995. Birth order and evolutionary psychology: A meta-analytic overview. Psychological Inquiry, 6, 75-80.
Sulloway, Frank J. 1996. Born to Rebel: Birth Order, Family Dynamics, and Creative Lives. New York: Pantheon; Vintage, 1997.
Sulloway, Frank J. In press. Birth order, sibling competition, and human behavior. In Paul S. Davies and Harmon R. Holcomb III, eds., The Evolution of Minds: Psychological and Philosophical Perspectives. Dordrecht and Boston: Kluwer Academic Publishers.
1. An important exception to this statement involves research on homosexuality, where birth order has been implicated in the build-up of antibodies to one of the minor histocompatibility antigens (Blanchard 1997). This apparent biological source of birth-order affects in psychosexual behavior applies only to a small portion of the population and appears to have no measurable influence on normal individuals.
2. I do not bother to address here Harris's claim that birth-order effects are less frequent in large studies because she bases this argument on data that are contaminated with errors.
3. Of course, it is inconsistent for Harris to code her results in terms of "studies" - purportedly to eliminate multiple findings that are not statistically independent--and then to fall back on the claim that these results cannot be tested statistically because they contain findings that are not statistically independent. It is also relevant to compare Harris's meta-analytic findings regarding Ernst and Angst's (1983) list of studies with her erroneous statement, in her 1995 article: "When the proper controls were used, no birth-order effects were found on personality" (p. 461). By Harris's own count, 49 significant confirming results are to be found in her own total of 179 controlled studies compiled from Ernst and Angst's tables, or more than ten times the rate of confirming results expected by chance.
4. Ways of resolving the file-drawer problem have attracted considerable discussion by statisticians, but Harris, having planted her seeds of doubt on this topic, chooses to pass over this literature. For example, Robert Rosenthal (1987) has suggested a "rough and ready guide" for determining whether a given number of significant findings in the published literature would be invalidated by unpublished null findings in file drawers. This test involves calculating the number of null findings that would have to exist in file drawers in order to cause the number of significant published findings to no longer exceed chance expectations. This number is 19S-N, where S is the number of significant findings and N is the number of nonsignificant findings. Specifically, Rosenthal's formula requires the existence of 1,244 null findings from controlled birth-order studies to invalidate the results in my own meta-analysis. Rosenthal considers a meta-analysis as passing the file-drawer test if the number of null findings needed for refutation is more than 5 times the total number of findings, plus 10 (in this case, 990 findings). In short, by Rosenthal's guide, these meta-analytic counts pass the file-drawer test. This form of the file-drawer test is considerably less powerful than one based on effect sizes. A conservative version of this alternative test involves estimating effect sizes based on p-values and setting z (the effect size for each finding), to 1.645 for all results that are significant at p<.05, and to 0 for all nonsignificant results. Based on this method, the number of null findings that are required to exist in file drawers in order to invalidate the significant meta-analytic totals in my own survey is 3,171, or 16 times the number of published findings. For Harris's totals, the required number of null findings in file drawers is 1,339, or 7.5 times the number of published findings. In actuality, the number of null findings that would be required to invalidate these results is considerably greater than either of these two conservative estimates.
5. Although self-report studies manifest more significant birth-order differences than would be expected by chance, many of these studies --even large ones --report small and nonsignificant effects. Judith Harris likes to cite Jules Angst's large birth-order study involving 7,582 college-age Swiss subjects, which found only one significant difference among the 12 scales he and his colleagues employed. These admittedly modest results are probably explained by the use of unanchored scales, the omission of the most relevant scales (for example, those related to conscientiousness), and problems with scale heterogeneity (that is, combining traits such as dominance and sociability when measuring general attributes such as extraversion--see Sulloway, in press). It is noteworthy that when these Swiss investigators sampled specific behaviors among their subjects, they obtained strikingly different results from those based on their self-report personality test. For example, laterborns were significantly more likely than firstborns to admit having experimented with drugs, including tobacco, alcohol, marijuana, tranquilizers, and hypnotics. In addition, firstborns were significantly more likely than laterborns to discuss their problems with their parents. Harris fails to cite these significant findings.
6. Firstborns and laterborns differed significantly on each of the six votes within the Convention; on the decision to sign, or not to sign, the protest petition; and on the composite scale for tough-mindedness.
7. Firstborn and laterborn scientists differed significantly on all five measures, which are similar, moreover, to measures of openness to experience employed on standard personality tests, such as the NEO PI-R (Costa and McCrae 1992).
8. These birth-order effects explain 4.2 percent of the variance in personality, after being controlled for age, sex, sibship size, and social class. This degree of influence is equivalent to a medicine that would increase one's chances of surviving a deadly disease from 40 percent to 60 percent, or a 50 percent increase over the base rate. It is worth noting that the shared family environment, which explains about 5 percent of the variance in personality, involves an even larger contribution to personality. Harris, who does not properly explain to her readers the real-world meaning of such modest "effect sizes," repeatedly describes the influence of the shared family environment as negligible, as with the following statement: "The data [from behavioral genetic studies] showed that growing up in the same home, being reared by the same parents, had little or no effect on the adult personalities of siblings. Reared-together siblings are alike in personality only to the degree that they are alike genetically" (1998:37). Added together, these two sources of environmentally explained variance (stemming from parents and siblings) total more than 9 percent, which is nearly a quarter of the entire amount of variance (40 percent) that is actually available for explanation. Based on these data alone, one could write a persuasive book about the influence of the family on personality, and these data are not even the whole story, because they do not include other within-family influences. In particular, parents have a much greater influence on the social attitudes and values of their offspring than they do on personality.
In a recent test of Harris's group socialization theory, Loehlin (1997) found that shared peer groups among late-adolescent twins explained only 2.6 percent of the variance in personality, whereas Harris's theory demands that it account for the bulk of the variance that is available for explanation (about 35 percent). Loehlin's study also demonstrated a role for parental treatment on personality, which, in his study, explained about 1 percent of the variance. Although Loehlin was not able to test for the role of shared peer groups in early and middle childhood, his findings can be extrapolated from another of his measures and are not particularly encouraging for Harris's theory. In short, peer groups seem to matter (certainly a reasonable proposition), but they do not appear to provide anything like the whole story about environmental influences, as Harris would have us believe.
9. The correlation between the two relevant sets of effects sizes (for siblings and spouses) is .73 (N=30 traits, p<.001).
10. The consistency of birth-order effects among roommates, compared with those found by direct sibling comparisons, is given by the correlation between these two sets of effect sizes--namely, .87 (N=25 traits, p<.001). With peer ratings, birth-order effects in my study are also statistically significant, although they are smaller than those effects observed among siblings, spouses, and roommates. This is to be expected because the reliability of peer ratings is substantially lower than for spouses and siblings (and presumably for roommates), thus attenuating correlations with birth order.
11. In addition to their findings about middle children, Salmon and Daly (1998) have also replicated my findings about birth order and attitudes toward radical change. They asked 100 middle-aged Canadian subjects, "Do you think that you are open to new and radical ideas (such as cold fusion)?" Of the firstborn respondents, 47 percent answered "yes" to this question, whereas 86 percent of the middle children answered in the affirmative, and 89 percent of the lastborns did so (partial r=.38, p<.001, controlled for age, sex, and sibship size).