1 Title: Does General Intelligence Deflate Standardized Effect Sizes of Cognitive Sex Differences?
7 (SEXNET's own) Marco del Guidice points out[^mdg] that in the presence of measurement error, standardized effect size measures like [Cohen's _d_](https://rpsychologist.com/d3/cohend/) will underestimate the "true" effect size. Recall that _d_ is the difference in group means, divided by the pooled variance. Thus, holding _actual_ sex differences constant, more measurement error means more variance, which means smaller values of _d_. Here's some toy Python code illustrating this effect:
9 [^mdg]: Marco del Guidice, ["Measuring Sex Differences and Similarities"](https://marcodgdotnet.files.wordpress.com/2019/04/delgiudice_measuring_sex-differences-similarities_pre.pdf)
13 from statistics import mean, variance
15 from numpy.random import normal, seed
17 # seed the random number generator for reproducibility of given figures,
18 # commment this out to run a new experiment
25 (len(X)*variance(X) + len(Y)*variance(Y)) /
30 def population_with_error(μ, σ, n):
33 def measurement_error():
35 return [trait() + measurement_error() for _ in range(n)]
38 # trait differs by 1 standard deviation
39 adjusted_f = population_with_error(1, 0, 10000)
40 adjusted_m = population_with_error(0, 0, 10000)
42 # as above, but with 0.5 standard units measurment error
43 measured_f = population_with_error(1, 0.5, 10000)
44 measured_m = population_with_error(0, 0.5, 10000)
46 smart_d = cohens_d(adjusted_f, adjusted_m)
47 print(smart_d) # 1.0193773432617055 — d≈1.0, as expected!
49 naïve_d = cohens_d(measured_f, measured_m)
50 print(naïve_d) # 0.8953395386313235
53 But doesn't a similar argument hold for non-error sources of variance that are "orthogonal" to the group difference? (Sorry, I know this is vague; I'm writing to the list in case any Actual Scientists can spare a moment to help me make my intuition more precise.) Like, suppose performance on some particular cognitive task can be modeled as the sum of the general intelligence factor (zero or negligible sex difference[^jensen]), and a special ability factor that does show sex differences. Then, even with zero _measurement_ error, _d_ would underestimate the difference between women and men _of the same general intelligence_.
55 [^jensen]: Arthur Jensen, _The g Factor_, Chapter 13
58 def performance(g, σ_g, s, n):
59 def general_ability():
61 def special_ability():
63 return [general_ability() + special_ability() for _ in range(n)]
65 # ♀ one standard deviation better than ♂ at the special factor
66 population_f = performance(0, 1, 1, 10000)
67 population_m = performance(0, 1, 0, 10000)
69 # ... but suppose we control/match for general intelligence
70 matched_f = performance(0, 0, 1, 10000)
71 matched_m = performance(0, 0, 0, 10000)
73 population_d = cohens_d(population_f, population_m)
74 print(population_d) # 0.7287587808164793
76 matched_d = cohens_d(matched_f, matched_m)
77 print(matched_d) # 1.018362581243161