Measuring familial sinistrality: Problems with dichotomous classification

Abstract

Familial sinistrality (FS), or familial left-handedness, is usually measured as a presence/absence dichotomy (FS+/FS−). Measuring FS as a dichotomy is problematic in at least two ways: (1) magnitude data are discarded and (2) there is an increased risk of violating statistical assumptions. We hypothesised that using a non-dichotomised FS measure would allow more sensitive statistical tests than using a dichotomised measure. The non-dichotomised measure examined was proportion of family that is left-handed, or Proportional FS (FSP). FS measures of 60 healthy adults were consistent with our hypothesis: (1) The effect size was 2.3 times larger when FSP was used than when Dichotomised FS (FSD) was used; (2) the observed relationship was statistically significant when FSP was used but not when FSD was used; and (3) statistical assumptions were violated when FSD was used but not when FSP was used. When cases were weighted to account for an increased likelihood of FS in large families, results were similar. Next, bootstrapping methods were used to estimate empirically the long-term differences in power between analyses incorporating FSP and those incorporating FSD. Using FSP was associated with long-term increases in both power and effect size of over 50% relative to using FSD. We conclude that using a proportional rather than a dichotomous measure of FS will enhance the accuracy of observed effect sizes, increase the sensitivity of statistical tests, and reduce the likelihood of violating statistical assumptions.