Statistical Analysis

To allow for seasonal differences in sleep duration (ie, shorter sleep duration in summer than winter), we accounted for the season during which the questionnaire was administered (summer versus winter). Summer was defined as May–October and winter as November–April, respectively, to be consistent with preexisting data available from NHANES, as this is how they defined summer and winter. After calculating the mean difference in sleep duration between the summer and winter months in males and females, we standardized sleep duration to the summer season by subtracting the mean sleep duration difference between winter and summer from winter sleep duration. Throughout this manuscript, sleep duration has been preadjusted for season for all comparisons between groups unless otherwise indicated.

To compare differences in sleep duration between Amish and NHANES, we estimated preadjusted sleep duration and other characteristics using linear models that adjusted for sex and age groups. In a parallel analysis we also categorized sleep duration into short (< 7 hours) and long (> 9 hours) duration and compared frequencies by the Pearson chi-square test and logistic regression to account for age and sex. All analyses are performed using R studio version 1.1.447 (The R Foundation, Vienna, Austria). Boxplots and forest plots were produced by the package ggplot2 in R studio (The R Foundation).

We estimated the heritability of sleep duration in the Amish using a linear mixed model that included the relationship matrix as a random effect. Heritability was defined as the proportion of the variance accounted for by the relationship matrix, after accounting for the effects of age group and sex. We also estimated the familial correlations in seasonally adjusted sleep duration among spouse pairs, siblings, and parent-child pairs. We used MMAP software (Jeff O’Connell, https://mmap.github.io/) to estimate heritability¹⁸ and Pearson correlations to estimate correlations among family members.