In the next step, we combined all 4 samples (ie, population, twin, full-siblings, and half-siblings) into 1 data set. From the population sample, we randomly selected the same number of individuals contained in the full-sibling analysis (n = 2 001 950). We performed 2 analyses. The first allowed all parameters for each sample to be independent within the samples and age intervals. The second modeled only the genetic resemblance, given evidence that resemblance for longevity in Sweden results only from genetic factors.19 Within each age interval, we assumed that this parameter reflected genetic resemblance for each sample: ie, 0 for the population, 0.25 for the half-sibling, 0.5 for the full-sibling, and 1.0 for the twin sample. We compared these models by AIC. If the second model fit better, we obtained an HR for the slope of the familial confounding within each age interval and an improved estimation of the association among all relatives, but especially MZ twins, where the data were sparsest. While this genetic slope (GS) is estimated using information from the general population and all the relative pairs,
