The dataset contained multiple members from some families; in such cases, data from family members were not truly independent observations. The SPSS procedure does not account for clustering within families. Therefore, the possible effect of clustering was assessed by using the Huber-White sandwich (Rogers, 1993) estimator available in the STATA version 7.0 (Stata Corp., College Station, TX) stcox command. Unequal selection probabilities, clustering, and weighting adjustments for non-response will result in design effects >1. However, no evidence of a design effect >1 was found, thus justifying the use of the SPSS procedure.
