In addition, in a multisite study of this nature, variation in ascertainment by site may add further complexity to the analysis. To identify loci that specifically affect SOC and BEH, the effects of age of ADI-R completion, gender, verbal/nonverbal status, and AGP site on the phenotypes were removed in the linkage analyses (Tables 2 and 3 in Supplement 1). However, one could argue that the effects of AGP site, identified here as a covariate, should not be removed. In a multisite genetic study, differences in a phenotype across sites could be caused by measurement error and/or by true differences in the severity of phenotypes among affected subjects. We suspect that the differences in SOC and BEH across the AGP sites might be caused by the latter, since all sites have demonstrated good reliability; of note, individual AGP sites have also recruited families from different clinical centers. Due to these arguments, we performed linkage analyses without the AGP site as a covariate (but with the adjustments for gender, verbal/nonverbal status, and age of ADI-R completion). Without the AGP site as a
