How to estimate the measurement error variance associated with ancestry proportion estimates.

To show how the variance of the measurement error (ME) associated with individual ancestry proportion estimates can be estimated, especially when the number of ancestral populations (k) is greater than 2. We extend existing internal consistency measures to estimate the ME variance, and we compare these estimates with the ME variance estimated by use of the repeated measurement (RM) approach. Both approaches work by dividing the genotyped markers into subsets. We examine the effect of the number of subsets and of the allocation of markers to each subset on the performance of each approach. We used simulated data for all comparisons. Independently of the value of k, the measures of internal reliability provided less biased and more precise estimates of the ME variance than did those obtained with the RM approach. Both methods tend to perform better when a large number of subsets of markers with similar sizes are considered. Our results will facilitate the use of ME correction methods to address the ME problem in individual ancestry proportion estimates. Our method will improve the ability to control for type I error inflation and loss of power in association tests and other genomic research involving ancestry estimates.