Table 2.
The 95th Percentiles of Association Statistics in the Absence of a Disease Locus (These Translate Directly to Thresholds for Genomewide Significance)[Note]
|
95th Percentile |
|||
| Scenario | Locus-Genome Statistic (Whole-Genome Score) | Locus-Genome Statistic (Strongest Marker in Genome) | Case-Control Statistic (Strongest Marker in Genome) |
| Null model (scenario 1) | −.1 | 2.7 | 3.7 |
| Mi varies between parents (scenario 2) | −.3 | 2.4 | 3.7 |
| λi varies between parents (scenario 3) | .1 | 2.8 | 3.7 |
| Disease locus (2-fold increased risk due to ancestry)a | .3–7.3 | 2.3–10.4 | 2.7–5.5 |
Note.— We performed 100 simulations with no disease locus for each of the three scenarios described in the text, for 200 cases and 200 controls with Mi∼20%±12% and λi∼6±2 studied in 2,147 of the markers described by Smith et al. (2004 [in this issue]), and analyzed the data with the disease association statistics. The 95th percentiles of simulations in the absence of a disease locus are approximately the same whether or not Mi and λi vary between parents. This indicates that substantial deviations from model assumptions are not likely to cause false positives in the MCMC analysis. For comparison, we also present simulations based on a real disease locus (twofold increased risk).
Values in this row are 5th–95th percentile ranges.