Fig. 1: Ratios between empirical type I error rates and the nominal α levels. | npj Genomic Medicine

Fig. 1: Ratios between empirical type I error rates and the nominal α levels.

From: A statistical perspective on baseline adjustment in pharmacogenomic genome-wide association studies of quantitative change

Fig. 1: Ratios between empirical type I error rates and the nominal α levels.

Upper panels (a and b): no measurement errors. Lower panels (c and d): white noise measurement error (normal relative error rate with mean zero). Dotted curves: baseline-unadjusted models. Horizontal dash line: the ratio (\(\alpha\) + 3*SE)/\(\alpha\) = 1 + 3*SE/\(\alpha \approx 1.67\), where SE is the margin of error calculated as \(\sqrt {\frac{{\alpha \ast (1 - \alpha )}}{n}}\), \(\alpha = 10^{ - 6}\) is the nominal level and \(n = 2 \times 10^7\) is the number of simulations. M1–M9 are defined in the “Methods” section.

Back to article page