Overview of false discovery rates.

<p>Both quantities assume a two-component <i>p</i>-value distribution: “null” <i>p</i>-values are uniformly distributed (with height <i>π</i>₀ <i>≤</i> 1), and “alternative” <i>p</i>-values that should peak at <i>p</i> = 0. The area of the null component with <i>p ≤ t</i> is simply <i>π</i>₀<i>t</i>, while the total area is the cumulative density function <i>F</i>(<i>t</i>). The total height at <i>t</i> is the density function <i>f</i>(<i>t</i>). The FDR is the proportion of the area with <i>p ≤ t</i> that corresponds to the null component. The lFDR parallels the FDR but is a ratio of densities (heights) rather than areas.