Review of the empirical calibration of P-value method

The empirical calibration of P-value method relies on a set of variables with no known association with the study outcome (‘negative controls’, NCs) to obtain a distribution of effect sizes under the null hypothesis (‘empirical null distribution’) in the available data.14 After being estimated from the data, the parameters of the null empirical distribution can be used to calibrate the test statistic and the P-value for the exposure of interest, assuming that bias arises from the same distribution (aka ‘exchangeability’ assumption). Full details on the method are reported in the Supplementary data available at IJE online.