Assessing Manipulated Data

Abstract

Statistics is not good at detecting manipulated data, but it can assess the consistence of the data with randomness. An example is given of a cholesterol reducing trial. The results consisted of 96 risk ratios, and often a 0, 1 or 9 was observed as final digit of the odds ratios, while the values 2–8 as final digits were virtually unobserved. We will test whether the observed frequencies of the final digits are compatible with equal frequencies, as expected, and we will use a Chi-square test for that purpose.