Evidence of Fraud in an Influential Field Experiment About Dishonesty is a blog post at Data Colada where researchers uncovered dishonest data meddling in a PNAS-published paper about… dishonesty.
A group of anonymous researchers found problems with the 2012 paper “Signing at the beginning makes ethics salient and decreases dishonest self-reports in comparison to signing at the end”. They brought their results to Uri Simonsohn, Joe Simmons, and Leif Nelson at Data Colada. The study is about people self-reporting odometer readings from their car. The paper’s Excel spreadsheet of the source data indicated mathematical malfeasance. Statistics lay out the case for faked figures but may make the layperson’s eyes glaze over. For them, the cherry on top is the examination of the Excel sheet’s fonts: Calibri, but also Cambria.
Let’s start with our claim that the Calibri and Cambria observations are near-duplicates of each other. What is the evidence for that? First, the baseline mileages for Car #1 appear in Calibri font for 6,744 customers in the dataset and Cambria font for 6,744 customers in the dataset. So exactly half are in one font, and half are in the other. For the other three cars, there is an odd number of observations, such that the split between Cambria and Calibri is off by exactly one (e.g., there are 2,825 Calibri rows and 2,824 Cambria rows for Car #2). Second, each observation in Calibri tends to match an observation in Cambria.