Reverse P-hacking: when non-significant results are preferred (dataset)

The data were analysed for the presence of selective reporting of non-significance and/or reverse P-hacking in the field of behavioural ecology.

On 1 February 2018, we searched for articles published between 1990 and 2018 in three leading behavioural ecology journals: Animal Behaviour, Behavioral Ecology, and Behavioral Ecology and Sociobiology, that had ‘experiment* AND control*’ in the ‘topic’ category in the ISI Web of Science database (1081 articles). We then searched these articles for tests of whether the mean value of confounding variables differed between treatment(s) and control groups where subjects were randomly assigned to groups. We only included confounding variables that involved measurements made on test subjects or groups of subjects (e.g. body mass, blood glucose, brood size).

To identify papers with suitable data, we first read Abstract, Methods and Results to see if there was any indication that the study was likely to include tests for confounding variables differing between control and treatment groups (i.e. studies that involved experimental manipulation). We recorded: the name of the confounding variable, P-values or statistical significance statements associated with tests for a difference, the sample sizes.