Increasing information gain in animal research by improving statistical model accuracy

Abstract Reduction of the numbers laboratory animals is one three pillars ethical animal research. Equivalently, information gain per should be maximized. A road towards this goal that barely taken in current research more accurate statistical modeling experiments. Here we show for a typical experiment (“open field test”) with outcomes are non-normally distributed count data, how can implemented and what achieved. We contrast state art – use confidence intervals based on null-hypothesis significance testing (NHST) –, Bayesian approach same underlying normal model, negative binomial model. find model leads to marked improvement knowledge gained experiment, especially small sample sizes. As experimental data violate assumptions simple, conventional models frequent, our findings have wider implications.