Catch Me If You Can: Testing the reduction of compound lotteries axiom in a tax compliance experiment

We test the Reduction of Compound Lotteries Axiom (RCLA) in an experiment on tax compliance. We disentangle the compound probability of audit and detection and test it against a more realistic situation with a probability of audit and a subsequent probability of detection. Various experiments have shown that abstract framing often leads to violations of the RCLA; our framed set-up also reveals statistically significant departures from the RCLA. We find that subjects comply more in the two-stage lottery set-up than they do in the one-stage equivalent and that violations are compatible with subjects applying different weighting functions to one-stage lotteries versus two-stage lotteries: inverse S-shaped - likelihood insensitive - in the former set-up; and significantly less insensitive in the latter. Violations substantially decline with subjects’ experience but not at the same rate for all groups of subjects.