1 Slot Machine Stopping Decisions: Evidence for Prospect Theory Preferences?

Abstract

This paper examines the stopping decisions of individual slot machine players using a unique casino dataset, asking whether the relationship between quitting behavior and winnings can be most easily reconciled using neoclassical expected utility theory (EU) or a prospect theory value function (PT) (Kahneman and Tversky, 1979). I assume that slot machine players have a utility of gambling (Conlisk, 1993) which motivates their visit to the casino, and preferences over winnings which may (PT) or may not (EU) depend on a reference-point level of winnings. Using the fact that outcomes of individual slot machine bets are independent draws, the central limit theorem is applied to assess whether gamblers quit approximately randomly as EU predicts, or whether quitting was concentrated disproportionately at some reference level of winnings as PT predicts. By comparing the data to simulations which match broad features in the data, statistical tests find substantial excess kurtosis (4 th moment) in final stopping distributions, thus rejecting that winnings are distributed normally and providing support for PT preferences. In addition, winnings distributions are positively skewed (3 rd moment) – this implies that people are more likely to quit with winnings above the reference point than below it, providing support for diminishing sensitivity of the PT value function’s gains and loss segments. 1 j2lien@ucsd.edu. Comments are welcome. I am grateful to my advisor Vincent Crawford for encouragement and guidance, Julie Cullen for invaluable input, and an anonymous casino which generously allowed me to use their data for this study. For helpful comments and conversations, I also thank Gray Calhoun, Rachel Croson, Chulyoung Kim, Craig McIntosh, Juanjuan Meng, Joel Sobel, and participants in the Micro theory lunch seminar. All errors are my own. 1

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 2 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.