Many studies have documented that people exaggerate the degree to which small samples are similar to the parent population and thus tend to believe in the independence of a binary sequence exhibiting negative autocorrelations. This belief is the so-called "gambler's fallacy". In this paper we introduce a method that enables us to test the bias of selection behavior for lotto games without using the exact distribution of the numbers chosen. There are two main findings: First, we show that the betting behavior of lotto players in the short term is strongly consistent with the gambler's fallacy. Second, consistent with the notion of Type Ⅱ gambler's fallacy, these same players tend to pick those numbers that have been drawn most frequently in the past. The latter may be due to the availability heuristic, since winning numbers having higher occurrence rates come to mind more easily.