We are all familiar with idea that the distribution of hands shown at showdown is different from the distribution of all hands dealt. The specter of this bias has confounded the poker botting community and lead many to eschew estimating hand ranges all together. While my scouring of the pokerai.org archives was not totally exhaustive, I don’t believe that anyone else has tried to quantify the bias in showed hands in a systematic way. In this analysis we will use a two econometric models, one created from a dataset revealing every hand, and the other from a dataset limited to hands showed at showdown, to predict a player’s hand range distribution in several different game states. By comparing the showdown equity of the match-up between an arbitrary hand and the ‘all hands’ and ’showed’ hand range distributions, we can estimate the bias in terms of its affect on showdown equity. In a meta-analysis of +45,000 game states, equity estimates derived from the dataset limited to hands showed at showdown were -1.34% ± 2.1% lower on average than those derived from the full dataset – indicating a slight upward bias in the strength of observed hands.

more))