### abstract ###
participants drank either regular root beer or sugar-free diet root beer  before working on a probability-learning task in which they tried to predict which of two events  would occur on each of  NUMBER  trials
one event e NUMBER  randomly occurred on  NUMBER  trials   the other e NUMBER  on  NUMBER 
in each of the last two blocks of  NUMBER  trials  the regular group matched prediction and event frequencies
in contrast  the diet group predicted e NUMBER  more often in  each of these blocks
after the task  participants were asked to write down rules they used for  responding
blind ratings of rule complexity were  inversely related to e NUMBER  predictions in the final  NUMBER  trials
participants also took longer to advance after incorrect predictions and before predicting e NUMBER   reflecting time for revising and consulting rules
these results support the hypothesis that an effortful controlled process of  normative rule-generation produces matching in probability-learning experiments   and that this process is a function of glucose availability
### introduction ###
in two-choice probability learning pl experiments  one event e NUMBER  has a better chance of occurring than the other e NUMBER   and event likelihoods are independent of responses
a participant's task is to predict which event will occur on each trial
although participants could maximize hits by exclusively predicting e NUMBER   they tend to approximately match prediction frequencies with outcome probabilities for many trials  CITATION
for example  if e NUMBER  occurs on  NUMBER  percent  of trials and e NUMBER  occurs on the other  NUMBER  percent  of trials  people usually predict e NUMBER  on about  NUMBER  percent  of their choices for hundreds of trials  CITATION
this probability-matching pm behavior is suboptimal because it produces hits on only  NUMBER  percent  of trials   NUMBER  x  NUMBER    NUMBER  x  NUMBER  instead of the maximal  NUMBER  percent  hit rate from always predicting e NUMBER 
