Covariation Assessments with Costly Information Collection in Audit Planning : An Experimental Study

In this paper we report the results of two experiments that investigate auditing covariation-estimate revisions when there are countervailing incentives for effectiveness and efficiency. Covariation estimates assess the degree of association between a 'Clue' that auditors might observe, and its potentially associated 'Condition.' We find that when participants choose which information to obtain from a contingency table (i.e., Cell Choice), they obtain too few cells of information in predictable patterns. Second, we find that when covariation estimation biases occur, they do so more due to this insufficient information collection rather than improper processing of collected information. Third, when participants collect information randomly and cannot choose particular cells (i.e., Sample-Size Choice), covariation estimation biases are significantly reduced. Finally, when participants are sensitized to the importance of all four cells of information, they collect more cells in subsequent Cell Choice, resulting in covariation estimate revisions that are more normative than those of non-sensitized participants. Overall, this evidence implies that when auditors estimate the association between an unfamiliar Clue-Condition pair, their estimates are more biased when they choose specific cells of information than when they choose sample size. Sensitization can reduce this bias and, therefore, could be important to auditor training and the design of decision aids