Abstract:
I distinguish two meanings of the term bias in the social sciences. In the first, biases are functional: they are necessary, and simultaneously enable and constrain perception and cognition. In the second, biases are viewed as errors and ideally should be reduced to zero. In the functional view, bias is value-neutral, neither good nor bad. This pragmatic perspective accepts that cognition must operate under conditions of uncertainty (rather than the certainty of a “small world”) and intractability (where the optimal solution cannot be calculated). Biases enable cognition to deal with these situations where probability theory offers no guidance, for instance, through intelligent heuristics. In contrast, the error view assigns a negative value to bias. It assumes that cognition deals with problems where the true state of the world is known with certainty – at least to some authority. This distinction has profound implications for research design: the two views lead not only to different answers, but also to different questions. Researchers who adopt the error view take the deviation between judgment and true state as the explanandum, not the judgment itself. As a consequence, the functional question – What does a bias achieve? – is virtually never asked, nor is the possibility considered that certain biases might lead to better judgments. This is one reason why less-is-more effects –conditions under which ignoring information yield more accurate inferences – were discovered only recently. Ultimately, views about the nature of bias can themselves become a bias in research on biases.