# Cognitive Biases

A cognitive bias is a directional departure from a normative model of reasoning (i.e., how one should reason). This is contrasted with the concept of noise, which is a departure from normativity in many directions. The existence of cognitive biases is a central concern for making better forecasts.

A cognitive bias is not necessarily a deviation from the correct answer (though it can be), but a deviation from normativity. For example, since Bayes Theorem is a normative model, any deviation from it would be considered a cognitive bias regardless of the true state of the world. Therefore, to say that someone is wrong because they exhibited a cognitive bias would be an example of the Fallacy Fallacy. A cognitive bias does not mean someone is wrong, it only means they did not reason according to how a normative model said they should have reasoned. Thus a cognitive bias is one source of bias, and is not the same as saying one is wrong. However, cognitive biases can often lead to being wrong. This distinction is important because in forecasting we often do not know what the correct answer is, but can still identify cognitive biases in our reasoning.

Since a cognitive bias is merely a directional deviation from normativity, and there is no widespread agreement on the set of models considered normative, a comprehensive list of biases is not possible even in theory. And indeed, in practice the list of biases is always growing as researchers discover new ways in which people do not reason as the researchers think they ought to reason. An example list of biases, and the normative model from which they deviate, is listed below.

- Status Quo Bias is a deviation from the Invariance Principle

- Conjunction Effect is a deviation from both Logic and Probability

- Base-Rate Fallacy is a deviation from Bayes Theorem

- Omission Bias is a deviation from Expected Utility

- Availability Bias is a deviation from the correct answer

The most common reason for the existence of a cognitive bias is that the individual is using a heuristic to reason because it is less cognitively demanding, and faster, than applying the normative model. In such a scenario, once the difference between their reasoning and the normative model is explained to the individual, they may agree that the normative model is superior. However, they may also reject that the normative model is superior and prefer their own normative model. For example, they may prefer a Frequentist approach as opposed to a Bayesian approach. They may even accuse the other individual of having a cognitive bias away from frequentism.

Deviations from normative models are not strictly bad for collective forecasting so long as the biases are not correlated. If all forecasters were to reason in the same way with the same information, then forecasts will be systematically biased. However, increases in variance in how people reason, and the information with which they reason, can lead to decreases in bias due to the Bias-Variance Trade-Off so long as the ways in which people reason still captures some signal. However, the central issue is that cognitive biases do tend to correlate with each other. The set of cognitive biases which have been named by researchers tend to not be isolated incidences, but systematic tendencies in the way in which people reason. Thus de-biasing is an important component of improving forecasts. Though research suggests that de-biasing mostly improves forecasts through decreasing noise.