Abstract
Confirmation bias is often used as an umbrella term for many related phenomena. Information searches, evidence interpretation, and memory recall are the three main components of the thinking process involved in hypothesis testing most relevant to investigations of confirmation bias; yet these have rarely been explored using a unified paradigm. Therefore, this paper examines how confirmation bias works in each of these three stages of reasoning, using four controversial topics. Participants (N = 199) first indicated their attitudes and then answered tasks measuring confirmation bias. The results showed that confirmation bias was most prevalent in information search as participants tended to search for information confirming their prior attitudes. During information interpretation, confirmation bias occurred only for more polarizing topics. On the other hand, our results did not show confirmation bias in memory recall, as there was no difference in recall of information confirming or disconfirming prior attitudes for any of the topics. Although our attitudes affect the way we process information, it seems the effect varies depending on the reasoning stage, and this can have implications for debiasing strategies.
Disclosure statement
No potential conflict of interest was reported by the authors.
Notes
1 Cognitive psychology views confirmation bias mostly as a systematic error (deviation from the normative model of reasoning) (Fiske & Taylor, Citation1991; Kahneman, Citation2003; Wason & Evans, Citation1974) (but see also Austerweil & Griffiths, Citation2011; Jern et al., Citation2014; Perfors & Navarro, Citation2009 ). Some other psychology fields view people’s hypothesis development processes as optimal (Klayman, Citation1995): for example, in social psychology confirmation bias is seen as an adaptive mechanism (Dardenne & Leyens, Citation1995; Hahn & Harris, Citation2014), and in evolutionary psychology it is a design feature that has evolved as a survival mechanism (Haselton et al., Citation2016; Mercier & Sperber, Citation2017).
2 We are aware that such a huge drop-out rate poses a great risk of self-selection. The main reasons participants did not complete the survey were the time it took (30 to 60 minutes) and cognitive resources. However, the end-of-survey feedback from those who finished it suggested the collected data was of good quality, as most responses were positive and participants declared they had put their best effort into the survey. It is possible that had everybody been incentivized to complete the survey the results might have been different, but it is doubtful that responses under such conditions would be of better quality and participants would perhaps answer randomly, obscuring the effects.
3 For the purposes of this study the actual words the participants typed were irrelevant, as this procedure served only to increase the ecological validity – so the process would resemble a real Google information search. The list of titles given to the participants was supposed to look like the results obtained from what they had typed in, rather than a pre-prepared list. Nonetheless, we analyzed their key words from this stage to examine whether confirmation bias could also be measured this way. There were no nonsensical words and the results were coded by the two independent coders.
4 Although Slovakia is generally a conservative country, as in other European countries there is no death penalty, very few schools require school uniforms, and most people do not own a gun.
5 Please note that this is an arbitrary method for calculating the direction of the bias; other ways could be used. Our inference of confirmation bias is based on the analysis of whether attitude (position on a given issue) affects this index – people with more positive attitudes should search for more positive information (their biased search index should be highly positive), while people with more negative attitudes should search for more negative information (their biased searched index should be highly negative). The same logic applies to the biased interpretation index and biased memory recall index described below.
6 Raw data are available at: https://osf.io/mnt7y/.