A confirmation bias in perceptual decision-making due to hierarchical approximate inference

Making good decisions requires updating beliefs according to new evidence. This is a dynamical process that is prone to biases: in some cases, beliefs become entrenched and resistant to new evidence (leading to primacy effects), while in other cases, beliefs fade over time and rely primarily on later evidence (leading to recency effects). How and why either type of bias dominates in a given context is an important open question. Here, we study this question in classic perceptual decision-making tasks, where, puzzlingly, previous empirical studies differ in the kinds of biases they observe, ranging from primacy to recency, despite seemingly equivalent tasks. We present a new model, based on hierarchical approximate inference and derived from normative principles, that not only explains both primacy and recency effects in existing studies, but also predicts how the type of bias should depend on the statistics of stimuli in a given task. We verify this prediction in a novel visual discrimination task with human observers, finding that each observer’s temporal bias changed as the result of changing the key stimulus statistics identified by our model. The key dynamic that leads to a primacy bias in our model is an overweighting of new sensory information that agrees with the observer’s existing belief—a type of ‘confirmation bias’. By fitting an extended drift-diffusion model to our data we rule out an alternative explanation for primacy effects due to bounded integration. Taken together, our results resolve a major discrepancy among existing perceptual decision-making studies, and suggest that a key source of bias in human decision-making is approximate hierarchical inference.

Party-related primacy effects in proportional representation systems: evidence from a natural experiment in Polish local elections

We study the primacy effects that occur when voters cast their votes because a candidate or party is listed first on a ballot. In the elections that we analyzed, there are three potential types of such effects that might occur when voters vote for (1) the first candidate listed on the ballot in single-member district (SMD) elections (candidate primacy); (2) the first party listed on the ballot in open-list proportional representation (OLPR) elections (party primacy); or (3) the first candidate on a party list in OLPR elections (list primacy). We estimated the party primacy effect (2) and established that there was no interaction between (2) and (3). A party primacy effect is especially difficult to estimate because parties’ positions on ballots are typically fixed in all multi-member districts (MMDs) and it is impossible to separate the first-position “bonus” from a party’s normal electoral performance. A rare natural experiment allowed us to estimate the primacy party bonus between 6.02 and 8.52% of all votes cast for the 2014 Polish local elections. We attribute the large size of such bonus to the great complexity of voting in the OLPR elections, especially the much longer ballots, voting in many simultaneous elections, and ballot design as a booklet rather than a sheet.

Candidate name order effects in New Hampshire: Evidence from primaries and from general elections with party column ballots

Research in a few U.S. states has shown that candidates listed first on ballots gain extra votes as a result. This study explored name order effects for the first time in New Hampshire, where such effects might be weak or entirely absent because of high political engagement and the use of party column ballots. In general elections (in 2012 and 2016) for federal offices and the governorship and in primaries (in 2000, 2002, and 2004), evidence of primacy effects appeared in 86% of the 84 tests, including the 2016 presidential race, when Donald Trump gained 1.7 percentage points from first listing, and Hillary Clinton gained 1.5 percentage points. Consistent with theoretical predictions, primacy effects were larger in primaries and for major-party candidates in general elections than for non-major-party candidates in general elections, more pronounced in less publicized contests, and stronger in contests without an incumbent running. All of this constitutes evidence of the reliability and generalizability of evidence on candidate name order effects and their moderators.

Nudge the Lunch: A Field Experiment Testing Menu-Primacy Effects on Lunch Choices

By way of a field experiment conducted at a university cafeteria this paper finds that placing a vegetarian option instead of a meat option at the top of a menu decreases the share of meat dishes sold by 11%. This translates to a 6% decrease of daily emissions due to food sales. Using data on payment method, we find that the result is most likely driven by non-students responding to the nudge.

The Primacy Effect in Impression Formation: Some Replications and Extensions

Individuals described as “fun, witty, and vicious” are typically rated more favorably than those described as “vicious, witty, and fun” despite the semantic equivalence of these statements. This is known as the primacy effect in impression formation. We tested whether these effects emerge from pragmatic inferences about communicative intentions (e.g., that communicators should relay the most important information first). Participants heard a list of descriptors, with the most positive adjective listed either first or last; they also learned either that (a) the list was compiled by a human (licensing the inference that the most important information should be conveyed first) or (b) randomly ordered by a computer (thus blocking such an inference). Across five experiments (total N = 2,882), we found support for a small primacy effect in impression formation, but found no evidence of a pragmatic explanation for primacy effects.

Language Proficiency Among Respondents: Implications for Data Quality in a Longitudinal Face-to-Face Survey

When surveying immigrant populations or ethnic minority groups, it is important for survey researchers to consider that respondents might vary in their level of language proficiency. While survey translations might be offered, they are usually available for a limited number of languages, and even then, non-native speakers may not utilize questionnaires translated into their native language. This article examines the impact of language proficiency among respondents interviewed in English on survey data quality. We use data from Understanding Society: The United Kingdom Household Longitudinal Study (UKHLS) to examine five indicators of data quality, including “don’t know” responding, primacy effects, straightlining in grids, nonresponse to a self-completion survey component, and change in response across survey waves. Respondents were asked whether they are native speakers of English; non-native speakers were subsequently asked to self-rate whether they have any difficulties speaking or reading English. Results suggest that non-native speakers provide lower data quality for four of the five quality indicators we examined. We find that non-native respondents have higher nonresponse rates to the self-completion section and are more likely to report change across waves, select the primary response option, and show straightlining response behavior in grids. Furthermore, primacy effects and nonresponse rates to the self-completion section vary by self-rated level of language proficiency. No significant effects were found with regard to “don’t know” responding between native and non-native speakers.

The Primacy Effect in Impression Formation: Some Replications and Extensions

Individuals described as “fun, witty, and vicious” are typically rated more favorably than those described as “vicious, witty, and fun” despite the semantic equivalence of these statements. This is known as the primacy effect in impression formation. We tested whether these effects emerge from pragmatic inferences about communicative intentions (e.g., that communicators should relay the most important information first). Participants heard a list of descriptors, with the most positive adjective listed either first or last; they also learned either that (a) the list was compiled by a human (licensing the inference that the most important information should be conveyed first) or (b) randomly ordered by a computer (thus blocking such an inference). Across five experiments (total N = 2,882), we found support for a small primacy effect in impression formation, but found no evidence of a pragmatic explanation for primacy effects.