Impression Bias Research Articles

Dataset Scale and Societal Consistency Mediate Facial Impression Bias in Vision-Language AI

Multimodal AI models capable of associating images and text hold promise for numerous domains, ranging from automated image captioning to accessibility applications for blind and low-vision users. However, uncertainty about bias has in some cases limited their adoption and availability. In the present work, we study 43 CLIP vision-language models to determine whether they learn human-like facial impression biases, and we find evidence that such biases are reflected across three distinct CLIP model families. We show for the first time that the the degree to which a bias is shared across a society predicts the degree to which it is reflected in a CLIP model. Human-like impressions of visually unobservable attributes, like trustworthiness and sexuality, emerge only in models trained on the largest dataset, indicating that a better fit to uncurated cultural data results in the reproduction of increasingly subtle social biases. Moreover, we use a hierarchical clustering approach to show that dataset size predicts the extent to which the underlying structure of facial impression bias resembles that of facial impression bias in humans. Finally, we show that Stable Diffusion models employing CLIP as a text encoder learn facial impression biases, and that these biases intersect with racial biases in Stable Diffusion XL-Turbo. While pretrained CLIP models may prove useful for scientific studies of bias, they will also require significant dataset curation when intended for use as general-purpose models in a zero-shot setting.

First Impression Bias: Evidence from Analyst Forecasts*

Abstract We present evidence of first impression bias among finance professionals in the field. Equity analysts’ forecasts, target prices, and recommendations suffer from first impression bias. If a firm performs particularly well (poorly) in the year before an analyst follows it, that analyst tends to issue optimistic (pessimistic) evaluations. Consistent with negativity bias, we find that negative first impressions have a stronger effect than positive ones. The market adjusts for analyst first impression bias with a lag. Finally, our findings contribute to the literature on experience effects. We show that a set of professionals in the field, equity analysts, apply U-shaped weights to their sequence of past experiences, with greater weight on first experiences and recent experiences than on intermediate ones.

National culture, online medium type, and first impression bias

ABSTRACT Relying on participants with full time work experience, this study investigates if the national culture of an individual, and the culture dimensions of uncertainty avoidance, in-group collectivism and power distance affect first impression bias reduction within a personnel decision context. This study also examines whether online video is superior to online text in reducing first impression bias in a hiring decision. We find that none of the above culture dimensions nor the type of medium used significantly affect first impression bias. However, we do find that national culture, operationalized as a dummy variable, reduces first impression bias to a significant degree.

Influence of service-entry waiting on customer’s first impression and satisfaction

PurposeThe purpose of this paper is to examine the influence mechanism of waiting time on customer satisfaction based on first impression bias, which explains how customers’ perceived service-entry waiting time (PSWT) influences their first impression of service staff and satisfaction in the context of online service. Furthermore, the moderating effect of three information formats (formal, informal and hybrid) of opening remark on the relationship between PSWT and first impression, and the moderating effect of perceived in-service waiting time (PIWT) on the relationship between first impression and customer satisfaction are investigated.Design/methodology/approachTwo studies were used to verify the research model. First, an experiment on prepurchase consulting services for cruise tourism products was designed, and 810 Chinese individuals have participated. Second, 20 interviews with e-commerce practitioners in China were conducted.FindingsThe results show that, first, PSWT negatively influences customers’ first impression of service staff. Second, customers prefer the hybrid format to present opening remarks, which not only conveys the respect of the staff but also fosters a relationship. Third, in-service waits are equally as important as service-entry waits in online service. When PIWT is longer, the positive influence of first impression on customer satisfaction is weakening, resulting in lower customer satisfaction.Practical implicationsThis study provides suggestions for online service enterprises to minimize the negative impact of waiting time and improve customer satisfaction through waiting time management.Originality/valueThis study provides a new perspective for exploring the mechanism of waiting time on customer satisfaction in online service context, and extends previous research related to waiting time by exploring the influence of waiting time in multiple service stages and expression modes of service staff.

First Impressions and Analyst Forecast Bias

We present evidence of first impression bias among finance professionals in the field. Equity analysts’ forecasts, target prices, and recommendations suffer from first impression bias. If a firm performs particularly well (poorly) in the year before an analyst follows it, that analyst tends to issue optimistic (pessimistic) evaluations. Consistent with negativity bias, we find that negative first impressions have a stronger effect than positive ones. The market adjusts for analyst first impression bias with a lag. Finally, our findings contribute to the literature on experience effects. We show that a set of professionals in the field, equity analysts, apply U-shaped weights to their sequence of past experiences, with greater weight on first experiences and recent experiences than on intermediate ones.

The Role of National Culture and Multimedia on First Impression Bias Reduction: An Experimental Study in US and China

Research problem: The purpose of the study was to explore the effects of national culture and medium type on first impression bias reduction in the context of job interview and performance appraisals. Research questions: (1) Does the evaluators' national culture affect first impression bias reduction? (2) Which medium type (text or multimedia) is more effective in reducing first impression bias? and (3) Does medium effect in reducing first impression bias vary with national culture? Literature review: The purpose of the literature review is to examine the current literature in the areas of national culture, medium type and first impression bias reduction, especially in a communication context. Researchers reviewed the discussions on first impression formation, and national culture theories, and the Media Richness Theory. For media, multimedia type was found to reduce first impression bias more than text medium type. For national culture, the literature review suggested that national culture dimensions (such as collectivism, assertiveness, and uncertainty avoidance) could possibly affect first impression bias reduction. Methodology: The researchers conducted a quantitative experiment with 407 students from a US university and a Chinese university, who majored in business-related disciplines. Researchers requested the treatment group participants to access performance bias cue. The participants then performed an initial appraisal of a manager based on Denison's leadership index. The participants then viewed the job interview of the manager, via different media, and performed the appraisal again. The performance appraisal data were collected via a website. The researchers used Analysis of Variance to analyze the data. Results and discussion: This study found that for our participants, national culture reduces first impression bias more than media. It identified that first impression bias reduction in US participants is significantly greater than that in Chinese participants, independent of media used. Regardless of national culture of media users, this study found that users using text medium were able to reduce first impression bias significantly more than users using multimedia. The implications of this study are that to reduce first impression bias in cross-cultural settings, managers and employees must adapt the channels and effort allocation for communication to national culture. Managers and researchers must understand that national culture is more important than media in reducing first impression bias. The limitations of the study were the use of student participants, and absence of national culture dimensions measurement. The study was conducted in only two national cultures. Future research could use employees as participants, measure national culture dimensions, and replicate the study in various national cultures.

Lateral OFC Activity Predicts Decision Bias due to First Impressions during Ultimatum Games

Despite the prevalence and potentially harmful consequences of first impression bias during social decision-making, its precise neural underpinnings remain unclear. Here, on the basis of the fMRI study using ultimatum games, the authors show that the responders' decisions to accept or reject offers were significantly affected by facial trustworthiness of proposers. Analysis using a model-based fMRI method revealed that activity in the right lateral OFC (lOFC) of responders increased as a function of negative decision bias, indicating a greater likelihood of rejecting otherwise fair offers, possibly because of the facial trustworthiness of proposers. In addition, lOFC showed changes in functional connectivity strength with amygdala and insula as a function of decision bias, and individual differences in the strengths of connectivities between lOFC and bilateral insula were also found to predict the likelihood of responders to reject offers from untrustworthy-looking proposers. The present findings emphasize that the lOFC plays a pivotal role in integrating signals related to facial impression and creating signal biasing decisions during social interactions.

Collective increase of first impression bias

AbstractWe propose a simple model of first impression bias (FIB), where agents tend to ignore features which contradict their initial view. We consider a population of agents which are all in contact with a media, communicating randomly chosen features of an object. In some cases, we observe on simulations that FIB is significantly more frequent when the agents interact with each other than when they are only in contact with the media. We design an analytical aggregated model of the global agent‐based model behavior, which helps to explain the higher number of FIB due to the interactions. © 2009 Wiley Periodicals, Inc. Complexity, 2010

The Role of Multimedia in Changing First Impression Bias

First impression bias refers to a limitation of human information processing in which people are strongly influenced by the first piece of information that they are exposed to, and that they are biased in evaluating subsequent information in the direction of the initial influence. The psychology literature has portrayed first impression bias as a virtually “inherent” human bias. Drawing from multimedia literature, this study identifies several characteristics of multimedia presentations that have the potential to alleviate first impression bias. Based on this literature, a set of predictions was generated and tested through a laboratory experiment using a simulated multimedia intranet. Half of the 80 subjects were provided with a biased cue. Subjects were randomly assigned to four groups: (1) text with first impression bias cue, (2) multimedia with first impression bias cue, (3) text without biased cue, and (4) multimedia without biased cue. The experimental task involved conducting a five-year performance appraisal of a department head. The first impression bias cue was designed to provide incomplete and unfavorable information about the department head, but the information provided subsequently was intended to be favorable of his performance. Results show that the appraisal score of the text with biased cue group was significantly lower than the text only (without biased cue) group. On the other hand, the appraisal score of the multimedia with biased cue group was not significantly different from the multimedia only (without biased cue) group. As a whole, the results suggest that multimedia presentations, but not text-based presentations, reduce the influence of first impression bias.

The goal of forming accurate impressions during social interactions: Attenuating the impact of negative expectancies.

Investigated the idea that impression formation goals may regulate the impact that perceiver expectancies have on social interactions. In simulated interviews, interviewers Ss were given a negative expectancy about one applicant S and no expectancy about another. Half the interviewers were encouraged to form accurate impressions; the others were not. As predicted, no-goal interviewers exhibited a postinteraction impression bias against the negative-expectancy applicants, whereas the accuracy-goal interviewers did not. Moreover, the ability of the accuracy goal to reduce this bias was apparently mediated by more extensive and less biased interviewer information-gathering, which in turn elicited an improvement in negative-expectancy applicants' performances. These findings stress the theoretical and practical importance of considering the motivational context within which expectancy-tinged social interactions occur.