Myopia for Statistical Correlations

How do distorted images of social groups arise - for example of women and men, immigrants and natives or ethnic minorities? Why do differences and inequalities sometimes appear more dramatic in the public debate than the statistics suggest - and in other cases rather trivialized? Prof. Dr. Hans Alves is investigating these questions in his ERC Consolidator project "The Attribute Frequency Model: Understanding and Mitigating Biases in Social Perception". He is developing a new model of social perception that traces many of these biases back to a simple, previously underestimated mechanism: the frequency of certain characteristics in our social environment.

The "attribute frequency model" assumes that we judge groups differently depending on whether we think about rare or frequent characteristics - such as unemployment (rare) or employment (frequent), top positions or average jobs, serious crimes or normal, mostly low-conflict everyday life. When we focus on rare characteristics, group differences usually appear larger and inequalities more extreme; when we think about frequent characteristics, the same groups often appear much more similar and inequalities smaller.

Concrete figures from Germany show how strong this change in perspective is: In 2020, the unemployment rate among immigrants was 14.4 percent, compared to 4.7 percent among German citizens. Immigrants were therefore more than three times as likely to be unemployed as German citizens - a very clear difference. However, if we look at the employment rates, we see that 85.6 percent of immigrants and 95.3 percent of Germans were employed; Germans were therefore only around 1.2 times more likely to be employed than immigrants. Depending on the perspective, the same social reality can therefore appear to be either very different (unemployment) or relatively similar (employment) - with the risk of overestimating or underestimating inequalities.

The project assumes that people hardly see through such statistical correlations in their everyday lives - a form of "meta-cognitive myopia". Rare characteristics are also particularly conspicuous: they signal problems (such as unemployment, serious crimes, rare serious illnesses) or extraordinary successes (such as top incomes, top positions) and therefore often dominate media reports and everyday conversations.

The ERC project investigates these assumptions using a broad mix of methods: online experiments with thousands of participants are planned, in which real statistics on income, wealth, the labor market, university careers or crime are presented in various formats. In addition, analyses of large data sets from official statistics and content analyses of media reports and social media will be used to examine whether public communication actually systematically focuses on rare extremes or rather frequent normal cases - and how this affects the assessment of inequality and the social situation. Finally, the team will test whether large language models (artificial intelligence) show similar bias patterns to humans.

A central aim of the project is to develop concrete recommendations on how information on social inequalities can be better presented in the future - for example, by simultaneously presenting rare and frequent characteristics or using key figures that are less susceptible to bias. The results are intended to support politicians, administrators, the media and organizations in conveying as realistic a picture of social differences as possible, reducing stereotypes and prejudices and focusing measures against actual injustices where they have the greatest impact on society.