Training machine learning algorithms for automatic facial coding: The role of emotional facial expressions' prototypicality

Automatic facial coding (AFC) is a promising new research tool to efficiently analyze emotional facial expressions. AFC is based on machine learning procedures to infer emotion categorization from facial movements (i.e., Action Units). State-of-the-art AFC accurately classifies intense and prototypical facial expressions, whereas it is less accurate for non-prototypical and less intense facial expressions. A potential reason might be that AFC is typically trained with standardized and prototypical facial expression inventories. Because AFC would be useful to analyze less prototypical research material as well, we set out to determine the role of prototypicality in the training material. We trained established machine learning algorithms either with standardized expressions from widely used research inventories or with unstandardized emotional facial expressions obtained in a typical laboratory setting and tested them on identical or cross-over material. All machine learning models' accuracies were comparable when trained and tested with held-out dataset from the same dataset (acc. = [83.4% to 92.5%]). Strikingly, we found a substantial drop in accuracies for models trained with the highly prototypical standardized dataset when tested in the unstandardized dataset (acc. = [52.8%; 69.8%]). However, when they were trained with unstandardized expressions and tested with standardized datasets, accuracies held up (acc. = [82.7%; 92.5%]). These findings demonstrate a strong impact of the training material's prototypicality on AFC's ability to classify emotional faces. Because AFC would be useful for analyzing emotional facial expressions in research or even naturalistic scenarios, future developments should include more naturalistic facial expressions for training. This approach will improve the generalizability of AFC to encode more naturalistic facial expressions and increase robustness for future applications of this promising technology.

Firing discrimination: Selective labor market responses of firms during the COVID-19 economic crisis

The speed of the economic downturn in the wake of the COVID-19 pandemic has been exceptional, causing mass layoffs-in Germany up to 30% of the workforce in some industries. Economic rationale suggests that the decision on which workers are fired should depend on productivity-related individual factors. However, from hiring situations we know that discrimination-i.e., decisions driven by characteristics unrelated to productivity-is widespread in Western labor markets. Drawing on representative survey data on forced layoffs and short-time work collected in Germany between April and December 2020, this study highlights that discrimination against immigrants is also present in firing situations. The analysis shows that employees with a migration background are significantly more likely to lose their job than native workers when otherwise healthy firms are unexpectedly forced to let go of part of their workforce, while firms make more efforts to substitute firing with short-time working schemes for their native workers. Adjusting for detailed job-related characteristics shows that the findings are unlikely to be driven by systematic differences in productivity between migrants and natives. Moreover, using industry-specific variation in the extent of the economic downturn, I demonstrate that layoff probabilities hardly differ across the less affected industries, but that the gap between migrants and natives increases with the magnitude of the shock. In the hardest-hit industries, job loss probability among migrants is three times higher than among natives. This confirms the hypothesis that firing discrimination puts additional pressure on the immigrant workforce in times of crisis.

Longitudinal analyses of gender differences in first authorship publications related to COVID-19

OBJECTIVE

Concerns have been raised that the COVID-19 pandemic has shifted research productivity to the disadvantage of women in academia, particularly in early career stages. In this study, we aimed to assess the pandemic's effect on women's COVID-19-related publishing over the first year of the pandemic.

METHODS AND RESULTS

We compared the gender distribution of first authorships for 42 898 publications on COVID-19 from 1 February 2020 to 31 January 2021 to 483 232 publications appearing in the same journals during the same period the year prior. We found that the gender gap-the percentage of articles on which men versus women were first authors-widened by 14 percentage points during the COVID-19 pandemic, despite many pertinent research fields showing near equal proportions of men and women first authors publishing in the same fields before the pandemic. Longitudinal analyses revealed that the significant initial expansions of the gender gap began to trend backwards to expected values over time in many fields. As women may have been differentially affected depending on their geography, we also assessed the gender distribution of first authorships grouped by countries and geographical areas. While we observed a significant reduction of the shares of women first authors in almost all countries, longitudinal analyses confirmed a resolving trend over time.

CONCLUSION

The reduction in women's COVID-19-related research output appears particularly concerning as many disciplines informing the response to the pandemic had near equal gender shares of first authorship in the year prior to the pandemic. The acute productivity drain with the onset of the pandemic magnifies deep-rooted obstacles on the way to gender equity in scientific contribution.