Veranstaltungsprogramm

In this paper, we try to accomplish four goals: First, we replicate the original Asch experiment with five confederates and one naïve subject in each group (N = 210). Second, in a randomized trial we incentivize the decisions in the line experiment and demonstrate that monetary incentives lower the error rate, but that social influence is still at work. Third, we confront subjects with different political statements and show that the power of social influence can be generalized to matters of political opinion. Finally, we investigate whether intelligence, self-esteem, the need for social approval, and the Big Five are related to the susceptibility of providing conforming answers. We find an error rate of 33% for the standard length-of-line experiment which replicates the original findings by Asch (1951, 1955, 1956). Furthermore, in the incentivized condition the error rate decreases to 25%. For political opinions we find a conformity rate of 38%. However, besides openness, none of the investigated personality traits are convincingly related to the susceptibility of group pressure.

Keywords: replication of Asch, group pressure, social influence, effect of monetary incentives

In diesem Projekt untersuchen wir, inwieweit ein sprachlicher gender bias in Stellenanzeigen zur beruflichen Geschlechtersegregation und damit zur Benachteiligung von Frauen auf dem Arbeitsmarkt beiträgt.

Wir definieren sprachlichen gender bias als Sprache, mit der Genderstereotype transportiert werden. Dies geschieht durch die Verwendung von Eigenschaftsworten, welche als stereotyp männlich oder weiblich angesehen werden, oder durch Worte, die üblicherweise mit einem Gender assoziiert werden. Den sprachlichen gender bias machen wir fest an Worten wie ‚unterstützend‘ oder ‚selbstbewusst‘. Genderstereotype in Stellenausschreibungen werden als Signale wahrgenommen und beeinflussen, für wie geeignet Frauen sich selbst für eine Stelle einschätzen, für wie geeignet Frauen von Arbeitgeber:innen eingeschätzt werden und welche Einschätzung Frauen durch die Arbeitgeber:innen erwarten.

Auf der Grundlage innovativer Big Data können wir den Umfang feststellen, in welchem Unternehmen auf dem deutschen Arbeitsmarkt sprachlichen gender bias verwenden und diesen mit der horizontalen und vertikalen Geschlechtersegregation des Arbeitsmarktes in Beziehung setzen. Wir nehmen an, dass die Formulierung einer Stellenanzeige die Stellensuche sowie die betriebliche Auswahl von Bewerber:innen beeinflusst. Auf der Angebotsseite lautet unsere Grundhypothese, dass sich Frauen weniger häufig bewerben, wenn Stellenanzeigen einen sprachlichen gender bias enthalten. Auf der Nachfrageseite nehmen wir an, dass Unternehmen weibliche Bewerber:innen als weniger geeignet für die Stelle einschätzen, wenn die Stellenanzeige einen sprachlichen gender bias enthält.

Als Datenbasis verwenden wir vor allem Stellenanzeigen, die auf der Website der Bundesagentur für Arbeit (BA Jobbörse) veröffentlicht werden. Die BA Jobbörse ist eines der größten Online-Stellenportale in Deutschland und bietet den Vorteil, dass es über nahezu alle Berufe und Qualifikationsstufen hinweg breit genutzt wird. Dadurch ist es möglich, die tatsächliche Verwendung von sprachlichem gender bias für fast den gesamten deutschen Arbeitsmarkt zu analysieren. Darüber hinaus reichern wir den Zusammenhang zwischen sprachlichem gender bias und beruflicher Geschlechtersegregation auf der Makroebene mit detaillierten Umfragedaten zu Einstellungsprozessen an, um die zugrundeliegenden Mikromechanismen aufzudecken.

Argument-based models of opinion dynamics model collective processes of deliberation by social exchange of pro and con arguments regarding a certain issue. Compared to most previous opinion models, they are based on a more complex -- multi-dimensional and multi-layered -- representation of opinions which is closely related to psychological theories of attitudes. Previous work has shown that argument communication theory is capable of explaining polarization under positive (assimilative) social influence in the presence of homophily. We extend these earlier models in two directions: (i.) based on experimental research on attitude defensive processing we introduce biased argument assimilation by which the probability to adopt new information depends on current opinions, and (ii.) we incorporate an external information source providing random arguments at a certain rate (as opposed to receiving arguments from others).

In this talk, we will study the collective behavior of this extended model and address the impact of biased processing and external information by systematic computational experiments. We will show that weak biased processing significantly accelerates group decision processes. As the bias increases, groups enter a meta-stable conflictual state of bi-polarization. This suggests that a certain level of biased processing may have evolved due to the selective pressures on a group’s ability to cohesively take joint action. This transition is preserved even under high levels of noise (external unbiased information) where the impact of peer influence plays only a minor role. Using data from a survey experiment we further show that specific combinations of noise level and strength of biased processing can reproduce empirical opinion data with high accuracy.

Factorial surveys are an increasingly popular method for studying hiring decisions. They have many advantages over field experiments as they are less costly, allow for more experimental variation, are ethically unproblematic, and can record more respondent information. However, it remains contested to what extent the intentions stated in them translate to real world behavior. To shed light on this issue, we study how the validity of factorial survey results is affected by (1) the social desirability of the topic under study, and by (2) characteristics of the survey respondent. We expect that a higher level of social desirability of the topic leads to a lower validity of factorial survey results. Concerning respondent characteristics, we expect that a higher personal disposition for social desirability and a lower level of engagement with the factorial survey situation lead to a lower level of validity.

In a sequential data collection design, we first send a fictitious application to employers (field experiment) for different apprenticeship positions in Germany (N = 3000). After eight weeks, we ask the same employers to rate vignettes of fictitious applicants (factorial survey). Here, the employers are aware of the fictitious nature of the applicants, but not of the exact goal of the study. We attempt to minimize hypothetical bias by developing vignettes that closely align with the actual applications in the field experiment. To vary the level of topic sensitivity, we first look at applicants’ ethnicity, which evokes a high level of social desirability bias. Second, we study hiring probabilities based on dropout from higher education. The dropout dimension elicits only a low level of social desirability as social norms are weak and using educational level as a hiring criterion can be justified as meritocratic principle by employers. Following the vignette ratings, we present additional survey questions. Here, we ask for demographic characteristics of the recruiters, their attitudes on a range of topics, and employ scales for socially desirability and attitudes towards surveys. This design allows us to assess how personal characteristics relate to the validity of vignette ratings. The experiments are currently in the field, so that results will be available for the meeting in late September.