Experimental Design
We run a stated-choice online experiment among newly enrolled undergraduate students at a large public University in Germany. Participants state, for a series of ficticious profiles of fellow students, whether or not they consider subjects with respective profiles as potential network partners. Profiles are described by characteristics (study field, gender, mother tongue other than German, main study motive, A-level GPA) and personality traits (conscientiousness, altruism, competitiveness, self-efficacy). In all dimensions, attributes are randomly drawn. 'Mother tongue other than German' is either empty or takes values 'Spanish', 'Turkish', or 'Arabian'. The dimensions A-level GPA and all traits take values 'above average=yes' or 'above average=no'. Participants state, for each profile separately, whether or not they consider a given profile as representing a potential network partner. The stated choices are incentivized: Subjects are informed that after the survey, they will be invited to an online networking event where they will be matched with fellow students in accordance with the (mutually stated) preferences over network partners. Subjects who truthfully state their preferences thus increase their chances to meet fellow students with preferred charcteristics. To test to what extent stated preferences reflect preferences over mating partners rather than network partners, three out of the nine profiles rated in total do not contain information on gender. Using only the subset of profiles without gender information, we will test to what extent preferences obtained from the set of complete profiles are robust to not revealing gender (and thus closing down the mating channel). Our main interest is understanding to what extent network preferences are shaped by homophily. Specifically, we plan to run estimations where the dependent variable will be an indicator for whether or not subjects consider a given profile as representing a potential network partner. The regressors will be differences between the subjects own characteristics and traits and those of the respective profile. Standard errors will be clustered at the level of the subject (participant).