The Talent Gap in Family Firms

The goal of our experiment is to quantify the preference of students to work for family firms when they graduate and how family ownership and control of a firm affects their decision. To do that, we are conducting an experimental survey following Colonelli et. al (2025) which is inspired by the non-deceptive resume rating design proposed by Kessler et al (2019). The methodology estimates preferences while avoiding deception. The students are asked to read synthetic job postings in which we randomize size, age, wage and degree of family ownership/participation in governance. The respondents are invited to report their interest in those job postings

The goal of our experiment is to quantify the preference of students to work for family firms when they graduate and how family ownership and control of a firm affects their decision. To do that, we are conducting an experimental survey following Colonelli et. al (2025) which is inspired by the non-deceptive resume rating design proposed by Kessler et al (2019). The methodology estimates preferences while avoiding deception. The students are asked to read synthetic job postings in which we randomize size, age, wage and degree of family ownership/participation in governance. The respondents are invited to report their interest in those job postings and respond 1) how interested they would be to receive a job offer for this job position, 2) how likely they think is that the company will offer them the position. There is no deception because the respondents know that the job postings are hypothetical. Also, we are giving the respondents incentive to truthfully respond because we inform them that we will use their responses to match them with real job posting using an AI tool (OpenAI).

2025-09-01

2025-10-15

Understand how likely they are to apply to work for family firms

The goal of our experiment is to quantify the preference of students to work for family firms when they graduate and how family ownership and control of a firm affects their decision. To do that, we are conducting an experimental survey following Colonelli et. al (2025) which is inspired by the non-deceptive resume rating design proposed by Kessler et al (2019). The methodology estimates preferences while avoiding deception. The students are asked to read synthetic job postings in which we randomize size, age, wage and degree of family ownership/participation in governance. The respondents are invited to report their interest in those job postings and respond 1) how interested they would be to receive a job offer for this job position, 2) how likely they think is that the company will offer them the position. There is no deception because the respondents know that the job postings are hypothetical. Also, we are giving the respondents incentive to truthfully respond because we inform them that we will use their responses to match them with real job posting using an AI tool (OpenAI). In the attached material we describe the survey flow.
We will send the survey to students of economics in Denmark.

To construct our sample surveys, we first use a data-scraping tool to gather job descriptions from a Danish website commonly used by undergraduate economics students at the University of Copenhagen (jobindex.dk). We filter these descriptions using an AI tool(ChatGPT) to ensure all jobs in the sample are related to economics, finance or business administration and are suitable for recent graduates in these fields.Next, we use the AI tool to analyze these descriptions and to determine the common structure of job postings as well as the information presented in each portion of the job description. Finally, we had AI generate several job descriptions, but randomly altered the quantities of firm size, age and ownership. Students are then asked to rate how likely they are to apply for the job, given the randomized qualities.

each student will rate 15 profiles

No

1500 students from 1 university

we expect 200-300 respondents

we will email 2000 students

Our simulation-based power calculation estimates how often the effect of family status on a hypothetical outcome (prob_apply) is detected as statistically significant (at the 5% level) across repeated samples of varying sizes. The regression model includes firm size, salary, and family status as predictors, with values drawn from the same distribution as the survey. The script computes the proportion of simulations where the p-value on family status falls below 0.05. This proportion approximates the statistical power for detecting that specific effect size at each sample size. In essence, this calculation helps determine the minimum number of observations required to detect a true effect of family status on the outcome with sufficient probability (e.g., ~80% power).