Worker Sorting and Gender-Inclusive Language in Job Ads

We collect data in an online survey-based discrete choice experiment in which respondents are asked to choose between job ads with exogenously varying attributes, following a similar setup as Maestas et al. (2023) and Nagler et al. (2023).

We study:
(i) whether job-seekers have any preferences over non-binary gender-inclusive language (NB-GIL) in vacancy postings
(ii) if yes, what is the willingness to pay to work for an employer that includes NB-GIL in their job ads
(iii) what worker or job characteristics drive these preferences

NB-GIL is not a standard job attribute (such as job flexibility or remote work that have been studied in the literature) but primarily a signal about the firm. As a result, we don’t ask about it directly in the experimental part of our survey, and include randomization whenever respondents come into contact with NB-GIL in our survey instructions.

We incentivize the survey by providing the respondents with non-public information from an online job board about wages and job attributes of recent vacancies based on their answers in the survey. We inform respondents about this reward at the start of the survey, which provides additional incentives to fill the survey accurately.

2025-01-22

2025-02-15

(1) Individual preferences over characteristics of job vacancies, including wage and NB-GIL language

(1) Individual preferences over characteristics of job vacancies:
(i) wage (drawn from a uniform distribution from -20% to +30% compared to average wages in this occupation based on real wage data from the online job board Indeed Germany)
(ii) NB-GIL language (binary)
(iii) number of vacation days (24, 26, 28, 30, 32)
(iv) flexible working (binary)
(v) subsidized lunch (binary)
(vi) free public transport ticket (Deutschlandticket) (binary)

We will measure these from stated preferences in our discrete choice experiment.

(1) Individual perceptions of employers offering NB-GIL
(2) Individual opinions on a range of political issues

(1) Individual perceptions of employers offering NB-GIL, measured as share of respondents who believe an employer that uses NB-GIL in a job ad :
(i) is young and dynamic
(ii) offers better perks and benefits
(iii) provides a more international work environment
(iv) has more women and minorities in leadership roles
(v) hires workers regardless of their political opinions and beliefs
(vi) is less focused on profits
(vii) offers meaningful work


We also ask respondents about their perceived chances to get a job offer by such a firm.

(2) Individual opinions on a range of political issues, measured as share of respondents who believe:
(i) In Germany, you can no longer express your opinion freely
(ii) Refugees only come to Germany to take advantage of our welfare state.
(iii) Women still face discrimination in the labor market and in the society.
(iv) Law enforcement is too lax when it comes to prosecuting climate protesters, such as "Last Generation".
(v) The elderly generation lives at the expense of the young.
(vi) It's important to use gender-inclusive language.

Individual perception of their political views, measured as a number on 0-100 scale for left- vs right-leaning.

Voting intentions for the next election, measured as share of respondents choosing one of 7 leading German political parties

The discrete choice experiment presents the respondent 7-8 pairs of hypothetical job vacancies and asks the respondent to choose which one they prefer more from each pair. The presented vacancies randomly vary any of the 6 dimensions listed in part (1) of question 3 above (all positions are indicated to be full-time). In particular, the random variation in wage allows us to estimate the willingness to pay (WTP) for these amenities listed in job ads. The job title and wage of the vacancies presented to each individual depend on their occupation and preferred job title, which the respondent chooses at the start of the survey.

In addition, we assign the respondents to 6 orthogonal treatment arms:
information about the content of vacancy postings: we show all respondents two postings to demonstrate what to expect in the experiment
(1) Vacancy with NB-GIL language shown first, then vacancy without NB-GIL language
(2) Vacancy without NB-GIL language shown first, then vacancy with NB-GIL language

Questions about perceptions about employers offering NB-GIL: respondents are asked to give their opinion about a hypothetical employer based on a shown vacancy posting
(1) Vacancy with NB-GIL and all amenities (lunch, Deutschlandticket, flexible work)
(2) Vacancy without NB-GIL and all amenities (lunch, Deutschlandticket, flexible work)
(3) Vacancy with NB-GIL and no amenities
(4) Vacancy without NB-GIL and no amenities

The respondents are also asked standard questions about their demographic characteristics and political opinions. We ask the respondents how they search for jobs to be able to match the results of our experiment to observational data on revealed preference for NB-GIL from an online job board.

The survey is incentivised: at the end of the survey, we display information about average pay and job attributes based on respondents’ answers. This information is taken from online job board Indeed.com and is not publicly available, and thus acts as an incentive for the respondent to fill the survey accurately.

Not available

Randomization done by a computer

Individual (choice between two job ads)

No

6000 individuals

Between 42,000 - 48,000 hypothetical choices. The exact number depends on the occupations selected by the respondents. Note that answers by170 resondents have been collected already as a part of piloting the study. We will pool this data with that from the pre-registered survey if the design survey doesn’t change substantially. We will separately present results from the pooled and pre-registered-only data.

Between 7,000 and 8,000 observations within each treatment arm.

We cannot provide minimum detectable effect sizes because we are the first to estimate the willingness to pay for the specific politically-charged language. However, our expriment setup is similar to Nagler et al. (2023) who ran similarly-sized experiments of 3,000 respondents. Our sample size is somewhat larger because preferences over specific language is likely to be more heterogeneous across groups of workers and we are interested in documenting this heterogeneity.

Pre-analysis plan