Language Comprehension and Labor Market Matching

How to cite

Russell, Andelyn. 2024. "Language Comprehension and Labor Market Matching." AEA RCT Registry. AEARCTR-0013822. https://doi.org/10.1257/rct.13822-1.0

Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.

Andelyn Russell*

* Primary investigator

Registry ID
AEARCTR-0013822
DOI
https://doi.org/10.1257/rct.13822-1.0
First registered
Last updated
Status
On going
Trial dates
2024-06-16 – 2024-12-15
Countries
India ()
JEL codes
O150, Z130
Registry record
http://www.socialscienceregistry.org/trials/13822

Abstract

As a medium of information transmission and symbol of group identification, language is central to economic transactions and decision-making. The economic implications of language are particularly salient in developing countries, where both greater linguistic diversity and lower GDP per capita growth suggest that language barriers create economic frictions. Yet there is limited causal evidence on the effect of language on economic outcomes due to the close correlation between language and ethnicity, as well as language learning and ability. To approach this gap, I partner with an online job board in India to examine the role of language in a key economic domain: labor market matching. In an online RCT, we measure the effect of providing additional language options for jobs posts on labor market matching.

Intervention

In a survey, first-time jobseekers are shown real jobs and provided the opportunity to apply. By default, jobseekers are shown the job posts in English. The intervention is to provide a Hindi option for viewing posts.

Experimental design

Jobseekers who are active on the job board and fulfill the following criteria are sent a message with a survey link:

- First-time jobseeker (no prior work experience on profile and has created an account on the job board no more than 60 days before receiving the survey)
- Ages 18-29
- Completed at least high school and no more than an undergraduate degree
- Lives in Delhi or Mumbai
- Hindi speaker (jobseekers who use the job board app in English and describe Hindi proficiency on their profile OR who use the job board app in Hindi)
- Interested in administration or sales jobs
- Have not completed this survey before

Upon opening the survey, jobseekers are informed that they will be shown real jobs in the department of their choice (administration or sales). If they choose to apply, their application will be sent to the recruiter. They are informed that companies wrote the job posts in English, and for some jobseekers, the posts have been translated into Hindi. Jobseekers are randomized to one of two experimental conditions:

- Control: jobseekers are shown job posts in English
- Treatment: jobseekers are shown job posts in English and can view a Hindi version of the job post as well

Treatment status is only revealed upon viewing the first job; the survey is identical across experimental conditions until that point.

At the job post level, we also randomize whether the post displays a message that the company is verified.

After viewing the job posts, jobseekers are asked about their language background, beliefs, and preferences. Half of jobseekers complete a short English comprehension test.

Design details

Randomization method
Randomization is conducted by the survey software.
Randomization unit
Randomization occurs at both the jobseeker and job post level: - Jobseeker level: -- Control (viewing job posts in English): 1/2 -- Treatment (viewing job posts in English plus a Hindi version): 1/2 - Job post level: -- Job post displays message that the company is verified: 1/4 -- Job post does not display this message: 3/4
Clusters
Because the number of clusters depends on jobseeker response rates to the surveys, it is not possible to predict the exact number of clusters. However, the target is 1,200 first-time jobseekers who assess at least 1 job.
Observations
Because the sample size depends on 1) the availability of jobs in each city and department and 2) how many jobs each jobseeker chooses to assess, it is not possible to predict the exact number of observations. Each jobseeker will assess up to 25 jobs.
Arms
It is not possible to predict the exact number of jobseekers within each experimental condition in advance. Upon opening the survey, jobseekers are randomized to one of two experimental conditions as described in “Randomization Unit.”
Primary outcomes
Jobseeker application decision. Jobseekers are shown jobs one at a time, and for each job, they are asked “Do you want to apply?” Responses are coded as 1=Yes, 0=No.
Secondary outcomes
We aim to measure the quality of matches between jobseekers and jobs. Conditional on applying to a job, we will measure the effect of the treatment on: - Recruiter clicking to reveal the jobseeker’s phone number: This is a proxy for “callbacks.” It takes a value of 1 if the recruiter clicked to view the jobseeker’s phone number, 0 otherwise. - Match score: For each job and jobseeker, the job board calculates a score reflecting the likelihood that the jobseeker is a good fit for the job. The score ranges from 0 to 1, and higher scores reflect better matches. - Distance between the jobseeker and job (in km): If obtainable, we will only analyze the treatment effect on this outcome for jobs that require working from the office. If there are extreme outliers, we anticipate winsorizing the data. It might not be possible to obtain the match quality outcomes for all jobseekers and jobs. For example, recruiters may withdraw their post from the job board, and the proxy for callbacks would not be available for these jobs. We will calculate treatment effects on these outcomes for all jobseekers and jobs for which the data can be obtained.

Cited by

No citation data available for this registration yet.

Keywords