Belief-updating about elderly’s cognitive abilities in the Czech labour market – Evidence from a machine learning-assisted field experiment

Last registered on October 31, 2025

Pre-Trial

Trial Information

General Information

Title
Belief-updating about elderly’s cognitive abilities in the Czech labour market – Evidence from a machine learning-assisted field experiment
RCT ID
AEARCTR-0017127
Initial registration date
October 28, 2025

Initial registration date is when the trial was registered.

It corresponds to when the registration was submitted to the Registry to be reviewed for publication.

First published
October 31, 2025, 8:27 AM EDT

First published corresponds to when the trial was first made public on the Registry after being reviewed.

Last updated
October 31, 2025, 8:54 AM EDT

Last updated is the most recent time when changes to the trial's registration were published.

Locations

Region

Primary Investigator

Affiliation
Masaryk University

Other Primary Investigator(s)

PI Affiliation
Masaryk University

Additional Trial Information

Status
In development
Start date
2025-10-30
End date
2025-11-30
Secondary IDs
Prior work
This trial does not extend or rely on any prior RCTs.
Abstract
We run a correspondence-field experiment in the Czech Republic to test whether employers reward cognitive abilities in mid/end-career applicants. We send fictitious applications to job advertisements, each vacancy receives one application only; that is, this experiment has a between-subjects structure and this choice is meant to reduce detectability concerns (Balfe et al., 2021). Moreover, while this choice usually limits statistical power, it addresses the concern that matched-pair (within-employer) correspondence may alter behavior and impact mechanism-specific discriminatory behaviors (Lahey, 2016).
Our applications consist in CVs, which are all equivalent in terms of education, work history, and English proficiency, while they have three randomized attributes. First, “Age,” disclosed implicitly via graduation year and evenly assigned from 40 to 55. The other two randomized characteristics are meant to capture different dimensions of cognitive abilities, at different margins. Second, “Certificate of Portuguese language knowledge” with three levels—None, CIPLE: A2 level or DAPLE: C1 level (i.e., certificate no or yes, and, if yes, how advanced); this signal means to capture verbal skills. Third, “Playing chess” as an analytical-skill signal with three levels—None, “Chess enthusiast”, or “Chess enthusiast ranked in in the first quartile of national ranking at Chess.com in 2024” (i.e., chess players no or yes, and, if yes, how good); this signal means to capture analytical cognitive skills and their intensity.
These randomized plausible signals are chosen to disclose different dimensions of cognitive skills and to be productivity-unrelated in most jobs within the Czech context. These features allow us to study how belief-updating about ability—rather than job-match (more on this below)—varies with age.
These fictitious applications will reach employers all over the country and for all kinds of jobs advertised through publicly available database of Public Employment Service of the Czech Republic.
While the above three attributes are randomized and orthogonal to employers’ needs, the reminder of the CVs closely fits employers’ needs. These CVs follow the usual Czech CVs structure, that is, one sheet, where the top panel lists job experience for all applicants, while the bottom panel reports additional qualifications, or hobbies, and a short cover letter-like paragraph. With the help of machine learning techniques, the top panel lists relevant, ad-fitting job experience from the last 10 years for all candidates (as such, the fit of each individual CV does not need to be tested with actual people); this strategy addresses the concern that older applicants would list longer job relevant experience. The short bio reports the usual, vague yet credible positive self-promotion (e.g., “I am reliable, proactive, and my previous employers appreciate my consistent performance”), and it explains that the candidate is seeking for a new job after having recently relocated to the area.
With the help of machine learning, we additionally assign candidates’ address. This address is composed of a randomly drawn, real town and street, as well as a fake number, between 5 and 20 km from the registered place of work or (if not available) address of the employer. While the combination of town and street are real, their combination with number returns a non-existent address—this is done to address potential ethical concerns.
Our applicants are all male. This choice is guided by various reasons; most importantly, concerns with statistical power in different occupations. While previous studies have showed that the response rate—when answering job ads—might generally be high in the Czech labour market, e.g., ~30%, it might vary a lot across occupations (Bartos et al., 2016).
The study wants to clarify whether job-irrelevant cognitive ability, move screening decisions differently for applicants aged 40-55, informing debates on late-career signalling and the limits in the increase of generic cognitive abilities.
External Link(s)

Registration Citation

Citation
Fumarco, Luca and Stepan Mikula. 2025. "Belief-updating about elderly’s cognitive abilities in the Czech labour market – Evidence from a machine learning-assisted field experiment." AEA RCT Registry. October 31. https://doi.org/10.1257/rct.17127-1.1
Sponsors & Partners

There is information in this trial unavailable to the public. Use the button below to request access.

Request Information
Experimental Details

Interventions

Intervention(s)
To measure the differential belief-update of elderly’s cognitive abilities, we conduct a randomized email correspondence test, in which each employer receives one email indicating the applicant’s age—between 40 and 55—and all the possible combinations of applicant’s job-irrelevant cognitive abilities, that is, Portuguese knowledge and chess. In other words, age, job-irrelevant cognitive ability proxies and their margin (either recency or ability level) are independently assigned.
Intervention (Hidden)
Intervention Start Date
2025-10-30
Intervention End Date
2025-11-30

Primary Outcomes

Primary Outcomes (end points)
Our pre-specified analysis estimates intent-to-treat effects of age, verbal and analytical skills on various outcomes, namely binary variables for any callback and interview invitation.
Primary Outcomes (explanation)

Secondary Outcomes

Secondary Outcomes (end points)
We measure the time to respond, in days.
We measure the courteousness of the responses.
Secondary Outcomes (explanation)

Experimental Design

Experimental Design
Main experimental design details are discussed in the abstract, here we summarize them. We send one job application per advertisement, thus using between-variation. CVs are randomized across age and job-irrelevant cognitive abilities. We signal age through the linear combination of current year of the experiment, graduation year, and year of graduation typical for required education level. We signal job-irrelevant verbal cognitive abilities through different levels of Portuguese knowledge. We signal job-irrelevant analytical cognitive abilities through different levels of chess skills. Attached documents illustrate a sample of CVs.
Experimental characteristics section discusses randomization details.
Fictitious job applicants have different combinations of the most frequent Czech male first names and family names.
We will apply for all relevant advertised job vacancies in the database of vacancies registered at the Public Employment Service of the Czech Republic. We exclude from the database jobs that are not posted by business companies.
Although our experimental design was informed by planning and reading of the literature, exogenous deadlines imposed some limitations. In particular, we could not run a perception check plan of our treatments (e.g., blinded raters scoring “verbal ability,” “analytical ability,” “general job-irrelevant”). Thus, we will conduct this test after the experiment and upload the results here.
After the experiment concludes, we will release mail sending and randomization procedure publicly.
Experimental Design Details
Randomization Method
Randomization occurs at the vacancy level, and the randomizing script will be posted, and timestamped.
Randomization Unit
Vacancy level
Was the treatment clustered?
No

Experiment Characteristics

Sample size: planned number of clusters
~11000 job advertisements
Sample size: planned number of observations
~11000 job advertisements
Sample size (or number of clusters) by treatment arms
Each employer receives one email.
The assignment of random treatment “age” is stratified by region, occupation and economic activity (defined by the highest ISCO and NACE codes), and suitability of the vacancy for third country (i.e., outside EU) citizens reported by employers. This is to ensure that age is distributed as uniformely as possible across those dimensions. The expected 1/15 probability that for each age (integer between 40 and 55) and realization varies between 707 and 775 per integer.
Portuguese knowledge is randomly assigned with 1/3 probability for each level, that is, no Portuguese mention ~3,950, Portuguese A1 certificate ~3,950, Portuguese C2 certificate in 2025 ~3,950.
Chess ability is randomly assigned with 1/3 probability for each level, that is, no chess mention ~3,950, being chess enthusiast ~3,950, well-ranked chess enthusiast ~3,950.
Assignment and level of the two cognitive ability signals are independent.
Minimum detectable effect size for main outcomes (accounting for sample design and clustering)
Assuming N=11,000, a baseline callback=0.12 (this is based on a worse-case scenario, with a callback rate slightly larger than Bertrand and Mullainathan (2004), Duguet et al. (2010), Carlsson and Rooth (2007)), alpha=0.05, and power=0.8, the minimum detectable difference is in the ballpark of ~2 percentage points, same goes for age.
Supporting Documents and Materials

Documents

Document Name
Advertisement specific CV - sample 5 CVs
Document Type
other
Document Description
File
Advertisement specific CV - sample 5 CVs

MD5: 0e4ef8659a00311d4382100fb0fc6d9d

SHA1: ad4e35f0e09be2302814b7c2075901317d7fa5f3

Uploaded At: October 28, 2025

There is information in this trial unavailable to the public. Use the button below to request access.

Request Information
IRB

Institutional Review Boards (IRBs)

IRB Name
Etická komise pro výzkum (Research Ethics Committee)
IRB Approval Date
2025-10-17
IRB Approval Number
EKV-2025-112
Analysis Plan

Analysis Plan Documents

Pre-analysis plan - clean (no track changes, sorry for that)

MD5:

SHA1:

Uploaded At: October 31, 2025

Post-Trial

Post Trial Information

Study Withdrawal

There is information in this trial unavailable to the public. Use the button below to request access.

Request Information

Intervention

Is the intervention completed?
No
Data Collection Complete
Data Publication

Data Publication

Is public data available?
No

Program Files

Program Files
Reports, Papers & Other Materials

Relevant Paper(s)

Reports & Other Materials