The Effect of Recommendation Letter Timing on Applicant Behavior

How to cite

Cecere, Grazia; Parasurama, Prasanna; Zhang, Jinny; Bapna, Sofia. 2025. "The Effect of Recommendation Letter Timing on Applicant Behavior." AEA RCT Registry. AEARCTR-0017483. https://doi.org/10.1257/rct.17483-1.0

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

Grazia Cecere* , Prasanna Parasurama , Jinny Zhang , Sofia Bapna

* Primary investigator

Registry ID
AEARCTR-0017483
DOI
https://doi.org/10.1257/rct.17483-1.0
First registered
Last updated
Status
In development
Trial dates
2025-12-23 – 2027-02-28
Countries
Private
Registry record
http://www.socialscienceregistry.org/trials/17483

Abstract

Applications—such as those to PhD programs—often require recommendation letters. We examine how the timing of this requirement affects application behavior. Specifically, potential applicants in the control group are informed that recommendation letters are required upfront (before their application is evaluated), whereas potential applicants in the treatment group are informed that recommendation letters are required only if they are shortlisted. We analyze application behavior across conditions and examine heterogeneity by gender and experience/quality.

Intervention

The study is set in the context of recruitment for an Engineering PhD program in Europe. Potential candidates are solicited via LinkedIn Recruiter InMail and are randomly assigned to one of two groups: control or treatment. The solicitation messages sent to all candidates are identical except for information about the selection process. The control and treatment groups receive the following information about the selection process.

The control group sees the following selection process:
** SELECTION PROCESS**
1. By midnight (CET) on February 15, submit application (CV, cover letter, transcript)
2. By midnight (CET) on February 15, letter writers submit two recommendation letters
3. Admission decision
Note: Recommendation letters are REQUIRED WITH THE APPLICATION.

The treatment group sees the following selection process:
** SELECTION PROCESS**
1. By midnight (CET) on February 15, submit application (CV, cover letter, transcript)
2. Shortlisting decision
3. If shortlisted, letter writers submit two recommendation letters
4. Admission decision
Note: Recommendation letters are REQUIRED ONLY IF SHORTLISTED.

Experimental design

The study is set in the context of recruitment for an Engineering PhD program in Europe.

Target candidates are identified using LinkedIn’s Recruiter tool (details under “Sample Size”) and solicited using LinkedIn's InMail.
(Note: LinkedIn’s Recruiter tool is a platform for talent acquisition. LinkedIn InMail is a premium messaging feature that lets recruiters send private messages directly to any member on the platform, using a limited number of monthly "credits" that come with a paid Recruiter subscription.)

Candidates are randomly assigned to one of two conditions: treatment or control.

Via LinkedIn InMail, each candidate receives information about the program (e.g., focus areas and campuses, program offerings, eligibility criteria, and the selection process), along with a unique link (i.e. unique to each candidate) to the application form. The InMail messages are identical across conditions except for information about the selection process (details under “Intervention”).

Due to LinkedIn InMail use limitations (constrained by LinkedIn Recruiter subscription credits, mentioned above), the InMails to candidates will be sent across three months (December 2025, January 2026, and February 2026). Candidates are randomly assigned to one of the three months (details under 'Randomization Method'); thus, we will identify upfront which month's mailing each candidate belongs to.

As noted, each candidate receives a unique URL, which limits them to a single application. The candidate’s name and LinkedIn ID are linked to the URL; if a different candidate uses the URL to apply, the observation will be discarded. A candidate may access the URL across different devices and resume the application where they left off if they do not complete it in one session.

Applications will not be accepted after the application deadline noted in the InMail.

We analyze application behavior (details under “End Points”) across conditions and examine heterogeneity by gender and experience/quality.

Design details

Randomization method
By a computer. Candidates will be solicited by LinkedIn InMail. Due to LinkedIn InMail use limitations (constrained by the LinkedIn Recruiter subscription) the InMails to candidates will be sent across three months (Dec 2025, Jan 2026, Feb 2026). First, candidates are randomly assigned to one of the three months using simple randomization. Then, candidates are block-randomized into two groups (control and treatment), by month, country and gender. Note: For country, we will have six categories. The first five correspond to the countries with the most potential candidates. The remaining countries are grouped into the sixth category.
Randomization unit
Individual
Clusters
N/A
Observations
~2100 candidates identified via LinkedIn's Recruiter tool (see 'Sample Size' for details.
Arms
2,086 candidates. This represents the complete set of potential candidates identified using the LinkedIn Recruiter tool based on the specified search criteria. The search was conducted on December 17, 2025. Search criteria: • Keywords: Research • Degree: Excluded Doctor of Philosophy and Doctorate; included Master of Engineering or Engineer’s degree • Graduation year: 2024–2026 • Seniority: Entry level • Field of study: Mechanical Engineering; Computer and Information Sciences and Support Services; Engineering; Electrical and Electronics Engineering; Computer Science; Industrial Engineering; Computational Science; Civil Engineering; Computer Engineering • Spoken language: English • Interest: Open to work • Locations: France, Spain, Switzerland, Malta, Italy, Portugal, Cyprus, Luxembourg, Belgium, Germany, Austria, Poland, Romania, Czechia, Slovakia, Slovenia, Bulgaria, Netherlands, Denmark, Sweden, Finland, Norway, Hungary, Greece, Estonia, Lithuania, Latvia, Ireland, and Croatia If a message cannot be sent to one of these candidates via LinkedIn’s InMail (e.g., because LinkedIn removes them from the candidate pipeline), that observation will be excluded from our sample.
Primary outcomes
Whether the solicited candidate started the application (binary)
Secondary outcomes
Whether the solicited candidate completed the application (binary). Whether the solicited candidate responded via LinkedIn InMail that they were interested (binary). Whether the solicited candidate was selected (binary); conditional on this data being made available by the recruiting institution.

Cited by

No citation data available for this registration yet.

Keywords