|
Field
Trial Status
|
Before
in_development
|
After
on_going
|
|
Field
Abstract
|
Before
This study explores how gender affects communication between borrowers and lenders. In partnership with a bank in Afghanistan, we invited micro-entrepreneurs to apply for small loans. The application collects standard financial information, and it also gives borrowers the opportunity to write a message to the loan officer explaining, in their own words, why their ability and willingness to repay can be trusted. Each applicant can provide one explanation addressed to a male loan officer and one addressed to a female loan officer. Providing financial information is mandatory, but writing a message is optional. By analyzing the narratives written by the same applicant, we assess whether individuals—and particularly female applicants—communicate differently when the message is directed to an opposite-gender loan officer. Next, we recruit loan officers to evaluate the applications and test whether the marginal value of borrower narratives is lower when they communicate with opposite gender. These findings will allow us to evaluate whether gender-based communication barriers contribute to disparities in access to finance and whether increasing the number of female loan officers could help reduce these gaps.
|
After
This study explores how gender affects communication between borrowers and lenders. In partnership with a bank in Afghanistan, we invited micro-entrepreneurs to apply for small loans. The application collects standard financial information, and it also gives borrowers the opportunity to write a message to the loan officer explaining, in their own words, why their ability and willingness to repay can be trusted. Each applicant can provide one explanation addressed to a male loan officer and one addressed to a female loan officer. Providing financial information is mandatory, but writing a message is optional. By analyzing the narratives written by the same applicant, we assess whether individuals—and particularly female applicants—communicate differently when the message is directed to an opposite-gender loan officer. Next, we recruit loan officers to evaluate the applications and test whether the marginal value of borrower narratives is lower when they communicate with opposite gender. These findings will allow us to evaluate whether gender-based communication barriers contribute to disparities in access to finance and whether increasing the number of female loan officers could help reduce these gaps.
AMENDMENT OF SEPTEMBER 11, 2026 — NEW VALIDATION STUDY AND LOAN-OFFICER EXPERIMENT
This amendment prospectively adds a narrative-validation study and a new loan-officer experiment to AEARCTR-0017067. These studies build on the earlier experiments and use applications and narratives from the original first-round borrower sample, not the later information-treatment round.
The neutralized narrative corpus was produced and frozen on August 18, 2026, using a locally run large language model to identify explicit recipient-gender expressions for automated removal. No manual narrative edits or subsequent narrative changes were made. At submission of this amendment, no participant in either new study has been recruited, consented, exposed to study materials, or provided an outcome response. No new officer treatment assignments have been generated. Survey batches have been prepared before participant recruitment.
The validation survey is planned for September 12, 2026. Native Dari speakers who can read Dari will guess the gender of the original intended addressee from displayed narratives. The new officer experiment is planned to start on September 19, 2026, with evaluations completed by October 3, 2026. Approximately 100 new loan officers will each evaluate 150 applications. Officers will be randomized between a regime using generic narrative headings and a regime whose headings truthfully disclose the gender of the narrative's original intended addressee. All displayed officer-study narratives will use the frozen neutralized texts. The primary officer outcome is the assessed probability of repayment, reported from 0 to 100 in increments of ten.
|
|
Field
Trial End Date
|
Before
June 01, 2026
|
After
October 03, 2026
|
|
Field
Last Published
|
Before
December 11, 2025 11:48 AM
|
After
September 11, 2026 04:20 PM
|
|
Field
Intervention (Public)
|
Before
This study examines how borrowers’ written communication in small-loan applications influences loan officers’ evaluations and actual lending outcomes at a partner bank. Eligible micro-entrepreneurs complete standardized, collateral-free loan applications that include a short narrative describing their ability and willingness to repay. Each applicant can provide one narrative addressed to a male loan officer and one addressed to a female loan officer. All applications are anonymized before review so that loan officers cannot see any borrower identifiers. Loan officers evaluate applications using the bank’s regular scoring procedures, and loan decisions are based on these evaluations. After the evaluation phase, officers complete a short follow-up survey about their review experience. Some details of the study design are temporarily withheld to preserve research integrity and will be made public once data collection and analysis are complete.
|
After
This study examines how borrowers’ written communication in small-loan applications influences loan officers’ evaluations and actual lending outcomes at a partner bank. Eligible micro-entrepreneurs complete standardized, collateral-free loan applications that include a short narrative describing their ability and willingness to repay. Each applicant can provide one narrative addressed to a male loan officer and one addressed to a female loan officer. All applications are anonymized before review so that loan officers cannot see any borrower identifiers. Loan officers evaluate applications using the bank’s regular scoring procedures, and loan decisions are based on these evaluations. After the evaluation phase, officers complete a short follow-up survey about their review experience. Some details of the study design are temporarily withheld to preserve research integrity and will be made public once data collection and analysis are complete.
AMENDMENT OF SEPTEMBER 11, 2026 — NEW VALIDATION STUDY AND LOAN-OFFICER EXPERIMENT
The extension tests whether loan officers evaluate differently the neutralized narratives originally written by the same borrower for male and female officers, and whether this source contrast changes when a heading explicitly identifies the original intended addressee's gender.
In Neutralized-N, both narrative sources are displayed under "Applicant's Message to Loan Officers." In Neutralized-A, a narrative originally written for male officers is displayed under "Applicant's Message to Male Loan Officers," and a narrative originally written for female officers under "Applicant's Message to Female Loan Officers." These headings follow the narrative's source, not the evaluating officer's gender. The corresponding narrative text is identical across regimes. Both regimes also include applications with no displayed narrative.
Applicant hard information, including applicant gender, is displayed and held fixed across versions of the same application. Each officer sees only one version of each applicant. A separate native-Dari-speaker survey will assess whether the original intended addressee's gender can be inferred from the neutralized passages.
|
|
Field
Intervention End Date
|
Before
June 01, 2026
|
After
October 03, 2026
|
|
Field
Primary Outcomes (End Points)
|
Before
Evaluation Score (Likert scale, 1 to 10)
|
After
Evaluation Score (Likert scale, 1 to 10)
AMENDMENT OF SEPTEMBER 11, 2026 — NEW VALIDATION STUDY AND LOAN-OFFICER EXPERIMENT
New officer experiment: assessed probability of full loan repayment at maturity, reported from 0 to 100 in increments of ten.
Validation survey: correctness of the respondent's guess about the gender of the narrative's original intended addressee.
|
|
Field
Primary Outcomes (Explanation)
|
Before
The primary outcome is the loan officer’s evaluation score, measured on a scale from 1 to 10, reflecting how likely they believe the applicant is to repay the loan by the maturity date, where 1 indicates “very unlikely to repay.”
|
After
The primary outcome is the loan officer’s evaluation score, measured on a scale from 1 to 10, reflecting how likely they believe the applicant is to repay the loan by the maturity date, where 1 indicates “very unlikely to repay.”
AMENDMENT OF SEPTEMBER 11, 2026 — NEW VALIDATION STUDY AND LOAN-OFFICER EXPERIMENT
The officer's repayment-probability assessment is analyzed in levels, with effects expressed in percentage points. Zero and 100 are valid responses; the response field has no default value. The earlier registration's 1–10 Likert-scale description does not apply to this new officer round.
For the validation survey, a response is coded 1 when the selected gender matches the original intended addressee and 0 otherwise. Correctness refers to addressee gender, not borrower gender. The resulting mean is the proportion of correct classifications. Failure to reject a chance benchmark will be described as absence of evidence of a difference, not as proof that recipient gender is completely undetectable.
|
|
Field
Experimental Design (Public)
|
Before
This study examines how written communication in loan applications affects how applicants are perceived by loan officers. We collaborate with a financial institution to collect anonymized loan applications that include a short open-ended section where borrowers describe their credibility and repayment ability. Professional loan officers evaluate these applications, and the bank’s lending decisions are based on these evaluations. To protect privacy, all applications are anonymized before review, and loan officers cannot see any borrower information that reveals gender or identity. The study aims to better understand how borrower communication influences officers’ assessments. More details of the study are temporarily withheld to preserve the integrity of the research and will be made public once the study concludes.
|
After
This study examines how written communication in loan applications affects how applicants are perceived by loan officers. We collaborate with a financial institution to collect anonymized loan applications that include a short open-ended section where borrowers describe their credibility and repayment ability. Professional loan officers evaluate these applications, and the bank’s lending decisions are based on these evaluations. To protect privacy, all applications are anonymized before review, and loan officers cannot see any borrower information that reveals gender or identity. The study aims to better understand how borrower communication influences officers’ assessments. More details of the study are temporarily withheld to preserve the integrity of the research and will be made public once the study concludes.
AMENDMENT OF SEPTEMBER 11, 2026 — NEW VALIDATION STUDY AND LOAN-OFFICER EXPERIMENT
The new officer experiment uses only first-round applications. Of 449 original applicants, 391 supplied at least one substantive narrative. This usable pool comprises 43 financed and 348 non-financed applicants. The 351 applicants with both narrative versions—181 men and 170 women—form the confirmatory sample for comparisons between narrative sources. The other 40 applicants have only one narrative. Three of the 46 originally financed borrowers supplied no usable narrative and are therefore outside the 391-applicant pool.
Approximately 100 new male and female loan officers will each review 150 applications: all 43 financed benchmark applications plus 107 applications randomly sampled without replacement from the 348 non-financed applicants. The non-financed sample is drawn separately for each officer. Officers who participated in the earlier officer experiments or were involved in the original financing decisions are not eligible for this new officer sample.
One available application version is randomly selected for each applicant–officer pair. Officers are independently of these packet draws assigned to one of the two heading regimes, with approximately equal allocation and no stratification. The applicant and version draws are not balanced by gender or other characteristics.
The validation survey is planned for September 12, 2026, with up to 100 participants and 50 passage judgments per participant. Participants will be recruited through research assistants' networks. Eligibility requires native Dari fluency and ability to read Dari; no geographic restriction is imposed. Loan officers, former partner-bank employees, and people involved in the earlier experiments are excluded. The officer round begins September 19, 2026, with completion planned for October 3, 2026.
|
|
Field
Randomization Method
|
Before
Randomization is conducted in the office using a computer program.
|
After
Randomization is conducted in the office using a computer program.
AMENDMENT OF SEPTEMBER 11, 2026 — NEW VALIDATION STUDY AND LOAN-OFFICER EXPERIMENT
Randomization is performed by computer. For each new officer, draw 107 non-financed applications without replacement, combine them with the 43 common financed applications, and independently draw one eligible version for each applicant. Allocate officers approximately equally between Neutralized-N and Neutralized-A without stratification, independently of the application and version draws. Randomly order each officer's packet. The validation survey uses the Stata program build_gender_neutralization_survey_assignments(5).do, with its five fixed stage-specific seeds and the 46+4 assignment procedure described in the validation subsection of Experimental Design (Hidden).
|
|
Field
Randomization Unit
|
Before
Borrower-level randomization:
The gender of the officer shown to applicants (“male officer” vs. “female officer”) is randomized at the individual borrower application level.
Officer–application-level randomization:
The narrative visibility condition (“Narrative Visible” vs. “Narrative Concealed”) is randomized independently for each officer–application pair.
Officer-level randomization:
In the post-evaluation survey stage, whether borrower gender is revealed in the officer’s feedback report (“Gender Revealed” vs. “Gender Hidden”) is randomized at the officer level.
|
After
Borrower-level randomization:
The gender of the officer shown to applicants (“male officer” vs. “female officer”) is randomized at the individual borrower application level.
AMENDMENT OF SEPTEMBER 11, 2026 — NEW VALIDATION STUDY AND LOAN-OFFICER EXPERIMENT
Loan-officer level: assignment to the recipient-unspecified or gender-specific-heading regime.
Applicant–officer-pair level: assignment of the displayed application version.
Within-officer packet: random selection of non-financed applications and random presentation order.
Validation survey: participant-specific selection of 46 distinct eligible applicants; target-version assignment balanced 23/23 within participant; positive-control pattern assignment in blocks of four participant IDs; and randomized passage order within participant. No new borrower-level intervention is conducted in this extension.
Officer–application-level randomization:
The narrative visibility condition (“Narrative Visible” vs. “Narrative Concealed”) is randomized independently for each officer–application pair.
Officer-level randomization:
In the post-evaluation survey stage, whether borrower gender is revealed in the officer’s feedback report (“Gender Revealed” vs. “Gender Hidden”) is randomized at the officer level.
|
|
Field
Was the treatment clustered?
|
Before
No
|
After
Yes
|
|
Field
Planned Number of Clusters
|
Before
Approximately 600 borrower applications (individual-level randomization, not clustered) and 50 loan officers (clustered randomization for post-evaluation survey stage).
|
After
Approximately 600 borrower applications (individual-level randomization, not clustered) and 50 loan officers (clustered randomization for post-evaluation survey stage).
AMENDMENT OF SEPTEMBER 11, 2026 — NEW VALIDATION STUDY AND LOAN-OFFICER EXPERIMENT
Approximately 100 new loan officers are the assignment clusters for the officer-level heading regimes. The separate validation survey is planned for up to 100 participants. The officer experiment reuses 391 existing applicant records; it does not recruit 391 new borrowers.
|
|
Field
Planned Number of Observations
|
Before
150*50=7500 application-officer
|
After
150*50=7500 application-officer
AMENDMENT OF SEPTEMBER 11, 2026 — NEW VALIDATION STUDY AND LOAN-OFFICER EXPERIMENT
Approximately 15,000 application–officer evaluations (approximately 100 officers x 150 applications), comprising approximately 4,300 financed-benchmark evaluations and 10,700 non-financed evaluations. The confirmatory source-comparison sample is the subset involving the 351 applicants who supplied both narratives.
The separate validation survey has up to 5,000 passage judgments (up to 100 participants x 50 passages): up to 4,600 neutralized-passage judgments and 400 original-positive-control judgments. These are repeated judgments, not 5,000 separate participants.
|
|
Field
Sample size (or number of clusters) by treatment arms
|
Before
Borrower-level treatment (unit = application; not clustered)
Officer gender shown = Male:
Planned invitations: 300 borrowers (≈ half female, half male overall sample stratified by borrower gender)
Expected completed applications: ≈250
Officer gender shown = Female:
Planned invitations: 300
Expected completed applications: ≈250
Note: Total invitations = 600 (300 female, 300 male). Expected completed applications ≈ 500 in total, with ≈ 250 per arm.
************
Officer–application condition (unit = officer–application evaluation; not clustered)
Narrative Visible: approximately 50 percent of officer–application evaluations
Narrative Concealed: approximately 50 percent of officer–application evaluations
Exact counts depend on the final number of evaluations per application. The split is designed to be 1:1 across all officer–application observations.
************
Post-evaluation survey treatment (unit = loan officer; clustered at officer level)
Borrower gender revealed in officer report: 25 clusters (officers)
Borrower gender hidden in officer report: 25 clusters (officers)
Total officers = 50, randomly assigned 25/25 across the two survey arms.
|
After
Borrower-level treatment (unit = application; not clustered)
Officer gender shown = Male:
Planned invitations: 300 borrowers (≈ half female, half male overall sample stratified by borrower gender)
Expected completed applications: ≈250
Officer gender shown = Female:
Planned invitations: 300
Expected completed applications: ≈250
Note: Total invitations = 600 (300 female, 300 male). Expected completed applications ≈ 500 in total, with ≈ 250 per arm.
************
Officer–application condition (unit = officer–application evaluation; not clustered)
Narrative Visible: approximately 50 percent of officer–application evaluations
Narrative Concealed: approximately 50 percent of officer–application evaluations
Exact counts depend on the final number of evaluations per application. The split is designed to be 1:1 across all officer–application observations.
************
Post-evaluation survey treatment (unit = loan officer; clustered at officer level)
Borrower gender revealed in officer report: 25 clusters (officers)
Borrower gender hidden in officer report: 25 clusters (officers)
Total officers = 50, randomly assigned 25/25 across the two survey arms.
AMENDMENT OF SEPTEMBER 11, 2026 — NEW VALIDATION STUDY AND LOAN-OFFICER EXPERIMENT
Officer-level heading regimes: approximately 50 officers in Neutralized-N and 50 in Neutralized-A, corresponding to approximately 7,500 application evaluations per regime if 100 officers each complete 150 evaluations. These are planned approximate counts, not fixed within-gender quotas.
For applicants with paired narratives, no narrative, N(M_i), and N(F_i) each have assignment probability 1/3. For single-narrative applicants, no narrative and the available narrative each have probability 1/2. Realized application-arm totals are not fixed or balanced within officers. The no-narrative display is identical across regimes.
Validation survey: up to 100 participant packets, each containing 46 neutralized passages (23 male-target and 23 female-target) and four original positive controls (two male-target and two female-target). These are within-participant passage assignments, not separate participant treatment arms.
|
|
Field
Intervention (Hidden)
|
Before
The intervention is designed to test whether (1) the informational value of borrower-written narratives in small-loan applications varies with the gender of the assigned loan officer, and (2) whether post-evaluation exposure to borrower gender information amplifies gender-based stereotypes among officers.
The study is implemented in partnership with a commercial bank in Afghanistan that issues standardized, collateral-free micro-loans to eligible small business owners. The intervention consists of two main stages:
Stage 1: Loan Application and Review Process
Eligible micro-entrepreneurs are invited by the partner bank to apply for a standardized, non-collateralized small loan with fixed, non-negotiable terms. Each application includes standard hard information (business characteristics, repayment history, etc.) and an open-ended narrative where applicants explain, in their own words, why the bank should trust their ability and willingness to repay. Each applicant can provide one narrative addressed to a male loan officer and one addressed to a female loan officer.
Each loan application is anonymized before being evaluated, meaning that all identifying details—including names, photos, and any gender-indicative information—are removed. Each application will have three versions:
Version 1. Narrative Concealed: The officer sees the application (hard information), but no narrative is shown.
Version 2. Narrative Written for Female Loan Officer Visible: The officer sees the application plus the borrower’s narrative addressed to a female officer.
Version 3. Narrative Written for Male Loan Officer Visible: The officer sees the application plus the borrower’s narrative addressed to a male officer.
Officers assign repayment-likelihood scores in all versions. These scores are used by the partner bank to make real loan approval decisions, ensuring that the evaluations carry economic consequences. By comparing each application’s mean scores across the visible and concealed conditions, we identify the marginal informational value of the narrative. By further comparing this marginal value when borrowers face an opposite-gender loan officer, we test whether the informational content of borrower narratives is lower when the assigned officer is of a different gender.
Stage 2: Post-Evaluation Survey (Officer Follow-up)
After all reviews and loan decisions are completed, loan officers participate in a short post-evaluation survey. Each officer receives a private report listing the applications they reviewed and the scores they assigned. For a randomly selected half of officers, the report also reveals the true gender of each borrower. For the other half, gender remains undisclosed.
The survey elicits officers’ beliefs about borrower credibility, communication ability, and repayment quality, as well as self-assessed confidence in their own scoring. Comparing responses between these two groups allows us to test whether exposure to borrower gender information ex post amplifies gender-stereotypical beliefs, particularly among male officers.
|
After
The intervention is designed to test whether (1) the informational value of borrower-written narratives in small-loan applications varies with the gender of the assigned loan officer, and (2) whether post-evaluation exposure to borrower gender information amplifies gender-based stereotypes among officers.
The study is implemented in partnership with a commercial bank in Afghanistan that issues standardized, collateral-free micro-loans to eligible small business owners. The intervention consists of two main stages:
Stage 1: Loan Application and Review Process
Eligible micro-entrepreneurs are invited by the partner bank to apply for a standardized, non-collateralized small loan with fixed, non-negotiable terms. Each application includes standard hard information (business characteristics, repayment history, etc.) and an open-ended narrative where applicants explain, in their own words, why the bank should trust their ability and willingness to repay. Each applicant can provide one narrative addressed to a male loan officer and one addressed to a female loan officer.
Each loan application is anonymized before being evaluated, meaning that all identifying details—including names, photos, and any gender-indicative information—are removed. Each application will have three versions:
Version 1. Narrative Concealed: The officer sees the application (hard information), but no narrative is shown.
Version 2. Narrative Written for Female Loan Officer Visible: The officer sees the application plus the borrower’s narrative addressed to a female officer.
Version 3. Narrative Written for Male Loan Officer Visible: The officer sees the application plus the borrower’s narrative addressed to a male officer.
Officers assign repayment-likelihood scores in all versions. These scores are used by the partner bank to make real loan approval decisions, ensuring that the evaluations carry economic consequences. By comparing each application’s mean scores across the visible and concealed conditions, we identify the marginal informational value of the narrative. By further comparing this marginal value when borrowers face an opposite-gender loan officer, we test whether the informational content of borrower narratives is lower when the assigned officer is of a different gender.
Stage 2: Post-Evaluation Survey (Officer Follow-up)
After all reviews and loan decisions are completed, loan officers participate in a short post-evaluation survey. Each officer receives a private report listing the applications they reviewed and the scores they assigned. For a randomly selected half of officers, the report also reveals the true gender of each borrower. For the other half, gender remains undisclosed.
The survey elicits officers’ beliefs about borrower credibility, communication ability, and repayment quality, as well as self-assessed confidence in their own scoring. Comparing responses between these two groups allows us to test whether exposure to borrower gender information ex post amplifies gender-stereotypical beliefs, particularly among male officers.
AMENDMENT OF SEPTEMBER 11, 2026 — NEW VALIDATION STUDY AND LOAN-OFFICER EXPERIMENT
The neutralization procedure uses Qwen3-30B-A3B-Instruct-2507, checkpoint mlx-community/Qwen3-30B-A3B-Instruct-2507-8bit, run locally using MLX-LM with temperature 0 and a maximum output length of 300 tokens. Each narrative is processed separately. The model receives the narrative text without separate applicant-gender metadata, identifiers, original-addressee labels, or the paired narrative. It returns deletion instructions for explicit recipient-gendered direct-address phrases. A deterministic routine applies the deletions and the documented minor spacing/punctuation normalization. No manual editing was performed. The corpus was frozen on August 18, 2026.
Both officer regimes use the same common narrative instruction: "Some applications include a written response from the applicant. Please evaluate each application using all information displayed in the application."
The common instructions do not disclose the existence of paired original narratives, neutralization, random reassignment, or hypothetical application versions. In Neutralized-A, the application-specific heading truthfully identifies the narrative's original intended addressee. No new gendered greeting is inserted. The no-narrative display in both regimes states: "No written response is displayed for this application."
|
|
Field
Secondary Outcomes (End Points)
|
Before
|
After
AMENDMENT OF SEPTEMBER 11, 2026 — NEW VALIDATION STUDY AND LOAN-OFFICER EXPERIMENT
Evaluation-level Brier loss among the 43 financed benchmark applicants.
|
|
Field
Secondary Outcomes (Explanation)
|
Before
|
After
AMENDMENT OF SEPTEMBER 11, 2026 — NEW VALIDATION STUDY AND LOAN-OFFICER EXPERIMENT
For officer j and financed applicant i, Brier loss is (p_ij - y_i)^2, where p_ij is the officer's 0–100 assessment divided by 100 and y_i records repayment of the full AFN 10,000 obligation by contractual maturity. Full on-time repayment is coded 1. Under the specified coding convention, late repayment, partial repayment, restructuring, waivers, and missing repayment information are coded 0.
Repayment outcomes are not yet known for all 43 financed benchmark applicants at registration. Brier loss and performance compensation will be calculated when the repayment information is compiled. Missing repayment information is coded 0 by convention; it is not evidence of an observed default. This accuracy outcome pertains to historically financed borrowers rather than all applicants.
|
|
Field
Building on Existing Work
|
Before
No
|
After
Yes
|