Differential Signaling Value of Attire in Social and Economic Evaluations

2026-02-09

2026-05-09

Evaluators’ first-order evaluations of profiles wrt.:
- Professional Prestige
- Access to Credit
- Willingness to form Friendships
- Willingness to Help

The primary estimands are treatment effects on evaluators’ first-order evaluations. Evaluations are elicited on a 7-point Likelihood Likert scale (-3 = "Extremely unlikely", ..., +3 ="Extremely likely"). Analyses are based on both the original scale and on deterministic probability-scales. Probability-scale representations are included to facilitate interpretation and comparability with later belief-elicitation experiments.

In addition to domain-specific estimates, analyses will assess whether treatment effects differ across domains using joint tests.

Evaluators’ second-order beliefs about other evaluators’ ratings wrt.:
- Professional Prestige
- Access to Credit
- Willingness to form Friendships
- Willingness to Help

Evaluators’ first-order evaluations of profiles wrt. personal attributes:
- Competence, Assertiveness
- Kindness, Trustworthiness.

Evaluators’ first-order evaluations of profiles wrt. the outcomes outside their assigned domain (e.g., loan officers also evaluate helping, friendship, and professional prestige).

Second-order belief outcomes elicit evaluators’ incentivized beliefs about how they believe other evaluators rate the same profiles. They will be elicited on a 7-point Likelihood Likert scale.

Beliefs about personal attributes and other outcomes will be elicited on a 7-point Likelihood Likert scale, similarly to the primary outcomes. The main analysis of these outcomes will be based on the pooled sample of all evaluators.

Profiles
The profiles subject to evaluation in this study are constructed from photographs of individuals residing in Dar es Salaam and their basic identifying information. Profiles were collected prior to this study. Each individual appears in two profiles that differ in attire, with one version featuring visually prominent (branded/accessorized) attire and the other plain attire. The same profiles are used across all outcomes.The target sample consists of approximately 655 profiles. Due to data quality or implementation constraints, a small number of profiles may be excluded prior to analysis. The same set of profiles is used across all outcomes.

Evaluations
Evaluations are conducted separately for four outcome domains, each with a distinct evaluator population:
1. Professional Prestige
Evaluators: business owners, managers, individuals affiliated with business associations, incubators, and other individuals in senior or decision-making roles within the local business community. We also consider individuals with high self-reported income.
Evaluators assess how likely they would be to agree to invite the profiled individual to an exclusive event.
2. Access to Credit
Evaluators: loan officers, credit analysts, and other loan professionals.
Evaluators assess how likely they would be to approve a loan application from the profiled individual.
3. Friendship
Evaluators: a community sample from Dar es Salaam.
Evaluators assess how likely they would be to initiate a friendship with the profiled individual after a first meeting.
4. Help
Evaluators: a community sample from Dar es Salaam.
Evaluators assess how likely they would be to offer help to the profiled individual in a brief street interaction.

Procedure
Each evaluator rates approximately 16 different profiles within their assigned outcome domain. Profiles are randomly assigned and shown in random order. No evaluator sees more than one profile of the same individual and no profile is shown twice to the same evaluator. The set of profiles shown to each evaluator is stratified by attire (visually prominent versus plain) and gender.
For each profile, evaluators provide:
• a first-order evaluation of the likelihood associated with the outcome domain, and
• a second-order belief about how other evaluators rated the same profile for the same outcome. Second-order belief responses are incentivized.
At the end of the survey, evaluators provide additional first-order evaluations for a randomly selected subset of previously seen profiles on four personal attributes: competence, assertiveness, trustworthiness, and kindness. For a different subset of previously seen profiles, evaluators are asked to provide evaluations for outcomes outside their assigned domain.

This design allows for comparison of treatment effects of visually prominent versus plain attire on evaluators’ assessments across different outcome domains.

Heterogeneity Analysis
I will examine heterogeneity in treatment effects, with particular attention to (i) gender of the profiled individual and gender of the evaluator, (ii) age of the profiled individual and age of the evaluator, and (iii) income-related characteristics of the profiled individual and the evaluator. Additional heterogeneity dimensions of interest include district, education level, occupation sector and marital status. In addition, while primary analyses will be conducted on the full sample of evaluators, results will also be reported for the subsample of evaluators who report being familiar with the clothing styles shown. All heterogeneity analyses are treated as exploratory.

Additional Measures
The survey additionally collects background information from evaluators, including familiarity with the attire shown, typical clothing expenditures, and whether the evaluator has met any individual appearing in the profiles before.

Not available

Randomization is conducted by a computer using a pre-specified algorithm. For each evaluator, the set and order of profiles are randomly assigned subject to the following ex ante constraints:

1. The gender × attire composition of profiles is balanced within each evaluator.
2. No evaluator is shown more than one profile version of the same individual and no profile is repeated in the main evaluations.
3. No evaluator is shown more than one profile featuring the same outfit.
4. Across evaluators, each profile appears at least once and approximately the same number of times.

Conditional on these constraints, profile assignment is random. The randomization algorithm was implemented prior to data collection and does not condition on evaluator characteristics or outcomes.

The unit of randomization is the profile evaluation within evaluator.
For each evaluator, individual profiles are randomly assigned and ordered, subject to pre-specified balance and coverage constraints. There is no randomization at higher levels.

No

Target: N = 1,000 evaluators, with approximately 250 evaluators per outcome.
Each evaluator assesses 16 profiles.
Final sample size may vary due to recruitment and incomplete surveys; the expected realized sample is 200–300 evaluators per outcome.

Target: N = 16,000 profile evaluations (1,000 evaluators × 16 profiles). At the time of preregistration, the profile set consists of 655 individuals featured in paired profiles. Final sample size may vary due to recruitment and incomplete surveys; the expected realized range is 12,500–19,500 evaluations in total.

Treatment is assigned within evaluator. Each evaluator views approximately 16 profiles, randomly ordered and cross-stratified by attire and gender, as specified above. I expect 50% of evaluated profiles to feature prominent and 50% to feature plain attire.

Power calculations are based on pilot data. Accounting for the clustered design with repeated ratings, the targeted minimum detectable effect size is 0.08 standard deviations of the outcome on both the 7-point evaluation Likert scale and its linear probability transformation. The unit of observation is an evaluator × profile rating.