Differential Signaling Value of Attire in Social and Economic Evaluations - Perceptions

2026-05-16

2026-08-16

- Participants’ beliefs about evaluation outcomes of their own profiles in four domains: professional prestige, credit access, ability to make friends, ability to receive help.
- Participants' preference and valuation to have one profile rather than the other evaluated, in each domain

Beliefs are elicited as the expected number of relevant evaluators, out of 100 evaluators, who would assign a positive evaluation to either profile (visually prominent vs. plain). Beliefs are elicited across all four outcome domains. In the main specification beliefs will be encoded as percentages. The main analysis will consider both levels of beliefs and differences between visually prominent and plain profiles within participants and domains. The primary focus of the analysis is on belief accuracy (misperception) of the relative attire effect in each domain. To measure these, we compare the effect of attire on the estimated probability of receiving a positive evaluation as realized in the preceding study, to the average prediction in this experiment, for all outcome domains.

In each outcome domain, I elicit participants’ preference and valuation (WTP) to have one profile rather than the other submitted for a bonus-relevant evaluation. Preferences are elicited with a simple binary choice (separate for each outcome domain) and valuation
is elicited using an incentive-compatible multiple price list (MPL), in the spirit of Becker–DeGroot–Marschak. In each row, participants choose between submitting the visually prominent profile or the plain profile for the evaluation. Across rows, an (unconditional) cash transfer attached to the less preferred profile varies, allowing identification of the monetary compensation required for participants to switch between profiles. Valuation is encoded as positive if in favor of the visually prominent attire profile and as negative if in favor of the plain attire profile. The primary valuation measure is the switching point (or implied interval) at which participants are indifferent between having the visually prominent or the plain profile evaluated, whereby positive evaluations are linked to bonus payments. Analyses will consider both the direction and magnitude of valuations for receiving the bonus for either of the two profiles. Valuations will be compared to potential realized evaluation earnings imputed from the evaluations data in the preceding study.

- Participants’ preference and valuation, to have one profile rather than the other displayed in a photo book which is not linked to evaluations or bonuses, in one randomly selected domain.

For one randomly chosen outcome domain, participants are informed that one of their profiles may be shown to evaluators in a photo book only (ahead of the evaluation of other participant’s profiles) without collecting evaluations or attributing bonuses. Preferences and valuations are elicited and measured analogously to the primary preference and valuation outcomes. The analysis will consider (i) any changes in preferences or valuation (binary outcome), (ii) changes in valuation relative to initial valuations, (always within the same outcome domain). These comparisons, are used to benchmark a decomposition of the initial valuation into a belief driven and an intrinsic preference driven component.

Participant Profiles

The study uses profile pairs of adult residents from Dar es Salaam, collected during a pre-study in January 2026. Each participant appears in two versions of their profile. Photos displayed on the profiles differ in attire, with one version featuring visually prominent (conspicuous) - and the other plain attire. Aside from photos, profiles also contain basic demographic information (age, residence, nationality and occupation sector), which is identical across the two profiles. These profiles form the basis for belief elicitation and valuation decisions, in this experiment.

Procedure

Participants complete the survey individually together with an enumerator. The survey consists of three main parts, plus additional measures to investigate mechanisms.


1. Belief Elicitation

For each outcome domain, participants predict how evaluators would assess each of their two profiles. Outcome and profile order is randomized. During belief elicitation, participants see the same profile- and evaluation screen previously shown to evaluators in the corresponding domain. Beliefs are elicited as the expected number of evaluators, out of a fixed reference group of 100 people, who would assign a positive evaluation to either profile. Belief responses are incentivized: participants are eligible for a bonus payment if their prediction is sufficiently close to the realized evaluation outcome for a randomly selected profile-domain combination, collected for a random subset of participants. In addition, participants are asked to briefly explain their reasoning for selected belief responses. These open-text responses will be analyzed and may be used to provide further evidence on mechanisms underlying belief formation.


2. Preferences and Valuations

For each outcome domain, the valuation elicitation directly follows the belief elicitation. Participants are informed that, in each outcome domain, one evaluator may evaluate one of their two profiles. If the evaluator gives a positive evaluation, participants receive a bonus
payment; if the evaluator gives a negative evaluation, they receive no bonus. In each outcome domain, I elicit participants’ preference and valuation to have one profile rather than the other submitted for this bonus-relevant evaluation. Preferences are elicited using a binary single-choice question and valuations are elicited using an incentive-compatible MPL, in the spirit of Becker–DeGroot–Marschak. In each row, participants choose between submitting the visually prominent profile or the plain profile for the bonus-relevant evaluation, with an additional cash transfer attached to one option and varying across rows. The preference or one row of the MPL is randomly selected for implementation, and both the evaluation bonus and any additional transfer are paid according to the choice made in that row. All MPLs are randomized along two dimensions: which profile appears in which column and whether relative transfers are ascending or descending. Individuals are free to choose different profiles for the bonus in different outcome domains.


3. Decomposition

For one randomly chosen outcome domain, participants are told that their profiles may not be part of evaluations and instead one of their profiles may be displayed to evaluators in a photo book ahead of evaluations of other participants. Here, no evaluations will be conducted and no bonus can be earned. The Preference and MPL elicitation in this task is conducted in an analogous manner to the initial MPL elicitation in the previous step.


Additional Measures
a) Cross-Domain Inference: To study how participants generalize information across domains or subgroups thereof, I elicit cross-domain inference. Participants are presented with randomly assigned truths about a person associated to one of the outcome domains (e.g., a person who easily gets invited to exclusive dinners with influential business people...) or wealth and are then asked to complete the sentence with three options referring to one of the other domains (e.g., ...easily gets loans approved/...gets loans approved just like other people/...struggles to get loans approved). These measures are used to assess whether participants rely on compressed or correlated representations of outcomes when forming beliefs. In particular, patterns in these responses as well as the broad set of beliefs may be compared to correlation patterns observed in evaluation data from the preceding study.
b) Domain Valuation: Participants report the subjective importance they attach to outcomes in each domain.
c) Memory and Associations: Participants report which groups of people they associate with the two types of attire.

These measures are used in exploratory analyses.


Heterogeneity Analysis

I will examine heterogeneity in beliefs, preferences and valuations, with particular attention to (i) gender, (ii) age, (iii) occupation sector, and (iv) income-related characteristics of participants. Additional heterogeneity dimensions of interest include district, education level, marital status, and clothing-related expenditures. In addition, I will explore whether participants whose observable characteristics more closely match those of their evaluators exhibit more accurate beliefs. All heterogeneity analyses are treated as exploratory.

Not available

The survey incorporates two cross-participant randomizations:
1. After completing the main belief and valuation tasks, each participant is randomly assigned to one of four outcome-specific “photo book-" conditions. This randomized assignment determines for which outcome domain each participant completes the no-evaluation photo book valuation task (see also secondary outcomes and “Decomposition” in experimental design).
2. At the end of the survey, there is also a randomized cross-domain inference task. In a series of questions, participants are randomly assigned information about one outcome domain and are then asked to draw inferences about another randomly paired outcome domain (see also “Additional Measures” in experimental design).


The survey additionally incorporates several independent randomizations at the respondent level. These govern the implementation of survey elements and are primarily implemented to balance potential framing, order, and implementation effects ex ante rather than as experimental treatments of substantive interest. Accordingly, the main specifications do not mechanically include all randomization indicators as controls. However, randomization indicators related to key elicitation procedures (e.g., MPL order and profile placement) will be incorporated in robustness analyses and may be included in alternative specifications where relevant.

In particular, I randomize the order of belief elicitation across outcomes, the order in which the two profiles are presented during belief, and preference elicitation, the structure of the MPL design during the valuation task, and the outcome domain selected for the control decomposition task. Finally incentives are based on random implementation of either the 100-people evaluation for beliefs or by implementing one of the preference or MPL choices made for one outcome for either the primary or the decomposition task. These orthogonal randomizations are designed to minimize order or framing effects, support incentive compatibility, and ensure that comparisons across tasks are not mechanically driven by survey implementation.

All randomizations are implemented at the individual respondent level.

No

Each respondent constitutes a single unit of randomization.

The target sample size for this experiment is 465-600 respondents. This target range exceeds the minimum sample size implied by the ex-ante power calculations of 465 required to detect belief–evaluation gaps in attire effects of 5–6 percentage points (at the 5% significance level) and allows for improved coverage of the original evaluation sample. Participant recruitment occurs in three phases. First, we focus on the on the pool of individuals who previously submitted profiles for the evaluation study and whose profiles were evaluated at least once in both attire conditions in all outcome domains, ensuring that belief elicitation can be directly matched to realized evaluation outcomes. Second, if the number of completed observations from this primary sample falls short of 465, recruitment is extended to individuals, who also submitted profiles for the evaluation study but who were not evaluated in all outcome domains for both attires. Third, if fewer than 465 completed observations are obtained after recruiting from this original sample, additional participants who had not previously submitted profiles and who were never evaluated will be approached.

The design does not involve treatment arms. Instead, several survey features are independently randomized at the individual level, as described above.

The sample size is determined based on the objective of detecting economically meaningful misperceptions in beliefs relative to observed evaluation effects. In particular, I consider it important to detect cases where participants perceive no effect, or an effect in the opposite direction, despite a non-zero realized evaluation effect. The power calculation focuses on professional prestige as the binding outcome domain. In the evaluation data, the effect of visually prominent relative to plain attire on the probability of receiving a positive prestige evaluation is approximately 5.7 percentage points. This is the smallest statistically significant evaluation effect among the main non-zero outcome domains. Using variance estimates from a pilot beliefs study, I estimate that a sample size of approximately 465 respondents is sufficient for the 95 percent confidence interval around a perceived effect of zero to be visually separated from the 95 percent confidence interval around the realized prestige effect.