Smile Inequality: Oral Health and Labor Market Outcomes

This study will implement three experiments: (i) a perception survey assessing public attitudes toward candidates with different visible traits, including oral health, (ii) a correspondence study analyzing callback rates based on oral health status in job applications, and (iii) an employer resume rating experiment where recruiters evaluate randomized candidate profiles and make hiring decisions.

Experiment 1: Perception Survey. Respondents are shown headshots randomized on six visual traits (e.g., poor oral health, smiling, obesity, darker skin tone, spectacles, and neutral expression). They rate photos on several traits, including confidence, trustworthiness, competence, health, suitability for a customer-facing role, socioeconomic status, etc. Profiles are also randomly assigned either a neutral blurb or a qualification-enhancing description.

Experiment 2: Correspondence Study. Real job applications are submitted to listings across India, randomized by headshot features (oral health, gender, caste, skin tone). Resume content remains constant across groups.

Experiment 3: Employer Rating Experiment. HR professionals are shown subsets of resumes that vary by oral health and other traits (e.g., gender, caste, education, marital status). For each profile, they make binary hiring decisions and provide Likert-scale ratings on customer preference, team fit, and employability. Monetary incentives are tied to their alignment with real-world hiring outcomes (from correspondence study) and perception ratings (from survey).

2024-10-10

2025-10-31

- Perceived employability and trait ratings in the perception survey
- Callback rate in the correspondence study
- Binary hiring decisions and Likert ratings in the employer experiment

- Perceived employability is the average of traits like confidence, trustworthiness, and competence ratings per headshot.
- Callback is a binary indicator: 1 if employer responds positively, 0 otherwise.
- HR ratings include binary and Likert scales on customer preference and team fit. Incentives reward accuracy compared to benchmark data from the other two studies.

- Trait-specific perception scores: attractiveness, wealth, etc.
- Heterogeneity by gender, caste, job type, marital status.
- HR justifications for hiring decisions (qualitatively coded)

- Trait ratings averaged across raters; analyzed separately by condition
- Heterogeneity tested via interactions of visual traits with applicant background attributes
- HR justifications coded into themes like professionalism or oral stigma

Three randomized experiments examine oral health bias:
(i) Headshot features are randomized between respondents in a perception survey.
(ii) Job applications vary only by the headshot shown; callbacks are tracked.
(iii) Hiring managers are shown randomized resumes and incentivized to rate them.

(i) Perception survey: headshot traits and text descriptions randomized at individual level
(ii) Correspondence study: resumes randomized by headshot and stratified by job type.
(iii) Resume rating: each HR professional sees a randomized subset of resumes in a within- and between-subject design

Computer-generated randomization

- Perception survey: photo (individual headshot)
- Correspondence study: applicant (resume)
- Employer experiment: profile (resume)

No

na

- Perception: ~400-500 participants * 6 headshots = 2,400 to 3,000 ratings - Correspondence: ~5,000 applications - HR Resume Rating: ~300-400 HR professionals * 12 profiles = 3,600 to 4,800 evaluations

- Perception: balanced across 6 visual features
- Correspondence: 2,500 control (good oral health), 2,500 treatment (poor oral health), cross-randomized with gender and caste
- HR Experiment: balanced combinations across oral health, qualifications, marital status, gender, caste

Accounting for the study design (between-subject, no clustering), I calculate Minimum Detectable Effect Sizes (MDEs) at 80% power and a 5% significance level for each main outcome: - Perception Study (Likert outcomes, scale 1–10): Assuming a standard deviation of 2 points (based on pilot data) and a sample of 400–500 respondents rating six headshots each (totaling ~2,400–3,000 ratings), the study is powered at 80% with a 5% significance level to detect a minimum difference of 0.25–0.30 points on a 10-point scale. This would capture moderate shifts in perceived traits like trustworthiness or employability due to oral health appearance. - Correspondence Study (binary outcome – callback): Assuming a baseline callback rate of 10% and 5,000 applications equally split across control and treatment arms, the study is powered at 80% with a 5% significance level to detect a minimum decrease of 1.2 percentage points. This translates to detecting a reduction in callbacks from 10% to 8.8% or lower for resumes featuring poor oral health appearance. - Employer Resume Rating Experiment (binary outcome – hiring decision): Assuming a baseline “yes” rate of 50% per resume and 300 to 400 HR respondents, each rating about 12 resumes (totaling 3,600 to 4,800 observations), the study is powered at 80% with a 5% significance level to detect a minimum effect size of approximately 5.5 to 6 percentage points. This corresponds to detecting a drop in hiring likelihood from 50% to 44%–45% for resumes with visible oral health issues.