TL;DR
Traditional economic models assume group identity (like gender) is a fixed, visible trait, but in the digital gig economy, it’s often a choice. This paper demonstrates that doctors strategically "mask" their gender in fields where they expect to face discrimination, only revealing it once they have accumulated enough positive patient reviews to prove their quality.
What Problem Does This Paper Address?
Economists have long used "statistical discrimination" models to explain why certain groups earn less. The theory suggests that when a buyer (patient) has little information about a seller (doctor), they rely on group averages—stereotypes—to guess quality. As more information (like reviews) becomes available, the "weight" of the stereotype should decrease.
However, the authors identify a major flaw in previous research: it assumes that gender or race is always visible. In many online markets, sellers can choose whether to post a photo or write a bio that reveals their identity. If a doctor knows they will be discriminated against, they might hide their gender. This "endogenous signaling" makes standard price-based tests for discrimination biased. The authors set out to create a new test for discrimination based on the choice to reveal one's identity.
Methods
The authors developed a theoretical framework where doctors choose the strength of their gender signal to maximize their expected price. They then tested this using a multi-step empirical approach:
- Gender Classification: To identify the "true" gender of doctors who masked their identity, the authors used:
- Strong Signals: Self-reported gender in biographies and profile pictures.
- Facial Recognition: The Baidu Face API to classify gender from photos.
- Name-Based Prediction: For doctors with no photo or bio, they used Gradient Boosting Decision Tree (GBDT) algorithms and the "ngender" Python package to predict gender based on Chinese naming conventions.
- Regression Analysis: They ran linear probability models to see how the probability of "strongly signaling" gender (using a photo or bio) changed based on the doctor's gender and their number of reviews.
- Cross-Field Comparison: They compared "male-dominated" fields (e.g., Surgery, Urology) with "female-dominated" fields (e.g., OB/GYN, Nutrition).
- Falsification Test: They analyzed doctors with "highly gendered names" (where masking is impossible) to ensure the observed trends weren't just due to general platform experience.
Datasets
- Good Doctor Online (GDO): The authors scraped data from China’s largest online healthcare platform (haodf.com).
- Primary Sample: 105,250 doctors from 139 top-tier Chinese hospitals (collected in 2020).
- Analysis Subsample: 35,810 doctors where gender could be reliably classified.
- Longitudinal Sample: A follow-up "recrawl" in 2025 of 2,196 doctors who were newcomers in 2020 to track how their signaling behavior evolved as they gained reviews.
Key Insights and Ideas
- Strategic Masking: Discrimination isn't just a fixed penalty; it's a hurdle that agents navigate. Doctors who belong to the "out-group" of a specialty (women in surgery or men in OB/GYN) intentionally provide less information about their gender when they are new to the platform.
- Symmetric Discrimination: The study finds that discrimination isn't exclusively "against women." Instead, it is against the minority gender in a specific field. Patients appear to use the gender "norm" of a department as a proxy for competence.
- The Information Threshold: As a doctor’s review count grows, the "signal-to-noise" ratio of their gender decreases. Once a doctor has a high rating based on hundreds of reviews, the risk of revealing their gender is lower because their individual reputation outweighs the group stereotype.
- The "Trust" Trade-off: While masking hides a "negative" trait (the stereotype), it also makes a profile look incomplete. Doctors eventually reveal their gender because a full profile with a photo and bio likely appears more trustworthy and "human" to prospective patients.
Key Results
- Male-Dominated Fields: Female doctors are 7.0 percentage points less likely than men to signal their gender strongly when they have zero reviews.
- Female-Dominated Fields: Male doctors are 7.9 percentage points less likely than women to signal their gender strongly initially.
- Closing the Gap: As reviews accumulate, the "masking" behavior disappears. In female-dominated fields, the signaling gap between men and women largely closes once a doctor reaches about 30 reviews.
- Price Effects: Initial price regressions showed a "female penalty" that diminished with reviews, but the authors proved this was partially an artifact of the signaling choice, reinforcing why their new "signaling-based test" is more accurate.
Limitations and Future Work
- Taste vs. Statistical Discrimination: While the results strongly align with statistical discrimination (which changes with information), the authors acknowledge they cannot entirely rule out "taste-based" discrimination (pure prejudice), though taste shouldn't theoretically change just because a doctor has more reviews.
- Effort Costs: The paper conjectures that there might be a small "effort cost" to uploading photos or writing bios, which might explain why some doctors wait, though this doesn't explain the gendered patterns across different specialties.
- Net Impact on Women: Although the discrimination is symmetric by field, the authors note that because the vast majority of medical specialties are male-dominated, statistical discrimination still poses a much larger aggregate barrier to women in medicine than to men.
- Future Research: The authors suggest applying this "signaling choice" framework to other gig economy platforms like Airbnb or TaskRabbit, where sellers also have discretion over how they present their race or gender.
Summary updated