Research

Recent changes in labor arrangements have increased interest in estimating and understanding the value of job flexibility. We leverage a large natural field experiment at Uber to create exogenous variation in expected market wages across individuals and over time. Combining this experiment with high frequency panel data on wages and individual work decisions, we document how labor supply responds to exogenous changes in expected market wages in a setting with virtually no restrictions on driver labor allocation. We find that there is i) systematic heterogeneity in labor supply responses both across drivers and within a driver over time, ii) significant fixed costs of beginning a shift, and iii) high rider demand when it is costly for drivers to work. These three findings motivate a model of labor supply with heterogenous preferences over work schedules, adjustment costs, and statistical dependence between market wages and the costs of driving. We recover the labor supply elasticities and reservation wages of this dynamic labor supply model via a combination of experimental estimates and other data moments. We then perform counterfactual analyses that allow us to examine how preference heterogeneity and adjustment costs influence the responses of workers' to wage incentives as well as infer drivers' willingness to pay for the ability to customize and adjust their work schedule. We also show that a static approach to the driver's dynamic problem delivers materially different estimates of workers' labor supply elasticities and their value of job flexibility.

We study the impact of Intimate Partner Violence (IPV) on various aspects of the victims’ lives throughout the course of violence, including their marital decisions, labor market performance, fertility decisions, and mental health. Our data consists of the universe of administratively reported IPV cases primarily from hospitals, police, and helplines in Taiwan from 2012 to 2018. We distinguish the violence effect and the report effect. We use later-treated victims as control groups under a staggered event-study framework. Among all victims who eventually report IPV to the officials, we find (i) the probability of divorce increases for both men and women after the report but not at the violence onset; (ii) women’s employment rate decreases after the onset of violence, but an incremental positive effect follows after reporting, particularly among young women; (iii) reduction in additional childbirth after reporting regardless of the victim’s gender; (iv) rising depression outpatient-visits after violence and reporting among women.

When the Regression Discontinuity (RD) design is employed to study treatment assigned by policy cutoffs, the identified local average treatment effect is often criticized for its lack of policy relevance due to treatment effect heterogeneity. Instead of assessing the effectiveness of the treatment, we focus on utilizing the RD design to improve the policy's targeting. We propose a method of choosing the variables to be included in the policy cutoffs by comparing their explanatory power (R^2) on the treatment effect heterogeneity with causal forest. The method is applied to the context of hyperlipidemia diagnosis assigned by the cholesterol reference range and its health benefits. Using administrative data from 6 million health checks in Taiwan, we document that the diagnosis reduces the short-term risk of complications such as strokes and heart failures by 0.174 percentage points (10.8%). By applying our method, we show that age stands out as the best single variable to be included in the cholesterol reference range to maximize the health benefits: for the oldest 20%, the health benefit is 3.543 times stronger.

Taken together, falling fertility rates and rising lifespans cause many of the world’s richest countries to face steep increases in age-related long-term care (LTC) needs that must be born by a comparatively shrinking native labor force. This paper provides novel, rich evidence on the costs families bear in order to manage the long-term care needs associated with the aging of their elderly members. Using rich administrative data, we find that health shocks to the elderly cause large families to grow larger: adult children are more likely to get married, and they have more children of their own. These responses are persistent over time, and they are consistent with the idea that LTC needs induce family members to substitute from formal employment into a mix of informal caregiving and home production. Members of smaller families, by contrast, experience sharp increases in mortality risk, in a way that is consistent with caregiving-related “deaths of despair.” Leveraging sharp policy changes in Taiwan that greatly increased access to international labor markets for formal live-in caregivers, we find that formal caregiver hiring amplifies positive fertility responses, whereas mortality responses vanish. These results suggest how immigration policy can causally improve native well-being and sustain longer-term native demographics.

How do adult children trade-off working and providing long-term care (LTC) to parents? How do international caregivers help with this trade-off? What are the effects of allowing international caregivers compared with other commonly implemented LTC policies? Using data from Taiwan, we first document that children are 4 percentage points less likely to participate in the labor market when parents’ LTC needs arise, with daughters, the less educated, and older children having the largest decreases in labor supply. We also find that a 2012 reform that allows more international caregivers significantly increases children's labor supply. Motivated by these findings, We then build and estimate a dynamic labor supply model, combining the descriptive evidence with an exogenous variation in caregivers' prices from a policy reform in Taiwan. The model features costs of returning to work, endogenous health processes, and unobserved heterogeneity in care and labor market skills. Model-based results suggest large costs of returning to work, especially for daughters and the less educated. By relaxing the current international caregiver regulations, permanent leaves from the labor market due to LTC is cut down by half and the welfare gain is 4 times larger than the currently implemented LTC tax deduction program..

Childhood disability has enormous impacts on family members. Limited by data, previous literature faces challenges in measurement and identification and focuses primarily on maternal labor outcomes. Using administrative records from the National Health Insurance Research Database from Taiwan, we study one of the most prevalent causes of childhood disability -- cerebral palsy (CP). We exploit its unexpected nature and our rich data to investigate a wide variety of impacts on family members. We examine 12,228 children diagnosed with CP and their families from 2000 to 2019. We take an event study approach to study various outcomes, including parents' labor supply, mental health, marital status, and fertility. We find that having a CP child decreases the mother's probability of work by 5.5 pp, and it increases divorce by 2.1 pp and depression by 25%. We also find a significant decrease in long-run fertility. These effects are larger for parents with worse socioeconomic conditions and for CP daughters.

How people respond upon receiving inheritances has been a crucial topic in the literature regarding wealth shock and inheritance tax. Among all outcomes, the discussion of Carnegie effect: how inheritances affect recipients' labor decisions has been estimated by different models using various data. The research introduce new event study designs using administrative data from the Taiwanese population to identify events of people receiving house inheritances. By comparing people who suffered from loss of parents without house inheritances to those also suffered from loss of parents but inherited houses, we obtain the sole average treatment effects of inheriting a house on labor decisions. Results are coherent to literature, showing that people decrease there labor income and probability to work significantly after receiving house inheritances, and the effects also spill over to spouses of the recipients.

This paper quantifies the trade-off between social-economic characteristics and physical attractiveness. We use the state-of-the-art facial recognition algorithm to extract features such as eye size, lip width, and nose height from more than 460,000 photos. We find that education is the most important factor when it comes to whom to send a Like through the mobile dating app. Given the same educational attainment, men are more likely to use facial traits, and women are more likely to use height as the criterion to determine to whom to send a Like. We find that male users prefer women with the following facial cues: larger eyes, shorter distance between two eyes, smaller nose width, longer hose length, smaller upper lip, and larger lower lip. By contrast, most of the males' facial traits are not statistically significant in explaining females' Like data, except that larger upper lip and wider nose are unattractive traits.

Leading recommender systems may recommend only a small fraction of users on the dating platform since the algorithms often exploit popularity and similarity that reinforce preference homogeneity and assortative mating in the marriage market. We apply a stylized matching model in economics to the existing algorithms to reduce inequality, and we evaluate the proposed method by a large-scale field experiment through a dating app. Our recommender improves predictive accuracy and reduces inequality, leading to substantially more matched couples than the other two competing algorithms. In particular, male users assigned to our novel recommender are four times more likely to receive females' responses. We improve several inequality measures: There are far fewer superstars promoted by our algorithm, and the distribution of the recommendation count is more even than the other two algorithms. Our algorithm also yields the highest coverage rate.