Topics of Research Interests

• • Determinants of Health and Labor Market Activity

  • Predictors of Chronic Disease and Cognitive Decline

Publications

1. "Predictive Effects of Obesity on Cancer Incidence in Middle-aged and Older Adults: A Panel Analysis Using KLoSA Data," with Hyun-E Yeom, Journal of Health Informatics and Statistics 2025 Aug, 50(3): 240-248. 

Objectives: Cancer remains the leading cause of death worldwide, and as the proportion of older adults increases, the public health burden from cancer is expected to grow. This study aimed to investigate predictors of cancer among middle-aged and older adults in Korea, with a particular focus on obesity. Methods: Using 15 years of data from Korean Longitudinal Study of Aging (2006 - 2020), we first applied the Least absolute shrinkage and selection operator (Lasso) to baseline data to identify relevant predictors of cancer incidence. Subsequently, we conducted the logistic regression analyses to estimate the effect of underweight, overweight, and obesity statuses on the probability of developing cancer, controlling for individual characteristics identified as relevant by the Lasso. Results: Following individuals who were cancer-free at baseline over a 15-year period, we found that obesity (BMI ≥ 25) at baseline was a significant predictor of cancer incidence occurring two to fourteen years later, even after adjusting for other predictive factors. Robustness checks confirmed that baseline obesity, even after controlling for obesity measured at various time points, remained a consistent predictor of cancer diagnosed two to twelve years later. Additionally, underweight status (BMI < 18.5) was also associated with a significantly increased risk of cancer incidence in later years. Conclusion: These findings highlight the predictive effects of obesity on cancer incidence in both short- and long-term perspectives and suggest that maintaining a healthy weight range may reduce cancer risk. The results underscore the importance of weight management as a key strategy in cancer prevention for aging populations. 

2. "Long-term Predictors of Cognitive Decline for Men and Women," Journal of Women and Economics, 21(1), April 2024, 1-21.

This study examines long-term risk predictors of cognitive decline among middle-aged and older adults in Korea, focusing on how predictors are different across men and women. Applying the Lasso methods, a standard machine learning approach, to the 10-year longitudinal data that follows individuals who were initially healthy in cognitive function from the Korean Longitudinal Study of Aging of Koreans, we aim to select only relevant risk predictors among an extensive set of individual characteristics that cover socio-demographic factors, economic status and activities, health status, and behavioral factors. We find heterogeneity in selected risk factors across men and women except for age and education level. For men, marital status, parental living status, and employment activities, among others, are estimated to predict cognitive decline. For women, handgrip strength, smoking, and cardiovascular diseases are estimated to affect the risk of cognitive decline. When the sample is stratified by employment activities, we also find heterogeneity in predictors across men and women. The long-term risk predictors for non-employed men and women are health-related factors. In contrast, marital status/parental living status and handgrip strength are key risk predictors for employed men and women, respectively. The findings are useful for classifying the risk group of cognitive decline at an early stage through selected risk predictors and for developing cognitive decline preventive programs that target the most vulnerable group at risk.

3. "Evaluating the Predictive Efficiency of Obesity-related Factors for Type 2 Diabetes: A Panel Study Using KoGES Data,"  with Hyun-E Yeom, Journal of Health Informatics and Statistics, 49(1), February 2024, 54-61.

Obesity is a global health concern, widely recognized for its association with the risk of developing type 2 diabetes. This study aimed to investigate the relative predictive efficacy of obesity-related indicators, including BMI, body fat percentage, and waist-to-hip ratio, in predicting the incidence of type 2 diabetes in Korean middle-aged to older adults. Using twelve years of panel data derived from the Korean Genome and Epidemiology Study, we conducted a logit regression analysis to evaluate the predictive efficiency of three obesity-related indicators while controlling for individual characteristics and behavioral risk factors. For initial screening, we examined the variation of three obesity indexes explained by age and male indicators. Among the three obesity indexes, about 11.3% to 49% of variations in body fat percentage and waist-to-hip ratio were explained away by age and male gender, but these two factors hardly explained BMI variation. We showed that all three obesity-related indicators were significant predictors of type 2 diabetes without other controls in the model. However, once controlling for socioeconomic and behavioral factors likely associated with health, we showed that only BMI remained a significant predictor of type 2 diabetes. We further showed that older age, male gender, smoking, and sleep duration were also significant predictors of type 2 diabetes, other than BMI. Our findings underscore the relatively superior predictive efficacy of BMI beyond the other obesity indicators for type 2 diabetes risk. The results indicate that a comprehensive assessment, combined with secure BMI monitoring, is imperative for preventing and early detecting type 2 diabetes, especially in populations with multiple risk factors. 

4. "Long-term Distributional Prediction of Cognitive Function," Seoul Journal of Economics, 37(1), February 2024, 75-98.

This study examines the long-term effects of diverse risk factors on the distribution of cognitive function measures, paying special attention to potential heterogeneities across different levels of cognitive function scores. It employs quantile regression techniques on a 10-year panel dataset from the Korean Longitudinal Study of Aging to assess the predictability of risk factors on cognitive decline. Findings indicate that factors such as age, education level, social interactions with close friends, and health status have more pronounced effects on cognitive function at lower quantiles of the Mini-Mental State Examination (MMSE) scores than at higher quantiles. This study also reveals that social interactions with parents, spouses, or close friends significantly predict cognitive function beyond age and education level, which are established nonmodifiable risk factors. It also identifies gender-specific predictors of cognitive function, namely, parental living status, marital status, and satisfaction with health and life for men and income and handgrip strength for women. The differential impact of these risk factors on MMSE score distribution suggests that interventions tailored according to the assessed cognitive function levels could be effective in identifying the cognitive decline risk group and implementing preventive measures. 

5. "Validation of the Korean Version of the Health Care Climate Questionnaire among Cancer Survivors,"  with Hyun-E Yeom and Jungmin Lee, Healthcare, 12(3), 323, January 2024, 1-10.

6. "Long-term Predictors of Cardiovascular Disease: A Machine Learning Approach," Journal of Economic Theory and Econometrics, 34(4), December 2023, 86-114.

This study investigates long-term cardiovascular disease (CVD) risk predictors for middle-aged and older adults in Korea. Using the Least Absolute Shrinkage and Selection Operator (Lasso) and the double-selection Lasso, this study provides novel evidence that Body Mass Index (BMI) is a single risk factor with long-term predictability for CVD odds ratio, selected apart from age, which is non-modifiable. The lasting effect of BMI on CVD risk remains robust and consistent across different methods and specifications that account for variable selection errors in high-dimensional logit regression and BMI’s time trends. In addition to the long-term predictive role of BMI in CVD risk, the disease burden associated with increased BMI is quantified by comparing the marginal effects of BMI to those of age across various groups. The marginal effect of elevated BMI is more pronounced in men than women and among the employed compared to the non-employed. Leading a healthy lifestyle through the control of BMI is a critical element for preventing CVD based on the empirical findings of the current study. 

7. "Examining the Heterogeneity by Employment Status in Dynamics between BMI and Cognition: A Longitudinal Cohort Study of the Korean Aging Population," with Hyun-E Yeom, Annals of Epidemiology, 87, November 2023, 1-8.

This study investigated the potential heterogeneity by employment status in the relationship between body mass index (BMI) trajectory and cognitive function among an aging population. We analyzed 2010–2018 cohort data from the Korean Longitudinal Study of Aging for individuals aged > 45 years (N = 4,889). We used quadratic terms, interaction terms, time-invariant unobserved fixed effects, and time-lag effects to estimate the dynamic and interactive relationships among study variables. The effect of BMI on cognitive function was heterogeneous based on employment status. For the non-employed group, the impact of BMI on cognitive function was demonstrated in an inverted U-shape with a turning point (BMI 25); a higher BMI (up to 25) was associated with higher cognitive function, but further increases beyond this threshold led to decreased cognitive function. For the employed group, the impact of BMI on cognitive function was non-significant. The nonlinear effect of BMI on cognitive function for the non-employed group was robust across various subgroups and specifications. The findings highlight the risk of obesity (BMI ≥ 25) on cognitive decline, particularly among non-employed individuals. This illuminates the critical role of labor activity in regulating the impact of BMI on cognitive function among an aging population. 

8. "Threshold Effects of Body Mass Index on Cognitive Function and Heterogeneity by Sex and Cardiovascular Risk Factors," with Hyun-E Yeom, Frontiers in Public Health, 10 July 2022.

Disclosing the underlying relationship between body mass index (BMI) and cognitive decline is imperative for cognitive impairment prevention and early detection. Empirical studies have indicated the risk of abnormal BMI leading to cognitive impairment. However, the relative risk of underweight or overweight on cognitive function is obscure. This study investigated the asymmetric causal effect of BMI on cognitive decline below and above an unknown threshold and the heterogeneity in the threshold level and the magnitude of the threshold effect due to sex and cardiovascular risk factors. Using 2010-2018 panel data from the Korean Longitudinal Study of Aging and a generalized method of moments estimator of the panel threshold model, we found that there was a threshold effect in the relationship between BMI and cognitive function. An increase in BMI below the threshold was associated with higher cognitive function, whereas a further increase in BMI above the threshold led to cognitive decline. The nonlinear pattern between BMI and cognitive function differed by sex and cardiovascular risk, appearing more distinctively within men or the cardiovascular risk group. The detrimental impact of being underweight or overweight on cognitive function is heterogeneous by sex or cardiovascular risk. For obese men or individuals with cardiovascular risk factors, maintaining adequate BMI should be highlighted to help prevent cognitive decline. 

9. "Age and Sex-specific Associations between Depressive Symptoms, Body Mass Index and Cognitive Functioning among Korean Middle-aged and Older Adults: A Cross-sectional Analysis," with Hyun-E Yeom, BMC Geriatrics, 22:412, 2022.

Although depression and body weight have been noted as important predictors of cognitive health, it remains unclear how age and sex influence the mechanism by which depressive symptoms and body weight are associated with cognitive functioning. This study examined whether and how the relationships between depressive symptoms and cognitive functioning mediated by body mass index (BMI) differ in terms of age and sex. A cross-sectional analysis of a large sample of population-based data (N = 5,619; mean age 70.73 [±9.07]), derived from the Korean Longitudinal Study of Aging, was conducted. Depressive symptoms were measured through the 10-item Center for Epidemiologic Studies Depression (CES-D) scale, and cognitive functioning was assessed with the Korean Mini-Mental State Examination (K-MMSE). The results showed that depressive symptoms were significantly associated with cognitive decline directly and indirectly through reduced BMI. The estimated coefficients indicated that a one standard deviation increase in CES-D scale was associated with about 0.9 decrease in K-MMSE score. However, the indirect relationship between depressive symptoms and cognitive function through BMI emerged only in men or individuals older than 70 years. The findings suggest that a careful assessment of BMI is warranted for early detection and prevention of cognitive decline related to depressive symptoms, particularly among older men.

10. "Heterogeneous Impacts of Body Mass Index on Work Hours," International Journal of Environmental Research and Public Health, 18(18), 2021.

This study examined how higher body mass index (BMI) affects the work hours of men and women and how the impact varies by gender and the value of BMI. Using a longitudinal dataset of 1603 British adults (men: n = 775; women: n = 828) and a panel threshold regression model, this study estimated that BMI has significant impacts on work hours but the pattern is different by gender and BMI groups. BMI is positively associated with work hours up to the estimated BMI threshold of 30, which corresponds to the clinical cutoff point of obesity; above this point, additional increases in BMI is associated with reduced work hours. The asymmetric nonlinear relationship between BMI and work hours was more evident among women, particularly female low-skilled workers. The results imply reduced work capacity and lower labor income for women with a higher BMI above an obesity threshold, highlighting a practical role of BMI’s obesity cutoff value. The findings of this study provide a new perspective regarding the economic burden of workplace obesity and point out the need to design gender-specific and BMI-based strategies to tackle productivity loss from obesity. 

11. "The Distributional Effect of Education on Body Mass," with Vince Daly, Seoul Journal of Economics, 34(2), May 2021.

We investigate the effect of education on mid-life obesity, with particular attention to potential heterogeneity across the Body Mass Index (BMI) distribution. Applying quantile regression methods to British men and women, we first find that childhood and parental BMI are critical determinants of obesity in middle age. We then establish that even when controlling for various weight-related factors in childhood and a potential endogeneity bias, a higher education level reduces the probability of being obese in middle age. We show that this education effect is obtained by a compression of the distribution of BMI (kg/m2) and a shifting of its center leftward toward a more healthy BMI range. We further show that income and physical activity are important channels of the education effect, and the significant effect of education at the upper quantile of the BMI distribution is neither a disguised income effect nor a healthy behavior effect. 

12. "Discussion on Selecting the Number of Breaks in the Pattern of Spread of COVID-19 (a reply to Zhao and Liang)," with Myung Hwan Seo and Hyun-E Yeom, International Journal of Infectious Diseases, 100, November 2020, 132.

13. "패널 자료를 이용한 교육이 비만에 미치는 효과 분석," 사회과학연구, 46(2), 2020, 149-166. 

This study investigates whether and to what extent education level is associated with obesity using a novel approach of quantile regressions. Based on the sample of 3,065 individuals from the Wisconsin Longitudinal Study and the details of educational attainment, body mass index (BMI), and family background in high school, we examine how education level is associated with the distribution of BMI in middle age. In order to control for genetic and childhood environmental factors that are likely associated with BMI in adulthood, we include sibling’s BMI and other family characteristics in the model. As a potential channel of education effect on BMI, we consider income effect by including parental income in high school and own middle-age income. We find that university education is significantly and negatively associated with BMI with a larger effect at the upper quantile of the BMI distribution. For sensitivity test, we consider alternative estimation approaches including instrumental variable quantile regression, unconditional quantile regression, and inequality test between lower and upper quantile points. The robust findings of education effect at the university level substantiate the important role of education in the prevention and treatment of obesity. 

14. "Estimating a Breakpoint in the Pattern of Spread of COVID-19 in South Korea," with Myung Hwan Seo and Hyun-E Yeom, International Journal of Infectious Diseases, 97, August 2020, 360-364. 

Amid the global coronavirus disease 2019 (COVID-19) crisis, South Korea has been lauded for successfully preventing the spread of this infectious disease, which may be due to the aggressive implementation of preventive policies. This study was performed to evaluate the pattern of spread of COVID-19 in South Korea considering the potential impact of policy interventions on transmission rates. A SIR (susceptible–infected–removed) model with a breakpoint that allows a change in transmission rate at an unknown point was established. Estimated trajectories of COVID-19 from SIR models with and without a breakpoint were compared. The proposed model with a break fitted the actual series of infection cases much better than the classic model. The estimated breakpoint was March 7, 2020 and the transmission rate dropped by 0.23 after the breakpoint. A counterfactual study based on our estimate indicated that the number of infected could have reached 2 500 000 compared to the peak of 8000 in the observed series. It is critical to consider a change in the transmission rate to evaluate the trajectory of spread of COVID-19 in South Korea. Our estimation and counterfactual experiments indicate that public health interventions may play a role in determining the pattern of spread of infectious diseases.

15. "Interactive Impact of Sleep Duration and Sleep Quality on the Risk of Developing Metabolic Syndrome in Korean Adults," with Hyun-E Yeom, Healthcare, 8(2), June 2020, 1-10.

Sleep quality is important for the normal functioning of hormonal and metabolic processes in the body; however, few studies have considered the effects of both sleep duration and sleep quality on predicting metabolic syndrome risk. We examined the interactive impact of sleep duration and sleep quality on the risk of developing metabolic syndrome using logistic regression analysis with a threshold based on hours of sleep. Data were collected from 411 adults in South Korea and, according to the estimated threshold of 6 hours of sleep (95% Confidence Interval, CI= 5-7 hours), participants were classified as short (<6 h) or adequate-long (≥6 h) sleepers. The two groups differed significantly on various health measures. While short sleepers were more likely than adequate-long sleepers to experience adverse health conditions, which increased their risk of developing metabolic syndrome, they were not influenced by sleep quality. For adequate-long sleepers, however, a decrease in sleep quality was associated with an increased risk of developing metabolic syndrome (odds ratio=1.24, 95% CI= 1.07-1.43). Our results suggest that both sleep duration and sleep quality are crucial determinants of the development of metabolic syndrome and that it is important to maintain at least 6 hours of sleep.

16. "Born to Be More Educated? Birth Order and Schooling," Review of Economics of the Household, 18(1), March 2020, 165-180.

In this study, I investigate the effect of birth order on schooling and its evolution over time. Using a rich dataset of siblings from the Wisconsin Longitudinal Study, I find that for both men and women, older siblings are likely to obtain more schooling than their younger siblings. I also find that the magnitude of the birth order effect is similar across two generations but find no evidence of inheritability of the birth order effect from parents to children. In an effort to disclose possible mechanisms for the observed birth order effects, I further examine how birth order is associated with various intermediate outcomes and the parental environment during the high school years, since it is circumstances during these years that best predict whether a person receives a college education. I find that parental expectations, children's own attitudes, academic performances and IQ scores in high school are significantly associated with birth order in ways that favor the first child.

17. "Estimation of Dynamic Panel Threshold Model using Stata," with Myung Hwan Seo and Sueyoul Kim, Stata Journal, 19(3), September 2019, 685-697.

•   Stata Command xthenreg 

We develop a Stata command xthenreg to implement the first-differenced GMM estimation of the dynamic panel threshold model, which Seo and Shin (2016, Journal of Econometrics 195: 169-186) have proposed. Furthermore, we derive the asymptotic variance formula for a kink constrained GMM estimator of the dynamic threshold model and provide an estimation algorithm. We also propose a fast bootstrap algorithm to implement the bootstrap for the linearity test. The use of the command is illustrated through a Monte Carlo simulation and an economic application.

18. "The Impact of Internet Gaming on Alcohol Consumption," with Dai-Jin Kim and Youngjo Lee, Journal of Mental Health Policy and Economics, 22, June 2019, 61-70.

We investigate the impact of internet gaming on alcohol consumption with data from South Korea. Using the gaming starting age and internet gaming club membership as instruments for internet gaming time, we show that men are less likely to drink alcohol as they engage more with internet games, while women are more likely to drink alcohol with longer hours of internet games. Building on these findings, we go on to explore potential factors that contribute to heterogeneity in gaming effects by gender with details on gaming behavior and preference. We find large disparities in types of gaming devices and playing partners between men and women and that these factors account for part of the different gaming effects. The opposite effects of internet gaming on alcohol consumption for male and female users are robust to alternative specifications and estimation methods. 

19. "Is There a Jump in the Transition?," with Myung Hwan Seo, Journal of Business and Economic Statistics, 35(2), April 2017, 241-249.

This article develops a statistical test for the presence of a jump in an otherwise smooth transition process. In this testing, the null model is a threshold regression and the alternative model is a smooth transition model. We propose a quasi Gaussian likelihood ratio statistic and provide its asymptotic distribution, which is defined as the maximum of a two parameter Gaussian process with a non-zero bias term. Asymptotic critical values can be tabulated and depend on the transition function employed. A simulation method to compute empirical critical values is also developed. Finite-sample performance of the test is assessed via Monte Carlo simulations. The test is applied to investigate the dynamics of racial segregation within cities across the US.

20. "The Long-Run Effect of Education on Obesity in the US," Economics and Human Biology, 21, May 2016, 100-109.

The proportion of obese population has been gradually increasing in the US over the past few decades. In this study I investigate how education is associated with Body Mass Index (BMI) in later stages of life. BMI, weight(kg)/height(m)2, is the principle measure used for classifying people as obese. Using sibling data and methods that take account of unobserved endowments and environment shared by siblings, I find that there is large variation in BMI between siblings and that education is negatively associated with BMI. One more year of schooling is associated with an estimated reduction of 0.15 in BMI. When considering different education levels, completing college education is associated with 0.7 reduction in BMI relative to high school graduation only. The significant effect of education on obesity that remains in the long-run has policy implications.

21. "Head Start, 4 Years after Completing the Program," Education Economics, 21(5), December 2013, 503-519.

In this study, I examine the effect of the Head Start program on children's achievements in reading and math tests during their first four years of schooling after completing the program. Using nationally representative data from the Early Childhood Longitudinal Study, I find large measurement error in the parental reports of Head Start attendance, which is new in the literature. Further I find that, after accounting for measurement error and potential selection bias, black Head Start children make significant progress towards third grade, whereas white and Hispanic children reap little gain from the program relative to their peers who were exposed to other types of program and care. 

22. "Catholic Schooling and Further Education," Economics Letters, 114(3), March 2012, 346-348.

Using new estimation methods and data, I find that Catholic schooling substantially increases years of schooling by 0.42 to 0.47.  The estimates are robust to various specifications that account for potential selection bias.

23. "Catholic Schools or School Quality? The Effects of Catholic Schools on Labor Market Outcomes," Economics of Education Review, 30(3), June 2011, 546-558.

This paper studies the effects of attending a Catholic high school on students’ labor market outcomes. Using data from the Wisconsin Longitudinal Study, I find that Catholic schooling is significantly associated with higher wages over the careers even after taking into account possible selection into Catholic schools with instruments. Using matched school quality data for public and Catholic schools, I further find that Catholic and public schools are different in various aspects of school quality measures and that these differences explain most of Catholic school effects. Among the school quality variables, teacher quality and the number of math courses taken are estimated to matter the most for students’ later earnings in the long run.