This document discusses several studies that estimate the effect of obesity on US healthcare costs. It first replicates the two-part model used in Cawley and Meyerhoefer (2012) on data from 2000-2005, finding similar results. It then modifies the model to use data from 2000-2012, adds further controls, and investigates time heterogeneity. The document also critiques the strengths and improvements needed in Kuo et al. (2015), which uses machine learning approaches and lifetime medical expenditures.
1. The Effect of Obesity on US Healthcare Costs
Enoch Chan
Neil Cho
Yuan Fei
Jessica Koh
May 15, 2015
Enoch Chan Neil Cho Yuan Fei Jessica Koh The Effect of Obesity on US Healthcare Costs May 15, 2015 1 / 12
2. Outline
Outline
1 Model Replication (Cawley and Meyerhoefer (2012))
Two-Part Model
Instrumental Variables
Replications Results
Limitations
2 Modifications
Final Results
3 Critique of Kuo, et al. (2015)
Strengths
Improvements
4 References
Enoch Chan Neil Cho Yuan Fei Jessica Koh The Effect of Obesity on US Healthcare Costs May 15, 2015 2 / 12
3. Model Replication (Cawley and Meyerhoefer (2012)) Two-Part Model
Two-Part Model (2PM)
Part 1: Logit
ln
pit
1 − pit
= β1BMIit + αt + Xitβc + it (1)
We obtain Pr(MedEx > 0)
Part 2: Gamma GLM with Log Link
ln(µYit
) = β1BMIit + αt + Xitβc + uit Yit ∼ Γ(α, β) (2)
Run on the subset of data where MedEx > 0
We obtain E[MedEx|MedEx > 0]
We obtain
E[MedEx|BMI, αt, Xc] = Pr(MedEx > 0) × E[MedEx|MedEx > 0] (3)
Enoch Chan Neil Cho Yuan Fei Jessica Koh The Effect of Obesity on US Healthcare Costs May 15, 2015 3 / 12
4. Model Replication (Cawley and Meyerhoefer (2012)) Instrumental Variables
Instrumental Variables
Motivation
Obesity may be endogenous
Possible mismeasurement error
Implementation
BMIit = β1BMIeld,it + β2BMI2
eld,it + αt + Xitβc + it (4)
Use two-degree polynomial of BMI of eldest biological child to
instrument for BMI
Use Obesity indicator of eldest biological child to instrument for
Obesity
We obtain ˆBMIit. Substitute ˆBMIit for BMIit in the two-part model.
Enoch Chan Neil Cho Yuan Fei Jessica Koh The Effect of Obesity on US Healthcare Costs May 15, 2015 4 / 12
5. Model Replication (Cawley and Meyerhoefer (2012)) Replications Results
Replication Results
Cawley and Meyerhoefer’s results:
IV: BMI: 149 (35), Obesity: 2741 (745)
Non-IV: BMI: 49 (9), Obesity: 656 (113)
Table: Marginal Effect Estimates from Two-Part Model (2000 to 2005 data)
IV (total expenditure) Non-IV (total expenditure)
VARIABLES (1) (2) (3) (4)
BMI 125.696** 68.848***
(40.306) (9.540)
Obesity 2062.187* 647.113***
(732.5577) (116.860)
Observations 40,421 40,421 40,421 40,421
Robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
We cluster standard errors at the household level
Enoch Chan Neil Cho Yuan Fei Jessica Koh The Effect of Obesity on US Healthcare Costs May 15, 2015 5 / 12
6. Model Replication (Cawley and Meyerhoefer (2012)) Limitations
Limitations
Instrumental variable validity
Genes that influence weight may also affect other unknown factors that
could affect residual medical costs
Non-longitudinal data
Unable to track respondents over time and estimate long-term effects
Mismeasurement in data
Generalizability of results
Limited to adults with biological children between ages of 11-20
Possible underestimation of results
Enoch Chan Neil Cho Yuan Fei Jessica Koh The Effect of Obesity on US Healthcare Costs May 15, 2015 6 / 12
7. Modifications
Modifications
1 Validity of the two-part model
Health expenditures are actual outcomes
The zeros in our data are “true zeros”
Frondel and Vance (2012): two-part model is most appropriate
2 Baseline model: expand time horizon to 2000-2012
Table: Marginal Effect Estimates from Two-Part Model (2000 to 2012 data)
IV (total expenditure) Non-IV (total expenditure)
VARIABLES (1) (2) (3) (4)
BMI 169.425*** 85.616***
(30.313) (7.234)
Obesity 3297.591*** 982.399***
(558.406) (97.053)
Observations 88,880 88,880 88,880 88,880
Robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
Enoch Chan Neil Cho Yuan Fei Jessica Koh The Effect of Obesity on US Healthcare Costs May 15, 2015 7 / 12
8. Modifications
Modifications
3 Further controls
Motivation: control for the effect of diseases with genetic causes that
are unrelated to obesity
Additional controls: health perception, insurance coverage
4 Investigating time heterogeneity
Motivation: increase in marginal effect of obesity when comparing the
2000-2012 sample with the 2000-2005 sample
Interaction terms between BMI/obesity and year fixed effects
Only 2012 showed a significant effect - topic for further research
Enoch Chan Neil Cho Yuan Fei Jessica Koh The Effect of Obesity on US Healthcare Costs May 15, 2015 8 / 12
9. Modifications Final Results
Final Results
Table: Marginal Effect Estimates from Two-Part Model with Further Controls
(2000 to 2012 data)
IV (total expenditure) Non-IV (total expenditure)
VARIABLES (1) (2) (3) (4)
BMI 93.136*** 35.979***
(29.527) (6.676)
Obesity 1956.029*** 293.408***
(552.047) (90.678)
Observations 88,880 88,880 88,880 88,880
Robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
Enoch Chan Neil Cho Yuan Fei Jessica Koh The Effect of Obesity on US Healthcare Costs May 15, 2015 9 / 12
10. Critique of Kuo, et al. (2015) Strengths
Strengths of Kuo, et al. (2015)
Use Machine Learning approaches to determine optimal model and
instrument
Box-Cox transformation, Markov Field
Refines definition of obesity into finer categories
Innovative use of Lifetime Medical Expenditures (LME)
Enoch Chan Neil Cho Yuan Fei Jessica Koh The Effect of Obesity on US Healthcare Costs May 15, 2015 10 / 12
11. Critique of Kuo, et al. (2015) Improvements
Improvements for Kuo, et al. (2015)
Instrumental variables
Validity: using average family weight as the instrument does not solve
the endogeneity problem
Implementation: unclear where and how IV is implemented
Implementation of LME
Calculation of LME per age group is a linear approximation of a
nonlinear relationship
Major difference between predicted marginal effect of obesity on LME
for men and women
Enoch Chan Neil Cho Yuan Fei Jessica Koh The Effect of Obesity on US Healthcare Costs May 15, 2015 11 / 12
12. References
References
Cawley, John and Meyerhoefer, Chad. “The medical care costs of
obesity: An instrumental variables approach.” Journal of Health
Economics, 2012, 31(1), pp.219-230.
Frondel, Manuel and Vance, Colin. “On Interaction Effects: The
Case of Heckit and Two-Part Models.” Ruhr Economic Papers, 2012,
309(1), pp.1-21.
Enoch Chan Neil Cho Yuan Fei Jessica Koh The Effect of Obesity on US Healthcare Costs May 15, 2015 12 / 12