Smoking, Drinking and Obesity

1
Smoking, Drinking and Obesity

• Hung-Hao Chang David R. Just
  Biing-Hwan Lin
• National Taiwan University Cornell
  University ERS, USDA
• Present at National Chung-Cheng
  University
• March, 2007

2
Background

• Smoking, Drinking and Obesity have caused serious
  public-health concern in the U.S.
• -- 65 of adults aged 21 and over were either
  overweight or obese. 30 of them were obese.
  Compared to 30 years ago, it increases almost
  50. (Hedley et al, 2004)
• -- Disease burden associated with obesity in
  the U.S is substantial. In 1995, the cost of
  obesity were US 92 billion, 10 of the total
  cost of illness.

3

• In 2000, tobacco smoking caused more than 400,000
  deaths. Smoking has been a leading preventable
  cause of mortality in the United States.
  Recently, anti-smoking has been an important
  policy in U.S.
• Evidence from public health has shown that
  drinking may be associated with smoking behavior.
  

4
Is smoking negatively associated with body
weight?
Smoking and obesity rates are two significant
trends over 30 years in U.S
From Gruber and Frakes (2006), Journal of Health
Economics.
5
Literature Review
6
What do we learn from previous studies?

• Association between body weight and unhealthy
  decisions
• The evidence whether the increased alcohol
  consumption contributes to body weight is mixed.
  However, it may be important to distinct the
  effects of drinking beer and liquor.
• Smoking tends to be negatively associated with
  body weight. However, the negative evidence has
  been re-investigated recently. (Chen et al, 2007.
  Gruber and Frakes 2006).

7
What may drive these inconclusive results?

• Interrelationship between unhealthy decisions
• Smoking and drinking are highly correlated.
  Failing to control for one in estimation may lead
  to serious bias. (Kenkel and Wang 1999).
• Conditional mean effect
• Most of the studies relied on the ordinary
  least squares (OLS). However, this method might
  not be sufficient in the context of obesity. (Kan
  and Tsai 2004).

8
Research Objectives

• Investigate the interrelationship among smoking,
  drinking beer, and drinking liquor. Determine if
  these decisions are jointly or independently
  determined.
• Identify factors that may affect each decision.
• Account explicitly for the effects of these
  decisions on body weight.
• Test if the effects of these decisions on body
  weight are heterogeneous (distinction between
  overweight and normal weight people).

9
Data

• Data from Continuing Survey of Food Intakes by
  Individuals (CSFII 1994-1996) is used. This data
  set is conducted by USDA.
• We exclude individuals under 20 years-old.
• The final sample size includes 3,409 adult of
  this survey.
• Body weight is measured as body mass index (BMI),
  weight in kilograms divided by height in meters
  squared.

10
Distribution of BMI in our selected sample
11

• According to the definition of the Center for
  Disease Control (CDC), overweight people are
  those whose BMI is greater than 25. If the BMI
  exceeds 30, the individual can be regarded as
  obese.
• In our sample, 45 are normal weight about 22
  are identified as obese.
• The distribution of BMI departs from the normal
  distribution.

12
Sample Statistics
13
(No Transcript)
14
Econometric Strategy

• Our econometric model contributes to previous
  studies
• in
• -- Smoking and drinking decisions are
  considered
• jointly.
• -- Account for endogeneity between drinking
  and
• smoking on body weight.
• -- Distinguish effects of these three
  decisions on
• different weight status (distribution of
  BMI).

15
Structure of the Empirical Analysis

• An innovative two-stage econometric model is
  proposed
• Stage 1 Three binary choices are specified
  smoking, drinking beer and drinking liquor. A
  tri-variate probit model is estimated to capture
  the correlations among these choices.
• Stage 2 A body weight equation is estimated to
  account explicitly for the endogenous choices. We
  estimate this quation with quantile regression
  method.

16
Stage 1 Modeling the joint decisions
(trivariate probit model)
17
Stage 1 (cont.)
Smoking Decision
Decision to drink beer
Decision to drink wine
18
Estimate the discrete choice model (MLE)

• The probability of regime (1,1,1)
• Log likelihood function of the entire eight
  regimes

where k12I1-1, k22I2-1, k32I3-1
19
Statistical Evidence of the Joint Decisions
20
Correlations between smoking and drinking

• Drinking beer and liquor is strongly associated
  (56).
• The decisions to smoke and to drink beer are
  significantly correlated (19). In addition, the
  correlation between drinking liquor and smoking
  is 16.
• This is consistent with the evidence of public
  health in terms of the gateway effect.

21
Other Determinants of Smoking and Drinking
22
(No Transcript)
23
Empirical findings

• Perception and knowledge of healthy food
  consumption decrease the likelihood to smoke.
• Low education and income lead to high chance to
  smoke, but low chance to drink wine.
• Male is more likely to smoke, and to drink beer.
• Job status increases the propensity to drink
  wine.
• Young generation has high probability to smoke.
• Other lifestyles also matter. If family members
  are on diet, they are less likely to smoke, and
  to drink beer and liquor.

24
How much we believe in our model
specification?-- Empirical results of
statistical tests
25
Findings

• If binary indicators are used, they are
  endogenous to the body weight. Therefore, there
  is a call for instruments (IV).
• When instruments are used, statistical tests show
  that the added restrictions are not rejected. In
  other words, our selected instruments are not
  over-identified.

26
Stage 2 Body Weight Equation

• The body weight equation is specified as
• To avoid endogeneity, predicted probabilities are
• used as instruments for Ij. Quantile regression
  is
• used to estimate this equation (Koenker and
  Bassett
• 1978).

27
Marginal change in the Qth quantile of BMI as a
result of change in X
28
Evidence of heterogeneous effects on BMI
29
Effect of smoking on BMI distribution
30
Effect of Drinking Liquor on BMI distribution
31
Effect of Drinking Beer on BMI distribution
32
Effects of other variables
33
Empirical findings

• A significant evidence supports the
  misspecification of using OLS. The effects are
  heterogeneous across the entire distribution of
  BMI.
• Smoking tends to be negatively correlated with
  BMI. However, it is insignificant over the entire
  distribution of BMI.
• Drinking beer tends to increase the body weight.
  However, this effect is not significant for obese
  people (above 85 percentile).

34

• Drinking liquor is found negatively associated
  with body weight. In addition, the decreasing
  effect is significant for obese people (75
  percentile).
• Knowledge of healthy food consumption decreases
  the risk of being overweight.
• Higher income leads to lower body weight.
• Race is also associated with body weight. Black
  have heavy weight than others, on average Asian
  are those with less weight.

35
Concluding and Policy Implications

• The discussion of smoking, drinking and obesity
  should be interpreted with caution. We have
  shown
• -- strong correlations between smoking, drinking
  beer and drinking liquor.
• -- heterogeneous effects of these decisions on
  BMI.
• The effect of smoking on body weight is found
  insignificant. As such, anti-smoking may not be
  the critical factor driving the increasing trend
  of body weight over 30 years.

36

• Drinking liquor is found negatively associated
  with body weight. Particularly, the effect is
  even stronger for normal weight people.
• Drinking beer tends to increase body weight
  regardless of the weight status. Beer drinkers
  are those in a higher risk of being overweight.
• Knowledge of healthy food consumption also have
  direct and indirect effects on body weight. A
  well-educated consumer has less likelihood of
  being overweight.