Download as pdf or txt
Download as pdf or txt
You are on page 1of 28

1

The Effect of Socioeconomic and Sociodemographic Variables on Obesity


- Using Pooled Regression and Pseudo Panel Approach

Lee, Choonsoo
1


Abstract

This study investigates the effect of SE and SD variables like income, occupation and its status,
education, age etc. on BMI using pooled regression and pseudo panel approach. The different effect
of occupation status between employees and others and regular and irregular employees on BMI is
investigated by analyzing the effect of occupation status on BMI to analyze the effect of employment
instability on obesity.
The fixed effect on BMI is estimated from pseudo panel model. The fixed effect of age on BMI
looks non-linear as age increase, esp. invert U-shaped. In terms of the fixed effect of sex BMI of men
is bigger than that of women but BMI of men is lower than that of women among cohorts who are
above 50s and below 3rd income percentile. Although the BMI of men increases as an income
increases, the effect of income on BMI of women is different by age level: The BMI of cohorts who
are below 30s decrease as an income increases, but BMI of cohorts who are above 40s increase as
income increase like BMI of men. At the end of the study the implication of this study is suggested.

. Introduction

The increase of prevalence of obesity has been threatening public health in Korea. Obesity is
related to the incidence of chronic diseases such as diabetes and hypertension. The direct association
with the risk of cardiovascular disease and obesity (Park et al., 2003), the higher risk of death from
diseases leading high mortality such as ischemic heart disease, cancer were reported in obese group in

1
PhD candidate, the Department of Food & Resource Economics, Korea University, Tel : +82-2-
3290-3481, E-mal: namakas@korea.ac.kr
2

Korea (Jee et al. 2005; Jee et al., 2006). In terms of costs of obesity, Jeong et al. (2002) reported that
'4,225 billion ~2,050 billion won' was estimated as socioeconomic costs of obesity in Korea although
this study did not count the cost from obesity prevention or treatment.
Most of studies about the effect of SE (socioeconomic) and/or SD (sociodemographic) variables on
obesity in Korea analyze the effect using cross sectional data like data from KNHNES (Korea
National Health and Nutrition Examination Survey). Due to the limitation of cross sectional data those
studies have limitation to analyze the long-run effect of those variables. This study analyzes the long-
run effect of SE and SD variables on obesity by using pseudo panel data.
Labor stability becomes more unstable after the Asian economic crisis of 1997. The extent of
employment instability as 'a greater possibility of losing a job and a declining possibility of re-
employment' was higher post 2000 compared to the period of before financial crisis and when
considering the status of workers, such an increase in employment instability can be characterized by
a greater possibility of unemployment for daily workers. At the same time there is an actual decline in
the number of jobs and the jobs that are being created are mostly irregular ones (Nam, 2005). People
who got stress very much secrete stress hormones like cortisol and catecholamine and hypersecretion
of these hormones negatively affect an obesity (Kang, 2006). Thus the increase of employment
instability might negatively affect an obesity. But there are few literatures which analyze the effect of
employment instability on obesity. This study analyzes the effect of employment instability on obesity
by analyzing the effect of occupation status on obesity.
The objective of this study is like that:
First, this study investigate the effect of SE (socioeconomic) and SD (sociodemographic) variables
like income, occupation and its status, education, age etc. on BMI (body mass index) using pooled
regression and pseudo panel approach. Especially using data are from KNHNES (include surveys in
1998, 2001, 2005, 2007, and 2008) the long-run effects of those variables on BMI and obesity are
investigated by pseudo panel approach.
Second, this study analyzes the effect of employment instability on obesity by analyzing the effect
of occupation status on BMI. The different of effect of occupation status on BMI is estimated.
Especially the distinction of effect between employees and others (including employer etc.) and
between occupation status of employees (regular employees and irregular employees) was clarified.
3

. Literature Review

A number of studies focused on relevant behavioral factors causing obesity. Diet and regular
physical activity, alcohol consumption and smoking, mental stress and marital status were considered
relevant factors related to obesity. Some of these studies counted socioeconomic factors in their
analysis as control variables. Most of the studies were cross-sectional studies and defective in finding
the causality.
The relationship of high total energy intake and obesity was not examined consistently. However,
the quality of meal such as the ratio of macronutrients could affect BMI (body mass index) or obesity.
Kim et al. (2007) identified the dietary diversity in obese adults without a control group and they
pointed out that the higher proportion of fat (22%) of total calories intake and frequent snacking were
considered as factors related to obesity. Shin et al. (2007) investigated the relationship between the
macronutrient intakes and obesity with four 3-day dietary records of 115 women for a year and find
out the unbalanced contribution of macronutrients to total energy intake affected BMI
Yoon et al.(2004) investigated the link between alcohol intake and metabolic syndrome with 1998
KNHANES consisting of 7962 adults (4365 women). With alcohol consumption, BMI and WC (waist
circumference) did not have linear association. But the odds ratio of having a large waist was 1.71
among heavy drinker. This result derived from adjusting for age, marital status, income, BMI,
smoking status and exercise and is valid because it used population based data in spite of secondary
data analysis. Paek and Hong (2006) found that alcohol intake has linear relationship with WC
although other relevant factors were not counted. In terms of smoking status, because the number of
case of smoking among women was not enough, no significant relationship was shown with obesity
measures in most of reviewed studies.
According to Park and Kim (2002), concerning mental stress, the odds ratio of visceral fat among
obese adults tended to be significantly lower by stress level whereas Kim et al. (2005) identifying the
associated factors with fat distribution in the abdomen among obese women found the significant
linear association between subcutaneous fat and stress level measured by the same method. Park and
Kim (2002) pointed out that marital status was associated with visceral fat obesity.
4

There are lots of studies which investigated related factors to obesity focusing on SEP
(socioeconomic position). The risk of obesity is inversely related to education (Kim and Kim 2007;
Yoon et al. 2006). Diverse SEP should be employed in the analysis to identify the pathway of SEP
leading to obesity but education tends to be easily adopted rather than other SEP measures such as
income or occupation. Yoon et al. (2006) found that the risk of obesity and abdominal obesity were
associated with education level inversely among women, whereas those were linked to income level
directly among men. Jung et al. (2002) reported that WCs among three groups by education level
were significantly different. The income had an inverse association with obesity among women. Paek
and Hong (2006) found that WC had inverse relationship with education among women aged over 35.
Bae et al. (1999) included the occupation as a socioeconomic factor in their study investigating the
factors related to BMI. High class for professionals and managers, middle class for sales and
administrators and low class for manuals were classified. In bivariate analysis between occupation
and BMI, BMI among low class was highest but after adjusting other factors, the relationship was not
significant.
Chung (2006) and Paek et al. (2006) identify how the effect of SEP on lifestyle factors leading to
obesity. Chung (2006) investigated the relative factors of abdominal obesity based on the 2001
KNHNES data. Smoking status, alcohol drinking, physical activity and the quality of meal were
adopted as lifestyle factors. Among women, according to simple logistic regression analysis, lower
education, lower poverty income ratio group, smoker, non alcohol drinker group and lower quality of
meal group had higher risk of abdominal obesity. As a result of multivariate logistic analysis, age and
education level showed significant associations with the risk of abdominal obesity. Alcohol drinking
had a significant inverse relationship with abdominal obesity whereas smoking was a directly
associated. There was no significant association with physical activity and the quality of meal.
Considering socioeconomic measures and behavioral factors simultaneously, educations still had a
significant effect on abdominal obesity. The effects of behavioral factors such as alcohol drinking and
smoking status were disappeared when SEP were included in the analysis. Paek et al. (2006) found
that there were significant effects between education and smoking, education and physical exercise,
income and alcohol intake, income and smoking on the risk of metabolic syndrome.
5

In this study many SE and SD variables used in precedent studies were used to analyze the effect of
occupation on obesity. Additionally this study use pseudo panel approach to investigate the long-run
effect of SE and SD variables unlike precedent studies which analyze the effect of SE and SD
variables using cross-sectional data.

. Definition of Variables

This study use data from the KNHNES in 1998, 2001, 2005, 2007, 2008. KNHNES was conducted
by the Korea Institute for Health and Social Affairs (KIHSA), the Korea Health Industry
Development Institute (KHIDI) and the Korea Center for Disease Control and Prevention (KCDCP).
Survey in 1998, 2001 was conducted by KIHSA and KHIDI, and survey in 2005 was conducted by
KIHSA, KHIDI, and KCDCP, and survey in 2007 and 2008 was conducted by KCDCP. Survey was
commissioned by the Ministry of Health and Welfare in response to the regulations in the National
Health Promotion Act. The KNHNES consists of four parts: the Health interview Survey, the Health
Behavior Survey, the Nutrition Survey, and the Health Examination Study. Totally 25,393 samples
which have all variables used in this study are used. Number of samples by years like that: 7,102 in
1998, 5,167 in 2001, 4,720 in 2005, 2,771 in 2007, and 5,633 in 2008. The reason why number of
samples in 2007 is small is that survey in 2007 is the sub-survey of 4th KNHNES including 2007,
2008, and 2009.

1. Independent variables

According to precedent literatures various SE and SD variables like income, sex, age, education,
marital status, smoking, drinking, and exercise affect obesity. This study classifies SE and SD
variables into 4 categories: household characteristics, personal occupation characteristics, personal SD
characteristics, and personal life style (table 1).
Household income and household size (number of member in household) are included in household
characteristics. Higher income increases possibility of having a higher qualitative meal and managing
6

health. Thus higher income may affect obesity negatively. Household income is used as criterion to
generate cohorts.
Household size represents the effect of family care and higher size may affect obesity negatively
because generally obese people are perceived negatively. And considering a negative relation between
stress and obesity higher size may affect obesity negatively because higher care lessens stress.
Although some studies consider the effect of household housing and asset, this study exclude these
variables. Because the effects of these variables can be captured by household income and these
variables are not available in 1998 and 2001.
Variables included in personal occupation characteristics are occupation and occupation status.
Because intensity of work, stress level and level of physical activity are different by occupations, the
effect on obesity may be different by types of occupation. In KNHANES occupation types are
different by years. There 8 types of occupation in 1998, 2001, 2005, but there are 7 types of
occupation in 2007 and 2008. To obtain data consistency occupation types are classified 6 types in
this study and are classified like that: 1) profession (public administration including soldiers and
professional scientific and technical services), 2) office (office job), 3) service (service and sales), 4)
agriculture (agriculture, fishery and skilled labor), 5) manual (simple manual labor including
assembling labor), 6) others (others including students, housewives, and unemployed person). As
worker in profession and office get more stress and are less active, they may have higher risk of
obesity than others. Workers in agriculture and manual who get less stress and are more active have
lower risk of obesity than others. Since workers in service are more active than others but get more
stress due to emotion work, the effect on obesity is not clear.
Occupation status is classified into 3 types: 1) regular (regular employee) 2) irregular (irregular
employee summing irregular labor and day worker) 3) others (others including employer and self-
employed worker, family worker who doesn't receive a salary, unemployed person, students,
housewives). Considering the employment instability employees may have higher risk of obesity. As
irregular employees get more stress than regular employees, the effect of irregular status on obesity is
worse than that of regular status.
Sex, age, level of education, and whether spouse or not are include in Personal SD characteristics.
As women are more interested in weights, risk of obesity of women may be lower than women. Sex is
7

used as criterion to generate cohorts. As aged people tend to be less active and get more nutrition
because of relatively higher income, age may have positive effect on obesity. But people who are
more than 60s tend to eat little. Thus the effect of age may be non-linear. Thus this study include age
and square of age as variables to analyze non-linear, esp. quadratic effect of age. Age is used as
criterion to generate cohorts. The level of education is classified 3 types, below high school, high
school graduate, and more than college graduate. As mentioned in Kim and Kim (2007) and Yoon et
al. (2006), the risk of obesity may be inversely related to education.
Some precedent literatures like Park and Kim (2002) investigate the effect of marital status on
obesity. In this study the marital status of sample is classified two types, spouse and spouseless
(including single, divorce, and separation by death) because the effect of marital status is resulted
from care and nutrition. In the terms of care and nutrition spouseless people may be same.
Smoking, alcohol drinking, frequency of having a snack a day, and frequency of eating out a day
are included as dummy variables which represent personal life style. According to Jung et al. (2002)
the risk of abdominal obesity adjusted for age and sex was higher among the group consuming over
500g alcohols per week. In terms of smoking status, because the number of case of smoking among
women was not enough, no significant relationship was shown with obesity measures in most of
reviewed studies. As in this study data from 5 years are used and many other variables are included as
control variables, the effect of smoking and drinking may be well estimated. As result of precedent
literatures, this study formulates hypotheses that the effect of drinking and smoking on the obesity is
positive and ambiguous respectively.
Data about frequency of having a snack and eating out a day are converted into dummy variables.
People who have a snack or eat out more than once a day is classified as having snack or eating out
person and other people are classified as others. Having snack and eating out may affect obesity
positively because people who have those behavioral intake more nutrition and energy but can't
sustain nutritional balance.

2. Dependent variables

8

Although variables which represent characteristics of obesity is various, generally BMI is used as
criterion of obesity. Thus this study use BMI as a dependent variable which represents obesity. BMI is
the ratio of weight (kg) which is numerator and heights square (m
2
) which is denominator.

Table 1. Characteristics of variables
Variable Characteristics
effect
on obesity
1

Household
characteri
-stics
number of people per
household
continuous variable -
Household income continuous variable -
Personal
occu
-pation
characteri
-stics
Occupation
Public administration / defense and
Professional Scientific and Technical Services
+
Office job +
Services and Sales +/-
Agriculture, fishery and Skilled labor -
Simple manual labor -
Others : unemployed person, student,
housewife etc.
base
Occupation status
Regular employee + small
2
Irregular employee : irregular and day laborer + big
2

Others :
-self employed and employer
-family worker who doesn't receive a salary
-unemployed person, student, housewife etc.
base
Personal
socio
-demo
-graphic
characteri
-stics
Sex
Men +
Women base
Age continuous variable, including age square + -
Level of education
Below high school base
High school graduate -
More than college graduate -
Spouse or spouseless
Spouse -
Spouseless : single, divorce, separating by
death etc.
base
Personal
life style
Smoking or not
Smoking +/-
Non-smoking base
Drinking or not
Drinking +
Non-drinking base
Frequency of having a
snack per day
Have a snack per day more than once +
Others base
Frequency of eating out
per day
Eat out per day more than once +
Others base
Note: 1. Hypothesis about the effect on obesity: + means positive effect, - means negative effect, and +/-
means ambiguous effect on obesity.
9

2. Big and small in the effect of occupation status means that the amount of bad effect of status on
obesity is like that: Regular < Irregular

. Model Specification

1. Pooled regression

The pooled OLS and probit models used in this study is equation (1). If error terms in pooled
regressions are heterogeneous unlike basic assumption, we can't trust the result of test. If 'H0 : error
term is homogenous' is rejected by Breusch-Pagan-Godfrey heterogeneity test, White's
Heteroskedasticity-Consistent Covariance matrix is used to make up for homogeneity assumption.

BNI
I,t
= b
0
+ b
m
T
m,I,t
+ c
k
X
k,I,t
+ e
I,t
(1)
18
k=1
4
m=1


where BMI is continuous variable, and T is time dummy which represent 2001, 2005, 2007, and 2008.
And X is vector of independent variables including occupation, occupation status, sex, age, age
2
,
level of education, spouse, smoking, drinking , having a snack, and eating out .

2. Pseudo panel approach

In this study pseudo panel approach is used to analyze the long-run effect of variables. In pseudo
panel model a set of C cohorts, each with a fixed membership that remains the same throughout the
entire period of observations, is defined. Each individuals observed in the survey belong to exactly
one cohort. Averaging the observation over individuals in each cohort, one gets

BNI

c,t
=
0
+
k
X

k,c,t
+ e
c,t ,
e
c,t
=
c,t
+ u
c,t
18
k=1
(2)
10


where BNI

c,t
is the average of BNI
I,t
over all individuals belonging to cohort c at time t. Unlike in
pooled regression dummy variables become continuous ones. To analyze cohorts fixed effect sex is
omitted from vector of independent variables and trend variable is included as independent variable to
control time effect of above one-way error component model. Value of trend variable by years is like
that: 1 (1998), 4 (2001), 8 (2005), 9 (2007), and 10 (2008). The assumption
c,t
=
c
is needed to
avoid an identification problem. This assumption is plausible if the number of observations in each
cohort is very large.
c,t
are most likely to correlated with the X
I,t
and a random effect specification
will lead to inconsistent estimates (Baltagi, 2008, p.211).
This study uses household income level, sex, and age level as criteria for cohort because those
variables are investigated by precedent literatures as variables which affect obesity and thus can
relatively sustain homogeneity in terms of obesity and BMI. Table 2 represents characteristics of
those variables.

Table 2. Criteria for Cohort
Criteria Level Characteristics
household
income
level
1

1 1st percentile (the lowest income)
2 2nd percentile
3 3rd percentile
4 4th percentile (the highest income)
sex
1 men
2 women
age level
1 20s, below 30 (19-29)
2 30s. more than 30 and below 40 (30-39)
3 40s, more than 40 and below 50 (40-49)
4 50s, more than 50 and below 60 (50-59)
11

5 more than 60
Note: 1. Household income level is classed by Household Equalization Income and lower level means higher
income.
2. Household equalization income = Household income / numbei of people by householu

To avoid identification problem the number of observations in each cohort should be very large.
The number of observations in each cohort is represented in appendix 3. Cohort number is in
appendix is generated like that: income100 + sex10 + age level. Average number of sample in
cohorts by years is 177.6 (1998), 129.2 (2001), 118.0 (2005), 69.3 (2007), 140.8 (2008). Although
average number of sample in cohorts of 2007 is small and some cohorts have few samples,
c,t
=
c

is assumed to avoid identification problem. Considering the number of majority of cohorts (exactly
56%) is more than 100 this assumption is not stubborn.

. Results

1. Pooled regression analysis

Table 3 represents the results of pooled OLS on BMI. As assumption of homogenous error term in
pooled regression is rejected by test, White's Heteroskedasticity-Consistent Covariance matrix is used
to estimate. Most variables have significant effect on BMI except manual, drinking, spouse, and snack.
Variables which affect obesity are like that: time dummy variable (except 2001), agriculture, income,
age square, sex, high school and college, and snack.
In terms of occupation professional, office, and service variables have positive effect on BMI but
agriculture has negative effect. But against expectation irregular and regular have negative effect on
BMI. The difference of effect on BMI is significant but that on obesity is not significant.
Size has positive effect on BMI but income affect BMI negatively in terms of household
characteristics. Age and age square are significant at 1% level and considering sign of each terms
relation between age and BMI is invert U-shaped. The BMI of men is bigger than women and highly
12

educated person represents less BMI . Smoking affect BMI negatively but eating out has positive
effect on BMI.


13

Table 3. The effect on BMI: Pooled regression result
Variable Coefficient t-Statistic
Intercept 18.235*** 79.723
Year
T2001 0.228*** 3.767
T2005 0.408*** 6.516
T2007 0.453*** 6.159
T2008 0.466*** 7.675
Occupation
Professional 0.338*** 3.647
Office 0.263** 2.502
Service 0.397*** 5.811
Agriculture -0.706*** -9.725
Manual 0.107 1.319
Occupation
status
Irregular -0.489*** -5.370
Regular -0.282*** -3.960
Household
Size 0.028* 1.755
Income -0.000*** -2.747
Personal
Socio
demo
-graphic
Age
Age 0.229*** 24.857
Age2 -0.002*** -24.520
Sex 0.444*** 8.629
Education
High school -0.647*** -11.129
College -0.970*** -13.558
Personal
life style
Smoking -0.309*** -6.388
Drinking 0.052 1.067
Spouse -0.016 -0.281
Snack -0.045 -1.012
Eating out 0.113** 2.217
R2 0.065
F-statistics
2
76.66***
Heteroskedasticity test: Breusch-Pagan-Godfrey test
3
8.162***
Wald test: Irregular = Regular
4
5.474**
Note:
1. *(**, ***) means that the value of coefficient is significant at level 10%(5%, 1%).
2. *(**, ***) means that the model specification is significant at level 10%(5%, 1%).
3. The value in heteroskedasticity test is f-statistics and *(**, ***) means that 'H0 : there exists
homoskedasticity' is rejected at significant level 10%(5%, 1%).
4. The value in Wald test is -statistics and *(**, ***) means that 'H0 : the effect of irregular = the
14

effect of regular' is rejected at significant level 10%(5%, 1%).


15

3. Pseudo panel analysis

As error component of pseudo panel model is most likely to correlated with the X
I,t
and random
effect specification will lead to inconsistent estimates, random effect model is not estimated. Only
period fixed effect is tested by F-statistics. Average of variables by cohorts and years in pseudo panel
regression is attached in appendix 1 and appendix 2, and Table 4 represents the result of pseudo panel
estimation. As trend is significant at 1% level, BMI tend to be increased.
In terms of occupation professional and agriculture affect BMI negatively. This result is
inconsistent with the result of pooled regression in which professional, office, and service have
negative effect on BMI.
Household income doesn't affect BMI but household size has negative effect on BMI. Although in
pooled regression income has a negative effect on BMI and size has a positive effect on positive effect
on BMI, in pseudo model the sign of effect on BMI of size is opposite.
The effect of education is consistent with the result of pooled regression. Highly educated cohort
have lower BMI. Like the result of pooled regression smoking has negative effect on BMI. Snack and
eating out do not have any significant effect on BMI.
Shortly, several estimation result of pseudo is inconsistent with the result of pooled regression. That
is very interesting but nettlesome problem. Considering the identification problem of pseudo panel
model inconsistence of result can come from shortage of pseudo panel data.


16

Table 4. The effect of BMI: pseudo panel approach
Variable Coefficient t-Statistic
Intercept 26.279*** 8.310
Trend 0.057*** 3.095
Occupation
Professional -2.200** -1.988
Office -0.445 -0.278
Service 0.131 0.149
Agriculture -2.074*** -2.721
Manual labor 0.968 0.972
Occupation
status
Irregular -0.827 -0.746
Regular -0.150 -0.160
Household
Income 8.16E-05 0.133
Size -0.368* -1.888
AGE 0.001 0.010
Education
High school -1.321** -2.080
College -2.098*** -3.069
Personal
lifestyle
Smoking -0.482*** -3.120
Drinking -0.312 -1.503
Spouse -0.186 -0.370
Snack -0.264 -1.099
Eating out 0.439 1.132
R-squared 0.888
F-statistic
2
19.749***
Period fixed effect test
3
6.118***
Wald test
4
: Irregular = Regular 0.494
Note:
1. *(**, ***) means that the value of coefficient is significant at level 10%(5%, 1%).
2. *(**, ***) means that the model specification is significant at level 10%(5%, 1%).
3. The value in Fixed effect test is f-statistics and *(**, ***) means that 'H0 : there's no fixed effect' is
rejected at significant level 10%(5%, 1%).
4. The value in Wald test is -statistics and *(**, ***) means that 'H0 : the effect of irregular = the effect
of regular' is rejected at significant level 10%(5%, 1%).

Table 5 represent fixed effect on BMI by cohorts, esp. age and sex effect. Net age effect can be
estimated by investigating difference of within-group in table 9. Fixed effect of age on BMI looks
non-linear as age increase, esp. invert U-shaped. Sex effect on BMI can be estimated by investigate
between sex within group in table 6. Although BMI of men is bigger than that of women generally,
17

BMI of men is lower than that of women among cohorts who are above 50s and below 3rd income
percentile.


18

Table 5. Fixed effect on BMI by cohorts: age and sex effect
Group
no.
Men Women
sex effect
(A-B)
Cohort no. income age BMI (A) Cohort no. income BMI (B)
Group
1
111 57.5 24.4 -0.499 121 63.6 -1.355 0.856
112 63.3 34.8 0.715 122 64.8 -0.268 0.983
113 58.5 44.8 0.172 123 58.4 -0.260 0.432
114 53.1 55.4 -0.288 124 52.6 -0.190 -0.098
115 42.5 70.0 -1.685 125 36.1 -1.162 -0.522
Group
2
211 129.1 24.5 -0.168 221 138.3 -1.006 0.838
212 143.6 34.8 1.119 222 146.3 -0.132 1.251
213 144.4 44.2 0.822 223 136.7 0.119 0.703
214 125.2 54.6 -0.208 224 117.2 -0.017 -0.192
215 109.8 67.3 -0.850 225 109.0 -0.542 -0.309
Group
3
311 212.0 24.8 0.241 321 219.7 -1.252 1.493
312 223.0 34.7 1.311 322 231.5 -0.486 1.797
313 227.9 44.3 1.093 323 225.6 0.118 0.976
314 209.6 54.3 0.130 324 203.7 0.331 -0.201
315 202.8 66.4 -0.385 325 208.0 -0.208 -0.177
Group
4
411 399.0 25.2 0.775 421 432.3 -1.418 2.194
412 380.6 34.8 1.959 422 417.1 -0.431 2.391
413 421.8 44.3 2.053 423 443.7 0.231 1.822
414 440.7 54.1 1.224 424 395.8 0.169 1.056
415 407.7 66.9 0.300 425 412.2 -0.072 0.372

Table 6 represent fixed effect on BMI by cohorts, esp. income effect. Generally the BMI of men
increase as income increase. But the effect of income on women's BMI is different by age level.
Although BMI of cohorts who are below 30s decrease as income increase, BMI of cohorts who are
above 40s increase as income increase like BMI of men.


19

Table 6. Fixed effect on BMI by cohorts: income effect
Group
no.
Men Women
Cohort no. income age BMI Cohort no. income age BMI
Group
5
111 57.5 24.4 -0.499 121 63.6 23.9 -1.355
211 129.1 24.5 -0.168 221 138.3 24.8 -1.006
311 212.0 24.8 0.241 321 219.7 24.9 -1.252
411 399.0 25.2 0.775 421 432.3 24.6 -1.418
Group
6
112 63.3 34.8 0.715 122 64.8 34.9 -0.268
212 143.6 34.8 1.119 222 146.3 34.5 -0.132
312 223.0 34.7 1.311 322 231.5 34.5 -0.486
412 380.6 34.8 1.959 422 417.1 34.8 -0.431
Group
7
113 58.5 44.8 0.172 123 58.4 44.6 -0.260
213 144.4 44.2 0.822 223 136.7 44.4 0.119
313 227.9 44.3 1.093 323 225.6 44.0 0.118
413 421.8 44.3 2.053 423 443.7 44.1 0.231
Group
8
114 53.1 55.4 -0.288 124 52.6 55.2 -0.190
214 125.2 54.6 -0.208 224 117.2 54.7 -0.017
314 209.6 54.3 0.130 324 203.7 54.2 0.331
414 440.7 54.1 1.224 424 395.8 53.7 0.169
Group
9
115 42.5 70.0 -1.685 125 36.1 70.2 -1.162
215 109.8 67.3 -0.850 225 109.0 68.3 -0.542
315 202.8 66.4 -0.385 325 208.0 68.3 -0.208
415 407.7 66.9 0.300 425 412.2 68.8 -0.072


. Conclusions and Implication

This study investigate the effect of SE and SD variables like income, occupation and its status,
education, age etc. on BMI using pooled regression and pseudo panel approach. And the difference of
effect on BMI between employees and others and between regular and irregular employees is
20

investigated by analyzing the effect of occupation status on BMI to analyze the effect of employment
instability on obesity.
In pooled regression model regular and irregular employee, income, highly educated, smoking have
negative effect on BMI, but professional, office, service, family size, and eating out have positive
effect on BMI. Age and BMI have non-linear relationship. Against expectation the difference of
effect on BMI between regular and irregular are not significant.
In pseudo panel model BMI tend to increase. And there exists inconsistence between pooled
regression and pseudo panel approach. In terms of occupation professional and agriculture affect BMI
negatively. The effect of office and service is not significant. This result is inconsistent with the result
of pooled regression in which professional, office, and service have negative effect on BMI.
Household income doesn't affect BMI but household size has negative effect on BMI. Although in
pooled regression income has a negative effect on BMI and size has a positive effect on positive effect
on BMI, in pseudo model the sign of effect on BMI of size is opposite. Like the result of pooled
regression smoking has negative effect on BMI but snack and eating out do not have any significant
effect on BMI and obesity ratio.
The fixed effect on BMI is estimated from pseudo panel model. Fixed effect of age on BMI looks
non-linear as age increase, esp. invert U-shaped. In terms of fixed effect of sex BMI of men is bigger
than that of women generally but BMI of men is lower than that of women among cohorts who are
above 50s and below 3rd income percentile. Although the BMI of men increase as income increase,
the effect of income on women's BMI is different by age level: BMI of cohorts who are below 30s
decrease as income increase but BMI of cohorts who are above 40s increase as income increase like
BMI of men.
The implication of this study is like that:
First, after controlling the effect of major variables, BMI of people tend to increase over time.
Considering that socioeconomic variables like income, occupation status etc. have significant effect
on BMI this trend can become permanent. Because recently inequality of income is deepening and
employment instability is increasing. Thus health problem related to the obesity should be considered
in terms of social structure.
21

Second, the negative effect of highly education on BMI is represented strongly. This result means
that education for lower educated people (tertiary education) may have good effect for health of those
people.


22

Reference

Bae, M. K., Lee, W. K., Song, C. H., Lee, K. M., and Jung, S. P. (1999), "The Factors Associated
with Body Mass Index of Adults", J. Korean Acad Fam Med, 20(7): 906-916.
Baltagi, Badi H. (2008), Econometric Analysis of Panel Data, 2008, 4th Edition, John Wiley & Sons,
Ltd.
Chung, H. R. (2006), "Prevalence of Abdominal Obesity and Associated Factors among Korean
Adults: The 2001 Korean National Health and Nutrition Examination Survey", Korean J. Nutr,
39(7): 684-691.
Hwang, J. Y., Ru, S. Y., Ryu, H. K., Park, H. J., and Kim, W. Y. (2009), "Socioeconomic Factors
Relating to Obesity and Inadequate Nutrient Intake in Women in Low Income Families
Residing in Seoul", Korean J Nutr, 42(2): 171-182.
Jee, S. H., Pastor-Barriuso, R., Appel, L. J., Suh, I., Miller, E. R., and Guallar, E. (2005), "Body Mass
Index and Incident Ischemic heart Disease in South Korean Men and Women", American
Journal of Epidemiology, 162(1): 42-48.
Jee, S. H., Sull, J. W., Park, J., Lee, S., Ohrr, H., Guallar, E., et al. (2006), "Body Mass Index and
Mortality in Korean Men and Women", The New England Journal of Medicine, 355(8): 779-
787.
Jeong, B. G., Moon, O. R., Kim, N. S., Kang, J. H., Yoon, T. H., Lee, S. Y., et al. (2002),
"Socioeconomic Costs of Obesity for Korean Adults", Korean J. Prev Med, 35(1): 1-12.
Jung, C. H., Park, J. S., Lee, W. Y., and Kim, S. W. (2002), "Effects of Smoking, Alcohol, Exercise,
Level of Education, and Family History on the Metabolic Syndrome in Korean Adults", The
Korean Journal of internal Medicine, 63(6): 649-659.
Kang, J. H. (2006), "Stress, Drinking and Obesity", Monthly Korea J., July, 2006: 168-171.
Kim, S. H., Kim, J. Y., Ryu, K. A., and Sohn, C. M. (2007), "Evaluation of the Dietary Diversity and
Nutrient Intakes in Obese Adults", Korean J. Community Nutr., 12(5): 583-591.
Kim, Y. H., Ou, S. W., Kim, Y. S., Chun, J. H., Yang, J., Yoon, Y. S., et al. (2005), "The Factors
Affecting the Fat Distribution in the Abdomen of Obese Women", Korean J. Obes, 14(1): 39-
46.
23

Kim, Y., & Kim, M. (2007), "Health Inequalities in Korea: Current Conditions and Implications", J.
Prev Med Public Health, 40(6): 431-438.
Nam, J. R. (2005), "Research on the Employment Instability and Its Causes", Korean J. of Labour
Economics, 28(3): 111-139.
Paek, K. W., Chun, K. H., Jin, K. N., and Lee, K. S. (2006), "Do Health Behaviors Moderate the
Effect of Socioeconomic Status on Metabolic Syndrome?", Annals of epidemiology, 16(10):
756-762.
Paek, K., and Hong, Y. (2006), "Health Behavior Factors Affecting Waist Circumference as an
Indicator of Abdominal Obesity", J. Prev Med Public Health, 39(1): 59-66.
Park, H., and Kim, P. (2002), "Lifestyle Factors Associated with Visceral Fat Accumulation by CT
Scan in Korean Obese Adults", Korean Journal of Obesity, 11(4): 337-348.
Park, H. S., Yun, Y. S., Park, J. Y., Kim, Y. S., and Choi, J. M. (2003), "Obesity, Abdominal Obesity,
and Clustering of Cardiovascular Risk Factors in South Korea", Asia Pacific Journal of
Clinical Nutrition, 12(4): 411-418.
Shin, M. H., Yoon, M. O., Nam, S. J., and Song, Y. M. (2007), "Relationship between the Source of
Energy Intake and Obesity in Korean Women Using the Average of Four 3-day Dietary
Records", J. Prev Med Public Health, 40(1): 45-50.
Yoon, Y. S., Oh, S. W., Baik, H. W., Park, H. S., and Kim, W. Y. (2004), "Alcohol Consumption and
the Metabolic Syndrome in Korean Adults: the 1998 Korean National Health and Nutrition
Examination Survey", Am. J. Clin Nutr, 80(1): 217-224.
Yoon, Y. S., Oh, S. W., & Park, H. S. (2006), "Socioeconomic Status in Relation to Obesity and
Abdominal Obesity in Korean Adults: a Focus on Sex Differences", Obesity, 14(5): 909-919.


24

Appendix 1. Average of variables by cohorts in pseudo panel regression
Cohort
no.
Cohort criteria
Dependent
variables
Occupation
Income Age Sex BMI Obe. ratio
Profess-
ional
Office Service
Agricul
-ture
Manual
labor
111 57.5 24.4 men 23.0 4.5 3.7 1.8 8.5 2.3 19.5
112 63.3 34.8 men 23.8 4.5 11.9 3.7 12.8 8.0 34.3
113 58.5 44.8 men 23.6 2.5 5.7 1.2 13.3 19.6 28.3
114 53.1 55.4 men 23.6 2.3 2.5 0.2 7.4 34.5 20.3
115 42.5 70.0 men 22.6 0.7 0.6 0.2 3.1 31.1 9.6
121 63.6 23.9 women 21.8 4.2 8.6 9.7 8.7 0.2 1.8
122 64.8 34.9 women 22.9 4.9 8.5 2.3 18.0 5.3 18.6
123 58.4 44.6 women 24.0 4.9 1.3 0.9 18.2 13.9 25.6
124 52.6 55.2 women 24.5 4.5 1.2 0.1 9.3 22.2 19.0
125 36.1 70.2 women 24.0 3.8 0.1 0.0 3.1 20.1 8.8
211 129.1 24.5 men 23.0 4.0 6.6 9.2 10.3 0.4 24.7
212 143.6 34.8 men 24.1 3.0 12.4 15.0 20.9 2.9 41.2
213 144.4 44.2 men 24.1 3.6 9.5 6.4 21.0 13.4 42.0
214 125.2 54.6 men 23.9 2.3 2.9 1.7 12.9 21.9 44.1
215 109.8 67.3 men 23.1 1.4 2.2 1.1 4.8 30.5 17.0
221 138.3 24.8 women 21.9 3.8 11.3 13.3 10.6 0.5 2.8
222 146.3 34.5 women 23.1 3.3 4.6 5.5 15.7 2.5 10.1
223 136.7 44.4 women 24.1 4.8 2.5 1.9 23.9 9.6 20.9
224 117.2 54.7 women 24.6 5.3 0.7 0.7 14.6 18.2 19.0
225 109.0 68.3 women 24.3 4.2 0.2 0.1 5.1 19.2 9.1
311 212.0 24.8 men 23.4 5.3 9.7 8.6 13.6 0.5 29.0
312 223.0 34.7 men 24.0 4.0 19.9 21.7 19.1 1.4 34.5
313 227.9 44.3 men 24.3 3.5 15.3 14.0 19.5 5.9 42.2
314 209.6 54.3 men 24.0 2.2 8.8 5.9 16.3 15.2 39.2
315 202.8 66.4 men 23.2 1.0 3.6 3.4 7.2 24.1 14.8
321 219.7 24.9 women 21.6 2.0 12.5 15.6 8.6 0.1 3.2
322 231.5 34.5 women 22.6 2.1 5.8 5.9 16.6 1.2 10.4
323 225.6 44.0 women 23.7 3.3 6.6 4.0 25.3 6.5 16.4
324 203.7 54.2 women 24.7 5.2 3.0 1.1 16.2 13.8 17.1
325 208.0 68.3 women 24.3 5.0 1.3 0.2 6.3 13.1 5.2
411 399.0 25.2 men 23.6 5.5 17.7 11.7 15.0 0.7 17.7
412 380.6 34.8 men 24.2 4.1 33.1 22.0 18.8 1.5 20.6
413 421.8 44.3 men 24.5 2.7 29.8 19.6 19.6 4.7 23.9
414 440.7 54.1 men 24.3 1.1 26.1 11.6 16.5 10.0 29.2
415 407.7 66.9 men 23.8 0.6 10.8 2.8 10.1 14.1 14.2
25

421 432.3 24.6 women 21.2 1.0 22.5 22.7 11.0 0.5 2.9
422 417.1 34.8 women 22.1 1.9 19.6 11.7 14.8 1.1 3.9
423 443.7 44.1 women 23.5 2.6 12.3 7.0 20.3 4.4 9.9
424 395.8 53.7 women 24.2 2.9 5.1 1.6 17.3 9.7 8.8
425 412.2 68.8 women 24.4 5.3 1.0 0.0 8.2 7.2 2.9
Whole
cohort
204.1 45.3 - 23.5 3.4 9.0 6.6 13.6 10.3 19.1
F-stat. 15.9*** 4462.0*** - 17.6*** 2.32*** 16.9*** 27.4*** 7.61*** 13.5*** 31.2***
Note:
*(**, ***) means that the difference of mean of variable by cohorts is significant at level 10%(5%, 1%).

Cohort
no.
Occupation status
House
-hold
size
Personal socio-demographic and life style characteristics
Education
Spouse
Smok
-ing
Drinking Snack Eating out
Irregu
-lar
Regular
High
school
College
111 10.7 15.1 3.2 65.1 31.1 10.6 53.9 67.3 56.1 44.0
112 18.2 24.2 3.7 59.0 26.2 51.8 60.9 74.3 52.6 44.0
113 10.0 13.3 3.8 39.0 14.3 71.2 55.4 70.0 45.1 27.0
114 11.0 5.4 3.0 19.8 6.1 81.1 54.5 63.3 47.2 17.5
115 4.6 3.4 2.5 13.6 3.8 91.1 47.1 46.4 47.7 6.9
121 7.9 18.8 3.7 61.2 34.8 27.0 13.8 40.1 65.2 38.5
122 13.3 18.9 4.2 56.5 22.1 73.6 23.2 28.7 66.2 14.3
123 21.1 9.2 3.4 33.3 4.2 66.9 18.1 29.8 57.3 15.7
124 13.5 8.5 2.7 7.4 1.5 72.0 18.3 20.1 50.2 7.7
125 6.6 1.7 2.0 2.3 0.3 45.3 24.8 16.4 44.5 5.4
211 14.4 32.2 3.5 57.4 38.5 20.0 57.9 69.4 64.8 51.6
212 10.2 53.8 4.0 56.5 36.5 82.6 55.3 72.6 59.0 45.6
213 13.4 29.4 4.1 48.9 16.8 91.8 57.0 66.3 55.6 43.8
214 18.1 17.4 3.3 26.9 5.7 90.3 53.8 58.5 58.8 26.2
215 8.5 7.6 2.8 21.6 8.7 90.8 49.6 46.7 53.9 13.4
221 10.4 23.9 3.9 57.8 37.4 42.6 20.6 40.6 73.0 34.1
222 12.1 12.8 4.1 62.5 21.5 90.1 18.6 34.5 69.1 12.2
223 16.9 14.1 3.8 40.9 6.3 85.4 19.6 29.2 66.9 18.9
224 12.2 9.3 3.0 12.9 1.4 76.7 17.8 19.8 60.1 9.7
225 6.0 3.5 2.9 6.1 0.8 52.3 21.0 14.3 53.1 7.0
311 13.2 39.5 3.4 57.0 40.4 24.9 53.0 70.5 63.2 52.0
312 6.7 67.0 3.7 42.6 52.5 86.3 56.2 72.8 61.4 54.4
313 7.0 46.5 3.9 47.5 29.8 92.7 53.4 67.1 60.3 44.3
314 11.5 26.8 3.4 37.2 10.5 95.1 46.5 52.9 56.1 34.2
315 4.6 8.7 3.4 30.0 17.9 91.8 40.4 47.6 58.1 15.3
26

321 7.6 28.5 3.7 48.3 50.0 43.6 20.2 36.2 76.0 39.2
322 9.3 17.5 4.0 58.2 33.8 93.8 17.6 32.2 72.8 15.3
323 13.5 18.4 3.8 48.6 14.2 93.6 16.7 28.9 70.2 14.1
324 8.5 13.5 3.3 20.6 5.3 84.2 18.1 23.3 68.7 12.2
325 6.3 1.1 3.6 10.7 3.7 50.5 23.1 14.6 53.9 7.0
411 8.3 45.9 3.4 41.9 57.4 20.6 51.3 65.4 67.2 57.0
412 2.5 67.0 3.4 30.4 68.0 84.5 53.0 70.1 64.1 59.5
413 3.0 53.4 3.7 36.1 52.4 96.2 51.6 63.2 58.3 49.9
414 5.5 44.0 3.3 35.8 34.5 96.6 46.0 58.0 61.9 43.4
415 4.2 12.0 3.3 28.0 29.1 88.7 39.5 42.1 66.5 21.8
421 9.8 43.8 3.6 39.9 59.1 35.3 17.1 39.3 74.1 42.8
422 5.9 28.4 3.9 38.8 57.6 94.0 14.5 31.8 76.3 20.1
423 6.1 19.6 3.6 47.9 32.4 96.1 17.0 26.5 71.7 20.8
424 5.0 11.6 3.3 34.6 13.5 86.2 17.2 19.1 73.9 16.1
425 2.0 1.8 3.7 14.4 7.2 48.6 20.6 14.2 64.7 8.8
Whole
cohort
9.5 22.9 3.5 37.4 24.7 70.4 35.4 44.6 61.6 27.8
F-stat. 4.88*** 49.2*** 22.6*** 30.6*** 34.8*** 56.2*** 2.73*** 3.89*** 1.15 11.5***
Note: *(**, ***) means that the difference of mean of variable by cohorts is significant at level 10%(5%, 1%).


27

Appendix 2. Average of variables by years in pseudo panel regression
Year
Cohort Criteria
Dependent
variables
Occupation
Income Age
Sex
ratio
BMI
Obe.
ratio
Profess
-ional
Office Service
Agricul
-ture
Manual
labor
1998 128.1 45.2 50.0 23.2 2.6 5.7 6.8 14.1 16.4 18.1
2001 174.0 45.2 50.0 23.5 3.2 7.8 6.7 15.8 7.3 18.7
2005 222.3 45.1 50.0 23.7 3.4 8.8 7.6 14.5 8.4 20.0
2007 230.0 45.5 50.0 23.7 3.8 12.0 6.4 11.1 7.6 17.9
2008 266.2 45.4 50.0 23.6 3.9 10.9 5.6 12.4 11.8 20.7
Whole
Periods
204.1 45.3 50.0 23.5 3.4 9.0 6.6 13.6 10.3 19.1
F-stat. 5.44*** 0.01 0.00 1.46 1.84 2.97** 0.40 2.76** 5.49*** 0.34
Year
Occupation
status
Household
size
Personal socio-demographic and life style characteristics
Education
Spouse Smoking Drinking Snack
Eating
out
Irregu
-lar
Regu
-lar
High
school
College
1998 7.7 20.7 3.6 33.9 19.5 74.3 35.9 48.0 70.1 18.5
2001 6.2 25.1 3.5 35.0 27.2 73.7 31.5 46.5 32.1 27.4
2005 11.5 23.6 3.3 40.6 25.1 68.7 26.0 76.9 63.6 38.5
2007 11.7 22.9 3.5 38.5 26.5 68.1 26.1 25.1 74.4 33.8
2008 10.4 22.5 3.4 39.2 25.1 67.3 57.3 26.5 67.9 20.6
Whole
Periods
9.5 22.9 3.5 37.4 24.7 70.4 35.4 44.6 61.6 27.8
F-stat. 7.18*** 0.31 3.40** 0.95 0.90 0.60 11.11*** 37.52*** 107.8*** 8.69***
Note *(**, ***) means that the difference of mean of variable by cohorts is significant at level 10%(5%, 1%).


28

Appendix 3. Number of samples by cohort and year
1998(7,102) 2001(5,167) 2005(4,720) 2007(2,771) 2008(5,633)
Cohort
no.
number
of
samples
Cohort no.
number
of
samples
Cohort no.
number
of
samples
Cohort no.
number
of
samples
Cohort no.
number
of
samples
111 107 111 60 111 27 111 23 111 40
112 77 112 67 112 42 112 15 112 27
113 97 113 56 113 49 113 16 113 33
114 108 114 67 114 67 114 28 114 43
115 276 115 208 115 224 115 162 115 300
121 89 121 61 121 33 121 16 121 41
122 97 122 86 122 63 122 20 122 45
123 116 123 105 123 85 123 26 123 46
124 170 124 109 124 111 124 50 124 99
125 441 125 347 125 362 125 250 125 539
211 112 211 95 211 59 211 40 211 77
212 185 212 154 212 119 212 52 212 87
213 154 213 130 213 117 213 56 213 82
214 122 214 81 214 71 214 58 214 106
215 158 215 96 215 120 215 100 215 214
221 165 221 128 221 94 221 48 221 138
222 216 222 234 222 189 222 81 222 178
223 184 223 137 223 133 223 75 223 159
224 162 224 98 224 108 224 84 224 172
225 193 225 144 225 141 225 127 225 266
311 185 311 103 311 71 311 31 311 65
312 287 312 165 312 110 312 91 312 141
313 224 313 163 313 154 313 74 313 149
314 144 314 71 314 107 314 40 314 121
315 108 315 71 315 80 315 59 315 98
321 265 321 142 321 129 321 62 321 107
322 330 322 208 322 198 322 146 322 280
323 248 323 169 323 208 323 88 323 227
324 144 324 89 324 111 324 67 324 151
325 126 325 80 325 92 325 56 325 122
411 157 411 115 411 65 411 38 411 62
412 275 412 169 412 147 412 78 412 162
413 189 413 176 413 168 413 92 413 158
414 131 414 110 414 104 414 63 414 118
415 65 415 57 415 56 415 44 415 91
421 225 421 182 421 106 421 59 421 138
422 324 422 245 422 215 422 123 422 243
423 213 423 217 423 226 423 109 423 245
424 142 424 99 424 99 424 80 424 146
425 91 425 73 425 60 425 44 425 117
mean 177.6 mean 129.2 mean 118.0 mean 69.3 mean 140.8
S.D 80.5 S.D 63.2 S.D 66.3 S.D 45.8 S.D 96.3
max 441 max 347 max 362 max 250 max 539
min 65 min 56 min 27 min 15 min 27
number of
cohorts
below
100
2

6
number of
cohorts
below
100
2

17
number of
cohorts
below
100
2

17
number of
cohorts
below
100
2

33
number of
cohorts
below
100
2

15
Note 1. Value in parenthesis is total number of sample by year.
2. Number of cohorts whose samples' number is below 100

You might also like