Available via license: CC BY 4.0
Content may be subject to copyright.
The Effect of Health-Related Behaviors on Income
in China: A Panel Data Analysis
Muyan Xie ( 1756314803@qq.com )
N/A
Research Article
Keywords: Income factors, health status, health-related activities, gender
Posted Date: September 28th, 2021
DOI: https://doi.org/10.21203/rs.3.rs-936625/v1
License: This work is licensed under a Creative Commons Attribution 4.0 International License.
Read Full License
The Effect of Health-Related Behaviors on Income in China:
A Panel Data Analysis
Muyan Xie
Abstract
Understanding the interaction between habits and income can be an efficient way
to economically improve income and this paper serves to figure out the effectiveness
of health-related income factors and gender differences in these factors using regres-
sions, interaction effects and fixed effect model estimations. I used China Health and
Nutrition Survey (CHNS) for 2000, 2004, 2006, 2009, 2011 and 2015 and expected
the study to be representative of the general population in China.
Results show clear effect of these factors, with that smoking can negatively asso-
ciate with income and tea, coffee, alcohol, and exercising be in positive association.
Also, gender differences are tested and interpreted. Extensions include discuss path-
way through which these factors influence income, urban residence’s interaction effect
with these factors and yearly differences.
Main findings are concluded for future possible uses, as policy makers for women’s
economical empowerment programs may refer to the statistics and find the area that
return the most such as drinking tea and quitting smoking.
Keywords: Income factors, health status, health-related activities, gender
Contents
1 Introduction 2
2 Literature Review 3
3 Research Questions 4
4 Methodology 5
4.1 Data.................................................... 5
4.2 Measures ................................................. 8
4.3 Methods.................................................. 10
4.3.1 OLS Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
4.3.2 Fixed Effects Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
4.3.3 Interaction Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
5 Results 11
6 Extension 17
7 Conclusion 23
7.1 Advantages and Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
7.2 Main Findings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
8 Acknowledgement 27
1
1 Introduction
Throughout history there is evidence for income inequalities for females.1Some people
may count the difference on the fact that women tend to work in different industries and
have less experience than men, while others may sum that the difference might be caused by
discrimination against women and fewer hours working at work.2 3
Also, what is worth noticing is that thorough researches have been conducted on factors
that might influence income, including education and health-related activities.4 5 Naturally,
people with living habits like refusing to smoke are found to earn more as a social award from
the society.6In addition to catering to social expectation, people with better habits can create
themselves a stronger body and thus the superior health status may lead them to be better
qualified to work more efficiently for longer time.
Quantitatively analyse the extend to which some factors like health, schooling and habits
can have on influencing income is useful. Also, intuitively, inspired by concluded sex differences
in immune responses, which lead females to respond more violently to several types of illness, I
designed to investigate deeper into different effectiveness of income factors across gender, trying
to figure out the sex’s degree of influence on income when we divide the determinant of income
into several smaller categories.7It is of great value if I can figure out if there is differences
across genders in the earning effects of improvement of habits, education or occupations, if
we are interested in economical empowerment for women and are trying to find a way to
most efficiently address the problem by investing in factors that are most crucial in influencing
income.8
Thus, this paper focus on analysing effect of each income factor both for the general working
population and each gender. I hope this paper can help to reveal gender income difference in
its most fundamental ways.
1Gender and Income Inequality. International Monetary Fund. Accessed August 12, 2021.
https ://www.imf .org/external/pubs/f t/sdn/2015/sdn1520inf o.pdf
2Tyson, Laura D’Andrea, and Ceri Parker. “An Economist Explains Why Women Are Paid Less.” World Economic Forum, March
2019, accessed August 12, 2021.
https ://www.weforum.org/agenda/2019/03/an −economist −explains −why −women −get −paid −less/
3Bleiweis, Robin. “Quick Facts about the Gender Wage Gap.” Center for American Progress, 2019, accessed August 12, 2021.
https ://www.americanprogress.org/issues/women/reports/2020/03/24/482141/quick −f acts −gender −wage −gap/
4Schultz, T.Paul. “Human Capital, Schooling and Health.” Economics & Human Biology 1, no. 2 (June 2003): 207–21, accessed
August 13, 2021. https ://doi.org /10.1016/s1570 −677x(03)00035 −2.
5Xiao, Yuxi, Haizheng Li, and Belton M. Fleisher. “The Earnings Effects of Health and Health-Related Activities: A Panel Data
Approach.” Applied Economics 47, no. 14 (2015): 1407–23, accessed August 13, 2021.
https ://doi.org/10.1080/00036846.2014.1000521
6Fox, Maggie. “Smokers Less Likely to Get Hired and Earn Less: Study.” NBCNews.com. NBC Universal News Group, April 12,
2016, accessed August 13, 2021.
https ://www.nbcnews.com/health/health −news/smokers −less −likely −get −hired −earn −less −study −n554321.
7Takahashi, Takehiro, and Akiko Iwasaki. “Sex Differences in Immune Responses.” Science 371, no. 6527 (2021): 347–48, accessed
August 13, 2021. https ://doi.org /10.1126/science.abe7199.
8Rubery, Jill, and Aristea Koukiadaki. “Closing the Gender Pay Gap: A Review of the Issues, Policy” International Labor Office,
2016, accessed August 13, 2021.
https ://www.ilo.org/wcmsp5/groups/public/ −−−dgreports/ −−−gender/documents/publication/wcms540889.pdf
2
2 Literature Review
I found that previous studies tried to correlate income and health-related activities. For
instance, previous study has shown that smoking and drinking are correlated with income
level.9 10 Interestingly, several prior studies have investigated the correlation between drinking
and income with no agreed conclusions and experts are still aruging about the effect of alcohol
drinking on income in psychological and empirical aspects.11 12 Whatever, currently, no paper
directly shows the correlation between coffee or tea drinking and income level. Studies, however,
show positive effects of coffee and tea have on productivity, an indicator of income level.13 14
For exercise, studies show a positive correlation between it and health 15 Moreover, good health
status has been established to be able to positively influence a person’s income. A longitudinal
data set starting from 1750, focusing on generations in US, showing that casual effects exist
between health situation and yearly income.16
Next, I focused on education and years of experiences, which have long been argued to be
in a casual relationship with income. In the famous Mincer Model, education and experiences
are found to be determinants of income.17 However, the Mincer Model only did the statistical
proof that these variables are of casual relationships. It was pointed out that the model did not
clarify the channels through which education and experiences might affect income in a certain
direction.18 Fortunately, previous research has shown that the model can be improved by
counting the effect of education on health improvement within schooling.19 This was achieved
9M. Christopher Auld, “Smoking, Drinking, and Income,” Journal of Human Resources XL, no. 2 (2005): pp. 505-518, accessed
August 14, 2021. https ://doi.org /10.3368/jhr.xl.2.505.
10 In ”Smoking, Drinking, and Income”, Christopher Auld pointed out that drinking is associated with 10-12 percent higher
income than drinking abstention, while smokers are associated with 24 percent lower income than non-smokers after correcting for
endogeneity
11 Jae-Hyung Lee, “The Influence of Alcohol Consumption on Income and Health: Empirical Evidence from a Panel of OECD
Countries,” Seoul Journal of Economics 26, no. 2 (2013). This article use panel data from 2001 to 2005 to do regression and find
statistically significant results that states moderate drinking is positively correlated with income level.
12 Magdalena Cerd´a, Vicki D. Johnson-Lawrence, and Sandro Galea, “Lifetime Income Patterns and Alcohol Consumption: Inves-
tigating the Association between Long- and Short-Term Income Trajectories and Drinking,” Social Science & Medicine 73, no. 8
(2011): pp. 1178-1185, https ://doi.org/10.1016/j.socscimed.2011.07.025.This article suggests that alcohol drinking is correlated
with lower income levels, as opposed to the results in Lee’s paper.
13 Tom M. McLellan, John A. Caldwell, and Harris R. Lieberman, “A Review of Caffeine’s Effects on Cognitive, Physical and
Occupational Performance,” Neuroscience & Biobehavioral Reviews 71 (2016): pp. 294-312, accessed August 15, 2021.
https ://doi.org/10.1016/j.neubiorev.2016.09.001.
14 Farrukh Afaq et al., “Health Benefits of Tea Consumption,” Beverages in Nutrition and Health, 2004, pp. 143-156, accessed
August 15, 2021. https ://doi.org /10.1007/978 −1−59259 −415 −310.
15 Jaana T. Kari et al., “Income and Physical Activity AMONG Adults: Evidence from Self-Reported AND Pedometer-Based
Physical Activity Measurements,” PLOS ONE 10, no. 8 (2015), accessed August 16, 2021.
https ://doi.org/10.1371/j ournal.pone.0135651.
16 Dora L. Costa, “Health and the Economy in the United States from 1750 to the Present,” Journal of Economic Literature 53,
no. 3 (January 2015): pp. 503-570, accessed August 17, 2021. https ://doi.org/10.1257/j el.53.3.503.
17 Jacob Mincer, Schooling, Experience, and Earnings (New York, New York: National Bureau of Economic Research, 1974).
18 Hall, Robert E. Journal of Political Economy 83, no. 2 (1975): 444-46. Accessed August 15, 2021.
http ://www.j stor.org/stable/1830936.
19 Damon Clark and Heather Royer, “The Effect of Education on Adult Mortality and Health: Evidence from Britain,” American
Economic Review 103, no. 6 (January 2013): pp. 2087-2120, accessed August 16, 2021. https ://doi.org/10.1257/aer.103.6.2087.
3
in 2003 by Schultz, who conducted quantified the economic return to schooling in terms of
human capital.20
When it came to the situation where I am examining gender differences in the income factors
stated above, I can find several relevant papers concerning gender and productivity. Education
is shown to be more beneficial to women than to men in terms of income growth by the channel
of improving one’s health.21 Biologically, health-related habits can impose a deeper influence
on females. Smoking is shown to be more detrimental to women because of the way in which
nicotine interact with sex hormones.22 23 Similarly, metabolism mechanism determines that
females are more vulnerable to the effect of heavily drinking alcohol and benefit more from
exercise.24 However, none of the papers about gender and habits uses longitudinal empirical
data to support their conclusion. Instead, they did biological experiments only. It is of high
value if we can use observed panel data to figure out the gender differences in several income
factors quantitatively.
To sum up, there is no perfect fit to the goal of correlating health-related activities to income
using unbiased statistics. The paper that almost did that job should be Y. Xiao et al. ’s research
in 2015.25. Still, it fails by using data set from urban families only. Also, in the data choice,
they drop all the observations with missing variables, a redundant step that lower the number
of observations the model could use to generate precise results. Finally, the paper did the OLS
regression for only two years, too short to observe notable patterns.26 For gender differences,
most of the researches are in an angle of biological and physiological differences between the
two genders, lacking the support of empirical evidence. In a nutshell, a new research should be
conducted to address all the above concerns.
3 Research Questions
Upon reviewing the articles, I found that a comprehensive assessment on income factors and
gender differences in these factors is missing. There are two missions for us to accomplish.
20 Guangjie Ning, “Can Educational Expansion Improve Income Inequality? Evidences from the CHNS 1997 and 2006 Data,”
Economic Systems 34, no. 4 (April 2010): pp. 397-412, accessed August 21. 2021.
https ://doi.org/10.1016/j.ecosys.2010.04.001.
21 Catherine E. Ross and John Mirowsky, “Gender and the Health Benefits of Education,” National Institute of Health 51, no. 1
(January 2010): pp. 1-19, accessed August 21, 2021. https ://doi.org/10.1111/j.1533 −8525.2009.01164.x.
22 Allen, Alicia M., Cheryl Oncken, and Dorothy Hatsukami. “Women and Smoking: The Effect of Gender on The Epidemiology,
Health Effects, and Cessation of Smoking.” Current Addiction Reports 1, no. 1 (2014): 53–60, accessed August 22, 2021.
https ://doi.org/10.1007/s40429 −013 −0003 −6.
23 Joni Hersch, “Gender, Income Levels, and the Demand for Cigarettes,” Journal of Risk and Uncertainty 21, no. 2 (2000),
accessed August 21, 2021. https ://doi.org/10.2139/ssrn.247822.
24 Richard Wilsnack, “Are Women More Vulnerable To ALCOHOL’S Effects?-Alcohol Alert No. 46-1999,” National Institute on
Alcohol Abuse and Alcoholism (U.S. Department of Health and Human Services, 2010), accessed August 21, 2021.
https ://pubs.niaaa.nih.gov/publications/aa46.htm.
25 Yuxi Xiao, Haizheng Li, and Belton M. Fleisher, “The Earnings Effects of Health and Health-Related Activities: A Panel Data
Approach,” Applied Economics 47, no. 14 (July 2015): pp. 1407-1423, accessed August 22, 2021.
https ://doi.org/10.1080/00036846.2014.1000521.
26 Jim Frost, “7 Classical Assumptions of Ordinary Least Squares (OLS) Linear Regression,” Statistics By Jim, May 19, 2021,
accessed August 23, 2021. https ://statisticsbyjim.com/regression/ols −linear −regression −assumptions/.
4
First, although paper had shown correlation between health-related activities and income, I
need to avoid the statistical bias in the model and establish causal relations. The can serve to
fully point out the monetary benefits of keeping some healthy habits and provide guidelines for
future social researchers to follow to make a powerful and persuasive advertisement for healthy
lives.
Secondly, previous studies, when came to the situation where females are paid less than
men, tend to assess the conditions in lights of discrimination and social expectations. It should
be useful if I can try to find gender differences in effectiveness of health-related income factors.
Getting the results, policy makers can decide in which, relating to education or alcohol absten-
tion, for example, can they most effectively invest in using their limited resources. Different
policies or incentives might be appropriate or effective for women compared to men.
Thus, considering the concerns, I am going to try to solve the following questions in my
paper:
1. What’s the correlation between health-related activities and income levels?
2. Does the effectiveness of health-related income factors differ by gender?
Much more refined than previous papers in this area, we discuss methods to mitigate biases
in the regression, include missing variables to reduce value of standard error, extend the data
set to 15 years and do interaction terms to explore gender differences. However, there is still
possible omitted variable bias from time-varying omitted factors.
4 Methodology
4.1 Data
We demand that the selected database to have health-related observations and income.
Also, the database need to be panel to come closer to establish casual effects. Thus, I choose to
acquire data from China Health and Nutrition Survey (CHNS). Conducted by UNC Carolina
Population Center China Project, the survey collect data from Mainland China from over 30,000
individuals for over 25 years, focusing on economics, nutrition and demographics features of
the country. To the latest update in 2015, the survey includes education, income, occupations,
health status and health related activities from 7200 households, covering population aging
from 2 to 110, which can be of much use to our present study. Randomly and anonymously
picking respondents, the survey can be reasonably used and be treated as representative of
Chinese households. 27
In table 1, descriptive statistics of the qualified individuals are provided.28 Over the years,
income of Chinese people significantly increases by about 20% to 80% each year.
27 Carolina Population Center University of North Carolina at Chapel Hill, “China Health and Nutrition Survey,” China Health
and Nutrition Survey - China Health and Nutrition Survey (CHNS) (UNC China Project, 2015),
https ://www.cpc.unc.edu/projects/china.
28 ”–” was used to indicate variables that are not observed for that year.
5
Table
1.
Descriptive
Statistics
by
Year
2000 2004
Variable Observation Mean Std. Dev. Min Max Observation Mean Std. Dev. Min Max
Annual Income 7611 8788.993 10003.3 62 5.332 157094.59 5426 10855 .437 1294 8.493 2.581 2 25000
Education and Experience
Schooling(years) 6536 7.962 3.792 0 18 5415 8.592 3.857 0 18
Experience(age - years schooling - 6) 6518 24.781 12.273 0 54 54 05 26.883 11.664 0 52
Current health status(% sample reporting)
Excellent or good 7611 0.57 0.495 0 1 5426 0.664 0.472 0 1
Fair 6300 0.25 0.433 0 1 5426 0.293 0.455 0 1
Poor 6300 0.033 0.179 0 1 5426 0.041 0.198 0 1
Habits(% sample reporting)
Smoking or not 6219 0.332 0.471 0 1 5252 0.333 0.471 0 1
Heavily drinking alcohol 7611 0.151 0.358 0 1 542 6 0 .171 0.377 0 1
Drinking tea 6300 0.406 0.491 0 1 542 6 0 .389 0.487 0 1
Adequately exercising 6300 0.095 0.294 0 1 542 6 0 .087 0.282 0 1
Drinking Coffee 6300 0.014 0.118 0 1 542 6 0 .024 0.152 0 1
Primary Occupation(% sample reporting)
Technical worker 7 583 0.073 0.26 0 1 5421 0.076 0.266 0 1
Manager 7583 0.043 0.203 0 1 542 1 0 .044 0.206 0 1
Office staff 7583 0.044 0.206 0 1 542 1 0 .041 0.198 0 1
Farmer 7583 0 .47 0.499 0 1 5421 0.344 0.475 0 1
Worker 7583 0 .17 0.376 0 1 5421 0.139 0.345 0 1
Other job 7 583 0.144 0.351 0 1 5421 0.137 0.344 0 1
Additional Information(% sample reporting)
Female 7611 0.463 0.499 0 1 542 6 0.4 78 0.5 0 1
Urban residents 6300 0.32 0.466 0 1 5426 0.306 0.461 0 1
2006 2009
Variable Observation Mean Std. Dev. Min Max Observation Mean Std. Dev. Min Max
Annual Income 5098 145 41.746 203 17.184 1.653 367000 5260 21628.179 34164.28 6.9 17 730000
Education and Experience
Schooling(years) 5090 8.933 4.102 0 18 5258 9.014 3.904 0 18
Experience(age - years schooling - 6) 5076 27.497 11.489 0 54 52 43 27.727 11.578 0 54
Current health status(% sample reporting)
Excellent or good 5098 0.675 0.469 0 1 -- -- -- -- --
Fair 5098 0.277 0.448 0 1 -- -- -- -- --
Poor 5098 0.043 0.202 0 1 -- -- -- -- --
Habits(% sample reporting)
Smoking or not 4940 0.329 0.47 0 1 51 24 0 .333 0.471 0 1
Heavily drinking alcohol 5098 0.18 0.385 0 1 5260 0.166 0.372 0 1
Drinking tea 5098 0.366 0.482 0 1 525 9 0 .362 0.481 0 1
Adequately exercising 5098 0.099 0.298 0 1 525 9 0 .098 0.298 0 1
Drinking Coffee 5098 0.024 0.153 0 1 525 9 0 .035 0.184 0 1
Primary Occupation(% sample reporting)
Technical worker 5 098 0.083 0.276 0 1 5260 0.086 0.28 0 1
Manager 5098 0.042 0.2 0 1 5260 0.043 0.203 0 1
Office staff 5098 0.051 0.221 0 1 526 0 0 .051 0.22 0 1
Farmer 5098 0.349 0.477 0 1 52 60 0.344 0.475 0 1
Worker 5098 0.157 0.364 0 1 52 60 0.162 0.368 0 1
Other job 5 098 0.171 0.377 0 1 5260 0.188 0.39 0 1
Additional Information(% sample reporting)
Female 5098 0.472 0.499 0 1 526 0 0.4 58 0.498 0 1
Urban residents 5098 0.315 0.464 0 1 5259 0.31 0.462 0 1
2011 2015
Variable Observation Mean Std. Dev. Min Max Observation Mean Std. Dev. Min Max
Annual Income 6728 273 90.852 33012.501 22.826 714000 6212 4165 7.325 92709.238 11.612 4528302
Education and Experience
Schooling(years) 6710 9.982 3.983 0 18 6205 10.719 3.87 0 18
Experience(age - years schooling - 6) 6696 27.039 11.733 0 54 61 87 2 6.257 11.934 0 54
Current health status(% sample reporting)
Excellent or good -- -- -- -- -- 6212 0.592 0.492 0 1
Fair -- -- -- -- -- 6211 0.361 0.48 0 1
Poor -- -- -- -- -- 6211 0.038 0.19 0 1
Habits(% sample reporting)
Smoking or not 6513 0.321 0.467 0 1 6092 0.27 0.444 0 1
Heavily drinking alcohol 6728 0.168 0.374 0 1 6212 0.121 0.326 0 1
Drinking tea 6726 0.416 0.493 0 1 -- -- -- -- --
Adequately exercising 6726 0.152 0.359 0 1 5696 0.091 0.288 0 1
Drinking Coffee 6726 0.083 0.276 0 1 -- -- -- -- --
Primary Occupation(% sample reporting)
Technical worker 6 726 0.111 0.314 0 1 6211 0.122 0.328 0 1
Manager 6726 0.058 0.234 0 1 621 1 0.05 0.219 0 1
Office staff 6726 0.072 0.258 0 1 6211 0.097 0.296 0 1
Farmer 6726 0 .24 0.427 0 1 6211 0.12 0.325 0 1
Worker 6726 0.168 0.374 0 1 62 11 0 .185 0.388 0 1
Other job 6 726 0.231 0.422 0 1 6211 0.217 0.412 0 1
Additional Information(% sample reporting)
Female 6728 0.466 0.499 0 1 6212 0.453 0.498 0 1
Urban residents 6726 0.391 0.488 0 1 5696 0.389 0.487 0 1
From year 2000 to 2015, years of education increase steadily, rising by about 34%, which can
be attribute to the emphasis to schooling in modern China and the fact that the government
is encouraging additional education of high school or college level during work.29
Health status is self-reported in the database. Through the years, the percentage of popula-
tion reporting themselves as in excellent of good health status is around 60%. The percentage
reporting fair health status varies from 25% to 30%. Very little people report their health as
poor over the years.
Observation rate of smoking decreases from 33% to 27%, probably as awareness of the
negative effects of smoking is spreading across the nation. Also, relative stable percentage,
about 40% and 3%, of the population report drinking tea or coffee.
Also, about 9% of the population report adequately exercising over the years. Percentage of
people who frequently drink alcohol increases at first but decreases then, varying from 15% to
18% and finally to 12%.
For sure, all variable summary reported here are processed first. I dropped income observa-
tions that are negative in the dataset because we are going to use percent change of income in
our future analysis and negative independent variable can pose a problem, which will not statis-
tically affect our results for less than 2% of the income observations in the original database in
negative. Also, I include only individual at working ages to exclude the confounding variables
of degrees of aging and status of being adolescents. For male, I only investigate individuals
with an age between 18 to 60. For females, I only investigate those with an age between 18
and 55.30 Next, the survey provide discrete dataset for schooling, listed in years of elementary
school, years of middle school and years of high school, etc. I calculate the schooling variable
in a continuous form by adding the years from each level of education. Chinese students start
elementary school from the age of 6, so I decide to calculate years of potential work experience
by deducting living years(ages) by years of education and 6. Next, I made dummy variables for
health status and get the percent of observation reporting corresponding health situation in a
chosen year. Smoking, drinking tea and drinking coffee are all in dummy variable form, with
value of 1 indicating ”yes” and value of 0 indicating ”no”.
Dummy variables for heavily drinking alcohol and adequate exercising have been specially
considered. The variable of heavily drinking alcohol is generated by the reported frequency of
drinking alcohol including wine and beer. It is defined as 1, or ”yes”, if the individual is female
and drinking one or more drink for more than five times a week. For male, heavily drinking
is set to the criteria for which the individual take two or more drinks for more than five times
a week.31 For exercise, an observation’s adequately exercise dummy variable is set to be 1, or
”yes”, if the individual is active in any set of sports combination summed in the survey. 32
29 Usually, people at work in China can get working benefits and may increase their education levels by attending state directed
Night Universities, which is credited with increasing the overall years of education by such a large extent in the dataset.
30 OCED Publication, “China: Pension System in 2018,” OCED for China, 2018, accessed August 25, 2021.
https ://www.oecd.org/els/public −pensions/P AG2019 −country −prof ile −China.pdf.
31 Defined under the standard set by United States Centers of Disease Control and Prevention
https ://www.cdc.gov/alcohol/f aqs.htm#moderateDrinking
32 In the CHNS dataset, multiple variables are created to reflect whether the individual is active in any sports. The researchers
7
Overall, rummaging through the survey’s dataset, we can see that occupations are extensively
reported, which may largely influence income. Due to the fact that we may need to use OLS
regression in our model, I decide to include the occupation variables and report them in dummy
variables.33
Holistically, rate of observing beneficial factors like education, experiences and good habits
increases over the years with observations of bad habits decreasing, leading us to have pre-
liminary proposition that health-related activities and health may play a role in determining
income level.34
4.2 Measures
I use 12 variables in all in our models in the next section. Here I will provide details of the
variables.
Independent Variables of Interest
The following variables are the health related activities we are interested in associating with
the income of the individual.
•Smoking: This dummy variable is generated by individual’s response to the question of
whether they smoke or not in the investigation year. Positive answer is included as 1 and
negative response is included as 0.
•Heavily Drinking Alcohol: This dummy variable is generated by self-reported frequency of
drinking alcohol and the standard mentioned in the previous section, Data.
•Drinking Tea: This dummy variable is generated by individual’s response to the question
of whether they drinking tea or not in the investigation year. Positive answer is included
as 1 and negative response is included as 0.
•Adequately Exercising: This dummy variable is generated by self-reported frequency of
doing various kinds of sports and the standard mentioned in the previous section, Data.
•Drinking Coffee: This dummy variable is generated by individual’s response to the question
of whether they drinking coffee or not in the investigation year. Positive answer is included
as 1 and negative response is included as 0.
Independent Control Variables
Some variables will affect income while have some relationship with health related activities.
As explained in the section of next section, I will include them in the OLS regression model.
report the results in combination of sports, commonly dance and gymnastics, field sports and walking, ball games or martial arts
and other unlisted sports.
33 For OLS, set variables’ coefficients will be biased if we exclude variables that have correlation with both the independent variable
and dependent variables. More information at Jim Frost, “Confounding Variables Can Bias Your Results,” Statistics By Jim, April
26, 2021, https ://statisticsbyjim.com/regression/confounding −variables −bias/. and section 4.3, methods
34 As previously explored in relationship between investment on health and income, by Cropper, M. L. “Health, Investment
in Health, and Occupational Choice.” Journal of Political Economy 85, no. 6 (1977): 1273–94, accessed August 26, 2021.
https ://doi.org/10.1086/260637
8
•Health Status: This variable is self-reported. Individual are asked to choose their current
health status during the investigation year to three different levels: Excellent or good,
fair, or poor. To better process the information, I made a dummy variable for each health
status, with 1 being in that status and 0 meaning not.
•Occupation: This variable is self-reported. Individual are asked to choose their current
occupation during the investigation year in six different options: Technical Worker, Man-
ager, Office Staff, Farmer, Worker or Other Job. To better process the information, I made
a dummy variable for each occupation option, with 1 being in that job and 0 meaning not.
•Schooling: This is a self-reported continuous variable indicating years of education the
individual has received up until the investigation year.
•Experience: This is a continuous variable measuring years of experience on the society,
calculated from the age and the years of education(Age−Schooling−6).
•Experience, square of: As argued in the Mincer Model, which studied income factors, it is
better to include the square of years of experience when studying any models in income
factors for the proposition that in standard human capital models, investment in training
at work decreases over time .35 This variable is generated by taking square of the variable
Experience.
•Gender: This dummy variable is included because income differ for gender for various
reasons.. This variable is 1 when the individual is a female while being 0 otherwise.
•Urban Residence: This dummy variable is included because income is different across loca-
tions. This variable is 1 when the individual is an urban resident while being 0 otherwise.
Dependent Variable
•Annual Income, natural log of: This variable is generated by taking the natural log of
the reported real annual income, which used 2006 as a base year. The annual income
was included by adding the following components of income together: wage, insurance,
investment, sales of household goods and governmental aids. In OLS regression models,
taking the natural log of income enables us to interpret the results in yearly percent
changes.
Additional Variables
Checking our variables, we find that some variables such as smoking, has missing observation,
being that the individual refuses to respond to the question. Thus, during the processing, I
combine each dummy variable with a counterpart, being the situation in which the variable’s
status is missing.
For example, for smoking, we have three separate recordings, with Smoking Y es being
the indicator of whether the individual is smoking or not, Smoking no being the 1 when the
individual does not smoke, Smoking M issing being the indicator of whether the individual
responded to the question and generate effective answers.
35 Thomas Lemieux, “The ‘Mincer Equation’ Thirty Years after Schooling, Experience, and Earnings,” Jacob Mincer A Pioneer
of Modern Labor Economics, 2000, pp. 127-145, accessed August 27, 2021. https ://doi.org/10.1007/0−387 −29175 −x11.
9
In comparison to prior research, this method enable us to maintain greater number of ob-
servations, which will help these standard error and make it smaller and obtain a sample size
two times larger than the previous studies.
4.3 Methods
4.3.1 OLS Model
I estimated the model using the Ordinary Least Squares regression, or OLS regression:
Yi=β0+β1Smokingi+β2Alcoholi+β3C of feei+β4Exercisingi+β5T eai+βKXK i +εi(1)
where Yiis the outcome variable, being the natural log of real annual income in this case.
Xni stands for the independent variables, indicating the five health-related activities of interest
like smoking, drinking alcohol, etc, and XKi being the pool of the other control variables, which
includes education, experience, health status and occupation.
I will do OLS regression for six separate times for each year and see the trending of the coef-
ficients in response to research question 1 of the effects of health related activities in earnings.
However, bias problem must be considered. Omitted variable bias occurs when unobserved
factors, such as qualities of planfulness or the degree to which the person is oriented to the fu-
ture, that in fact have influence on income while having correlation with health-related activities
like smoking, are not included in the model.36
Still, methods exist for us to utilize to mitigate the bias. Fixed effect panel data estimation
may help to mitigate this bias, however, differences by gender cannot be analyzed using fixed
effect model because gender is constant.
4.3.2 Fixed Effects Model
The unbalanced database is, fortunately, longitudinal. Thus, I can apply fixed effect models
to the panel database in order to try to lessen the omitted variable bias.37
I used the following equation to write the fixed effect model:
Yit =β1Smokingi+β2Alcoholi+β3C of feei+β4Exercisingi+β5T eai+βKXK i +αi+εit (2)
where Yit is the real income of person i in year t, αibeing the fixed effect of being the target,
which will not vary over the years but is different across people. Xit is independent variables
36 referred from Barreto; Howland (2006). ”Omitted Variable Bias”. Introductory Econometrics: Using Monte Carlo Simulation
with Microsoft Excel. Cambridge University Press. If something in ε, the error term, correlates with the dependent variable, which
in this case is the log of annual income, and one independent variable, then the coefficient of that particular independent variable
in the OLS model will be biased for that the model will count the effect on the omitted variable on that correlated independent
variable
37 Wikipedia Society, “Fixed Effects Model,” Wikipedia (Wikimedia Foundation, June 22, 2021), accessed August 28, 2021.
https ://en.wikipedia.or g/wiki/F ixed ef fects model.
10
at the year tand εit is the idiosyncratic error term
For convenience, I applied entity-demeaned OLS regression. From equation(2), we can
deduce the following equation:
Yit −
1
T
T
X
t=1
Yit =β1 Xit −
1
T
T
X
t=1
Xit!+. . . + εit −
1
T
T
X
t=1
εit!(3)
or simply:
˜
Yit =β1˜
Xit +. . . + ˜uit (4)
where ˜
Yit stands for the difference between the log of real annual income of the year tand
the average of the log of annual income each year in the range of the panel database for the
designated individual i.˜
Xit, on the other hand, means similarly as ˜
Yit. The αicoefficients are
not all individually estimated in this entity demeaned estimation. However, they are controlled
for and they can represent things like a person’s innate ability, patience and furture orientedness.
The only difference is that ˜
Xit measures the independent variables.
Using the Entity-demeaned OLS regression, we are able to eliminate the effect of unobserved
variables that are constant over time, affect the income, and correlate with health oriented
activities. Applying this model, we can improve our regression significantly in evaluation of
question of effectiveness of health-related income factors.
4.3.3 Interaction Effects
In research of the effectiveness of each income factors across genders, I utilized the interaction
effect model in the OLS regression, calculating the interaction term values of the income factors
when they are tested across genders.38 39 This coefficient represents the differences in the effects
of the health-related variables between females and males and I estimate regressions for each
year in the data set.
5 Results
Table 2 reports the OLS cross sectional regression results from year 2000 to year 2015, ana-
lyzing the effect of health-related activities and other independent income factors on earnings.
In regard of our main area of investigation, several health-related activities can have effects
on income. Most notably, tea drinking is suggested to have positive effects on income, varying
from 5% in 2009 to 10.1% in 2000. coefficients on smoking, on the other hand, imply that
smoking has negative effects on income and the results are statistically significant.40 Drinking
38 Wikipedia Society, “Interaction (Statistics),” Wikipedia (Wikimedia Foundation, July 9, 2021), accessed August 29, 2021.
https ://en.wikipedia.org/wiki/I nteraction(statistics).
39 the interaction terms stands for the differences in the extent of the income factors for the two genders, with β1stands for the
unique effect of the variable it represent when the other interacted variable in in β3, or β2is 0 and vice versa
40 β=−0.065, P < 0.1 in 2000, β=−0.117, P < 0.01 in 2004, β=−0.064, P < 0.05 in 2009 for smokers as relative to nonsmokers
11
Table
2:
Regression
Results
by
Year
2000 2004 2006 2 009 2011 2015
VARIABLES Coef. Std. Err. Coef. Std. Err. Coef. Std. Err. Coef. Std. Err. Coef. Std. Err. Coef. Std. Err.
Education and Expe rience
Schooling 0.032 0.005 0.034 0.006 0.045 0.006 0.046 0 .005 0.039 0.005 0.042 0.006
*** *** *** *** *** ***
Experience 0.055 0.005 0.031 0.006 0.027 0 .005 0.037 0.005 0.019 0.004 0.019 0.005
*** *** *** *** *** ***
Experience, square -0.001 0.000 0.000 0.000 0.000 0.000 -0.001 0.000 0.000 0.000 0.000 0.000
*** *** *** *** *** ***
Health Status
Excellent or good 0.283 0.095 0.121 0.082 0.319 0.079 --- --- --- --- 0.492 0.095
*** *** ***
Fair 0.145 0.097 0.004 0.084 0.256 0.081 --- --- --- --- 0.364 0.096
*** ***
Missing(Health Status) 0.325 0.125 0.412 0.371 0.052 0.167 --- --- --- --- 0.071 0.185
***
Habits
Smoking
Yes -0.065 0.035 -0.117 0.040 -0.043 0.039 -0.041 0.0 37 -0.064 0.031 0.038 0.040
* *** **
Missing(Smoking) 0.163 0.176 0.956 0.616 -0.978 0.386 -0.966 0.637 0.1 82 0.081 0.334 0.191
** ** *
Heavily Drinking Alcohol
Yes 0.038 0.035 0.061 0.041 0.088 0.041 0.130 0.040 0.027 0.033 -0.071 0.049
** ***
Missing(Alcohol) 0.022 0.073 -0.123 0.113 0.189 0.123 0.29 4 0.141 0.029 0.096 0.0 18 0.276
**
Tea Drinking
Yes 0.101 0.028 0.074 0.032 0.116 0.030 0.050 0.030 0.092 0.023 --- ---
*** ** *** * ***
Missing(Tea) 0.143 0.199 0.525 0.302 0.547 0.183 0.191 0.29 5 0.403 0.346 --- ---
* ***
Adequately Exercise
Yes 0.043 0.042 -0.012 0.048 0.130 0.039 0.05 0.041 0.10 6 0.028 0.080 0.044
*** *** *
Missing(Exercise) -0.0 81 0.049 -0.076 0.180 --- --- --- --- --- --- 0.587 0.286
* **
Coffee Drinking
Yes 0.023 0.108 0.256 0.064 0.209 0.066 0.119 0.064 0.204 0.035 --- ---
*** *** * ***
Missing(Coffee) -0.034 0.159 0.267 0.261 -0.320 0.266 -0.286 0.249 0.16 2 0.261 --- ---
Occupation
Technical Worker 0.189 0.052 0.324 0.050 0.364 0.050 0.340 0.047 0.338 0.036 0.2 93 0.041
*** *** *** *** *** ***
Manager 0.208 0.055 0.174 0.063 0.373 0.065 0.335 0.068 0.333 0.046 0.259 0.060
*** *** *** *** *** ***
Office Staff 0.135 0.061 0.285 0.055 0.275 0.053 0.202 0.060 0.201 0.041 0.2 22 0.043
** *** *** *** *** ***
Farmer -0.759 0.040 -0.918 0.047 -0.594 0.047 -0.494 0.043 -0.465 0.040 -0.491 0.055
*** *** *** *** *** ***
Worker 0.123 0.040 -0.003 0.0 46 0.217 0.044 0.101 0.043 0.151 0.032 0.135 0.042
*** *** ** *** ***
Missing(Occupation) -0 .644 0.078 -0.767 0.053 -0.816 0.060 -0.615 0.059 -0.550 0.049 -0.845 0.049
*** *** *** *** *** ***
Additional Information
Being Female -0.096 0.035 -0.222 0.040 -0.184 0.037 -0.092 0.0 36 -0.149 0.030 -0.114 0.035
*** *** *** ** *** ***
Living in Urban 0.107 0.031 0.189 0.032 0.228 0.031 0.104 0.030 0.093 0.025 0.150 0.03 0
*** *** *** *** *** ***
Constant 7.730 0.132 8.395 0.127 8.122 0.127 8.662 0.099 9 .153 0.081 9.024 0.137
*** *** *** *** *** ***
Observations 5,279 5,227 4,918 5,107 6,484 5, 559
R-squared 0.251 0.26 0.286 0.207 0.218 0.236
Adjusted R^2 0.251 0.26 0.286 0.207 0.218 0.236
Note: *** p<0.01, ** p<0.05, * p<0.1
coffee can have positive influence in income. Coefficients on drinking coffee suggests as relative
to non coffee drinkers, coffee drinkers earn 11.9% to 25.6% more. Exercising is also positively
associated with earnings, increasing income by around 10% over the years. Interestingly enough,
heavily drinking alcohol is associated with higher income, from 2.7% to 13.0%, though the
statistical evidence is not substantiated over years and coefficients are not constantly positive.
Highly similar to results generated in the classical Mincer Model, one of experience have
almost same effect on increasing income as one year of education. As time goes, however, the
effect of schooling (β= 0.042, P < 0.01) began to outweigh that of experience (β= 0.019, P <
0.01) by 2.3 % as indicated in the regression results for 2015.
As for self-reported health status, the coefficients are always positive. Holding other inde-
pendent variables constant, excellent or good health status is correlated with 28.3% to 49.2%
higher income than poor health, as there is some evidence in year 2000, 2006 and 2015, all
being statistically significant. This sufficiently shows that good health is important to earning
highly.
Also, occupation plays a key role in determining income, as testified in the statistically
significant coefficients along the years. Though occupation can be largely associated with
education, differences other than years of education can be manifested in the occupation the
individual finally take as a result of class ranking and curriculum quality. Comparing with
holding ”other jobs” like being a driver, being a farmer indicates much lower income. However,
this phenomenon fades in recent years, as the effect changed from -91.8% in 2004 to -49.1% in
2015. Besides, all other jobs like being technical worker or office staff, will correlate with 13.5%
to 29.3% higher than ”other jobs”.
Finally, gender and location of residence can be earning factors too. Being a female while
holding all other independent variables constant is associated with around 11% lower income
and living in cities is associated with 9.3% to 22.8% more income than rural residents.
Table 3 shows the results from the fixed effect model, which eliminated the bais generated
by individual differences that does not vary over the years.
The identifying variations for the effect of health-related activities on income are the changes
in behaviors within a person over time. For example, the effect of smoking comes from people
who change their smoking behavior. Answering the research question of effects of health-related
activities, the results show that smoking and alcohol drinking do have association with income,
but is statistically insignificant (β= 0.020, P = 0.15 and β= 0.037, P = 0.21). Tea drinking
can have a positive effect, 4.6%, on income (β= 0.046, P < 0.05). Adequate exercisers can
have 3.0% more income than those who exercise insufficiently (β= 0.030, P < 0.05). Finally,
coffee drinking associates with 3.3% more income than nondrinkers, with p-value equals 0.10,
while containing 0 in its standard error boundaries.41
Notably, the effects of gender and urban residence can’t be investigated using the new model
due to the fact that no observations in the survey ever changed their gender and no people move
41 Thus, the results should be viewed with caution
13
Table
3:
Fixed
Effect
Regression
Results
Variables Year 2000 - 2015 Variables Continue
Coef. Std. Err. Coef. Std. Err.
Education and Experience Coffee Drinking
Schooling 0.078 0.006 Yes 0.033 0.033
*** *
Experience 0.104 0.006 Missing(Coffee) -0.031 0.126
***
Experience, square -0.001 0.000 Occupation
*** Technical Worker 0.078 0.039
Health Status **
Excellent or good 0.254 0.056 Manager 0.090 0.041
*** **
Fair 0.178 0.056 Office Staff 0.053 0.035
***
Missing(Health Status) 0.564 0.057 Farmer -0.417 0.033
*** ***
Habits Worker 0.111 0.031
Smoking ***
Yes 0.020 0.031 Missing(Occupation) -0.609 0.035
***
Missing(Smoking) -0.181 0.151 Additional Information
Being Female --- ---
Heavily Drinking Alcohol
Yes 0.037 0.024 Living in Urban --- ---
Missing(Alcohol) -0.004 0.059 Constant 6.127 0.142
***
Tea Drinking
Yes 0.046 0.019
** Observations 32574
Missing(Tea) 0.369 0.127
*** R-squared 0.218
Adequately Exercise
Yes 0.030 0.021 Adjusted R^2 0.218
**
Missing(Exercise) -0.041 0.062
Note: *** p<0.01, ** p<0.05, * p<0.1
between towns and rural villages.42 Thus, I use ”—” as placeholders in the table.
Education here is shown to have statistically significant effect on earnings, which is definitely
expected. Also, as predicted in the Mincer Model, years of experience have a effect similar to
(in case large than) that of education. One years of experience can even generate 2.6% more
income than one year of education.
Compared to those in poor health status, holding all other independent variables constant,
better health statuses are associated with higher earnings (β= 0.254, P < 0.01 and β=
0.178, P < 0.01). Occupation, similarly, can largely influence one’s income, as in accordance
with the results in the cross-sectional data analysis.
Table 4 is designed to answer the question of gender differences in the earning effects of
health-related activities. While applying the interaction effects principles, I used cross-sectional
OLS regression instead of fixed effect estimation, preserving the significance of gender in the
model at the expense of keep constant individual unobserved variable bias. Gender, as indicated
above, cannot be included in the fixed effect model due to the fact that no one in the sample
changed their sexual identity and thus gender will be omitted in the panel data model. Thus we
have to apply OLS regression by year. Also, I include other independent variables in concern
of omitted variable bias.
In the table, only coefficients on the interaction terms of interest are reported for conciseness
and clarity.43 In 2015, the interaction term of tea drinking and coffee drinking are nonexistent
for sake of no observations of these two activities in that year.
Females are more sensitive to the negative effects of smoking on wages. At year 2004, the
interaction term of smoking and sex shows females are more vulnerable to smoking in affecting
income than to men (β=−0.209, P < 0.05) and similarly in 2015 (β=−0.344, P < 0.01). The
interaction term of smoking and sex is -0.209 in 2004, showing that females are more vulerable
to smoking in affecting income than to men. In particular, for women, being a smoker relative
to a nonsmoker, is associated with 27.1%(-6.2% + -20.9% = 27.1%) decrease in wages, whereas
for men, being a smoker relative to a nonsmoker is associated with 6.2% lower wages
In terms of drinking alcohol, the interaction is only significant in 2006 (β= 0.231, P < 0.01),
indicating that for females, being a alcohol drinker relative to a non-drinker, is associated
with 30.4% increase in wages. Throughout the cross sectional regression, the interaction term
coefficients of alcohol and gender do change sign, indicating the effect is not stable over the
years.
For tea drinking, females are shown to benefit more than men. In 2000, 2006, 2009 and
2011, the interaction term is positive. The interaction is statistically significant in 2000 (β=
0.039, P < 0.1) and 2011 (β= 0.136, P < 0.01).
42 This can be caused by the fact that whenever the target change their location during the surveys years, the researchers will lose
track of them and cannot include them in the reported database.
43 As explained before, the interaction term is the difference between the effect of the independent variable on females and the
effect generally when I pool the two genders together for analysis.
15
Table
4.
Interaction
Terms
for
Gender
2000
2004
2006
2009
2011
2015
VARIABLES Coef. Std. Err. Coef. Std. Err. Coef. Std. Err. Coef. Std. Err. Coef. Std. Err. Coef. Std. Err.
Health related activities
Smoking * Female -0.151 0.110 -0.209 0.173 -0.011 0.149 0.111 0.133 -0.025 0.150 -0.344 0.148
** ***
Heavily Drinking Alcohol *
Female
-0.035 0.092 -0.035 0.190 0.231 0.110 -0.022 0.115 -0.004 0.095 0.020 0.162
***
Tea Drinking * Female 0.039 0.056 -0.016 0.063 0.016 0.061 0.011 0.059 0.136 0.046 --- ---
* ***
Adequately Exercise * Female -0.013 0.097 -0.056 0.098 0.193 0.074 -0.051 0.079 -0.059 0.055 -0.273 0.254
*** ** ***
Coffee Drinking * Female 0.119 0.218 0.072 0.128 0.073 0.132 -0.168 0.123 -0.122 0.068 --- ---
*** *
Constant 7.726 0.133 8.388 0.128 8.145 0.128 8.667 0.100 9.162 0.083 8.996 0.137
*** *** *** *** *** ***
Observations 5,279 5,227 4,918 5,107 6,484 5,559
R-squared 0.253 0.261 0.288 0.208 0.219 0.237
Adjusted R^2 0.253 0.261 0.288 0.208 0.219 0.237
Note: *** p<0.01, ** p<0.05, * p<0.1. The model was run using other independent variables in concerns of omitted variable bias. This table only choose to not include the other
variables for clarity. The omitted independent variables are available in previous cross-sectional OLS result list.
It is complicate for effect of adequately exercising. The term is positive in 2006 (β=
0.193, P < 0.01) and negative in 2011 (β=−0.059, P < 0.05) and 2015 (β=−0.273, P <
0.01). And for drinking coffee, females tend to benefit less, as the interaction term in 2009
(β=−0.168, P < 0.01) and 2011 (β=−0.122, P < 0.1) is negative and statistically significant.
6 Extension
In goal of furthering the understanding of the research questions, I conducted the interaction
effects of place of residence, fixed effect model of levels of the different health-related activities
with occupation and/or health status removed, and include dummy variables of each year in
an independent fixed effect model.
Table 5 shows the interaction results of the urban residence and health-related activities.
For smoking, drinking alcohol or tea or coffee, the interaction effect with the urban residence
is not so noteworthy for the fact that the coefficients generated can vary in sign and is not
constantly statistically significant around one particular value.
For adequately exercising, the interaction effect with urban residence is notable for the
coefficients are all positive and statistically significant in 2009 (β= 0.159, P < 0.05) and 2011
(β= 0.116, P < 0.01), which can be possibly caused by the fact that urban residents experience
less violent physical activities during work than rural villagers and thus active exercise can make
a huge difference in physical fitness.44
44 Guo, Xinyan. “Research on Perceived Profiles and Stages of Exercise Behavior Change in Urban Residents.” Proceedings of the
2015 International Conference on Education Technology, Management and Humanities Science, 2015, accessed September 1, 2021.
https ://doi.org/10.2991/etmhs −15.2015.86
17
Table
5.
Interaction
Term
for
Urban
Residence
2000 2004 2006 2009 2011 2015
VARIABLES Coef. Std. Err. Coef. Std. Err. Coef. Std. Err. Coef. Std. Err. Coef. Std. Err. Coef. Std. Err.
Health related activities
Smoking * Urban -0.034 0.062 0.022 0.068 0.048 0.064 -0.030 0.062 -0.049 0.051 -0.040 0.070
*
Heavily Drinking Alcohol *
Urban
-0.093 0.072 -0.223 0.085 -0.044 0.077 0.081 0.078 -0.020 0.065 -0.129 0.100
*** * **
Tea Drinking * Urban 0.148 0.058 0.209 0.063 -0.036 0.059 -0.029 0.059 -0.092 0.045 --- ---
** *** **
Adequately Exercise * Urban 0.041 0.083 0.136 0.093 0.065 0.077 0.159 0.080 0.116 0.058 0.113 0.088
** **
Coffee Drinking * Urban -0.002 0.217 0.197 0.150 0.112 0.145 -0.308 0.131 -0.101 0.077 --- ---
* ** *
Constant 7.760 0.133 8.434 0.126 8.112 0.128 8.658 0.099 9.128 0.083 9.017 0.137
*** *** *** *** *** ***
Observations 5,279 5,227 4,918 5,107 6,484 5,559
R-squared 0.253 0.262 0.286 0.209 0.219 0.237
Adjusted R^2 0.253 0.262 0.286 0.209 0.219 0.237
Note: *** p<0.01, ** p<0.05, * p<0.1. The model was run using other independent variables in concerns of omitted variable bias. This table only choose to not include the other
variables for clarity. The omitted independent variables are available in previous cross-sectional OLS result list.
Table 6 compares three cases. In case 1, the model is regressed using levels of habits without
health status or occupation.45 Case 2 regresses the model together with health status and
Case 3 regresses the model together with occupation. Case 1 removes only individual constant
fixed effect(FE) while keeping endogeneity in error term remains. Case 2 eliminates individual
constant FE and unobserved occupation related characteristics. Case 3, on the other hand,
minimizes individual FE and unobserved health related characteristics.46 Using this model,
we can glean some insight into the bias due to proxies while noticing the effect of difference
levels of health-related habits, verifying our assumptions about interaction between health or
occupation with health-related activities.
Rummaging through the result lines in education and experience lines, health status but
not occupation will have influence on their returns. As the same scenario in Mincer Model, one
year of experience can have similar or even strong positive effects on earnings, with returns to
education and experiences all highly statistically significant with any proxies.
But for health-related activities listed, health, surprisingly, does not have a huge correlation
with them.47 The only observable difference when including health status as a proxy is on
exercising, with one hour of exercising varying up to 50% from that resulted without health
status proxy.
On the other hand, still surprisingly, occupation proxy does have influence on activities like
alcohol consumption, tea drinking, etc. This is quite noteworthy, but rational if we take social
expectation and professional norms into consideration. Tea drinking needs time and money and
only those with a light, intellectual and well-paid job can afford it. Similarly, alcohol drinking
can be directly influenced by occupation. For instance, managers might drink more through
socialization and farmer may drink for relaxation.
As for each health-related variable, the controlled dummy variable is level of 0, indicating
no occurrence of the behavior, we can properly interpret the results as the level at which the
behavior will cause statistically significant effects on the income.
Smoking and exercising are measured in continuous variable form for sake of original struc-
ture of database from CHNS. While continuous measure of smoking is insignificant, one ad-
ditional hour of exercising per week is associated with 1.8% higher income in case 1 (β=
0.018, P < 0.01).
For alcohol and tea drinking, only the highest levels are associated with significant influences
in income relative to 0 level. People who drink alcohol everyday, the highest level in the
regression, are associated with 8.1% higher income than non-drinkers, which is possibly caused
by the fact that higher incomers will socialize a lot and this can help increasing wages. Drinking
45 Here, levels of habits indicate the different amount of dose a individual will take in that activities. Each level is annotated by
L n, for example, L1 of heavily drinking alcohol means drinking alcohol almost every day
46 Previous studies have shown that occupation dummies can be precise proxy for unobserved job-related characteristics, as shown
in J. de BEYER and J. B. KNIGHT, “The Role of Occupation in the Determination of Wages,” Oxford Economic Papers 41, no.
1 (January 17, 1989): pp. 595-618, accessed September 1, 2021. https ://doi.org/10.1093/oxf ordj ournals.oep.a041916.
47 as further established in the simple correlation results in the extension section of Yuxi Xiao, Haizheng Li, and Belton M. Fleisher,
“The Earnings Effects of Health and Health-Related Activities: A Panel Data Approach,” Applied Economics 47, no. 14 (July
2015): pp. 1407-1423, accessed September 3, 2021. https ://doi.org/10.1080/00036846.2014.1000521.
19
Table 6: Fixed Effect Regression Results
Variables
Overall
With
Health
Status
Proxy
With
Occupation
Proxy
Coef.
Std.
Err.
Coef.
Std.
Err.
Coef.
Std.
Err.
Education
and
Experience
Schooling
0.110
0.005
0.078
0.006
0.111
0.005
***
***
***
Experience
0.152
0.006
0.114
0.006
0.141
0.005
***
***
***
Experience, square -0.001 0.000 -0.001 0.000 -0.001 0.000
*** *** ***
Habits
Smoking
(Hundred
Stick/Week)
0.001 0.000 0.001 0.000 0.000 0.000
Heavily
Drinking
Alcohol
L1 Almost every day 0.081** 0.036 0.079** 0.035 0.061* 0.035
L2 3-4 times a week 0.073* 0.038 0.066* 0.037 0.058 0.037
L3 Once or twice a week 0.048 0.030 0.050* 0.030 0.037 0.029
L4 Once of twice a month 0.040 0.030 0.032 0.030 0.025 0.030
L5
No
more
than
once
a
month
0.039
0.037
0.019
0.037
0.036
0.036
Missing
Data
0.079
0.064
0.028
0.065
0.077
0.061
Tea
Drinking
L1
Almost
every
day
0.046**
0.022
0.048**
0.022
0.038*
0.022
L2
4
-
5
times
a
week
0.077*
0.043
0.068
0.043
0.066
0.042
L3
2
-
3
times
a
week
0.059*
0.035
0.044
0.035
0.053
0.035
L4 No more than once a week 0.032 0.055 0.021 0.054 0.048 0.054
L5 2-3 time a month 0.018 0.069 0.011 0.070 0.005 0.070
L6 Only once in a month 0.117 0.151 0.110 0.150 0.142 0.152
L7 Virually no 0.064 0.109 0.047 0.110 0.063 0.107
Missing Data
-
0.082
0.094
0.189**
0.096
-
0.075
0.095
Adequately
Exercise
(Hour/Week)
0.018*** 0.005 0.009** 0.005 0.015*** 0.004
Coffee Drinking
L1 Almost every day 0.059 0.097 0.065 0.099 0.050 0.096
L2 4-5 times a week -0.134 0.140 -0.146 0.133 -0.136 0.150
L3
2
-
3
times
a
week
0.042
0.066
0.039
0.067
0.017
0.066
L4
No
more
than
once
a
week
-
0.038
0.055
-
0.017
0.056
-
0.024
0.052
L5
2
-
3
time
a
month
0.101
0.084
0.124
0.084
0.110
0.081
L6
Only
once
in
a
month
0.066
0.106
0.088
0.107
0.094
0.104
L7
Virually
no
0.078
0.080
0.068
0.078
0.038
0.085
Missing
Data
-
0.093
0.093
0.128
0.096
-
0.064
0.094
Health Status
Excellent or good 0.257 0.056
***
Fair 0.178 0.056
***
Missing(Health Status) 0.582 0.058
***
Occupation
Technical Worker 0.083 0.039
**
Manager
0.102
0.041
**
Office Staff
0.061
0.035
*
Farmer
-
0.400
0.033
***
Worker 0.117 0.031
***
Missing(Occupation) -0.631 0.035
***
Constant
4.928
0.102
5.799 0.143 5.236 0.102
***
*** ***
Observations 32421 32421 32403
R-squared 0.173 0.185 0.208
Adjusted R^2 0.173 0.185 0.208
Note: *** p<0.01, ** p<0.05, * p<0.1; Urban residence and sexual identity is omitted in the chart for they are not regressed in the fixed effect
analysis
due
to
the
fact
of
low
mobility
and
constant
sexual
identity.
tea, as always, is more than a simple health improvement activity in China. People may drink
tea for mental wellness and recovery.48 Thus, it is reasonably that professions like manager,
which are associated with high income normally, can create more chance for the individual to
drink tea, as drinking tea every day associates with 4.6% higher income.
Holistically, mild level of any form of health-related activities will not generate statistically
significant influence on income.
Table 7 shows the fixed effect results when I make dummy variable for each year and include
year as a proxy in the regression model.
The benefit is to include them in the panel data regression to control for the effect of any
factors that vary over time but are the same for all individuals, potentially some economy-wide
trends in wages aren’t due to health behaviors. This way the coefficients on the health-related
variables are arguably more likely due to a change in health behavior and not some omitted
factor like a general trend in wages in the Chinese economy during this time period.
Compared with table 3, the new regression using years as dummies, in fact, yields little
variation for coefficients of health-related activities, which provide evidence that general trends’
influence, as a bias, is mitigated already.
Surprisingly, the results are noteworthy for coefficients of education. The education coeffi-
cient change from β= 0.078 in the original FE model to β= 0.037, indicating some factors
that vary over time, same for all individuals and have association with education. As pro-
posed before in the section of Literature Review, the national government has increasingly
invest in education and reshape the workforce by providing university course at night and free
skill-related online courses.
48 Michael Hurwitz, “Why Do Chinese People Love Tea So Much?,” Yoyo Chinese, 2018, accessed September 4, 2021.
https ://yoyochinese.com/blog/learn −mandarin −chinese −tea −culture −why −is −tea −so −popular
21
Table
7:
Fixed
Effect
Regression
Results
With
Year
as
Proxy
Variables Year 2000 - 2015
Variables
Continue
Coef. Std. Err. Coef. Std. Err.
Education
and
Experience
Coffee
Drinking
Schooling 0.037 0.034 Yes 0.030 0.033
*
Experience 0.061 0.034 Missing(Coffee) -0.081 0.130
*
Experience, square
-
0.001
0.000
Occupation
*** Technical Worker 0.081 0.039
Health Status **
Excellent or good 0.252 0.056 Manager 0.093 0.041
*** **
Fair
0.177
0.056
Office Staff
0.053
0.034
***
Missing(Health Status) 0.291 0.103 Farmer -0.424 0.033
*** ***
Habits Worker 0.107 0.031
Smoking
***
Yes 0.020 0.031 Missing(Occupation) -0.595 0.035
***
Missing(Smoking) -0.067 0.175 Additional Information
Being Female --- ---
Heavily
Drinking
Alcohol
Yes 0.034 0.024 Living in Urban --- ---
Missing(Alcohol) 0.013 0.058 Year
2004 0.058 0.135
Tea
Drinking
2006 0.220 0.201
Yes 0.043 0.019 2009 0.582* 0.309
** 2011 0.717* 0.373
Missing(Tea) 0.266 0.157 2015 0.749 0.523
*
Adequately
Exercise
Constant
7.377
0.959
Yes 0.026 0.021 ***
** Observations 32574
Missing(Exercise) -0.048 0.062 R-squared 0.220
Adjusted R^2 0.220
Note:
***
p<0.01,
**
p<0.05,
*
p<0.1
7 Conclusion
7.1 Advantages and Limitations
In this research, I improved previous study on health-related activities’ effects on income
by including the missing variables in the model, which will significantly decrease the standard
error for sake of more observations included in the database.
Also, for the first time in researches, I try to delve deep into the gender differences on the
influences of health-related activities on income, which can be of much use in understanding
the underprivileged status of females in the job market.
This study used interaction terms for residence as well, in order to show the effect of res-
idence in income. Also, I removed health status and occupation in the extension section in
comparison in order to show the effect of unobserved health characteristics and unobserved job
characteristics, trying to precisely contributing the influences.
Further, all health-related activities are assigned into new continuous or dummy variables in
order to capture the effect of different levels of habits. From the results, we can find the effect
of consuming one more unit in the continuous variable and find the lowest level at which the
habits can show statistical significant effect on earnings.
While this study may offer some quite new findings, which I summarize below, we also
acknowledge its limitations. Health status, an important independent variable in the model, is
self-reported and each individual can have differing standard for this variable.
Finally, there are most likely some unobserved, individual specific and correlating with health
habits independent variable that vary through time. For example, people may have differing
levels of family influence, which is not reported and differing among the population. Family
influence, or family control over the individual, definitely correlates with observations of health-
related activities and have influence on income. For this, I cannot use statistical methods like
panel data estimation to eliminate the bias by not including this time-varying factor in the
regression.
7.2 Main Findings
In basic OLS regression by year and fixed effect estimation, drinking tea, adequately exer-
cising and drinking coffee all will improve the individual’s income, by 1% to 11% while alcohol
drinking, counter-intuitively, is associated with income in cross-sectional study but this effect
doesn’t carry out to the fixed effect model. Smoking is detrimental to income, but the statistics
are insignificant in fixed effect model.
Most notably, in the comparison model that determine whether health-related activities
affect health through health status, current health situations are shown to not be pathways
for the influence of exercising, tea or coffee, the statistically significant independent variables.
Benefits can be caused by reduced mental stress, however.49 The health-related activities can
49 Mayo Clinic, “Exercise and Stress: Get Moving to Manage Stress,” Mayo Foundation for Medical Education and Research,
August 18, 2020, accessed September 10, 2021.
https ://www.mayoclinic.org/healthy −lifestyle/stress −management/in −depth/exercise −and −stress/art −20044469
23
be determined jointly with health status too. Finally and most trustfully, direct effects of the
health-related activities may not manifest immediately after the action. It may take some and
thus the proxy of health status will appear to have no correlation with these activities.
Surprisingly, occupation yield statistically significant correlation with health-related activi-
ties. It was explained in the results section that these might be achieved due to socialization.
The most innovate and groundbreaking results are in the gender interaction parts. Women
benefit more from drinking tea and less from drinking coffee, as supported by investigations in
female biological differences.50 Also, females benefit more from quitting smokings.
Lastly, only top levels(doing everday) of the habits will generate statistically significant ef-
fects on earnings, indicating further that mild levels of habits have no correlation with earnings,
which is great in supporting that health-related activities are determinants of income, for if in-
come level can determine different level of health-related habits, all possible levels should be
statistically significant enough to yield correlation with income, which didn’t occur here.
Reference List
1. Heien, D. “The Relationship between Alcohol Consumption and Earnings.” Journal of Studies on Alcohol, 1996.
https://www.jsad.com/doi/10.15288/jsa.1996.57.536.
2. Medgyesi, M´arton, and Istv´an Gy¨orgy T´oth. “Income, Wealth, Employment, and Beyond.” Europe’s Income,
Wealth, Consumption, and Inequality, 2021, 177–217. https://doi.org/10.1093/oso/9780197545706.003.0005.
3. Inder, Brett. “Estimating Long-Run Relationships in Economics.” Journal of Econometrics 57, no.1-3 (1993):
53–68. https://doi.org/10.1016/0304-4076(93)90058-D
4. Jaccard, James, and Robert Turrisi. Interaction Effects in Multiple Regression. Thousand Oaks, New York: Sage,
2003.
5. Arrow, Kenneth J. The Theory of Discrimination. Princeton, N.J.: Princeton University Press, 1971.
6. Pro ject, Borgen. “Importance of Girls’ Education around the World.” The Borgen Project. Borgen Project,
September 15, 2019. https://borgenproject.org/importance-of-girls-education/.
7. Notanubun, Zainuddin. “The Effect of Organizational Citizenship Behavior and Leadership Effectiveness on Public
Sectors Organizational Performance: Study in the Department of Education, Youth and Sports in Maluku Province,
Indonesia.” Public Organization Review 21, no. 1 (2020): 1–18. https://doi.org/10.1007/s11115-020-00475-4.
8. Pitarakis, Jean-Yves. “On the Bias of the OLS Estimator in a Nonstationary Dynamic Panel Data Model.”
Statistics & Probability Letters 38, no. 2 (June 1998): 145–50. https://doi.org/10.1016/s0167-7152(97)00165-x
9. Abadir, Karim M. “Ols Bias in a Nonstationary Autoregression.” Econometric Theory 9, no. 1 (March 1993):
81–93. https://doi.org/10.1017/s0266466600007349
10. Blackburn, McKinley L., and David Neumark. “Are Ols Estimates of the Return to Schooling Biased Downward?
Another Look.” The Review of Economics and Statistics 77, no. 2 (May 1995): 217. https://doi.org/10.2307/2109861
11. Sapra, S.K. “Asymptotic Properties of Restricted Maximum-Likelihood Estimator of Parameters in Censored Re-
gression Model.” Econometric Theory 4, no. 3 (August 1988): 535–36. https://doi.org/10.1017/s0266466600013463
50 Allen, Alicia M., Cheryl Oncken, and Dorothy Hatsukami. “Women and Smoking: The Effect of Gender on The Epidemiol-
ogy,Health Effects, and Cessation of Smoking.” Current Addiction Reports 1, no. 1 (2014): 53–60, accessed September 11, 2021.
https ://doi.org/10.1007/s40429`013`0003`6
24
12. Neumark, David, and Sanders Korenman. “Sources of Bias in Women’s Wage Equations: Results Using Sibling
Data.” The Journal of Human Resources 29, no. 2 (1992): 379–450. https://doi.org/10.3386/w4019
13. Gupta, Anviti, and Alok Gupta. “Women Empowerment Initiatives and Socio-Economic Development.” Women
Empowerment and Well-Being for Inclusive Economic Growth, January 2021, 106–19.
https://doi.org/10.4018/978-1-7998-3737-4.ch007
14. Kabeer, Naila. “Gender Equality and Women’s Empowerment: A Critical Analysis of the Third MILLENNIUM
Development Goal 1.” Gender & Development 13, no. 1 (May 2005): 13–24.
https://doi.org/10.1080/13552070512331332273
15. Malatesta, Victor J., Robert H. Pollack, Terri D. Crotty, and Lelon J. Peacock. “Acute Alcohol Intoxication and
Female Orgasmic Response.” The Journal of Sex Research 18, no. 1 (February 1982): 1–17.
https://doi.org/10.1080/00224498209551130
16. Tekin, Erdal. “Employment, Wages, and Alcohol Consumption in Russia.” Southern Economic Journal 71, no. 2
(October 2004): 397. https://doi.org/10.2307/4135298
17. Frijters, Paul, and Aydogan Ulker. “Robustness in Health Research: Do Differences in Health Measures, Tech-
niques, and Time Frame Matter?” Journal of Health Economics 27, no. 6 (May 2008): 1626–44.
https://doi.org/10.1016/j.jhealeco.2008.06.003
18. Cavalli, Lodovica, and Alessandra Tavani. “Coffee Consumption and Its Impact on Health.” Beverage Impacts on
Health and Nutrition, February 2016, 29–47. https://doi.org/10.1007/978-3-319-23672-8 3
19. Gravelle, Hugh, and Matt Sutton. “Income, Relative Income, and Self-Reported Health in Britain 1979-2000.”
Health Economics 18, no. 2 (February 2009): 125–45. https://doi.org/10.1002/hec.1354
20. Hershbein, Brad, Melissa Schettini Kearney, and Luke Pardue. “College Attainment, Income Inequality, and Eco-
nomic Security: A Simulation Exercise.” AEA Papers and Proceedings, May 2020. https://doi.org/10.3386/w26747
21. Cunningham, Cindy M., and Robert D. Mohr. “Using Tools to Distinguish General and Occupation-Specific Skills.”
Journal for Labour Market Research 53, no. 1 (December 16, 2019). https://doi.org/10.1186/s12651-019-0256-1
22. Huikari, Sanna, Hanna Junttila, Leena Ala-Mursula, Timo J¨ams¨a, Raija Korpelainen, Jouko Miettunen, Rauli
Svento, and Marko Korhonen. “Leisure-Time Physical Activity Is Associated with Socio-Economic Status beyond
Income-Cross-Sectional Survey of the Northern FINLAND Birth Cohort 1966 Study.” Economics & Human Biology
41 (May 2021): 100969. https://doi.org/10.1016/j.ehb.2020.100969
23. Hurley, Jeremiah, Emmanouil Mentzakis, and Marjan Walli-Attaei. “Inequality Aversion in Income, Health, and
Income-Related Health.” Journal of Health Economics 70 (July 2020): 102276.
https://doi.org/10.1016/j.jhealeco.2019.102276
24. Boon, Niels. “Health Potential for Functional Green Teas?” International Journal for Vitamin and Nutrition
Research 78, no. 6 (December 1, 2008): 275–81. https://doi.org/10.1024/0300-9831.78.6.275
25. Reardon, Sean F., and Kendra Bischoff. “Income Inequality and Income Segregation.” Inequality in the 21st
Century, January 15, 2018, 426–35. https://doi.org/10.4324/9780429499821-73
26. Wheeler, Christopher H. “Urban Decentralization and Income Inequality: Is Sprawl Associated with Rising In-
come Segregation across Neighborhoods?” FRB of St. Louis Working Paper, no. 2006-037A (May 25, 2006).
https://doi.org/10.20955/wp.2006.037
27. Hogan, Richard. “Class, Race, Gender, and Life Chances.” Inequality and Stratification, 2015, 175–96.
https://doi.org/10.4324/9781315662794-17
28. Khan, Naghma, and Hasan Mukhtar. “Tea Polyphenols for Health Promotion.” Life Sciences 81, no. 7 (July 26,
2007): 519–33. https://doi.org/10.1016/j.lfs.2007.06.011
25
29. Gao, Yanyan. “Urban Bias, Rural-Urban Income Gap and Agricultural Growth: The Resource-Diverting Effect
of Rural-Urban Income Gap in China.” SSRN Electronic Journal, 2010. https://doi.org/10.2139/ssrn.1857269
30. Zhang, Yingqiang, and Tor Eriksson. “Inequality of Opportunity and Income Inequality in Nine Chinese Provinces,
1989–2006.” China Economic Review 21, no. 4 (December 2010): 607–16.
https://doi.org/10.1016/j.chieco.2010.06.008
31. Xiao, Yuxi, Haizheng Li, and Belton M. Fleisher. “The Earnings Effects of Health and Health-Related Activities:
A Panel Data Approach.” Applied Economics 47, no. 14 (2015): 1407–23.
https://doi.org/10.1080/00036846.2014.1000521
32. Basu, Deepankar. “Bias of Ols Estimators Due to Exclusion of Relevant Variables and Inclusion of Irrelevant
Variables.” Oxford Bulletin of Economics and Statistics 82, no. 1 (November 19, 2019): 209–34.
https://doi.org/10.1111/obes.12322
33. Khan, Naghma, and Hasan Mukhtar. “Tea and Health: Studies in Humans.” Current Pharmaceutical Design 19,
no. 34 (2013): 6141–47. https://doi.org/10.2174/1381612811319340008
34. Ridgeway, Cecilia L. “Interaction and the Conservation of Gender Inequality: Considering Employment.” American
Sociological Review 62, no. 2 (April 1997): 218. https://doi.org/10.2307/2657301
35. Wooldridge, Jeffrey M. Econometric Analysis of Cross Section and Panel Data. Cambridge, Mass, Mass.: MIT
Press, 2011.
36. Conti, Gabriella. “Early Endowments, Education, and Health.” Academia.edu, 2011.
https://www.academia.edu/5594807/Early endowments education and health
37. Kenkel, Don, and Ping Wang. “Are Alcoholics in Bad Jobs?” National Bureau of Econoc Research working paper,
1999. https://doi.org/10.3386/w6401
38. “A Literature Review of the Costs of Smoking and Drinking.” The Costs of Poor Health Habits, 1991, 137–42.
https://doi.org/10.4159/harvard.9780674422261.c9
39. Duflo, Esther. “Women Empowerment and Economic Development.” Journal of Economic Literature 50, no. 4
(2012): 1051–79. https://doi.org/10.1257/jel.50.4.1051
40. Gabbuti, Giacomo. “Labor Shares and Inequality: Insights from Italian Economic History, 1895–1970.” European
Review of Economic History 25, no. 2 (2020): 355–78. https://doi.org/10.1093/ereh/heaa010
26
8 Acknowledgement
Special thanks here toward Professor Mark Foley. His guidance ignites me and leads me step
by step to our final goals. He leads through a vigorous and valuable memory of research. I also
would like to thank my parents for providing support me all the way through my academical
life.
27
