View metadata, citation and similar papers at core.ac.

uk brought to you by CORE

provided by Research Papers in Economics

Tobacco Chewing, Smoking and Health Knowledge:

Evidence from Bangladesh

Shiferaw Gurmu Mohammad Yunus

Georgia State University Bangladesh Institute of Development Studies

Abstract
Unlike the substance abuse studies in developed countries, tobacco consumption and its
adverse effects in developing countries are poorly studied. The objective of this paper is to
identify which factors influence individuals’ decision to smoke cigarettes, chew tobacco and
their knowledge about the health hazards of tobacco use. To allow for the potential
correlation among smoking tobacco, chewing tobacco, and health knowledge, we estimate a
trivariate probit regression model using household survey data from Bangladesh. For both
chewing tobacco and smoking, the results show how the probabilities of uninformed tobacco
user and uninformed nonuser vary across different demographic groups.

Citation: Gurmu, Shiferaw and Mohammad Yunus, (2008) "Tobacco Chewing, Smoking and Health Knowledge: Evidence
from Bangladesh." Economics Bulletin, Vol. 9, No. 10 pp. 1-9
Submitted: December 2, 2007. Accepted: May 16, 2008.
URL: http://economicsbulletin.vanderbilt.edu/2008/volume9/EB-07I10014A.pdf
 1. Introduction

It is recognized that tobacco consumption is harmful to health, and is one of the

main causes of death worldwide. According to Gajalakshmi et al. (2000), eight out of ten
smokers now live in developing countries, and the prevalence of tobacco consumption
has been rising in most low- and middle-income countries. Much of the existing literature
on tobacco consumption and control has largely focused on developed countries (Baltagi
and Levin 1986, Becker and Murphy 1988, Becker et al. 1994, Kenkel 1991, Jones 1989,
Mullahy 1997, and Chaloupka and Warner 1999). Existing studies for developing
countries have focused on smoking patterns and trends using aggregate data. By
contrast, very little is known about the link between consumer characteristics, policy-
based factors and tobacco use in developing countries; exceptions include Jones and
Kirigia (1999) and Gurmu and Yunus (2004). While Jones and Kirigia (1999) used
household survey data from South Africa to identify which factors influence individual
women’s choice to smoke cigarettes, Gurmu and Yunus (2004) used survey data from
Bangladesh to assess the extent of the use of smoking- and chewing-tobacco using
generalized bivariate negative binomial regression model. The absence of studies on
individual smoking behavior in most other developing countries may largely be attributed
to lack of household level data on tobacco consumption.
Unlike in developed countries where cigarette smoking is common, both
smoking- and chewing-tobacco are prevalent among tobacco users in many developing
countries in Africa and Asia. This paper analyzes individual tobacco consumption
behavior using household survey data from Bangladesh. The objective is to identify
which factors influence individuals’ decision to smoke cigarettes, chew tobacco and their
awareness about the health hazards of tobacco use. To allow for the potential correlation
among smoking tobacco, chewing tobacco, and health knowledge, we estimate a
trivariate probit regression model. Section 2 presents description of data and empirical
methodology, followed by results and conclusion in sections 3 and 4.

2. Data and Methods

2.1 Data

We use data from the Tobacco Prevalence Survey (TPS) in Bangladesh sponsored
by the World Health Organization in 2001; see Yunus, (2001) for details. The Survey was
conducted in two administrative districts, Chittagong and Rangpur, of paramount interest
for tobacco production and consumption in the country. While the former is the center for
smuggling of foreign brands of cigarettes, the latter is a major tobacco-growing region.
Data on daily consumption of smoking- and chewing-tobacco along with other
socioeconomic and demographic characteristics and parental tobacco consumption habits
were collected from respondents of 10 years of age and above. Our analysis is based on
sample size of 15,000 individual respondents.
Table 1 shows definition of variables as well as their means and standard
deviations. The dependent variables are dichotomous variables regarding daily use of
smoking tobacco, chewing tobacco, and knowledge about health hazards of tobacco use.
Most of the users of tobacco in TPS data are daily users; only about 4 percent of males

1
and 3 percent of females are occasional users of tobacco (Yunus 2001). The
percent of daily users of tobacco products are 24.3% and 13.4% for smoking- and
chewing-tobacco, even though 85.9% of the respondents are aware of the health hazards.1
Regarding the health hazards of tobacco use, summary statistics reported in Yunus (2001)
show that respondents are aware of respiratory diseases (30%), lung cancer (33%), heart
diseases (17%), and stroke (5%).2 The typical respondent is a Muslim, married, in
his/her early thirties, lives in rural area, and has about 7 years of formal schooling.
Although the country is mostly agrarian, only around 11% of the respondents were
related to agricultural occupation in either doing agricultural operations on their own
farms or working as agricultural wage laborers. More than one-half of the fathers and
slightly less than two-thirds of the mothers of the respondents use or have used tobacco in
some form or other.

2.2 Model

A consumer is said to be addicted to a good, if it involves reinforcement,

tolerance and withdrawal. However, a rational consumer also considers the future
negative consequences of harmful behavior given the state of health knowledge.
Consumers become aware of the consequences of the products they consume through
print and electronic media. Following Mullahy and Portney (1990), Kenkel (1991) and
Jones and Kirigia (1999), we employ a static utility maximization framework, where
utility is assumed to be a function of smoking tobacco (S), chewing tobacco (C), the state
of knowledge of negative health consequences of using tobacco (K), and a composite
non-addictive good.3 The state of the health knowledge depends on the level of
consumption of goods (including smoking- and chewing-tobacco), a host of demographic
factors, and unobserved individual characteristics. The solutions to the individual’s utility
maximization problem provide reduced forms for the choice variables.
Given the limitations of the survey data with no extraneous information on prices
and lack of suitable instruments, we use the following reduced forms for smoking- and
chewing-tobacco, and health knowledge:

S = S(X, µ) (1)

C = C(X, µ), (2)

K = K(X, µ), (3)

1
The observed joint percentage frequencies for tobacco consumption and health knowledge, f(smoking,
chewing, knowledge), are f(+, +, +) = 1.9, f(+, 0, +) = 18.7, f(0, +, +) = 8.6, f(0, 0, +) = 56.7, f(+, +, 0) =
0.7, f(+, 0, 0) = 3.0, f(0, +, 0) = 2.2, and f(0, 0, 0) = 8.3, where + denotes current users of tobacco products
and/or awareness about health hazards associated with tobacco use and 0 represents nonusers.
2
In this paper, we use a binary outcome measure of awareness of health risks of tobacco use because of
lack of access to individual-level data disaggregated by type of risks. This dummy variable measures only
the relative risks; it may or may not imply absolute risks since there are no details in the survey of the risks
respondents face personally. For a related literature, see, for example, Schoenbaum (1996), who
investigated whether smokers understand the mortality effects and magnitudes of smoking.
3
Kenkel and Chen (2000) address the question of whether and how information about the risks of smoking
can influence consumers’ use of tobacco.

2
where X is a vector of observable factors and µ represents unobserved individual
characteristics. The presence of unobserved heterogeneity, µ, implies that the error terms
of empirical versions of the reduced forms 1 to 3 will share a common component and
can be expected to be correlated with each other. This is exploited in our empirical model
specification, where we use a trivariate probit model to allow for correlation between the
unobserved determinants of smoking- and chewing-tobacco and health knowledge.
Specifically, for observation i (i =1, ..., n), we assume that (Si, Ci, Ki) has a trivariate
normal ( xis β s , xic β c , xik β k ; ρ sc , ρ sk , ρ ck ) distribution, where the β ’s and ρ ’s are unknown
vectors or scalars of mean and correlation parameters, respectively.

3. Results

Table 2 presents sets of coefficient estimates and absolute t-ratios from the
trivariate probit model, along with the value of the maximized log-likelihood function
and the estimates of the correlation parameters, denoted by rho. The estimated negative
and statistically significant correlation between tobacco consumption measures suggests
that smoking and chewing tobacco are substitutes, as expected. The remaining
correlations are insignificant.
Regarding factors affecting probability of using tobacco and health knowledge,
Table 2 shows that male respondents are more likely to smoking tobacco while women
are more likely to use chewing-tobacco. This is in line with the custom of the country,
where adolescent female smokers are reprimanded, but females are encouraged to chew
betel leaves with nuts as mouth fresheners. While the result for urban indicates no
difference in smoking behavior, urban residents are less likely to use chewing-tobacco
than their rural counterparts are. There is no evidence of significant family income effect
on the likelihood of tobacco consumption in either form. On average, individuals that are
more educated are less likely to use tobacco in either form, a finding that is consistent
with that of Jones and Kirigia (1999). Respondents in urban areas and those with more
education are more likely to be aware of health hazards of tobacco use. On average,
parental use of tobacco seems to reduce the probability of consuming tobacco.
The results from the trivariate probit model can be used to identify groups of
individuals that may be used in improving consumer information to reduce tobacco
consumption. There are four groups, but the first two are of primary interest: (a) those
individuals who may be more likely to be at risk of consuming tobacco (uninformed non-
user), (b) those users of tobacco who may be more likely to respond to information
shocks or health education (uninformed user), (c) informed tobacco-user, and (d)
informed nonuser. Tables 3 and 4 present the sample means of the probabilities
associated with (smoking, health knowledge) and (chewing tobacco, health knowledge),
respectively.4 The results are tabulated for the whole sample as well as selected
determinants of choice probabilities, including benchmark groups for indicator variables.
Observe that, in tables 3 and 4, the probability of uninformed tobacco user is lower than
the probability of uninformed nonuser, except for male, agricultural wage labor and
4
Estimated average probabilities for smoking and chewing tobacco as well as for tobacco use and health
knowledge are available upon request from the authors. The latter are computed from the distribution of
(smoking tobacco + chewing tobacco, health knowledge).

3
business occupation in Table 3. The proportion of uninformed user of tobacco relative to
informed user is much greater among individuals who did not complete elementary
school. Respondents in rural areas seem to be the most at risk of becoming a smoker.
Those in agricultural-labor occupation are more likely to respond to health education.
Overall, the proportion of ‘informed’ seems pretty high, suggesting on the face of
it little or no role for government intervention. The data preclude our providing a definite
answer to the question of the effectiveness of intervention strategies such as health
education and tobacco taxes. For example, our measure of risk of tobacco use places all
risks into one category. Previous studies provide evidence suggesting that people have a
tendency to overestimate the risks from smoking, and learn about the health risks of
smoking through both experience and acquisition of information (Viscusi 1991; Viscusi
and Hakes 2008). We also focus on smoking or tobacco chewing participation decision,
not on the quantity of tobacco consumed.5

4. Conclusion

This paper has investigated the socio-economic determinants of the probability of

tobacco chewing, smoking and knowledge of the health risks of tobacco use. There is
strong evidence that the effects vary across different socio-economic groups and by types
of tobacco use. The cross sectional nature of our data and lack of information on key
explanatory variables (e.g., prices) precludes detailed analysis of participation,
consumption and quitting behavior, along with the effectiveness of tobacco control
policies. Prior literature mostly using US data have explored the adequacy of risk beliefs,
suggesting that people may have more information about some risks, may understand
gradients in risk but not the absolute risks, and make decisions about whether to smoke
based on their risk perceptions.6 Future research directions might broaden the analysis
for developing economy presented here to account for differential risks of tobacco use.

References

Baltagi, B H. and D. Levin (1986) “Estimating Dynamic Demand for Cigarettes using
Panel Data: The Effects of Bootlegging, Taxation, and Advertising Reconsidered”
Review of Economics and Statistics 68, 148-155.

Becker, G.S. and M. Grossman, and K. M. Murphy (1994) “An Empirical Analysis of
Cigarette Addiction” American Economic Review 84, 396-418.

Becker, G. S. and K. M. Murphy (1988) “A Theory of Rational Addiction” Journal of

Political Economy 96, 675-700.

Chaloupka, F. J. (1991) “Rational Addictive Behavior and Cigarette Smoking” Journal of

Political Economy 99, 722-742.

5
For example, Viscusi and Hakes (2008) provides evidence that cigarette taxes influence the amount of
cigarette demanded, not the discrete smoking status decision.
6
For example, see Viscusi and Hakes (2008) and references there in.

4
Chaloupka, F.J. and K. E. Warner (1999) “The Economics of Smoking” National Bureau
of Economics Research working paper 7047.

Gajalaksmi, C.K., P. Jha, K. Ranson and S. Nguyen (2000) “Global Patterns of Smoking
and Smoking –Attributable Mortality” in Jha P. and F.J. Chaloupka (eds.) Tobacco
Control in Developing Countries, Oxford: Oxford University Press, 11-39.

Gurmu, S. and M. Yunus (2004) “An Empirical Analysis of Tobacco Consumption in a

Developing Economy” In G.N. Muuka (ed.) International Academy of African Business
and Development Conference Proceedings, Kentucky: IAABD, Volume 5, 358-364.

Jones, A. M. (1989) “A Double-Hurdle Model of Cigarette Consumption” Journal of

Applied Econometrics 4, 23-39.

Jones, A. M. and J.M. Kirigia (1999) “Health Knowledge and Smoking among South
African Women” Health Economics 8, 165-169.

Kenkel, D.S. (1991) “Health Behavior, Health Knowledge and Schooling” Journal of
Political Economy 99, 287-305.

Kenkel, D. and L. Chen (2000) “Consumer Information and Tobacco Use” in Jha P. and
F.J. Chaloupka (eds.) Tobacco Control in Developing Countries Oxford, Oxford
University Press, 177-214.

Mullahy, J. (1997) “Instrumental-Variable Estimation of Count Data Models:

Applications to Models of Cigarette Smoking Behavior” Review of Economics and
Statistics 79, 586-593.

Mullahy, J. and P.R. Portney (1990) “Air Pollution, Cigarette Smoking, and the
Production of Respiratory Health” Journal of Health Economics 9, 193 – 205.

Schoenbaum, M. (1996) “Do Smokers Understand the Mortality Effects of Smoking?

Evidence from the Health and Retirement Survey” American Journal of Public Health
87, 755-759.

Viscusi, W.K. (1991) “Age Variations in Risk Perceptions and Smoking Decisions”
Review of Economics and Statistics 73, 577-588.

Viscusi, W.K. and J.K Hakes (2008) “Risk Beliefs and Smoking Behavior” Economic
Inquiry 46, 45-59.

Yunus, M. (2001) Craving for Nicotine: A Study on Tobacco Prevalence in Bangladesh,

World Health Organization, Regional Office South-East Asia, accessed at
http://w3.whosea.org/EN/Section1174/ section1462/pdfs/surv/sentinelbangladesh.pdf

5
 Table 1. Definition of Variables and Summary Statistics (Sample Size n = 15,000)

Variable Description Mean Standard

Deviation
Smoke = 1 if smoking tobacco used daily 0.243 0.429
Chew = 1 if chewing tobacco used daily 0.134 0.341
Knowledge = 1 respondent is aware of health hazards
of tobacco use 0.859 0.348
Region =1 if Rangpur resident 0.500 0.500
Urban =1 if urban resident 0.374 0.484
Age Age in years 30.65 14.97
Education Years of formal schooling 6.876 4.70
Muslim = 1 if religion is Islam 0.789 0.408
Income Monthly family income ‘000 Tk. 7.496 10.30
Male =1 if male 0.549 0.498
Married =1 if married 0.576 0.494
Agri-Labor =1 if agriculture labor occupation 0.112 0.315
Service =1 if service occupation 0.127 0.333
Business =1 if business occupation 0.131 0.338
Self Employed =1 if self employed or household chores 0.306 0.461
Student =1 if student 0.263 0.440
Father use =1 if father uses tobacco 0.537 0.499
Mother use =1 if mother uses tobacco 0.646 0.478

6
Table 2. Coefficient Estimates from Trivariate Probit for Smoking-, Chewing-Tobacco and
Health Knowledge (n = 15,000)

Variable Smoking Tobacco Chewing Tobacco Knowledge

Constant -3.849*** (30.57) -2.943*** (21.24) 0.674*** (6.31)
Region 0.143*** (4.40) 0.496*** (14.49) -0.388*** (13.16)
Urban -0.009 (0.26) -0.112*** (3.02) 0.280*** (8.41)
Age 1.129*** (18.67) 0.884*** (15.36) 0.065 (1.23)
Age-squared -0.113*** (17.45) -0.064*** (10.28) -0.010 (1.63)
Education -0.882*** (9.02) -0.070 (0.71 0.539*** (5.60)
Education-squared 0.331*** (4.81) -0.314*** (3.82) 0.208*** (2.66)
Muslim -0.107*** (3.09) -0.077** (2.10) 0.104*** (3.12)
Income 0.079 (0.54) -0.012 (0.06) -0.614*** (3.86)
Male 1.979*** (36.29) -0.554*** (10.84) 0.007 (0.20)
Married 0.078* (1.87) -0.077 (1.60 0.203*** (4.78)
Agri-Labor 0.080 (1.23 0.329*** (3.64 0.063 (0.91)
Service -0.092 (1.42) 0.236** (2.50) 0.047 (0.63)
Business 0.144** (2.31) 0.236*** (2.59) -0.012 (0.17)
Self Employed -0.166** (2.40) 0.295*** (3.47) -0.069 (1.04)
Student -0.566*** (7.73) -0.586*** (4.49) 0.132** (2.02)
Father Use -0.111*** (3.20 -0.035 (0.94) -0.136*** (4.30)
Mother Use -0.084*** (2.61) -0.260*** (7.27) -0.137*** (4.45)

Rho (Smoking, Chewing) -0.248*** ( 10.97)

Rho (Smoking, Knowledge) -0.006 (0.27)
Rho (Chewing, Knowledge) 0.002 (0.08)
Log Likelihood Function -14750.9
Note: *, **, and *** denote statistical significance at the 10%, 5%, and 1% level, respectively. Figures
within parentheses are absolute values of t-ratios. The excluded occupational category is wage labor.

7
 Table 3: Average Probabilities for Smoking Tobacco and Health Knowledgea

Variable Informed Informed Uninformed Uninformed

(#Observations) Smoker Non- Smoker Non-smoker
smoker
(Ps11) (Ps01) (Ps10) (Ps00)
Full Sample (15000) 0.169 0.689 0.035 0.106
Region (7500) 0.172 0.646 0.048 0.134
Urban (5604) 0.144 0.776 0.014 0.067
Rural (9396) 0.185 0.637 0.048 0.300
ESC (8466)b 0.129 0.794 0.010 0.066
Less than ESC (6534)b 0.222 0.553 0.067 0.158
Muslim (11830) 0.168 0.696 0.033 0.103
Male (8241) 0.296 0.574 0.059 0.007
Female (6759) 0.015 0.829 0.005 0.150
Married (8636) 0.239 0.612 0.049 0.100
Unmarried (6364) 0.076 0.794 0.016 0.115
Agri-Labor (1677) 0.458 0.321 0.140 0.081
Service (1900) 0.247 0.688 0.021 0.044
Business (1972) 0.412 0.474 0.063 0.052
Self-employed (4593) 0.055 0.761 0.018 0.166
Student (3942) 0.020 0.790 0.002 0.099
Father Use (8053) 0.196 0.645 0.044 0.115
Mother Use (9688) 0.165 0.682 0.038 0.116
Sample size (n) 3091 9793 550 1566
a
The underlying probabilities are computed for each individual using estimates from the trivariate probit
model for smoking tobacco, chewing tobacco and health knowledge. Since smoking tobacco and health
knowledge follow a bivariate normal distribution, the probabilities are computed for each respondent using
the univariate and bivariate normal cumulative distribution functions as follows:
Ps11 = Φ ( xsi β s , xki β k , ρ sk ) , Ps 01 = Φ ( xki β k ) − Ps11 , Ps10 = Φ ( xsi β s ) − Ps11 , and
Ps 00 = 1 − Ps 01 − Ps10 − Ps11 .
b
ESC denotes elementary school complete or higher.

8
 Table 4: Average probabilities for Chewing Tobacco and Health Knowledgea

Variable Informed Informed Uninformed Uninformed

(#Observations) Chewer Non- Chewer Non-chewer
chewer
(Pc11) (Pc01) (Pc10) (Pc00)
Full Sample (15000) 0.105 0.754 0.029 0.112
Region (7500) 0.129 0.690 0.043 0.138
Urban (5604) 0.084 0.837 0.012 0.067
Rural (9396) 0.118 0.704 0.039 0.139
ESC (8466)b 0.064 0.860 0.007 0.070
Less than ESC (6534)b 0.158 0.616 0.058 0.167
Muslim (11830) 0.101 0.763 0.026 0.110
Male (8241) 0.074 0.796 0.018 0.111
Female (6759) 0.142 0.702 0.042 0.113
Married (8636) 0.156 0.695 0.040 0.109
Unmarried (6364) 0.036 0.834 0.014 0.116
Agri-Labor (1677) 0.160 0.619 0.054 0.167
Service (1900) 0.068 0.867 0.006 0.058
Business (1972) 0.093 0.792 0.016 0.098
Self-employed (4593) 0.197 0.619 0.059 0.125
Student (3942) 0.003 0.896 0.004 0.100
Father Use (8053) 0.133 0.708 0.041 0.117
Mother Use (9688) 0.101 0.745 0.032 0.121
Sample size (n) 1677 11308 439 1576
a
The underlying probabilities are computed for each individual using estimates from the trivariate probit
model for smoking tobacco, chewing tobacco and health knowledge. Since chewing tobacco and health
knowledge follow a bivariate normal distribution, the probabilities are computed for each respondent using
the univariate and bivariate normal cumulative distribution functions as follows:
Pc11 = Φ ( xci β c , xki β k , ρ ck ) , Pc 01 = Φ ( xki β k ) − Pc11 , Pc10 = Φ ( xci β c ) − Pc11 , and
Pc 00 = 1 − Pc 01 − Pc10 − Pc11 .
b
ESC denotes elementary school complete or higher.