This article is an open-access article distributed under the terms and conditions of the Creative Commons Attribution license (http://creativecommons.org/licenses/by/3.0/).

Using aggregated panel data taken from three waves of the Indonesian Family Life Survey (1993–2000), this article tests the myopic addiction behaviour of cigarette demand. Sensitivity analysis is done by examining a rational addiction behavior of cigarette demand. The results provide support for myopic addiction. The short- and long-run price elasticities of cigarette demand are estimated at −0.28 and −0.73 respectively. Excise taxes are more likely to act as an effective tobacco control in the long-run rather than a major source of government revenue.

Economic models of addiction can be divided into three groups: imperfectly rational addiction, rational addiction and myopic addiction [

Indonesia is a significant contributor to the global burden of disease from tobacco-related illnesses. The prevalence of smoking among males age 15 years and above increased from 53.4% in 1995 to 63.2% in 2001 and to 63.1% in 2004. Among adult females, the prevalence of smoking has also increased from 1.7% in 1995 to 4.5% in 2004. Overall, cigarettes consumed increased from 33 billion in 1970 to 217 billion in 2004. With the fourth largest population in the world, Indonesia in 2002 ranked as the fifth largest consumers of cigarettes (182 billion) behind China (1.7 trillion), USA (463 billion), Russia (375 billion), and Japan (299 billion) [

Increasing the price of tobacco is one of several strategies used to curb tobacco consumption [

Evidence on tobacco consumption behavior from Indonesia is rare. A prior study using data from the 1999 national socio-economic survey [

There have been no complete estimates for Indonesia of demand response to changes in taxes on tobacco products. These estimates (short-run and long-run) are crucial for both health and revenue policy. Furthermore, understanding addictive behavior in developing countries is an important regulatory question since the tobacco industry has focused on the developing world as a target for new customers. To the best of the authors’ knowledge, there is no study that estimates for Indonesia a dynamic specification (including addictive behavior) of cigarette demand with complete information on demand price elasticities. This paper will investigate such a model for cigarette demand and use the model to estimate both short- and long-run price elasticities of cigarette demand in Indonesia.

The addiction model developed here is tested using pooled individual panel data from three waves of the Indonesian Family Life Survey (IFLS) spanning the period 1993–2000. Although aggregate panels have less variability relative to individual level data, and this has been regarded as a weakness of time series studies [

Most of the existing studies on price elasticities in developing countries do not develop model dynamics that contain rational addiction. Studies that develop myopic addiction include Tansel [

In this study, we estimate a myopic addiction for cigarette demand model in the form:
_{i}_{t}_{it}_{1}) on current cigarette consumption (_{it}

We also perform a sensitivity analysis that examines myopic versus rational addiction. Empirical specifications for rational addiction are as follows:

Statistical significance of the coefficient (_{2}) on lead consumption (

Applying ordinary least squares (OLS) in _{it}_{i}_{it}_{it}_{−1} in _{it}_{−1} and _{it+}_{1} in _{it}_{−}_{1}_{it}_{−1} and _{it}_{+1} in _{it}_{−1} and _{it}_{+1} then measured with error. Measurement error in either dependent or independent variables leads to biased OLS coefficient estimates.

We explored two groups of econometric specifications in our analysis. The first group includes estimators ignoring the endogenous-regressors problem, including OLS, fixed effects (FE), and random effects (RE). The second group includes estimators that treat error-in-variables and unobservable heterogeneity: two-stage least squares (2SLS), fixed effects two-stage least squares (FE2SLS), random effects two-stage least squares (RE2SLS), and generalized methods of moments (GMM).

Diagnostic tests helped us select the most appropriate estimator. We tested for endogeneity in lagged consumption by performing a Hausman specification test. If lagged consumption is found to be exogenous, we opt for either OLS, FE or RE. To test the appropriateness of an individual-effects estimator instead of regular OLS, we use the Breusch-Pagan Lagrange Multiplier test [

However, if lagged consumption is indeed endogenous, we suggest instruments (and maintain that they are valid) and use estimators from Group II. To evaluate whether there may be a bias from weak instruments, first we run the OLS regressions for the lagged consumption _{it}_{−1}:
_{i}

The Indonesian Family Life Survey (IFLS) was carried out by the RAND Corporation in conjunction with Indonesian researchers and various international agencies in 1993 (IFLS1), 1997 (IFLS2) and in 2000 (IFLS3). IFLS is a panel and the sampling scheme for the first wave (IFLS1) is determines the sample in subsequent waves. IFLS1 sampling is stratified on provinces and urban/rural location, then randomly sampled within these strata. The IFLS sample included 13 of Indonesia’s 26 provinces which contained 83% of the 1993 population. Within each of the 13 provinces, enumeration areas (EAs) were randomly chosen from a nationally representative sample frame used in the 1993 national social economic survey (SUSENAS), yielding 321 EAs with oversampling in urban areas and in smaller provinces to facilitate urban-rural and Javanese–non-Javanese comparisons. Within an EA households were randomly selected based upon 1993 SUSENAS listings and twenty (thirty) households were selected from each urban (rural) EA.

IFLS1 contacted a total of 7,730 households and obtained a final sample size of 7,224 households. In IFLS2 6751 of the original 7,224 households (93.5%) were relocated and re-interviewed. In IFLS3 the re-contact rate was 95.3% of IFLS1 households. Nearly 91% of households are complete panel households interviewed in all three waves. IFLS contains measures of smoking behavior from individuals aged 15 and above. Frankenberg and Karoly [

Explanatory variables include the number of cigarettes smoked as recorded in previous (lag) waves, which we take as a measure of the effects of past cigarette consumption on current marginal utility of cigarette consumption. For testing the rational addiction model we also included the number of cigarettes smoked in the next (lead) wave.

We included measures of cigarette and alcohol prices at the time of the interview. The IFLS formats these questions differently across waves. For IFLS2 and IFLS3, cigarette prices were constructed from individual responses to the following question: “

Other time-varying explanatory variables included a monthly income proxy (from expenditure recall data), expressed as a real value with 2,000 CPI data. To obtain a per-equivalent adult measure of consumption, all income proxy data was adjusted for family size. The natural log of age was included to control for age-related health problems from smoking. A dummy variable indicating the presence of children less than 14 years of age was also included; we assumed that individuals might moderate tobacco consumption when small children are present. We also included a dummy variables indicating employment status.

For Group II estimators (discussed above) an instrument is needed. Appropriate instrumental variables in our context will play an important role in determining past cigarette consumption (a potentially endogenous variable) but will not affect current consumption (the dependent variable) except through past consumption. For the myopic model, we instrumented the lagged consumption with lagged cigarette prices. Other dummy variables, which we consider to be proxies for wealth or economic stability, were also included as potential instruments for lagged consumption: dwelling walls are brick; dwelling floor is permanent; dwelling is owned or being bought (1/0); and individual’s religion.

For the rational addiction model, we also included one lead of tobacco price as an instrument for future cigarette consumption. We avoided lagged values of cigarette consumption as instruments for lead consumption due to concerns about serial correlation in the errors.

We first checked for endogeneity of the lagged consumption variable in the demand equation using Durbin-Wu Hausman and Hausman-Wu statistical tests. This is a Likelihood Ratio (LR) test distributed as a ^{2} with 1 degree of freedom. The value of the test was about 7.9 with a p-value of 0.005. We thus rejected the null hypothesis of exogeneity (

A further consideration is to choose estimators among the alternatives in Group II. The Pagan and Hall’s test [_{it}

A number of tests were employed to test the relevancy, validity and orthogonality requirements of the instruments. A reduced form regression of the suspected endogenous variable, the lagged consumption, on the full set of instruments was estimated using OLS. The coefficients on the instruments in first-stage least squares _{it}_{−1} equations are given in

The ^{2} shows that the models explained a high proportion (12 percent) of the variation for lagged consumption. ^{2} and Shea Partial ^{2}. A gap between the Partial ^{2} and Shea partial ^{2} in our study considerably a small, suggesting the model is well-identified [

The proposed instruments also passed the over-identification tests. The Hansen-

The effect of price was always significant at 1 percent level. Cigarette prices had a negative effect on smoking, whilst alcohol prices had a positive and significant effect on smoking, suggesting that alcohol and cigarette are substitutes. Although cigarette consumption was found to be positively associated with income, the finding was insignificant. The low and insignificant income elasticity here (0.015) is not uncommon in the contexts of pooled models and developing countries. Blecher [

The last row of the _{it}_{it}_{2} / (1–_{1}). The short-run and long-run price elasticities, evaluated at the mean, were −0.28 and −0.73, respectively. The findings that the long-run price elasticity, in absolute value, exceeds the short-run one is in line with both theoretical expectations and empirical findings. For _{it}_{it}_{3} /(1–_{1} – _{2}), and the implied discount factor and the discount rate: _{2}/β_{1}_{1}/β_{2}

This study aimed to investigate the demand for cigarettes through a myopic addiction model and to use this model to estimate the price elasticity of cigarette demand in Indonesia. Sensitivity analysis was done by examining a rational addiction model. We explored several empirical approaches using panel data and selected the most appropriate techniques given both endogeneity of regressors and behavior of the error terms (_{it}_{it}

Results suggest that variables we suspected might be endogenous are indeed endogenous in our model. The Group II estimators are appropriate for handling endogeneity. To select among the four alternatives, we adopted the Pagan and Hall statistic tests for testing unknown heteroskedasticity. Since the use of either a random- or fixed-effects model can be justified by a Wu-Hausman test, we used this test to assess whether regression parameters characterizing the random outcome variable stay constant across all cross-sectional units for all time periods [

We confirm that Indonesian smokers are myopic addicts. Estimates yield a positive coefficient on lagged consumption that is highly statistically significant in both

Price increases had a negative and significant impact on cigarette consumption. The elasticity estimates suggest that a 10 percent increase in cigarette prices would lead to a 2.8 percent decrease in cigarette consumption in the short run and 7.3 percent decreases in the long run. Although our elasticity estimates seem low given other developing country estimates, they are comparable to other pooled data studies from developed countries. Pooled data studies have shown elasticities which are consistently lower than those in time series studies. The estimates from pooled data of the impact of price on cigarette consumption in developed countries (Baltagi and Levin [

The effect of cigarette price on consumption has important implications for revenue and health policies. Increasing cigarette prices via excise taxes can control tobacco use and at the same time raise government revenue. Our study finds cigarette demand is inelastic (with a price elasticity less than one), suggesting the percentage increase in prices would always be larger than the consumption response. However, we also estimate demand to be more elastic in the long-run than in the short-run. A long-run price elasticity of −0.73 indicates that tobacco taxation can be an effective tool to reduce cigarette consumption for Indonesia. Empirical studies have shown that tax increases are regarded as the most effective tool for tobacco consumption reduction and have been especially effective among young people and people with low incomes [

Should taxes be calibrated to be revenue maximizing or calibrated to maximize the reduction in tobacco consumption? Addiction in our study is a combination of habit and preference adaptation, starting with subjective individual preferences: individual smokers reveal that they gain net utility (or satisfaction) from tobacco consumption. For public health priorities, our model indicates that “addiction” can be conditioned and does respond to incentives. So for both revenue and health policymakers, we provide a methodological innovation for analyzing how tobacco tax policy can maximize revenue given public health goal or

Our elasticity estimates suggests that price increases brought about by higher taxes would cause government revenue to increase as the proportionate change in prices would exceed the proportionate change in consumption. However, if is thought that tax policy exploits true addicts, whose demand elasticity is very high, to compensate for revenues lost to decreases in consumption among smokers who are only addicted in the technical economist’s sense, then optimal tax rates would tend to be underestimated by our economic models. Our models also may be insensitive to heterogeneity in the population of smokers with respect to elasticities. Thus, improved estimates of price sensitivity at different income levels and knowledge of the extent to which tobacco users underestimate health harm in the Indonesia context can better inform tobacco control efforts.

The tax revenue generated from this policy could logically be used to increase public health budgets as smokers impart negative externalities to non-smokers and to public health. However, given that the demand for cigarettes is more elastic in the long-run, further excise tax increases are more likely to act as a tobacco control mechanism in the long-run rather than as a constant source of government revenue. Future increase in the tax will induce some smokers to quit and prevent others from becoming regular or persistent smokers. They also will reduce the number of ex-smokers returning to cigarettes and will reduce consumption among continuing smokers. Empirical evidence from South Africa shows that a doubling of the real price of cigarettes between 1993 and 2003 would reduce consumption by a quarter in the short term [

The effects of the price of alcohol call for further research. Higher alcohol prices lead to higher cigarette consumption in our empirical specifications. This positive cross-elasticity could indicate that alcohol and cigarettes are substitutes, which might contradict evidence from developed countries for their complementarity. Our finding may be due to a lower proportion of true addicts among a low- income population. We are unable to test this assumption within our models and empirical specifications. Another natural concern is other tobacco-related substitutes. If demand for cigarettes falls, does this mean that the demand for other tobacco products will increase? Future research should gather information about potential substitution of other tobacco products, such as kreteks, bidis, and hand rolled cigarettes for manufactured cigarettes. These substitution issues have an obvious bearing on tobacco control policy and its effectiveness.

This work has been supported by the National Institute of Health/Fogarty International Center, under grant number Ro 1-TW065938. Authors would like to thanks Teh-Wei Hu, Zheng-Zhong Mao, Hai-Yen Sung, Michael Ong, Anita Lee and Eliza Tong for their comments and supports. The authors are grateful to the RAND Corporation for providing us with the IFLS data. The authors are thankful to Jon Jellema for editorial support.

Definition of variables used in the models and its descriptive statistic.

_{t} |
Current cigarette consumption (ln) | 2.207 | 0.746 |

_{it}_{−1} |
One lag cigarette consumption (ln) | 2.203 | 0.760 |

_{t} |
Current price cigarette (ln) | 4.623 | 0.701 |

_{t} |
Current price alcohol (ln) | 8.612 | 1.263 |

Ln-exp | Monthly per-capita income (ln) | 11.156 | 1.004 |

Working | 1 if working, 0 otherwise | 0.582 | 0.493 |

Ln-age | Individual age (ln) | ||

Child14 | 1 if children aged ≤14, 0 otherwise | ||

Instruments (z) | |||

_{t}_{−1} |
One lag price cigarette (ln) | 4.280 | 0.520 |

Wall | 1 if dwelling wall is brick, 0 otherwise | 0.588 | 0.492 |

Floor | 1 if dwelling floor is permanent, 0 otherwise | 0.155 | 0.362 |

Hhown | 1 if dwelling is owned/bought, 0 otherwise | 0.805 | 0.396 |

Moslem | 1 if Moslem, 0 otherwise | 0.871 | 0.335 |

Summary statistics test: endogeneity and instrumental tests.

_{it}_{−1} | |

Wu-Hausman | 7.91^{***} |

Durbin Wu Hausman (DWH): ^{2}(1) |
7.91^{***} |

a. Heteroskedasticity^{2}(11) | |

Pagan-Hall general test statistic | 66.52^{***} |

Pagan-Hall test with assumed normality | 125.30^{***} |

White/Koenker n^{2} test statistic |
75.57^{***} |

Breusch-Pagan/Godfrey/Cook-Weisberg | 148.77^{***} |

b. Overidentifying: | |

Sargan (2SLS) | 4.03 |

Basmann (2SLS) | 4.03 |

Hansen- |
4.01 |

c. Orthogonality: | |

^{2}(1) |
0.37 |

First-stage regression of the lagged (_{it−1}

Coef. | SE | |
---|---|---|

Pc_{t} |
−0.0559^{***} |
0.020 |

Pa_{t} |
0.2253^{***} |
0.012 |

Ln-exp | −0.027 |
0.014 |

Ln-age | 0.3470^{***} |
0.031 |

If Child14 | 0.2775^{***} |
0.050 |

If working | 0.0995^{***} |
0.027 |

Excluded instruments: | ||

_{t}_{−}_{1} |
−0.0033 | 0.022 |

If dwelling wall is brick | −0.0976^{***} |
0.020 |

If dwelling floor is permanent | −0.1715^{***} |
0.029 |

If dwelling is owned or being bought | −0.0744^{***} |
0.024 |

If Moslem | −0.1404^{***} |
0.029 |

Constant | −0.3265 | 0.213 |

^{2} |
0.121 | |

Shea partial ^{2} |
0.016 | |

Partial ^{2} |
0.016 | |

Test of | ||

All instruments, F(11, 4107) | 61.04^{†} | |

Excluded instruments, F(5, 4107) | 18.04^{†} |

Significance at 1% level and

5%; SE is robust standard errors.

GMM estimation results: myopic vs. rational addiction models.

Lagged consumption (_{t−1} |
0.625 |
0.509 |

Lead consumption (_{t+1} |
n.a |
0.112 [0.157] |

Price cigarette (_{t} |
−0.275 |
−0.135 |

Price alcohol (_{t} |
0.143 |
0.159 |

Per–capita income (Ln) | 0.015 [0.012] | −0.01 [0.019] |

Individual age (Ln) | −0.098 |
0.053 [0.063] |

If child14 exist | 0.048 [0.044] | 0.097 [0.071] |

If working | 0.094 |
0.029 [0.039] |

Constant | 1.089 |
0.094 [0.437] |

Observations | 5696 | 1783 |

R–squared | 0.27 | 0.34 |

Short–run price elasticity | −0.275 | −0.135 |

Long–run price elasticity | −0.733 | −0.356 |

Discount factor | n.a | 4.54 |

Discount rate | n.a | 3.54 |

Note: Robust standard errors in [brackets];

* significant at 10%;

**significant at 5%;

***significant at 1%. The short-run price elasticity is the coefficients estimates of cigarette price, _{3}_{it}_{it}_{3} /(1–_{1} – _{2}); and the implied discount factor is _{2}/β_{1}_{1}/β_{2}