Economic and Demographic Determinants of Canadian Households Use and Spending on Tobacco

ABSTRACT - This paper reports the results of an investigation into the determinants of Canadian household tobacco expenditures. The inverse hyperbolic sine (IHS) double-hurdle model was used to accommodate zero observations in the sample as well as the non-normality of the error terms. The findings indicate that the double-hurdle model which allows separation of tobacco use participation and consumption level is superior to the single-decision Tobit model. The profile of a household most likely to purchase tobacco as well as the profile of a household that is relatively a heavy tobacco user are presented.


Mohamed Abdel-Ghany and J. Lew Silver (1998) ,"Economic and Demographic Determinants of Canadian Households Use and Spending on Tobacco", in E - European Advances in Consumer Research Volume 3, eds. Basil G. Englis and Anna Olofsson, Provo, UT : Association for Consumer Research, Pages: 312-317.

European Advances in Consumer Research Volume 3, 1998      Pages 312-317


Mohamed Abdel-Ghany, University of Alabama, U.S.A.

J. Lew Silver, University of Alabama, U.S.A.


This paper reports the results of an investigation into the determinants of Canadian household tobacco expenditures. The inverse hyperbolic sine (IHS) double-hurdle model was used to accommodate zero observations in the sample as well as the non-normality of the error terms. The findings indicate that the double-hurdle model which allows separation of tobacco use participation and consumption level is superior to the single-decision Tobit model. The profile of a household most likely to purchase tobacco as well as the profile of a household that is relatively a heavy tobacco user are presented.

Tobacco is consumed in the forms of manufactured cigarettes, "roll-your-own" tobacco, cigars, and pipe tobacco. Household expenditure on tobacco constituted about 1.8% of total household spending in Canada in 1992 (Statistics Canada 1992). This expenditure is of public policy importance for three major reasons. First, federal and provincial taxes on tobacco provide a substantial amount of government revenue. How are the changes in income and demographic structure likely to affect consumer spending on tobacco? Second, due to medical and statistical evidence of the damaging effects of tobacco use, there is a societal concern regarding health matters associated with the level of tobacco consumption. Reduction of tobacco use is now an important objective of government health policy. Third, given that tobacco and alcohol are habit-forming subtances, quite often they are consumed by the same individuals. It follows that a public policy for one good cannot be independent of the other (Goel and Morey 1995).

Most of the earlier studies regarding tobacco consumption have been based on American aggregate time series (Fujii 1980; Hamilton 1972; Houthakker and Taylor 1970; Lyon and Simon 1968). The use of micro-data, however, allows examination of the effects of numerous household characteristics that are not available in aggregate time series. One of the problems associated with the use of such data, however, is that the sample often contains a significant number of observations that report zero expenditures. Consequently, econometric techniques not accounting for zero expenditures lead to biased and inconsistent estimates (Maddala 1983).

The double-hurdle model is now established as superior than the Tobit model (Tobin 1958) in dealing with the zero expenditures. It features two separate stochastic processes of participation and consumption and therefore, it allows for examining the determinants of both participation and consumption decisions and provides more meaningful insights on consumer behavior than the Tobit model (Cragg 1971). The decision to use tobacco is not only an economic decision but also is influenced by social as well as health considerations. Consequently, many individuals would not use tobacco even if their income increased substantially or the price dropped dramatically. The double-hurdle model has been applied to consumer demand and in particular the demand for tobacco (Atkinson, Gomulka and Stern 1984; Blaylock and Blisard 1992; Jones 1989a; Jones 1989b; Jones 1992).

In the above mentioned studies, the double-hurdle model has been estimated with normally distributed errors. However, a violation of such an implicit assumption is apt to cause inconsistency of the parameter estimates (Robinson 1982). Recently, Yen (1995) and Yen and Jensen (1996) have used the inverse hyperbolic sine (IHS) as an alternative transformation of the dependent variable to accommodate for non-normality distribution of error terms in the double-hurdle model.

The interpretational advantages of double hurdle models over Tobit models can not be overstated in studies of this kind. As an example of this, household income (other factors unchanged) is likely to be an important determinant in both the decision of whether or not to use tobacco and in how much tobacco to consume. But it is also likely that the signs on these effects will differC fewer smokers are likely to be realized in higher income households, while higher expenditures should be realized by higher income households if tobacco is a normal good (in the economist’s sense) in consumption or as higher income households purchase higher quality or more pricey tobacco products. A double hurdle modeling approach allows for such a divergence of the signs of the participation and consumption effects, whereas within a Tobit framework does not. Obviously, such a disaggregation of the income effect on tobacco use is of value to policy makers, as programs to curb tobacco use can either be aimed at prevention or reduction of consumption.

The purpose of this paper is to use the IHS double-hurdle model to investigate whether the decision to consume tobacco by Canadian households (participation decision) has economic and demographic determinants different from the level of consumption (consumption decision). The analysis is applied to cross-sectional data.


Source of data

Data for this study are from the Canadian 1990 Survey of Family Expenditures collected for the Family Expenditure Surveys Section, Household Surveys Division, Statistics Canada (Statistics Canada 191). This survey was designed to provide information for persons living in private households in 17 metropolitan areas of Canada. A household is defined as a person or persons living in one dwelling unit. In most cases, households of two or more are occupied by those related by blood, marriage or adoption. The survey provides extensive national household expenditure data for Canada.


The double-hurdle model features a participation equation Xa+m and an expenditure equation Yb+e such that E=Yb+e, if Xa + m>0 and Yb+e>0; =0 otherwise, where X and Y are vectors of explanatory variables, a and b are vectors of parameters, and m and e are the error terms.

The dependent variable is the yearly household expenditures on tobacco. The independent variables used in the model are outlined in Table 1. The same variables were included in both the participation and consumption equations with the following exception: In the participation equation a dummy variable was included that had a value 1 if the household received social assistance. The social assistance variable was replaced in the consumption equation by the amount of government income received by the household.



We hypothesize that there is a positive correlation between household spending on tobacco and alcohol because both alcohol and tobacco demand have been found to have a strong habit-persistence effect (Goel and Morey 1995). It is hypothesized that there is a negative relationship between years of formal education and spending on tobacco. More educated individuals should be more cognizant of the damaging effects of tobacco use on health.

Regional location of residence is hypothesized to affect expenditures on tobacco because regional dummy variables may serve as proxies for tobacco price differentials. Also, population in some regions may be tolerant of the use of tobacco due to differences in culture and tradition (Blaylock and Blisard 1992). Ethnicity is hypothesized to capture differences in cultural factors as well as tastes and preferences. Also, individuals with different occupations are hypothesized to have different tendencies of using tobacco.

It is hypothesized that as the number of children in the household increases, spending on tobacco decreases. Blaylock and Blisard (1992) suggested that as the number of children in a household increases, the possibility increases that the household includes some individuals who have allergic reactions to smoke or who object to the use of tobacco as a form of social behavior. Marital status, gender, age, income, and wealth are believed to affect the use of tobacco.

The following transformation on some variables were employed in order to arrive at the achieved estimations: Age was divided by 100, all income series were divided by $10,000, and all expenditures were divided by $1,000.

Characteristics of the Sample

Sample statistics for all variables for the entire sample and for subsamples containing current tobacco users and nonusers are presented in Table 2. Forty-six percent of the total sampled households contained current tobacco users.


Estimation of the IHS double-hurdle model was carried out using the Newton-Raphson algorithm. The parameter estimates of the model are shown in Table 3. The IHS parameter (q) is significantly different from zero. Thus the standard double-hurdle model is rejected when tested against the IHS alternative, suggesting that the assumption of normally distributed errors used in previous studies of tobacco consumption leads to misspecification of the model.



Houshold expenditure on alcohol was a highly significant determinant on both the decision to use tobacco and the level of tobacco consumption. As is well known, in some individuals both alcohol and tobacco are addictive substances, and inclusion of a measure of alcohol usage as an explanatory variable in our model in this sense provides a measure of the predisposition toward tobacco use of individuals in sampled households. As shown in Table 3, expenditures on alcohol have a large positive effect on both the probability of tobacco consumption and the level of consumption among users. In fact, other factors equal, households containing alcohol users will increase spending on tobacco products by approximately $0.18 for every additional dollar spent on alcohol.

Unlike those of most of the explanatory factors used in this study, the signs of coefficients on household income and the number of children present differed across decision steps. Other factors equal, higher income households are found to be less likely to contain tobacco users but expenditures increased with income among households where users were present. Both of these effects are statistically significant. Similarly, the number of children present in the household has almost no discernible effect on "participation" (though it is a positive effect) and a highly significant negative effect on consumption. As we said earlier, a delineation of such effects is impossible using Tobit analysis and provides strong justification for the use of double hurdle modeling in this study.

It is interesting that investment income effects on both the probability of tobacco use and on tobacco consumption are statistically insignificant while increases in government transfer income appear to lead to (highly significant) increases in tobacco consumption. A priori one would expect wealthier households to be less likely to contain tobacco usersC notice from Table 3 that the sign of investment income in the participation equation is negativeCand given a supposition that tobacco is a normal good, expect the coefficient of all income variables in the consumption or expenditure equation to be positive. One supposition for the negative and insignificant measured effect on investment income in consumption is that wealthier households are, on average, those with more highly educated individuals. As the variables indicating higher levels of education have negative and highly significant coefficients, we suspect that the investment income effect is confounded with these higher education effects.



By the same token, the measured effect of transfer income on consumption may to some extent be confounded with the effects of the lower-level education measures. One might expect individuals with less education to have higher levels of consumption and these individuals are more likely to be receiving public assistance income. In Table 3 it is seen that transfer income is positive and highly significant, an expected result, and the effects of variables, Education 2 and Education 3, are seen not to be significantly different from that of the least educated reference group.

The remaining explanatory variables in the consumption and participation equation are primarily used as controls for effects found to be important in prior studies of this subject (see, for example, Blaylock and Blisard, 1992). And signs and significance of many of these effects are as might be expected a priori and are consistent with results in earlier works. We control for within-country regional effects, for example, using dummy variables for six provinces, finding a significant difference between the province effects shown in Table 3 and the reference province (British Columbia) only in the two indicated instances. Similarly, neither gender, occupation nor ethnicity appear to have much bearing, if any, on tobacco use or consumption in these data.

As shown in Table 3, the more important control variates are found to be those having to do with traits of individual consumers, actual or potential. Clearly, the number of adults present in the household and marital status have rather large and significant effects on bth participation and consumption. These effects are all positive as well, a possible indication that tobacco users in the same household tend to reinforce behaviors in one another.

The effects of age reported here must be interpreted with care. As we said earlier, age was divided by 100 in the estimation stage of this study. Such rescaling is often necessary in maximum likelihood estimation to ensure convergence of the iterative numerical procedure, especially when there is a mixture of continuous and qualitative measures on the right-hand-side of the model. Thus, to correctly evaluate the effect of age on, say, tobacco consumption, one must make this evaluation at Age/100. Because of the inclusion of the quadratic age-squared effect, the correctly evaluated effect of age on consumption is then, 8.839 - 11.985 (Age/100), or the partial derivative of consumption with respect to age. Evaluated at an age of 44 years, the approximate mean age of tobacco consuming individuals in our data, this effect is, approximately, $3.57. Keep in mind, however, that this effect must be weighted by the change in the probability that a household is a consumer when age changes (approximately, 0.008 in this sample), so an approximate measure the complete effect of a one-year increase in age on tobacco consumption in this sample of households is 0.02856, or about $28.56 per household per age-year. This, if correct, would certainly not be good news to officials administering the Canadian health care system, as it translates into an approximate increase of just over eleven packs of cigarettes per year per household other factors constant and assuming a per pack price of cigarettes of $2.50. This extended inference is made, of course, assuming that all users in the sample are cigarette smokers, which, undoubtedly, they are not.


The findings of this study indicate that the transformation of the dependent variable using the inverse hyperbolic sine (IHS) to accommodate for the non-normality distribution of the error terms is warranted. The results also lend credence to the superiority of the double-hurdle model over the single decision Tobit model. The results clearly illustrate the importance of a two-step approach to tobacco consumption modeling.

Parameter estimates of the double-hurdle model showed that the effects of some variables such as income and the presence of children in the household were different in each decision step. It provided more information than would have the Tobit model regarding the unique role of each variable in participation and consumption decisions. This additional information may be valuable for understanding consumer behavior in the tobacco market and for developing public awareness programs.

The profile of a household most likely to purchase tobacco is a household that purchases alcohol, has less educated household head, less income, receives government social assistance, contains married couples, the head of the household is a female and is Canadian born, has larger number of adults, and is located in Manitoba or Saskatchewan. A profile for a household that is relatively a heavy tobacco user includes the following characteristics: a household that uses alcohol, has more income, receives more government social assistance, contains more adults, less children, with married couple, located in Atlantic, Quebec, or Ontario, with a household head who has less education, is Canadian born or belongs to Ethnic Group 2, and has an occupation that is not in the teaching field.


Atkinson, A. B., J. Gomulka and N. H. Stern (1984), Household Expenditure on Tobacco 1970-1980: Evidence From the Family Expenditure Survey, ESRC Programme on Taxation, Incentives and the Distribution of Income, London School of Economics, Discussion Paper No. 57.

Blaylock, J. R. and W. N. Blisard (1992), "U.S. Cigarette Consumption: The Case of Low-Income Women," American Journal of Agricultural Economics, 74, 698-705.

Cragg, J. G. (1971), "Some Statistical Models for Limited Dependent Variables With Applications to the Demand for Durable Goods," Econometrica, 39, 829-844.

Fujii, E. T. (1980), "The Demand for Cigarettes: Further Empirical Evidence and Its Implications for Public Policy," Applied Economics, 12, 479-489.

Goel, R. V. and M. J. Morey (1995), "The Interdependence of Cigarette and Liquor Demand," Southern Economic Journal, 62, 451-459.

Hamilton, J. L. (1972), "The Demand for Cigarettes: Advertising, the Health Scare and the Cigarette Advertising Ban," Review of Economics and Statistics, 34, 401-410.

Houthakker, H. and L. D. Taylor (1970), Consumer Demand in the United States 1929-70, Cambridge: Harvard University Press.

Jones, A. M. (1989a), "A Double-Hurdle Model of Cigarette Consumption," Journal of Applied Econometrics, 4, 23-39.

Jones, A. M. (1989b), "The UK Demand for Cigarettes 1954-86, A Double-Hurdle Approach," Journal of Health Economics, 8, 133-141.

Jones, A. M. (1992), "A Note on Computation of the Double-Hurdle Model With Dependence With an Application to Tobacco Expenditure," Bulletin of Economic Research, 44, 67-74.

Lyon, H. J. and J. L. Simon (1968), "Price Elasticity of the Demand for Cigarettes in the United States," American Journal of Agricultural Economics, 50, 888-895.

Maddala, G. S. (1983), Limited Dependent and Qualitative Variables in Econometrics, Cambridge: Cambridge University Press.

Robinson, P. M. (1982), "On the Asymptotic Properties of Estimators of Models Containing Limited Dependent Variables," Econometrics, 50, 27-41.

Statistics Canada (1992), Family Expenditures in Canada 1992, Household Surveys Division, Statistics Canada, Ottawa, Ontario, Canada.

Statistics Canada (1991), 1990 Survey of Family Expenditure Public Use Microdata File. Family Expenditure Survey Section, Household Savings Division, Statistics Canada, Ottawa, Ontario, Canada.

Tobin, J. (1958), "Estimation of Relationships for Limited Dependent Variables," Econometrica, 26, 24-36.

Yen, S. T. (1995), "Alternative Transformations in a Class of Limited Dependent Variables Models: Alcohol Consumption by U.S. Women," Applied Economics Letters, 2, 258-262.

Yen, S. T. and H. H. Jensen (1996), "Determinants of Household Expenditures on Alcohol," The Journal of Consumer Affairs, 30, 48-67.



Mohamed Abdel-Ghany, University of Alabama, U.S.A.
J. Lew Silver, University of Alabama, U.S.A.


E - European Advances in Consumer Research Volume 3 | 1998

Share Proceeding

Featured papers

See More


Making Sense of Spontaneity: In-The-Moment Decisions Promote More Meaningful Experiences

Jacqueline R. Rifkin, Duke University, USA
Keisha Cutright, Duke University, USA

Read More


The Experiential Advantage: A Meta-Analysis

Evan Weingarten, University of California San Diego, USA
Joseph K Goodman, Ohio State University, USA

Read More


O12. When do People Waste Time? Testing a Mechanism for Parkinson’s Law.

Holly S Howe, Duke University, USA
Tanya Chartrand, Duke University, USA

Read More

Engage with Us

Becoming an Association for Consumer Research member is simple. Membership in ACR is relatively inexpensive, but brings significant benefits to its members.