Cigarette and Tobacco Consumption: Have Anti-Smoking Policies Made a Difference?
by PETER BARDSLEY , NILSS OLEKALNS PETER BARDSLEY and NILSS OLEKALNS [*] The consumption of cigarette and tobacco products in Australia is modelled using the rational addiction theory of Becker and Murphy, augmented by data on advertising, regulatory intervention, and demographic factors. Over the past 35 years, price (including tobacco taxes), real income, and demographic effects explain most of the variation in tobacco consumption. Advertising by tobacco companies has had a relatively small direct effect on consumption. Work-place smoking bans and health warnings on cigarette packs have had a relatively minor impact, while antismoking advertising and bans on electronic media advertising have had no detectable direct effect. I Introduction In this paper we examine cigarette and tobacco consumption in Australia over the period from 1962/63 to 1995/96. Our concern is to understand the effect of government policies aimed at reducing cigarette and tobacco consumption. [1] In order to isolate the effects of government intervention, it is necessary to understand all the other factors that have also influenced consumption. The approach that we adopt is based on Becker and Murphy's (1988) rational addiction model. In Australia, as in many other countries, smoking has been a major public health concern in recent decades. It is the leading cause of preventable morbidity and premature death in Australia and is responsible for over half of all drug-caused hospital bed days (Winstanley et al. 1995). Approximately 15 per cent of all recorded deaths in Australia can be attributed to tobacco related causes such as lung cancer, coronary heart disease and chronic bronchitis and emphysema.[2] Habitual smokers (those who consume at least one packet a day) have their average life expectancy reduced by around six years (Taylor 1993). In response to these public health concerns, a variety of anti-tobacco public policies have been implemented in Australia. These policies include a mixture of taxation measures, provision of public health information and health warnings, and prohibitive actions. In addition, public funds have been used for a number of years to sponsor anti-smoking advertising campaigns. The amount of money devoted to these campaigns has been quite small relative to the amount spent on pro-tobacco advertising by the tobacco companies themselves. [3] Nevertheless, the anti-smoking campaigns have achieved a reasonably high profile, particularly since the gradual phasing in of pro-tobacco advertising bans. Advertising by tobacco companies is now almost totally prohibited, the only exceptions being point-of-sale advertising and incidental advertising associated with tobacco companies' sponsorship of major sporting events. Other government interventions include a requirement that cigarette and tobacco packaging carry prominent warnings of the dangers of smoking, and a ban on smoking in many public places (including on aircraft, on public transport, and in all government office buildings). Fiscal initiatives have also been important: around 65 per cent of the retail price for a pack of cigarettes is now accounted for by the Federal excise duty and State licence fees (Winstanley et at. 1995). As a consequence, the real price of cigarette and tobacco products has increased by over 175 per cent over our sample period. In this paper we focus on the effect of these policies and quantify the relative contribution that each intervention has made to changes in cigarette and tobacco consumption. We also consider the possible countervailing effect of advertising and promotion by tobacco companies. Tobacco producers have argued that industry advertising has little impact on the total size of the market, and that the main effect of their advertising is on market share (Stutchberry 1992). If this claim is true, then the public health case for advertising bans would be considerably weakened. To study these questions we use aggregate data for the cigarette and tobacco market in Australia spanning the past 35 years. The data include information on prices, taxes and consumption as well as advertising expenditure by tobacco sellers, expenditure on anti-smoking advertising and education, and major regulatory changes. We also consider the influence of the changing age structure of the population over the sample period. The analytic framework that we use is Becker and Murphy's (1988) model of rational addiction. This model takes into account, in a straight-forward way, the intertemporal dependence of preferences that is characteristic of addiction. It is sufficiently general to accommodate a wide range of dynamic behaviours associated with addictive consumption (Chaloupka 1992). The Becker-Murphy model has now been successfully applied to a number of addictive commodities, including tobacco, gambling and caffeine (Chaloupka 1991, Mobilia 1993, Olekalns and Bardsley 1996). The fact that tobacco-delivered nicotine is physiologically addictive reinforces the potential usefulness of the Becker-Murphy framework to model cigarette and tobacco consumption. [4] Our empirical results confirm that the rational addiction model is an attractive approach to modelling tobacco consumption. [5] We find significant price, income and demographic effects, both in the short and long terms, as well as evidence of the backward and forward consumption linkages predicted by the theory. We are able to detect and quantify the effect of industry advertising and of the major public health policy interventions. Taking into account all these factors, we are able to track actual consumption quite closely over the 35-year period of our data. Our paper is organized as follows: the Australian tobacco market is described in Section II. The Becker and Murphy model is outlined in Section III and we describe our data, estimation technique and results in Section IV. The policy implications are discussed in Section V. II Tobacco in Australia The history of tobacco consumption in Australia shows several distinct phases (see Figure 1). [6] Consumption prior to 1939 was relatively static with a slight increase in the years following World War I offset by a fall in demand during the Depression. Consumption jumped noticeably under war-time conditions between 1940 and 1945, when cigarettes were readily available to the military, and then grew strongly in the post-war decades. Consumption peaked in the 1960s, and then began to fall quite rapidly after about 1975. There are a number of factors which may have contributed to this fall in consumption. During the late 1950s and early 1960s it gradually became apparent that there is a connection between smoking and health, culminating in the 1964 statement by the US Surgeon General. In response to these health concerns, State and Federal taxes on tobacco increased steadily. [7] As a result, the real price of cigarettes increased, rising quite rapidly after 1982. As well as increasing taxes, the government has progressively restricted advertising by the tobacco industry. [8] A ban on cigarette and tobacco advertising in the electronic media was phased in over the period September 1973 to September 1976, with allowances made for 'accidental or incidental' exposure of tobacco products as a result of the sponsorship of sporting or cultural events. By April 1996, even this loophole was closed. [9] Over time, the prohibition has been extended to other forms of advertising. The print media were targeted in legislation enacted in 1989 while advertising on billboards, illuminated and other outdoor signs has been banned since the beginning of 1996. Point-of-sale advertising is still allowed. Government funded anti-smoking campaigns date back to the period 1972 to 1975 when a Federal government initiative, the 'National Warning Against Smoking,' was implemented. Since 1983, anti-smoking campaigns have been conducted with funding from both State and Federal governments. The most notable of these campaigns have been carried out by the Quit organization, which provides a variety of resources (primarily educational) to encourage a reduction in tobacco consumption. This is reinforced by high-profile print and television advertising campaigns, and sponsorship of sporting programs. However, these expenditures have been quite small relative to the amounts that have been spent by the tobacco companies on advertising. In real per capita terms advertising expenditure has, on average, been more than 20 times higher than anti-smoking expenditure. In addition to placing restrictions on advertising, the government has directly regulated cigarette smoking in various other ways. From 1972, a mandatory health warning has been required on cigarette packs. Federal workplaces have been required to be smoke-free since 1986, and smoking has been banned on aircraft since 1987. Demographic effects are also likely to have had an important influence on consumption. A casual examination of Figure 1 suggests that the decline in consumption over recent years may be due in art to the diminishing size of the population who became heavily addicted to tobacco during and after World War II. Over our sample period, the proportion of the total adult population who were of adult age in 1940 has fallen from being half the population to now comprising less than 10 per cent of the population. In addition, we know that smoking propensities vary across ages and genders and that this has changed over time. In 1945, 72 per cent of men and 26 per cent of women in Australia smoked; by 1992 these proportions were 28 per cent and 24 per cent respectively (Winstanley et al. 1995). III The Theoretical Model Becker and Murphy (1988) have proposed a simple model for the consumption of addictive commodities, which is characterized by linear dynamics for an addictive stock variable, and rational forward looking behaviour by a utility maximizing agent. The consumer's instantaneous utility is [U.sub.t] = U([c.sub.t], [S.sub.t], [y.sub.t]), where [c.sub.t] is current consumption of the addictive good, [S.sub.t] is a stock variable measuring the degree of addiction, and [y.sub.t] is a composite non-addictive consumption good. [S.sup.t] enters the utility function as a proxy for the health effects induced by the addictive consumption history and is assumed to evolve over time according to a simple investment equation: [S.sub.t+1] = (1 - [delta]) [S.sub.t] + [c.sub.t] (1) where [delta] is the rate at which the addiction depreciates over time. Using the non-addictive composite commodity as the numeraire, and denoting the price of the addictive good as [P.sub.t], the consumer is subject to a lifetime budget constraint [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (2) where [beta] is the discount factor. The individual's problem is to choose [[{[c.sub.t]}.sup.[infinity]].sub.1], [[{[S.sub.t]}.sup.[infinity]].sub.1] and [[{[y.sub.t]}.sup.[infinity]].sub.1] to maximize the discounted stream of current and future utilities [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (3) subject to the budget constraint and the dynamics of the addictive stock. Following Becker and Murphy, we approximate the utility function in a neighbourhood of the steady state by a quadratic in order to linearize the estimating equation and write [U.sub.t] = [[alpha].sub.y][y.sub.t] + [[alpha].sub.c][c.sub.t] + [[alpha].sub.s][S.sub.t] + [U.sub.yy]/2 [[y.sup.2].sub.t] + [U.sub.cc]/2 [[c.sup.2].sub.t] + [U.sub.SS]/2 [[S.sup.2].sub.t] + [U.sub.yS] ([y.sub.t][S.sub.t]) + [U.sub.cS] ([c.sub.t][S.sub.t]) + [U.sub.yc] ([y.sub.t][c.sub.t]). (4) The marginal utilities with respect to the addictive good and the addictive stock are: [U.sub.c] = [[alpha].sub.c] + [U.sub.cc] [c.sub.t] + [U.sub.cS] [S.sub.t] + [U.sub.yc] [y.sub.t] (5) [U.sub.S] = [[alpha].sub.S] + [U.sub.SS] [S.sub.t] + [U.sub.yS] [y.sub.t] + [U.sub.cS] [c.sub.t]. (6) If we assume that current and past consumption of the addictive good has no effect on the marginal utility derived from consuming the composite good, these equations imply. [10] [c.sub.t] = [[xi].sub.0] + [[xi].sub.1][c.sub.t-1] + [beta] [[xi.sub.1] [c.sub.t+1] + [[xi].sub.2] [P.sub.t] + [[xi].sub.3] [P.sub.t-1] + [beta] [[xi].sub.3] [P.sub.t+1]. (7) where [[xi].sub.1] [greater than] 0, [[xi].sub.2] [less than] 0, and [[xi].sub.3] [greater than] 0. The positive sign for [[xi].sub.1] follows from the reinforcement effect of addictive consumption ([U.sub.cS] [greater than] 0). Positively signed coefficients on past and future prices are less intuitive given the complementarity of consumption across periods. The explanation is that [[xi].sub.3] measures the marginal effect on current consumption holding consumption in periods t - 1 and t + 1 constant. This implies that other factors must be operating to offset the effects of past and future price changes on, respectively, past and future consumption. The relevant complementarity therefore relates to these offsetting effects, and gives rise to the positive value for [[xi].sub.3]. Equation (7) nests a number of different behaviours. A non-addicted consumer responds only to current price and perhaps to other exogenous factors such as income or advertising. An addicted but myopic consumer ignores the implications of current decisions for his future consumption, and only responds systematically to current and past information. An individual who is addicted and rational takes these implications into account. [11] The consumption dynamics are revealed by the roots of the characteristic equation [beta][[xi].sub.1] [[lambda].sup.2] - [lambda] + [[xi].sub.1]. (8) The roots of this equation, which should be real and positive, are given by [[lambda].sub.1] = 1 - [(1 - 4[xi].sub.1] [beta][[xi].sub.1]).sup.0.5]/2[[xi].sub.1] (9) and [[lambda].sub.2] = 1 + [(1 - 4[[xi].sub.1] [beta][[xi].sub.1]).sup.0.5]/2[[xi].sub.1] (10) A stable solution exists if [[lambda].sub.2] is greater than unity and [[lambda].sub.1] is less than unity. The smaller root describes the effect of anticipated shocks to future consumption, and the larger root relates to past consumption (Sargent 1987, chap 9.8; Chaloupka 1991). The long-run elasticity of demand at price P is [[epsilon].sup.lr] = [([[xi].sub.2] + [[xi].sub.3] + [beta][[xi].sub.3])/(1 - [[xi].sub.1] - [beta][[xi].sub.1])]P/c (11) where c is the steady-state level of consumption given by C = [[xi].sub.0] + ([[xi].sub.2] + [[xi].sub.3] + [beta][[xi].sub.3])P/1 - [[xi].sub.1] - [beta][[xi].sub.2] (12) (i) Advertising and Information As in many other countries, there has been a long-standing debate in Australia about the effect of advertising on aggregate tobacco consumption. The most contentious issue is whether advertising by tobacco companies expands the total size of the market or whether it merely encourages already committed smokers to switch their allegiance across brands. The available international evidence is mixed, with various studies finding that advertising increases cigarette consumption while other studies find little or no impact (Duffy 1996). Some Australian research has concluded that there is no evidence of a systematic link between cigarette consumption and advertising (Johnson 1986) although Clements et al. (1985) concluded that the ban on electronic media advertising in Australia reduced consumption by around 2 to 3 per cent per annum after 1976. McLeod (1986) has since suggested that this effect was probably short lived. [12] In this paper, we assume that advertising by the tobacco industry changes the marginal utility of consumption, making tobacco consumption more attractive at the margin. We assume that second-order effects are negligible, so that this can be modelled as a shift in the utility function parameter [[alpha].sub.c] in Equation (5). It can be shown that this parameter enters in a linear way into the term [[xi].sub.0] of Equation (7). [13] It enters other terms of this equation only through any effect it may have on the marginal utility of income; we assume this to be negligible. We thus conclude that the effect of advertising is to shift the constant term in Equation (7), leaving the other terms stable. It is Likely that advertising may have some lingering effect. We allow for this by introducing a decaying stock of advertising. It is not so clear how to account for the effect of anti-smoking advertising and health education. Some of this advertising may act on the marginal utility of consumption in a way that is similar to industry advertising. [14] However it is likely that an important effect is to change consumers' perceptions of the long-term health costs of smoking. We assume that this can be captured by a shift in the utility function parameter [alpha].sup.S] in Equation (6). Once again, we assume that second-order effects are negligible. Just as for [[alpha].sub.c], this parameter enters Equation (7) by shifting the constant term [[xi].sub.0]. We thus conclude that in either case, whether it affects the marginal utility of consumption or the perception of long-term health effects, the effect of antismoking advertising and education can also be accounted for by an additive shift in Equation (7). (ii) Demographic Effects and Aggregation The rational addiction model applies to a representative consumer. Our data are generated by a population with a changing demographic structure, and we know from survey evidence that gender and age differences exist in the propensity to consume cigarette and tobacco products (Australian Bureau of Statistics 1989/90). As a result, changes in the demographic structure of the population may be expected to lead to variations in measured per capita consumption. Furthermore, the individuals in the population change from year to year. In particular, the generation of smokers who became heavily addicted to tobacco during and after World War II gradually disappear from our population as they grow older and die. There are two issues to consider. First there is a sample selection issue caused by people entering or dropping out of the population due to births, deaths and migration; then there is an aggregation issue caused by the heterogeneous demographic composition of the population. Let us put aside for a moment the first problem and consider the aggregation issue. Let I be a demographic class in the population at year t. For example, I might be females between 20 and 30 years of age who are alive at date t. Let [n.sup.I] be the number of individuals in class I, and let n = [[sigma].sub.I][n.sup.I]. We assume that all individuals in a demographic class have identical utility functions, and that individuals in different classes differ only in the shift parameters [[alpha].sub.c] and [[alpha].sub.S] of their utility functions. They may, for example have been exposed to different histories of advertising and health education, and this advertising and education may have been targeted to different demographic groups. The coefficients in Equation (7), other than the intercept, are by this assumption constant across individuals and across classes. The intercept may vary between demographic classes; we denote it by [[[xi].sup.I].sub.0]. We now take the basic single agent relationship equation (7), and average it over all the decision making agents present in the population at date t. This will give us our aggregate equation [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (13) where the [[[xi].sup.I].sub.0] are demographic effects, the [X.sub.j] are influences on consumption such as incomes and advertising, and [epsilon] is a mean zero error recording price forecasting and consumption planning errors. Thus, under these assumptions we can incorporate demographic population effects by an additive shift in the constant term. Now let us turn to the second problem, and consider the effect of people entering and leaving the population. Migration [15], births and deaths mean that the population in a particular age/gender group in the current period is not necessarily the same as the corresponding group of people who made consumption decisions in the previous period, nor is it the exact same group who will make consumption decisions in the future. For the younger age groups, net migration means an increasing number of group members across adjacent time periods. Death means that numbers are likely to be decreasing across adjacent time periods for the older age groups. Where net migration is the dominant factor, measured consumption in period t -- 1 is too low relative to current consumption. This is because new arrivals form part of the pool of current consumption but their consumption history is not recorded in the data for previous periods. Where mortality is the dominant factor, measured consumption in period t + 1 will be too low r elative to current consumption since deaths mean that the consumption histories of some individuals are recorded in the current data but their consumption decisions will not feature in the data in subsequent years. We correct for these complications by introducing adjusted measures of lead and lagged consumption: [16] [c.sub.+] = (1 - [[sigma].sub.I] ([n.sup.I]/n) ([[n.sup.I].sub.+] - [n.sup.I]/[n.sup.I]) ([[c.sup.I].sub.+] - [c.sub.+]/[c.sub.+])) [c.sub.+] [c.sub.-] = (1 - [[sigma].sub.I] ([n.sup.I]/n) ([[n.sup.I].sub.-] - [n.sup.I]/[n.sup.I]) ([[c.sup.I].sub.-] - [c.sub.-]/[c.sub.-])) [c.sub.-] Notice that the adjustment depends on the relative size ([n.sup.I]/n) of the demographic class I, on the net growth rate ([[n.sup.I].sub.+] - [n.sup.I]/[n.sup.I]) (due to deaths and migration) of the class, and on ([[c.sup.I].sub.+] - [c.sub.+]/[c.sub.+]), which measures how different consumption is in this class compared to the population average. Once this adjustment is made, we have the aggregate estimating equation which incorporates both demographic effects and the effects of agents entering and leaving the population [c.sub.t] = [[sigma].sub.I] [[[xi].sup.I].sub.0] [n.sup.I]/n + [[xi].sub.I] [c.sub.t-1] + [beta][[xi].sub.t][[xi].sub.t+1] + [[xi].sub.2] [P.sub.t] + [[xi].sup.3] [P.sub.t-1] + [beta][[xi].sub.3] [P.sub.t+1] + [[[sigma].sup.k].sub.j=1] [[theta].sub.j] [X.sub.j] + [[epsilon].sub.t]. (14) For future reference, note that the implied steady state consumption is c = [[sigma].sub.I] [[[xi].sup.I].sub.0] [n.sup.I]/n + ([[xi].sub.2] + [[xi].sub.3] + [beta][[xi].sub.3]) P + [[[sigma].sup.k].sub.j=1] [[theta].sub.j][[X.sub.j]/1 - [[xi].sub.1] - [beta][[xi].sub.1] (15) IV Data, Estimation Technique and Results (i) The Data A number of questions have to be resolved before using equation (14) for estimation. The first is what to include in the list of exogenous variables. As well as income, we use information on pro- and anti-smoking advertising, a measure of the age structure of the population and data on various government interventions. Specifically, the variables are: * [y.sub.t]: real per capita household disposable income * [propyoung.sub.t]: the proportion of the adult population under the age of 45 * [workban.sub.t]: the proportion of indoor workers reporting smoking bans at their primary place of work * [ads.sub.t]: real per capita advertising expenditure in the main media by tobacco companies * [anti.sub.t]: real per capita expenditure by anti smoking organisations * [warning.sub.t]: a dummy variable taking the value 1 from the date when health warnings were first printed on cigarette and tobacco packaging * [tvban.sub.t]: a dummy variable taking the value 1 from the date at which cigarette and tobacco advertising was banned in the electronic media. The case for including an income measure is fairly self-evident. However, since there is a possibility that cigarette and tobacco products are inferior goods, we have no prior expectation for the sign of the coefficient on income. The inclusion of the age structure of the population can be justified by survey evidence indicating that the proportion of cigarettes consumed in Australia is smaller for young people than it is for old people (Australian Bureau of Statistics 1989/90). [17] Therefore, we would expect to see a negative coefficient on this variable. [18] Workplace smoking bans mean that workers must delay consumption, or in some cases consume only at the inconvenience of leaving the building. This might be expected to act directly on the addictive stock. Bans may also be interpreted to some degree as an increase in the effective price of consumption, taking into account time and inconvenience. We expect that this coefficient would be negative. Finally pro- and anti-smoking advertising would be expected to enter the regression with, respectively, a positive and a negative coefficient, while the health warnings and the electronic media ban should both enter with negative coefficients. A decision must also be made about the treatment of the depreciation rate for the addictive stock, [delta]. One option is to restrict [delta] to some hypothesized value. The alternative is to allow [delta] to be unrestricted. Restricting the value of [delta] imposes a constraint on the coefficients in equation (14). This may help to offset some of the problems associated with any multicollinearity caused by our use of time-series data. In our estimation, we impose the value [delta] = 1. This means that the entire addictive stock is depleted within the course of one year. This rapid depreciation is consistent with the available physiological evidence. [19] Since the imposition of the restriction means that the respective coefficients on lagged and future price are equal to zero, its validity is relatively easy to test (and we do so later in the paper). [20] The possibility of multicollinearity also leads us to impose another restriction, this time on the value of [beta]. This parameter links the coefficients on past and future consumption (see equation (7)). We experimented with a range of values without any there being any significant change to the results. Those that we report are for a value of [beta] = 1/(1.02). Once again, we test the validity of this restriction later in the paper. Equation (14) is estimated using time-series data for Australia. The data are yearly and the sample period extends from 1962/63 to 1995/96. Figure 2 shows plots of the quantitative variables used in the analysis and a detailed description of all of the data and its sources is in Appendix A. Here, we provide some brief details. The consumption of cigarettes and tobacco is measured by real expenditure (in 1989/90 dollars) and is converted to a per capita basis by dividing by the adult population (aged 15 years or more). Price is measured by the implicit price deflator for cigarette and tobacco products. The data on advertising by cigarette companies come from the Commercial Economic Advisory Service of Australia. The data comprise annual expenditure on advertising in the main electronic and print media for the category "smoking accessories". [21] Unfortunately, there is not a corresponding series relating to the price of advertising. This means that the advertising data have to be converted into real terms using the consumer price index. The adult population is then used to derive per capita series. Since the effect of advertising is likely to accumulate over time, we estimate the stock of advertising, [[alpha].sub.t] using [[alpha].sub.t] = (1 - [[delta].sup.[alpha]]) [[alpha].sub.t-1] + [[alpha].sub.t] where [[delta].sup.[alpha]] is the depreciation rate and [[alpha].sub.t] is advertising expenditure in period t. Measuring the advertising stock requires an estimate of the initial stock. We set this equal to zero. Data on the expenditure of anti-smoking organizations come from Hill and Scollo (forthcoming). These data relate to the estimated expenditure on anti-smoking education aimed at adults. This includes educational and counselling activities and the activities of the Quit campaign. [22] The data are converted to a real, per capita basis using the consumer price index and the adult population. To be consistent with our treatment of the tobacco companies' advertising activities, we estimate the stock of anti-smoking advertising, [z.sub.t], given by [z.sub.t], = (1 - [[delta].sup.z]) [z.sub.t-1] + [z.sub.t], where [[delta].sup.z] measures the depreciation rate and [z.sub.t] is expenditure. Choices have to be made about the values of, respectively, [[delta].sup.[alpha]] and [[delta].sup.z]. A sensitivity analysis over a variety of different values for [[delta].sup.[alpha]] and [[delta].sup.z] (the results are in Appendix B) reveals that the stock of pro-tobacco advertising is always a significant explanator of consumption regardless of the assumed depreciation rates while the anti-tobacco advertising stock is only significant for two choices of depreciation rates, but in both of these cases the coefficient has the incorrect sign. [23] The results that we discuss below assume that [[delta].sup.[alpha]] = [[delta].sup.z] = 0.75. [24] Dummy variables are used to capture any significant effects on consumption arising out of the health warnings on cigarette and tobacco packaging and the advertising prohibitions. These dummies take the value 1 starting from the period in which the various bans were first introduced. Workplace smoking bans are measured by data taken from Chapman et al. (1997) on the percentage of indoor workers reporting smoking bans at their workplace. This set of explanatory variables is quite comprehensive. However, there are some possible influences on cigarette and tobacco consumption that are not amenable to direct measurement. In particular, an argument can be made that there has been a heightened awareness of the health risks associated with smoking over the past 30 years. Whilst anti-smoking education programs could have had a role in this process, this growing antismoking sentiment may also have been part of a gradual attitude shift towards more health oriented behaviours, e.g., jogging, dieting etc. Later in the paper, we experiment with including a linear time trend in the analysis which may go some of the way towards capturing this change in community attitudes. (ii) Estimation Technique A generalized methods of moments (GMM) instrumental variables estimator is used to obtain the parameter estimates. GMM is appropriate in this instance since the use of leads and lags gives rise to the possibility of serial correlation in the residuals. The need for an instrumental variables procedure arises from the endogeneity of past and future consumption; three lags and three leads of [P.sub.t] are used as instruments for [c.sub.t-1] and [c.sub.t+1]. Prices are suitable instruments in this context since the optimal consumption in any period depends on the past history and expected future course of prices. We also include leads and lags of the Federal and State excise tax rates as additional instruments. Movements in these tax rates are responsible for a significant proportion of changes in the price of cigarette and tobacco products and announcements of changes in these rates are well publicized. Hansen's (1982) J test of the over-identifying restrictions implied by the instruments is used as a portmantea u specification test of the model (see Davidson and MacKinnon 1993, p. 616). (iii) The Results The results are reported in Table 1. [25] The first column shows the estimates for the basic Becker-Murphy model, having only the contemporaneous price and lagged and lead consumption as explanatory variables. The results for the expanded model, which includes the other exogenous influences on consumption, are in the second column and third columns. The parameters on lagged and lead consumption have been restricted as described above. These results provide strong support for the Becker-Murphy specification. Consumption is affected by both past and future expenditures, indicating that the cigarette and tobacco products are addictive and that the consequences of this addiction for future consumption are explicitly recognized. The contemporaneous price also has the expected sign. The results from the expanded model, shown in Table 1 as 'expanded model 1', support most of the conjectures that we made above. Consumption affected by income and the coefficient indicates that cigarette and tobacco products are normal goods. The age structure of the population matters for aggregate consumption with a younger population likely to smoke less. The workplace smoking bans and the health warnings on cigarette and tobacco packaging have both contributed to the decline in consumption. And, contrary to the arguments of the tobacco industry, advertising does expand the size of the market. However, anti-smoking advertising and the electronic media ban are both insignificant. The third column of the table, 'expanded model 2', shows the results when the two insignificant variables are omitted and these results are similar to those obtained when the variables are included. [26] All three specifications imply stable dynamics and the values for the long-run elasticities are plausible. [27] We experimented with a variety of other specifications in order to check the robustness of our results. The first of these was to estimate the expanded model allowing [delta] to be unrestricted. Although both lead and lagged price were significant, the implied long-run price elasticity was positive and the over-identifying restrictions were rejected, leading us to reject this specification. We also relaxed the coefficient restrictions on lagged and lead consumption and conducted a Wald test for the validity of the restriction. The test statistic was 0.23 which means that we cannot reject the restriction. [28] V Policy Impact Tobacco consumption in Australia peaked in 1969/70. Since that date consumption has fallen by about 60 per cent, the price of tobacco products (including taxes) has increased by 174 per cent, real income has increased by 39 per cent, and the 'baby boomers' have moved through into middle age. Government has increasingly intervened to discourage tobacco consumption. How important have these factors been in explaining cigarette and tobacco consumption? The Becker-Murphy model describes a dynamic process in which consumption adjusts over time towards a steady state value. This steady state is the long-run value at which consumption would settle if prices, income and all other influences were henceforth kept fixed through time. As policy shifts the underlying parameters, the location of the steady state will change. Movements In the observed data are thus due to a combination of drift towards the steady state and shifts of the steady state. From a policy point of view the steady state is more important than the transient adjustment dynamics, and we concentrate here on explaining how and why this steady state has shifted over time. [29] Figure 3 shows the result of simulation experiments (using Equation (15)) which illustrate how changes in the various explanatory variables have shifted the steady state consumption through time. In each case, one explanatory variable is singled out and allowed to take its actual historical values. All other explanatory variables are held frozen at their 1963 values. The upper panel in this Figure shows the cumulative effect of income changes, of price changes (including taxes), and of demographic changes. Each of these has had an important effect. Consumption has been driven up by rising real income, and it has been driven down by tax and price increases, especially during the last decade. Consumption is sensitive to the age structure of the population, increasing as the population ages. [30] The lower panel of Figure 3 shows the effect of advertising, workplace bans and health warnings. It is immediately apparent that the effect of these influences is an order of magnitude less than the effects of income, price and demographics. Contrary to tobacco industry views, industry advertising (holding prices fixed) does appear to have increased consumption. In 1972 up to 7 per cent of aggregate consumption may be attributed to advertising. However we are unable to detect any significant effect of the form of advertising. Beginning in 1973, advertising in the main electronic media was virtually banned, but the coefficient related to this event is not significant. The impact of industry advertising falls to zero after 1993, when most advertising was banned. The effect of advertising must be interpreted with considerable care. It can be argued that, as well as shifting demand, industry advertising may also increase price (by facilitating product differentiation). This price effect (which is held constant at zero in this simulation) would be expected to act in the opposite direction to the demand shift, tending to reduce consumption. Our analysis of advertising impact holds prices fixed, so it provides an upper bound on the effect of advertising. Therefore, it is not clear whether the total effect of advertising, when any price effects are included, has been to increase consumption significantly. [31] Health warnings on cigarette packs have caused a statistically significant but very small decrease in consumption. Smoking bans in public spaces and in the workplace have reduced consumption by about 5 per cent since they were introduced in 1989. Anti-smoking advertising (the Quit campaign) and education have had no detectable direct effect on aggregate consumption; we discuss below whether they may have had an indirect effect. One message is clear. Compared to price, income and demographic influences, the effect on aggregate consumption of industry advertising and of direct regulatory intervention has been small. Figure 4 shows the combined effect of all these explanatory variables. The predicted steady state tracks the actual values quite closely. Figure 5 shows estimates of short- and long-run price elasticities. These elasticities are calculated at the steady state consumption value at each date. The short-run price elasticity appears to be about --.2 throughout the first 20 years of the data, but then as prices increase and consumption falls it rapidly becomes more elastic. It is natural to ask whether this is a real effect or an artefact imposed by the choice of functional form. The functional form that we have used, following Becker and Murphy and Chaloupka (1991), is a flexible quadratic approximation to the true utility function which should fit well close to the steady state saddle point. In fact, as is clear from Figure 4, consumption throughout the period does appear to have remained quite close to the steady state value. We thus conclude that the shifting elasticity estimate is unlikely to be driven by the functional form. There is, in fact, evidence of a change in market structure consistent with a change in elasticities. Product differentiation has increased, pack sizes have increased, a series of price wars has been observed, and margins appear to have decreased (Winstanley et al. 1995). These changes are consistent with a decrease in oligopoly rent in the industry and an increase in demand elasticity. The increase in elasticity may be associated with regulatory bans on product-differentiating advertising. Figure 5 shows that the estimated income elasticity has also increased throughout the sample period. [32] VI Conclusion In this paper we have studied cigarette and tobacco consumption in Australia over the period 1962/63 to 1995/96. There is strong support for the Becker-Murphy rational addiction model of aggregate consumption over this period. Current consumption is affected by past and anticipated future consumption, and price and income are both significant. All of the sign restrictions implied by the model are satisfied. The steady state consumption implied by the model tracks actual consumption quite closely. The estimated model is consistent with quite rapid decay of the addictive stock. We cannot reject the hypothesis that the addictive stock has fully depreciated within a year. However this short-lived physical addiction, when coupled with the consumption feedback, drives a long-lived economic addiction. Long-run elasticities are around 2.7 times as large as short-run elasticities. We find that most of the variation in consumption has been driven by price (including taxes), by income and by demographic effects. Our model suggests that, other factors being held constant, consumption will rise as the population ages and as real incomes rise. This suggests that if current tobacco tax and regulatory policies are held constant, then consumption may begin to rise again in the future. The effects of industry advertising and regulatory intervention are relatively small. Advertising bans reduce consumption, but the effect is small and may be over-stated if price effects are considered. Health warnings on cigarette packs reduce consumption by a detectable but very small amount. 'Quit' anti-smoking education and advertising has had no detectable direct effect (but see below). Work-place smoking bans have reduced consumption by about 5 percent since their introduction, but there is limited scope for further reductions as around 66 per cent of the population is now subject to work-place bans. Virtually all of the reduction in tobacco consumption can thus be attributed to tobacco taxes. Income growth and demographic effects have tended to increase consumption, and direct regulatory intervention has had a very small effect. It is tempting to conclude that direct government interventions (other than tobacco taxes) have been ineffective, but we consider that such a conclusion may be premature. It could be argued that anti-smoking education and other interventions, while having only a small direct effect, may have created a climate in which government has been able to raise tobacco taxes to unprecedented high levels. It is also important to remember that we are considering only aggregate consumption. There may be particular groups within the population where these interventions are more effective than is evident in aggregate data. The implication for Australia is that if tobacco consumption is to be held at current levels, taxes on tobacco must remain high. As incomes increase and the population ages, even higher taxes may be required to counteract income and demographic effects. Regulatory restrictions of advertising and consumption can be expected to play a secondary role. Whether it is socially desirable to restrict tobacco consumption is of course another question, which we do not consider here. It would be interesting to know whether these conclusions apply to other economies and to other societies. Tobacco-related public health issues are important in Australia and other developed countries. They are even more important throughout the less developed world. If our findings apply to these societies, then income growth and demographic transition may lead to a significant worldwide increase in tobacco consumption and in tobacco-related public health problems. APPENDIX A The Data Nominal Expenditure on Cigarette and Tobacco Products National Accounts: National Income, Expenditure & Product, Australian Bureau of Statistics, cat. no. 5206.0, various issues. Implicit Price Deflator, Cigarette and Tobacco Products National Accounts: National Income, Expenditure & Product, Australian Bureau of Statistics, cat. no. 5206.0, various issues. Until 1970/71, constant price expenditure data are only available for the combined group 'cigarettes, tobacco and alcohol'. After 1970/71, separate constant price data are available for 'cigarettes and tobacco'. The pre-1970/71 implicit deflator is adjusted by multiplying by the ratio of nominal cigarette and tobacco expenditure to nominal expenditure on cigarettes, tobacco and alcohol. Consumption of Cigarette and Tobacco Products This is measured by real expenditure, calculated by dividing nominal expenditure by the implicit price deflator. Relative Price of Cigarette and Tobacco Products Implicit price deflator for cigarette and tobacco products divided by the All Groups: Capital Cities Consumer Price Index, Australian Bureau of Statistics, cat, no. 6401.0, various issues. Nominal Household Disposable Income National Accounts: National Income, Expenditure & Product, Australian Bureau of Statistics, cat, no. 5206.0, various issues. Real Household Disposable Income Nominal household disposable income divided by the consumer price index. Population Population aged 15 and over. Commonwealth of Australia Year Book, various issues, Australian Government Publishing Service. As well as giving our aggregate adult population series, to derive the various per capita measures used in the analysis, this source also gives us the age/gender breakdown in order to calculate the values for [n.sup.I] used in the adjustments we make to lagged and lead consumption. Advertising The data are taken from Commercial Economic Advisory Service of Australia (1991) and relate to advertising expenditure in the main electronic and printed media for the category 'smoking accessories', comprising cigarettes, tobacco, cigars and miscellaneous (i.e. lighters, papers, etc.). Expenditure on Anti-Smoking Activities The data are from Hill et al. (in submission). Smoking Propensity by Different Demographic Groups Our adjusted measures of lead and lag consumption,[c.sub.+] and [c.sub.-], require values for [[c.sup.I].sub.+] and [[c.sup.I].sub.-] the average consumption in population class I. We calculate this using information from an Australian Bureau of Statistics survey (Australian Bureau of Statistics 1989/ 90). This survey relates only to the State of New South Wales. However, as New South Wales is the most populous State in Australia, and as it seems unlikely that there would be wide variation in smoking propensities across States, we assume that the results of this survey can be generalized across the Australian population. This survey reports the average number of cigarettes smoked by males and females for various age groups (18-24 years, 25-34 years etc.). From this, we can calculate the proportion of total cigarette consumption accounted for by the various age/gender groups and hence we can calculate [[c.sup.I].sub.+]and [[c.sup.I].sub.-s]. APPENDIX B Sensitivity of the Results to Different Assumptions About [[delta].sup.a] and [[delta].sup.z] The table shows the significance levels at which the null hypotheses that, respectively, pro- and anti-tobacco advertising have no effect on cigarette and tobacco consumption can be rejected, for different assumed values of [[delta].sup.a] and [[delta].sup.z]. The expanded specification, equation (14), is used to generate the results. A compltete list of all the paramater estimates, t-ratios etc. can be obtained from the authors. | |
[[delta].sup.a] [[delta].sup.z] p-level p-level
[ads.sub.t] [anti.sub.t]
0.25 0.25 0.00 0.49
0.50 0.00 0.48
0.75 0.00 0.54
0.50 0.25 0.00 0.40
0.50 0.04 0.11
0.75 0.00 0.69
0.75 0.25 0.00 0.0 [*]
0.50 0.00 0.0 [*]
0.75 0.00 0.16
[[delta].sup.a] [[[epsilon].sup.lr].sub.96] J Statistic
0.25 -3.248 7.10
-3.216 7.07
-3.361 7.56
0.50 -3.285 10.37
-2.719 20.82
-3.159 7.75
0.75 -3.299 8.96
-3.414 7.25
-3.451 7.22
| | The J statistic is Hansens's (1982) test of the overidentifying restrictions. [[[epsilon].sup.lr].sub.96] is the estimated long-run elasticity in 1996. * indicates that the coefficient has the wrong sign. REFERENCES Alchin, T. (1991), 'A Survey of Econometric Studies of Tobacco Advertising,' University of Western Sydney Working Paper 91/02. Australian Bureau of Statistics (1989/90), State of Health in New South Wales 1989/90, cat. no. 4330.1, Australian Government Publishing Service, Canberra. Becker, G.S. and Murphy, K.M. (1988), 'A Theory of Rational Addiction,' Journal of Political Economy 96, August, 675-700. Becker, G.S., Grossman, M. and Murphy, K.M. (1994), 'An Empirical Analysis of Cigarette Addiction,' American Economic Review 84(3), June, 396-418. Chaloupka, F. (1991), 'Rational Addictive Behaviour and Cigarette Smoking,' Journal of Political Economy 99, August, 722-42. Chaloupka, F. (1992), 'Clean Indoor Air Laws, Addiction and Cigarette Smoking,' Applied Economics 24, 123-205. Chapman, R., Borland, R., Scollo, M., Brownson, R., Dominello, A. and Woodward, S. (1997), 'The impact of workplace smoking bans on declining cigarette consumption in Australia and the USA', paper presented at the 10th World Conference on Smoking and Health, Beijing, August. Clements, K.W., McLeod, P.B. and Selvanathan, E.A. 1985), 'Does Advertising Affect Drinking and Smoking?', Discussion Paper, Department of Economics, University of Western Australia. Commercial Economic Advisory Service of Australia (1991), Advertising Expenditure in the Main Media, St Leonards. Davidson, R. and MacKinnon, J.G. (1993), Estimation and Inference in Econometrics, Oxford University Press. Duffy, M. (1996), 'An Econometric Study of Advertising and Cigarette Demand in the United Kingdom,' International Journal of Advertising 15, 262-84. English, D.R., and Holman, C.D.J. (1995), The Quantification of Drug Caused Morbidity and Mortality in Australia, Commonwealth Department of Human Services and Health, Canberra. Glass, Richard, M. (1995), 'Reply: Caffeine Dependence Syndrome,' Journal of the American Medical Association 273(8), May 10, 1419. Hansen, L.P. (1982), 'Large Sample Properties of Generalised Method of Moments Estimators,' Journal of Econometrics 50, 203-38. Hill, D., White, V. and Scollo, M. (forthcoming), 'Smoking Behaviours of Australian Adults in 1995', Medical Journal of Australia. Hu, T., Sung, H., and Keeler, T. (1995), 'Reducing Cigarette Consumption in California: Tobacco Taxes vs an Anti-Smoking Media Campaign,' American Journal of Public Health 85(9), 1218-23. Hughes, J.R., Higgins, S.T. and Bickel, W.K. (1994), Nicotine Withdrawal Versus Other Drug Withdrawal Syndromes: Similarities and Dissimilarities,' Addiction 89, 1461-70. Johnson, L.W. (1986), 'Advertising Expenditure and Aggregate Demand for Cigarettes in Australia,' International Journal of Advertising 5, 45-58. Mcleod, P.B. (1986), 'Advertising Bans, Tobacco and Cigarette Consumption,' Economics Letters 20, 391-6. Mobila, P. (1993), 'Gambling as a Rational Addiction,' Journal of Gambling Studies 9(2), 121-51. Olekalns, N. and Bardsley, P. (1995), 'Caffeine Dependence Syndrome,' Journal of the American Medical Association 273(8), 10 May, 1417-18. _____ and _____ 1996), 'Rational Addiction to Caffeine: An Analysis of Coffee Consumption,' Journal of Political Economy 104(5), October, 1100-04. Roemer, R. (1993), Legislative Action to Combat the World Tobacco Epidemic, 2nd edn, World Health Organisation, Geneva. Sargent, T.J. (1987), Macroeconomic Theory, 2nd edn, Academic Press, San Diego. Stutchbury, M. (1992), 'Cigs Up, Smoking Down,' Australian Financial Review, 14 October. Taylor, R. (1993), 'Risks of Premature Death from Smoking in 15 Year Old Australians,' Australian Journal of Public Health 17(4), 358-64. Townsend, J. (1996), 'Price and Consumption of Tobacco,' British Medical Journal 52(1), 132-42. US Surgeon General (1988), The Health Consequences of Smoking. Nicotine Addiction. A Report of the Surgeon General, US Department of Health and Human Services, Public Health Service, Office on Smoking and Health, DHSS Publication No (CDC) 88-8406. Winstanley, M., Woodward, S. and Walker, N. (1995), Tobacco in Australia: Facts and Issues, Victorian Smoking and Health Program. (*.) We would like to thank Peta Arnott for research assistance and Michelle Scollo, Victorian Smoking and Health Program, for her extremely helpful comments, advice and assistance with some of the data used in this paper. We also thank an anonymous referee and an editor of this journal for their comments on an earlier draft and the Australian Research Council for financial assistance. Any errors in the paper are our responsibility. (1.) We do not ask whether these policies are a good idea. In this paper we do not consider the welfare economics of tobacco regulation. (2.) The figure of 15 per cent is based on English et al.'s (1995) calculation of 18 920 tobacco related deaths in 1992 and a total number of registered deaths in that year of 123 660 (Dx Database, Variable DCRQ.UN80DTHAUS) (3.) See Figure 2. (4.) The US Surgeon General declared tobacco delivered nicotine to be addictive in 1988 (US Surgeon General 1988). (5.) The usefulness of the rational addiction approach is not always accepted by health practioners. See Olekalns and Bardsley (1995) and the editorial response by Glass (1995). (6.) The data shown in Figure 1 are based on customs and excise records. (7.) Levying a tax on tobacco to try to curtail its consumption is not a new policy. For example, James I of England, concerned by the rapid spread of tobacco consumption, raised the tobacco tax from 2 to 82 pence per pound. This resulted in tobacco becoming more expensive than silver (Townsend 1996). (8.) Advertising bans are widely used throughout the world. For example, in 1993 27 countries had totally banned tobacco advertising, with another 77 countries having some form of restriction in place (Roemer 1993). (9.) With a few notable exceptions, such as the Australian Formula 1 Grand Prix. (10.) See the technical appendix to Becker and Murphy (1988) for details. (11.) As an example, consider an individual who expects consumption of cigarettes and tobacco to be relatively lower in the next period because of an anticipated price increase. This would have no implications for the current behaviour of a non-addicted or a myopically addicted individual. However, an addicted and rational individual will recognize that the price increase will raise the future cost of current consumption and will make the appropriate adjustments to current consumption. A rational consumer will also take notice of new information about future health effects. (12.) Alchin (1992) surveys this literature. (13.) See the technical appendix to Becker and Murphy (1988). (14.) Some Quit advertising has emphasized the dirty and disgusting nature of smoking. Some of this advertising has also emphasized the anti-social effect of smoking, and the disapproval of others. (15.) Migration had a significant effect on the demographic structure of the Australian population study period. (16.) The `+'("-") subscript indexes period t+1 (t-1). A more detailed description of these adjustments can be found in the working paper version of this paper, available at the Department of Economics, University of Melbourne, web site (http://www.ecom.unimelb.edu.au/ecowww/). (17.) Based on this survey, only 43 per cent of aggregate cigarette consumption is by people aged under 45 years. (18.) The same survey indicates that, on average, men consume more cigarettes than women. However, the ratio of males to females has remained fairly constant over the sample period. (19.) withdrawal symptoms from nicotine generally begin within 2 to 12 hours of cessation, peak after 2 to 3 days and last, on average, only 3 to 4 weeks. Some symptoms such as cravings and hunger or weight gain, can persist for up to 6 months (Hughes et al. 1994). (20.) Other studies also impose this restriction, e.g. Becker (1994). (21.) This refers to cigarettes, tobacco, cigars and miscellaneous (i.e. lighters, papers, etc.) (22.) Specifically, expenditure by the various Quit campaigns together with expenditure by government health departments, cancer and health charities and health promotion foundations. (23.) The sensitivity analysis is based on estimation of equation (14). These results are discussed in more detail below. (24.) A complete set of results for the other depreciation rates can be obtained from the authors. The estimated parameters for the non-advertising variables are fairly robust to these different possible values for [[delta].sup.[alpha]] and [[delta].sup.z]. (25.) Since the data may be non-stationary, some care needs to be exercised when interpreting the standard errors. (26.) The insignificance of anti-smoking advertising in Australia contrasts with recent evidence for California where an anti-smoking media campaign was judged to have reduced consumption (Hu et al. 1995). However, the scale of the advertising campaign in California was much larger than that attempted in Australia. It may be the case that there is a threshold effect in anti-smoking advertising. The insignificance of anti-smoking education in our results does not, therefore, necessarily imply that an optimal policy response is to reduce expenditure on these campaigns. (27.) The increase over time in the value for the long-run elasticity probably reflects changes in the structure of the Australian tobacco market over the sample period. We return to this point in Section IV. (28.) We also experimented by including a linear trend. This led to a rejection of the over-identifying restrictions. In addition, we used an instrument for advertising to allow for its possible endogeneity. Following Duffy (1996), the Treasury bill rate was used as an instrument (Duffy also used a price index for advertising media but this is not available for Australia). The results were very similar to those reported in Table 3 and can be obtained from the authors. We also considered a number of other different specifications. Pro- and anti-smoking-advertising were entered in levels (as opposed to a per capita basis) but the dynamic stability conditions were not satisfied and the over-identifying restrictions were rejected. We also measured the respective advertising series as flow variables but both were insignificant. In neither case were the variables statistically significant. We also used the unadjusted lag and lead consumption series in conjunction with the full set of exogenous variables. This resulted in the coefficient on future consumption being larger than the coefficient on lagged consumption. Finally, we considered possible interaction effects between anti-tobacco advertising and price (on the grounds that high prices may increase the effectiveness of the advertising) and between anti- and pro-tobacco advertising (on the grounds that the effectiveness of one might be higher when the other is low). We did not find evidence of any significant interaction effects. All of these results can be obtained from the authors. (29.) The long-run dynamics reported in the paper track the actual data very closely (see Figure 4), so it appears that the short-run dynamic effects are quite small. One can solve the differential equation to track the short-run dynamics, but as the evolution of the system is governed by a saddle point singularity, one expects difficulties with numerical stability. Simulation experiments confirm this. Starting at any date one can track the short-run dynamics for up to four or five years at most before numerical instability sets in. (30.) The Becker-Murphy model predicts an increase in consumption with age (other effects being corrected for). (31.) This question is being examined using a different data set. A full analysis of this point would also take into account the possibility that elasticities may have changed (see below), reducing the incentive to differentiate products. (32.) These increases in the elasticities towards the end of the sample are not due to instabilities in the estimated coefficients. To test for instability, the basic model was estimated with the addition of three variables, [dum.sub.t][c.sub.t-1], [dum.sub.t][c.sub.t+1] and [dum.sub.t][p.sub.t] where [dum.sub.t] is a dummy variable taking the value 1 from 1990 onwards. Only [dum.sub.t] [P.sub.t] was significant suggesting possible instability in the coefficient on price. However, [dum.sub.t] [P.sub.t] was insignificant when added to the expanded model and from this we conclude that the coefficient on price is stable. | |
GMM Estimates
Variable Basic Model Expanded Model 1 Expanded Model 2
Constant 18.285 [***] 2627.395 [***] 2217.992 [***]
[c.sub.t-1] 0.493 [***] 0.188 [***] 0.244 [***]
[c.sub.t+1] 0.484 [***] 0.185 [***] 0.239 [***]
[p.sub.t] -0.108 [***] -2.137 [***] -1.667 [***]
[y.sub.t] . 0.010 [***] 0.009 [***]
[workban.sub.t] . -9.63 [***] -0.875 [***]
[propyoung.sub.t] . -38.535 [***] -33.173 [***]
[ads.sub.t] . 1.205 [***] 2.144 [***]
[anti.sub.t] . -3.171 .
[warnt.sub.t] . -8.029 [***] -5.049 [**]
[tvban.sub.t] . -1.921 .
[R.sup.2] 0.990 0.991 0.992
J 6.620 7.218 7.233
[[lambda].sub.1] 0.797 0.195 0.260
[[lambda].sub.2] 1.230 5.224 3.924
[[[epsilon].sup.lr].sub.63] -0.797 -0.581 -0.581
[[[epsilon].sup.lr].sub.79] -0.783 -0.607 -0.575
[[[epsilon].sup.lr].sub.96] -1.970 -3.451 -3.018
| | Note: (***.), (**.) and (*.) denote significance at the, respectively, 1%, 5% and 10% levels. J is Hansen's (1982) test statistic for over-identifying restrictions. [[[epsilon].sup.lr].sub.63], [[[epsilon].sup.lr].sub.79] and [[[epsilon].sup.lr].sub.96] are the long-run elasticity estimates using, repectively, the data values for 1963, 1979 and 1996. The restrictions [delta] = 1 and [[beta].sub.1] = 1.02 * [[beta].sub.2] have been imposed. The expanded models assume [[delta].sup.a] = [[delta].sup.z] = 0.75. |
|