The Effect On Tobacco Consumption Of Advertising Bans In Oecd Countries


Tobacco consumption data for 22 countries for 27 years were analysed and the six countries with ADVERTISING BANS found to have INCREASED consumption.

This document contains the main part of the article. The Appendix is in a separate 36 kbyte document. You can click on any heading below, to skip straight to any part of the whole article.

[Published as Stewart, M J, 1993. The effect of advertising bans on tobacco consumption in OECD countries. International Journal of Advertising. 12, 155-180.]

The Effect on Tobacco Consumption of Advertising Bans in OECD Countries

Michael J Stewart

Abbey Management Services, London, UK


Data on annual tobacco consumption per adult, and real price, in 22 out of the 24 OECD countries for the 27 years from 1964 to 1990 are presented. 414 of the 594 figures are based on private final consumption expenditure on tobacco in current and constant prices submitted by national governments' statistical offices and published in OECD National Accounts. For countries and years where the OECD has not published this data, sources from each country have been used to complete the data set.

By 1990, six out of the 22 countries had implemented a ban on all forms of tobacco advertising. Two simple analyses of the consumption data show no negative effect on consumption. A regression equation explaining consumption in all countries over all years, taking into account price, the changing age profile of each country, long term trends, and any advertising ban, is reported. This indicates that the average effect on per capita tobacco consumption of advertising bans has been a small increase. This increase is not quite statistically significant, but clearly refutes the belief that advertising bans have appreciably reduced consumption. Reasons why an increase might in fact be expected are discussed.


Although restrictions on tobacco advertising exist in all 24 OECD countries, only six countries had legislated for a complete ban by 1990. These are: Iceland (first full year in force 1972), Norway (1976), Finland (1979), Portugal (1984), Italy (1984), and Canada (1989). (Italy actually introduced a ban in 1962, but it was not effectively enforced before the Act of 1983.)

Advocates of a European Community-wide ban on tobacco advertising have given the impression that tobacco consumption has been reduced in those countries which have banned advertising.


The primary data source used for this analysis is a single publication, National Accounts Detailed Tables Volume 2, published each year by the OECD (e.g. OECD, 1991a). In this the components of private final consumption expenditure are itemised by category, one of which is tobacco, both in current prices and in constant prices. Dividing expenditure in current prices by expenditure in constant prices, gives a tobacco current price index. If this is divided by the Gross Domestic Product (GDP) price index, we get a tobacco real price index.

The expenditure figures at constant prices are themselves an index of tobacco consumption, but are expressed in each country's own currency. For comparability across countries, the expenditure figures at 1985 prices are divided by the retail price per 20 cigarettes (assumed to contain 20 grams of tobacco), in local currency in 1985, to yield a consumption figure in tonnes. This figure for each year is then divided by the estimate for the population aged 15 and over in that country in that year from United Nations data sets (1991).

A Worked Example

To illustrate these calculations consider Canada in 1985 and 1986. OECD 1991a (page 11) gives private final consumption expenditure in millions of Canadian dollars at current prices as 5922 and 6433 for the two years, and 5922 and 5497 at 1985 prices. Thus consumption fell by 7.2 per cent, but the money spent rose by 8.6 per cent, so the price of tobacco must have risen by 17.0 per cent. OECD 1991b (page 139) gives the GDP price index in 1985 as 100.0 and in 1986 as 102.3. If tobacco prices rose 17.0 per cent while general prices rose only 2.3 per cent, then the real price of tobacco rose by 14.4 per cent. For comparability across countries, real prices of tobacco have been converted to 1990 US dollars, but this does not affect the percentage changes from one year to the next. The tobacco real price figures for Canada used in this study, and shown in the Appendix (see Table 5) [36 kbytes], were $1.76 in 1985 and $2.01 in 1986, which thus reflect the 14.4 per cent increase in real price implied by the OECD National Accounts data.

While total consumption fell by 7.2 per cent, the adult population, according to United Nations (1991), rose from 19,920,000 to 20,159,000, so the per capita consumption fell by 8.3 per cent from 297.3 1985 Canadian dollars to 272.7. For convenient comparison across countries, these figures can be converted to grams of tobacco by dividing by the 1985 price of a gram of tobacco. Industry sources indicate a typical retail price in 1985 was 1.827 Canadian dollars for 20 cigarettes i.e. .09135 per gram. Dividing 297.3 and 272.7 by this conversion factor gives 3255 and 2985, the per capita tobacco consumption figures used in this study, and shown in Table 4 in the Appendix.

Note that the OECD figures really provide an index of tobacco consumption and of real price. It would have been possible to use the figures arbitrarily scaled to, say, 100.0 in 1990. The scaling actually used, based on an estimate of 1985 retail price in each country, merely converts these indices into the more familiar form of grams of tobacco per adult, and the price of 20 cigarettes in 1990 US dollars, without affecting the annual percentage changes within each country.

Non-OECD Data

OECD data has been used for all countries and years where it has been published. Unfortunately, some countries have failed to provide private final consumption expenditure for tobacco completely, others have not provided it as far back as 1964, and yet others are a year or so in arrears. Rather than omit these countries from the study, the best available data directly from the countries in question has been used.

In fact, before it was realised that the OECD had published data for most countries retrospectively back as far as 1964, a complete data set over the 27 years from 1964 to 1990 had been assembled in this way. This 'direct' data set has now been used only to fill in the gaps in the OECD data as follows.
Japan 1964-1989
Australia 1964-1969 and 1989
New Zealand 1964-1989
Denmark 1964-1965
Iceland 1973-1976 and 1988-1989
Ireland 1964-1969 and 1988-1989
Netherlands 1964-1968
Portugal 1964-1989
Spain 1964-1989
Switzerland 1964-1989

As the latest available OECD figures are for 1989, the 1990 figures used in the study were in all cases obtained directly from each country. The trend in the 'direct' figures between 1989 and 1990 was applied to the 1989 OECD figure in order to obtain the best possible 1990 figure.

Only two of the six countries with advertising bans are represented by non-OECD data: Iceland for six years, and Portugal for all years. The Iceland data were supplied by ATVR, the government importing monopoly. The data for Portugal from 1970 to 1990 were supplied by Tabaqueira, the Portuguese government monopoly. The figures for 1964 to 1969 were interpolated from OECD (1985). The full data set used in this analysis is shown in Tables 4 - 8 in the Appendix.


The primary conclusion of this study is based on a rather complicated form of analysis. Lest it be thought that this was because the raw data imply the opposite conclusion, two simple analyses are presented first.

Comparing consumption before and after a ban

Even calculating per capita tobacco consumption from data published by the OECD over 27 years involves a certain amount of manipulation since earlier constant price expenditures expressed in, say, 1970 prices have to be converted to the 1985 prices used in the latest volume, and adult population figures for 24 countries for 27 years obtained. So the first analysis simply looks at some raw data. The current edition of OECD (1991a), which reports private final consumption expenditure on tobacco for the years 1977 to 1989, will be available in any commercial library. Table 1 presents the completely raw data, exactly as they appear in that publication, for the three countries which have introduced an advertising ban during those years. (Portugal failed to submit data to the OECD).

Canada introduced a ban at the beginning of 1989 so there is only a single year of OECD data showing a 2 per cent drop in consumption. But consumption had been falling at an average of 3 per cent per year since 1981. Note also that the OECD tobacco expenditure at current prices rose by 16 per cent from 7258 in 1988 to 8387 in 1989, implying a 18 per cent increase in the price of tobacco. If an advertising ban had really reduced consumption by 7 per cent, as the TSB report asserts is the expected effect, we might expect to have seen a greater reduction than 2 per cent.

Finland banned advertising during 1978, yet the OECD figures show that consumption from 1979 to 1989 averaged 3644, about 7 per cent above the 1977 figure.

Italy introduced an effective ban on advertising in 1983, yet the consumption during 1984 to 1989 averaged 4116, about 4 per cent above the 1982 figure.

These observations merely indicate that if advertising bans do reduce tobacco consumption by a perceptible amount, then some relatively complicated analysis is needed to show that they do, since it is not obvious from the raw consumption data.

Comparing ban countries with the rest of the OECD

To estimate the effect of an advertising ban we need to compare actual consumption with what we would expect consumption to have been without any intervention. A simple expectation would be that per capita consumption would have followed the same pattern as in the 16 OECD countries without bans. Table 2 shows the consumption in the ban countries expressed as a percentage of the consumption they would have had if consumption had increased or decreased by the same percentage as in the rest of the OECD.

Only in Canada was consumption below expectation, and that may have been due more to the real price being 18 per cent higher than in the year before the ban. The population weighted average of all the figures in the table is 103.3. So, as a simple matter of arithmetic, the average adult in OECD countries with a tobacco advertising ban consumed 3.3 per cent more than if consumption trends in those countries had followed those in the rest of the OECD.

It is not suggested that this is a satisfactory way of estimating the effect of advertising bans on consumption, but the table does suggest that if consumption in these countries was significantly reduced by the bans then some explanation is called for as to why these countries increased their consumption relative to the rest of the OECD.


In order to obtain the best estimate of the effect of advertising bans on consumption, a series of pooled cross- sectional time series regression models of per capita tobacco consumption in each of the 22 countries, in each of the 27 years, were estimated. A 'dummy variable', having a value of one if a ban had been in force in that country in that year, and otherwise a value of zero, was one of the explanatory variables. The regression coefficient for this variable would thus represent the effect of the advertising ban, whilst the other explanatory variables would allow for the effect on consumption, in both ban and non-ban countries, of such factors as the price of tobacco and the age profile of the population.

The form of the regression equation was that of a single global constant multiplied by a series of multiplicative factors each representing the influence of a particular determinant of tobacco consumption. The model was estimated by non-linear regression using the algorithm due to Marquardt (1963). The use of non-linear regression permits a flexible structure in which each factor may be dependent on any number of parameters, and each parameter can be either estimated from the data or fixed at a value estimated from outside the model. By making small changes to the computer program, a particular parameter such as the price elasticity can be allowed to be different in each country, or constrained to be the same.


This analysis was conducted on data spanning the 27 years from 1964 to 1990. This is considerable period, and it should be remembered that the world was appreciably different in 1964. It was the time of Khrushchev, de Gaulle, Johnson, Wilson, Franco, Salazar, Papandreou, Erhard and Nehru; the time of the Vietnam War and Beatlemania, the year Gaston Roelants won a gold medal at the Tokyo Olympics. More to the point, in most OECD countries cigarettes were advertised on television, plain cigarettes outsold filter tipped, and no country yet required health warnings on packets or in advertisements. Per capita disposable income was lower in Japan than in Spain, and Portugal's was only $2,200 in 1990 US dollars. Inflation in the OECD as a whole was 2.6 per cent, unemployment less than 4 per cent.


In order to get the best estimate of the effect of an advertising ban, we need to determine what other factors affect consumption, estimate the sensitivity or 'elasticity' of consumption to each variable, and use the estimate to allow for the effect the variable would have had on consumption both before and after an advertising ban. A regression equation of the type described allows all of these to be done simultaneously. The following variables were examined to see how they affected consumption.

Tobacco real price

The private final consumption expenditure figures from the OECD provide a direct measure of changes in the actual price paid per gram of tobacco. By dividing these figures by the GDP price index in each country we get a measure of 'real' price changes.

Most of the annual change in tobacco price is due to changes in taxes of one form or another, and therefore reflects political decisions. In some countries, taxes have been kept more or less in line with inflation, and so there are only small changes in real price. In several countries with government monopolies, actual prices have remained unchanged for as long as eleven years. This will result in a gradual reduction in real tobacco price as general prices rise. This may well then be followed by a large rise, such as the 23 per cent increase in real price in Japan in 1976. In other countries, governments may sharply increase taxes, ostensibly for the purpose of reducing consumption on health grounds. Examples of such increases in real price occurred in 1976 in Finland (+24 per cent), in 1982 in Ireland (+16 per cent) and Italy (+12 per cent), in 1983 in the USA (+16 per cent) and Australia (+11 per cent), in 1986 in New Zealand (+33 per cent), in 1988 in Greece (+24 per cent), and in 1989 in Canada (+13 per cent). Such dramatic increases may herald a sustained policy of increasing taxes ahead of inflation as in the USA, Canada, New Zealand, Australia and Norway, or real prices may subsequently be allowed to fall again as in Ireland and Italy.

In about half of OECD countries the real price of tobacco was greater in 1990 than in 1964, but only one country, New Zealand, has increased prices in line with disposable incomes. In all other countries the number of minutes a worker must work in order to buy 20 cigarettes has fallen, in 11 countries by more than a half. Under such circumstances one might expect that price elasticity would have become less over the period.

The age profile

When comparing tobacco consumption over time or between countries it is usual to divide consumption in tonnes by the adult (i.e. aged 15 and over) population. As a measure of smoking intensity this is an improvement on dividing by the total population, as those under 15 do not consume much tobacco. But it assumes that all members of the adult population, whether 15-19, 40-44, or 80+, consume at the same rate. This is implausible, since both young adults, and the elderly, are known to smoke less. A country in which age groups smoke at constant rates which differ from group to group will show changes in the consumption per adult if the proportions of its population in different age groups alter over time.

To make an appropriate adjustment for this effect we need to know the relative consumption rates at different ages. These figures will vary somewhat from country to country, and also from year to year, and accurate information on them is not to be had. As a first approximation therefore, figures have been used relating to the UK, averaged over the period 1966 to 1981 (Wald and Nicolaides-Bouman, 1991), shown in Table 3. These weights were then applied to the proportion of the adult population in each age group, in each year, in each country. These 8,316 population figures were obtained from United Nations (1991).

The resulting adjustment factors are shown in theAppendix (Table 8). The population profiles of all OECD countries except Ireland changed between 1964 and 1990 in such a way as to reduce the average consumption per adult even if adults of each age had not changed their consumption. This is because an increasing proportion of adults are now reaching the ages at which they reduce their tobacco consumption. The effects are not large, but would have caused reductions in per adult consumption rates in the advertising ban countries ranging from -1.1 per cent in Iceland, to -3.7 per cent in Portugal.

Disposable income

The effect on consumption of changes in real per capita private final consumption expenditure (from OECD 1991b) in 1990 US dollars (shown in Table 6 in the Appendix) was examined. Whereas the real price variable exhibits considerable variation, both up and down, real income has tended to follow a relatively smooth upward trend, with only minor perturbations during recessions. This tends to make it difficult to disentangle the effect of income from that of all the other slowly changing determinants of smoking.

In addition, the general pattern of consumption in the whole OECD data set, is that as real per capita income rises from the $3000 or $4000 1990 US dollar levels typical in most EC countries in 1964 to around the $8000 level, tobacco consumption rises, presumably as more women, teenagers, and pensioners can afford to smoke more. As income rises beyond this level, it may be that societies start to worry about the health effects, and an increasing proportion of the middle classes realise they can give up smoking without too much trouble, so that consumption thereafter declines with increasing income.

Different countries are at different stages of economic development and therefore of this trajectory of tobacco consumption. Thus the richest countries such as Canada and the USA have spent most of the 27 years going down, the poorest such as Greece and Portugal have spent them mainly going up, and the middling countries such as Austria, Belgium, West Germany, and the Netherlands, have gone up and then come down. There are nevertheless sufficient exceptions to this rule for it to be inadequate as the basis of a global model covering all OECD countries. France is an obvious exception which has behaved as if it were much poorer than it is. And all the other English speaking countries, UK, Ireland, Australia and New Zealand, started their declines earlier than their disposable incomes warranted.

For these reasons therefore no effect of disposable income per se was detected, and the variable is not explicitly included in the regression model.

Unemployment rate

Unemployment rates were obtained from OECD (1991c, 1991d and 1991e), and are shown in Table 7 in the Appendix. Absolute unemployment rate was not found to explain tobacco consumption, but there was a suggestion that the annual increase in the unemployment rate caused a reduction in consumption. The effect was not statistically significant, and so is not included in the final model.

Female workforce participation

The proportion of the female population aged 16-64 going out to work, which rose substantially in many countries, was not found to have any effect on tobacco consumption.

Other variables

As well as the influence of economic and demographic variables, there is a host of 'cultural' variables which go to make up the attitude a society has towards smoking in general. Each step towards the anti-smoking society, from prohibition of smoking on buses to anti-smoking teaching in schools, from requiring a health warning in tobacco advertisements to politicians avoiding smoking on television, may individually have little effect on a particular year's tobacco consumption, but they contribute to a gradually increasing social pressure on people not to smoke, and thus ultimately consumption in reduced.

To represent this effect and also the positive effect on consumption of gradually increasing prosperity at the earlier stages of economic development, the model allowed each country to have its own individual trend. This was a two parameter, quadratic, trend, i.e. the trend itself was not required to be constant over the whole 27 years; it could be steeply up at the beginning, levelling off to a peak and then declining towards the end. The peak could be at any time during the period, indeed the formulation allows any smooth curve over the 27 years. The only constraint is that it be smooth, with no large changes between consecutive years, since it is there to represent the combined effect of all variables that change slowly.

Intrinsic consumption level in each country

The variables described above are intended to explain variation in per capita tobacco consumption in each country over time. They are not expected to explain why the average consumption varies across countries. For example, the average consumption over the 27 years in Ireland was 25 per cent higher than in Spain, despite the real price of tobacco being almost three times as high, and average disposable income being lower.

Accordingly the model contains a parameter for each country to represent its intrinsic level of tobacco consumption.


The regression model considered most satisfactory by the author will be described and the results from it presented. Then a number of variations on the 'base' regression will be described, and how those variations affected the estimated effect of advertising bans reported.

The base regression had an R-squared of 99.21 per cent, i.e. over 99 per cent of the variation in the 594 consumption figures about the global mean of 2748 grams per adult per year was explained. The standard error of estimate was 85 grams, which means that two thirds of the residuals (the difference between the actual consumption figure and the model's prediction of that figure) were less than 85 grams, or about 3 per cent.

Real price of tobacco

This factor incorporated a parameter for each country's price elasticity, and a parameter allowing a constant percentage change per year. This latter parameter was significant (t=3.1, where the 't-value' is the regression coefficient divided by its standard error) and indicated a 2.6 per cent reduction in price elasticity per year, averaged across all countries. Thus if a particular country was estimated to have a price elasticity of -0.5, that would be the price elasticity applying in the middle year (1977), with price elasticity trending from -0.67 in 1964 to -0.33 in 1990.

A preliminary regression had shown that the price elasticities estimated for Australia and Portugal were not statistically significant, and as the parameters were positive, they were set to zero and not estimated in the base regression. The price elasticity estimates, with their standard errors in parentheses, for the other 20 countries, in 1977, are shown below. To obtain the 1990 elasticity they should be multiplied by 0.66.

Canada         -0.37 (0.12)   
USA       -0.29 (0.03) 
New Zealand    -0.25 (0.25)   
Japan     -0.18 (0.03) 
Austria        -0.34 (0.32)   
Belgium   -0.61 (0.24) 
Denmark        -0.29 (0.50)   
Finland   -0.45 (0.31) 
France         -0.23 (0.11)   
W Germany -0.54 (0.10) 
Greece         -0.35 (0.13)   
Iceland   -0.32 (0.97) 
Ireland        -0.30 (0.18)   
Italy     -0.39 (0.08) 
Netherlands    -0.69 (0.07)   
Norway    -0.49 (0.41) 
Spain          -0.16 (0.06)   
Sweden    -0.45 (0.43) 
Switzerland    -0.83 (0.27)   
UK        -0.55 (0.10)



Two parameters were estimated for each country, and as they have no direct interpretation, unlike the price elasticity parameters, they are not tabulated here. For illustration, the two figures for Canada were -0.010769 (0.001873) and -0.000613 (0.000242). The trend factor was calculated as: 1 - 0.010769y - 0.000613y²

where y is the year minus 1977, which ensures that the factor has the value of 1 in 1977. So the value of this factor in 1990 would be: 1 - 0.010769 x 13 - 0.000613 x 169 = 0.756

Thus the model would expect, other things being equal, that consumption in 1990 would be only 75.6 per cent of what it was in 1977 because of the combined effect of many gradually changing variables for which there are no data to include in the model explicitly.

Intrinsic consumption level in each country

A single parameter for each country was estimated. This represents the level of expected consumption in each country, relative to one, arbitrarily chosen as the USA, if identical conditions apply to all countries. The numerical values are of little interest, depending as they do on the scaling of the index of tobacco consumption derived from the OECD private final consumption expenditure data. They varied between 0.43 for Portugal and 1.41 for the Netherlands.

The age profile

This factor was calculated from the 14 parameters in Table 3. The parameters were fixed at those values, not estimated by fitting to the consumption data. The values of this factor, which are shown in Table 8 in the Appendix, varied between 0.954 for Sweden in 1986 (when the average age of an adult was 38), and 1.059 for Japan in 1970 (when it was 29).

The constant

The global constant for the regression was 4003, with a standard error of 11. This represents the expected consumption in the USA if all the other factors were exactly 1. It is of no interest in itself, but is needed to multiply by the product of all the explanatory factors in order to get the actual consumption level predicted by the model for each year for each country.

Advertising ban

This parameter was estimated as 3.80, with a standard error of 2.57. In other words, for a year in a country in which an advertising ban was in force, the model expects per capita consumption to be 3.8 per cent higher than if no ban is in force, after allowing for all the other factors. The 't value' is 1.48. This is less than 1.65, which is the critical value for statistical significance at the 95 per cent level, so we cannot say that advertising bans have been proved to increase consumption.

Another way of interpreting this result is to say we can be 95 per cent confident that the true effect is within 1.96 times the standard error of the estimated coefficient. This gives upper and lower confidence limits of +8.84 per cent and -1.24 per cent. On the basis of this regression, there is only one chance in 40 that the advertising bans reduced consumption by more than 1.24 per cent.

The base regression in graphical form

What this regression has done is to take the actual pattern of consumption in each country over the 27 years, and break it down into components: the effects due to price, to the age profile, to the long-term trend, and, in the ban countries a 3.8 per cent increase during the years of the ban. The difference between the product of the four explanatory factors, and the actual consumption, is the unexplained residual. Figures 1 to 6 [30k bytes] show all these variables, stacked on top of each other, for the six ban countries. Each layer of the sandwich shows only the variation in the variable from its lowest value during the 27 years to its highest, i.e. they are graphs with 'suppressed zeroes'. An idea of the vertical scale can be gained from the height of the advertising ban column, which in each case represents the estimated 3.8 per cent increase.

Technical note on autocorrelation

One of the assumptions on which the regression formulae are based is that the residuals are uncorrelated. The degree of autocorrelation is measured by the Durbin-Watson statistic. The Durbin-Watson statistics for the Ordinary Least Squares (OLS) version of the above regression equation, calculated separately for each country, averaged 1.06. (The advertising ban coefficient was 4.08 (2.06)). This indicates the presence of autocorrelated residuals, which would mean that the indicated standard errors would be under-estimates of the true ones.

Accordingly, a Generalised Least Squares (GLS) regression was run in which the simultaneously estimated first-order autocorrelation coefficient, for each country, was used to adjust the expected dependent variable, as in Judge et al., (1988), page 406. These autocorrelation coefficients varied from 0.03 in Ireland to 0.85 in Norway. The Durbin-Watson statistics from the GLS regression averaged 1.78, with a lowest value of 1.13. The critical Durbin-Watson statistic for a regression with 27 periods and 5 parameters is 1.08. (This regression involved 110 parameters, or 5 per country.) Thus after this adjustment for autocorrelation, there is no significant autocorrelation in the residuals from any of the 22 countries. The regression standard errors, which are greater in the GLS regression than the OLS one, can therefore be interpreted with confidence. All the detailed results presented above are from the GLS regression.


The precise formulation of the regression described above was a matter of judgment, and other analysts might consider slightly different formulations to be more satisfactory. Partly to address this issue, and partly to show how much the estimated effect of advertising bans depends on the precise formulation of the model, a number of variations on the base model were estimated. In each case, only a single change from the base regression has been made.

Global price elasticity

In this regression, instead of allowing a different price elasticity to be estimated and used in each country, a single global value was estimated. This was -0.312, with a standard error of 0.018 (t=17.2). As with the base regression, there was a downward trend in the elasticity, in this case of -2.9 per cent (0.9) per year. Thus the best estimate of average tobacco price elasticity in the OECD varied from -0.407 in 1964 to -0.182 in 1990. R-squared fell from 99.21 per cent to 99.09 per cent. The estimated effect of the advertising bans was scarcely changed, at +3.4 per cent (2.6).

No price effect

In this regression, the price factor was removed completely, so that the regression took no account of price at all. This reduced the R-squared from 99.21 per cent to 98.64 per cent. The estimated effect of the advertising bans was still positive, at +2.5 per cent, with an increased standard error of 3.1. That removing the allowance made for price movements has this effect is expected, because governments that introduce advertising bans tend also to increase real tobacco prices ahead of other countries. If price is not allowed for, the reduction in consumption due to these price increases may be falsely attributed to the advertising ban.

This regression was not a plausible representation of reality, as the extreme statistical significance of the price elasticity in the previous regression shows that price has an important negative effect on consumption. But it does demonstrate that the base result is not dependent on implausibly high price elasticities offsetting a substantial reduction in consumption due to advertising bans.

No age profile effect

The elaborate procedure for adjusting consumption to reflect changes in the age compositions of countries may give the impression of a factor introduced in order to obtain a desired result. So a regression was estimated with this factor completely excluded. In fact it made very little difference, with the estimated effect of advertising bans slightly greater, at +3.9 per cent (2.6), without the age profile adjustment.

No weighting used

In any cross-sectional regression the question arises as to how to weight the errors in each country to get the best global estimates. The 22 countries in the data set have 1977 adult populations varying from 167,510,000 in the USA to 160,000 in Iceland. This is a ratio of 1047 to 1. Not only do the data for the larger countries represent many more people, they are probably measured more accurately as larger countries can afford more elaborate government statistical offices. For these reasons one would wish to give more weight to the USA's data than to Iceland's. On the other hand, if weights in proportion to populations are used, the influence of smaller countries on the results would be negligible.

As a compromise, weights proportional to the square roots of the populations have been used. If the populations of the countries are considered as different-sized samples from a single population then the use of the square roots of the sample sizes as weights can be shown to be the most efficient statistical estimator (Hanushek and Jackson 1977, page 152). Wallace and Silver (1988) use square roots of populations as weights in a regression across the 50 states of the USA.

So far as the ban countries are concerned, this procedure yields weights as follows: Canada 1.032, Finland 0.478, Iceland 0.098, Italy 1.614, Norway 0.435, and Portugal 0.644. These weights have been used for all the regressions reported so far. For comparison, the base equation was re-estimated with equal weights for each country. The estimate of the effect of advertising bans became +1.0 per cent with a standard error of 2.3.

Population weights used

For completeness, the base regression was repeated with each country's population as its weight. This raised the estimate of the effect of advertising bans to +4.6 per cent, with a standard error of 4.2.


A proposal to ban tobacco advertising throughout the European Community is currently under discussion. If we want to know what the likely effect on consumption of such a ban would be, the best way is to see what has happened in countries, similar to EC member states, that have already introduced bans. The OECD includes all EC member states, and 11 other, rather similar countries. Membership of the OECD involves each country's government statistical office submitting national accounts in detailed standardised format, and such data includes consumers expenditure on tobacco at current and constant prices. No more authoritative data on tobacco consumption exists.

When that data is analysed in as neutral and unbiased a way as possible, it does not show any negative effect of advertising bans on tobacco consumption. Indeed it suggests, but does not prove, that they may have had the opposite effect to that intended.

To the layman it seems plausible that advertising bans might reduce consumption, and plausible that they might not have any effect. But that they would actually increase consumption seems unlikely. But there is at least one mechanism by which they might have that effect.

It is not now possible to ban only tobacco advertising; what would be banned would be tobacco advertisements incorporating a health warning. (The mean year of introduction of health warnings in cigarette advertisements in the 16 OECD countries without advertising bans was 1979.) Whereas tobacco advertisements themselves may have no effect on tobacco consumption (and they are not designed to do so, only to attract smokers from other brands), the health warning is there to remind smokers of information already received from numerous sources concerning the risks associated with smoking. Health warnings in advertisements may thus have some deterrent effect. If so, then abolishing 'tobacco advertising' would have a positive effect on consumption. The minimum of 10 per cent of the space of each advertisement now required, in the EC, to be devoted to a health warning is the equivalent of millions of ECUs of on-going anti-smoking advertising across the Community.

That countries which continue to have government-backed health warnings widely disseminated through tobacco advertisements have reduced their tobacco consumption, relative to countries which have stopped doing so, is not really so implausible.




Hanushek, E.A. and Jackson, J.E. (1977) Statistical Methods for Social Scientists. New York: Academic Press

Judge, G.J., Hill, R.C., Griffiths, W.E., Lutkepohl, H., and Lee, T. (1988) Introduction to the Theory and Practice of Econometrics. New York: Wiley.

Lee, P.N. (1975) Tobacco consumption in various countries. London: Tobacco Research Council.

Marquardt, D.W. (1963) An algorithm for least-squares estimation of nonlinear parameters. Journal of the Society for Industrial and Applied Mathematics, 11, 431-441.

OECD (1985) Measuring Health Care 1960-1983. Paris: OECD.

OECD (1991a) National Accounts Detailed Tables, Volume II 1977-1989. Paris: OECD.

OECD (1991b) National Accounts Main Aggregates, Volume I 1960-1989. Paris: OECD.

OECD (1991c) National Statistics 1960-1989. Paris: OECD.

OECD (1991d) Labour Force Statistics 1969-1989. Paris: OECD

OECD (1991e) OECD Economic Outlook 49 July 1991. Paris: OECD

Senbaishi (1978) The History of the Monopoly, Volume 5. Tokyo: Japan Tobacco & Salt Public Corporation.

Senbaishi (1990) The History of the Monopoly, Volume 6. Tokyo: Japan Tobacco Inc.

Tabacalera (1990) Series historicas de consumo de tabaco elaborado 1957-1988. Madrid, Spain: Tabacalera.

United Nations (1991) Machine-readable Data Sets - The 1990 Revision Global Population Estimates and Projections, IBM-PC diskette, Sex and Age, 1990. New York: United Nations.

Wald, N. and Nicolaides-Bouman, A (1991) UK Smoking Statistics, 2nd Edition. Oxford: Oxford University Press.

Wallace, T.D. and Silver, J.L. (1988) Econometrics: An Introduction. Massachusetts, USA: Addison-Wesley.

Biographical Note

Michael J Stewart is a consultant specialising in econometric analysis of the sales effects of advertising of consumer goods. He read psychology and physics at Liverpool University. After four years in the marketing services department of H J Heinz (UK) Ltd, and eight years managing the market analysis unit of Beecham Products Ltd, he set up Abbey Management Services in 1981. He has published papers in the Journal of the Market Research Society, Admap and The International Journal of Advertising, and presented papers to ESOMAR. His special interest is incorporating the, often under-estimated, effects of weather in econometric models.

Table 1 - Tobacco expenditure at constant prices published by OECD

          Canada         Finland        Italy
          million        million        billion
          1985           1985           1980
          dollars        markkaa        lira

1977      6356           3410           3493
1978      6202           3397 100       3469
1979      6482           3548 104       3770
1980      6552           3490 103       3857
1981      6694           3319  98       3914
1982      6683           3462 102       3946
1983      6414           3605 106       3990 100
1984      6107           3785 111       4121 103
1985      5922           3462 102       4215 106
1986      5497           3660 108       4294 108
1987      5257           3889 114       4016 101
1988      5290 100       3850 113       3994 100
1989      5185  98       4017 118       4038 101

Back to text reference to Table 1

Table 2 - Consumption trends after bans compared with rest of OECD

Canada         -0.37 (0.12)   USA       -0.29 (0.03)
New Zealand    -0.25 (0.25)   Japan     -0.18 (0.03)
Austria        -0.34 (0.32)   Belgium   -0.61 (0.24)
Denmark        -0.29 (0.50)   Finland   -0.45 (0.31)
France         -0.23 (0.11)   W Germany -0.54 (0.10)
Greece         -0.35 (0.13)   Iceland   -0.32 (0.97)
Ireland        -0.30 (0.18)   Italy     -0.39 (0.08)
Netherlands    -0.69 (0.07)   Norway    -0.49 (0.41)
Spain          -0.16 (0.06)   Sweden    -0.45 (0.43)
Switzerland    -0.83 (0.27)   UK        -0.55 (0.10)

Back to text reference to Table 2

Table 3 - Relative tobacco consumption per adult by age group

     Non-Ban   Consumption relative to Non-Ban Countries    
     Countries ICE  NOR  FIN  ITA  POR  CAN 

1971 3220      100  
1972 3309       99
1973 3407       96
1974 3457      107
1975 3411      104  100
1976 3470      104   96
1977 3397      100   97
1978 3330      103   95  100
1979 3383      103   98  102
1980 3332      105  103  101
1981 3306      111   92   96
1982 3224      113   85  102
1983 3139      121   88  108  100  100
1984 3135      122   91  113  103   98
1985 3112      110   99  104  105   98
1986 3041      114  106  112  108  101
1987 3004      117  108  120  102  103
1988 2907      117  112  122  104  107  100
1989 2904      112  113  127  105  107   97
1990 2868      112  115  124   98  112   92

Average        109  100  111  104  104   95

Back to first text reference to Table 3

Back to second text reference to Table 3Back to start of article

Back to Publications page

Back to Home page

Copyright © 1996 Abbey Management Services ( Email:mikestrt@dircon.co.uk)

FORCES is supported solely by the efforts of the readers. Please become a member or donate what you can.

Contact Info
Forces Contacts
Media Contacts
Links To Archived Categories

The Evidence
Inside Forces
About Forces
Book case