Relative Mortality Risk and the Decision to Smoke

Joan Costa Joan Rovira

EEE 87

Octubre, 2000

http://www.fedea.es/hojas/publicado.html

Joan Costa a,b and Joan Rovira ^b

^a LSE-Health, London School of Economics and Political Science, Cowdray House, Houghton Street, London WC2A 2AE, UK

^b Research Group on Economics of Social Policy, Departament de Teoria Econòmica, Universitat de Barcelona, Spain.

Abstract

This paper examines the role of relative mortality risk perceptions as explaining the smoking decision making under some degree of uncertainty and imperfect information. Unlike other previous studies we argue that the heterogeneity of mortality risk perceptions may justify under some circumstances the provision of public information on smoking mortality attributable risk. Using a representative survey for Spain in 1997 we find that even though mortality relative risk perceptions explain the decision to smoke, smokers do not show overestimated risk perceptions as prior research suggest. Moreover, individuals are sensible to alternative risk perception measures.

Contact Address:

Joan Costa i Font Grup de Recerca en Economia de la Política Social Dept. Teoria Econòmica Universitat de Barcelona Dalmases, 61 3er 2ª, 08017 Barcelona (Spain) Tel/Fax: 932121205 E-mail: J.Costa-Font@lse.ac.uk

1. INTRODUCTION

The decision to smoke constitutes a typical decision under some degree of uncertainty about its consequences. Although smoking has been proven to be responsible of some share of attributable mortality a share of the population still smokes. A stream of research focuses on examining the existence of systematic biases on the understanding of the risks individuals face when undertaking risky activities (Viscusi, 1984). Essentially, studies elicit subjective judgements of risks people perceive from particular activities (Lichtenstein, 1978), and typically risk associated with some particular hazard. The study of risk perceptions aims to identify the model that fits to forecast human behaviour in order to help the performance of public policy i.e information campaigns. A clear example of an application of this source of studies concerns the risks implied in the consumption of hazardous goods as fat food, cigarettes and alcoholic beverages. The appropriateness of these studies becomes clear whenever the interest is focused on examining the individual decision making under both imperfect information and the presence of uncertainty on the occurrence of some future hazard dependent on current individual behaviour.

Risk perceptions and behavioural studies applied to risks of smoking have tended to measure disease-related risk perceptions, typically lung cancer as a proxy of other smoking health risks (Viscusi, 1990, 1992 and Liu and Hsieh, 1995; Hu et al 1995; Hsieh, 1996)¹. Results from these studies conclude that smoking risks perceptions tend to be overestimated, both by non smokers and smokers. However, one of the main criticisms to this source of studies relies on the fact that they partially contend a measure of smoking risks. Its far known that smoking may cause other diseases such as other sources of cancer, lung diseases and heart attacks and its responsible of a higher morbidity rates that may rise mortality rates. Moreover, disease-related measures are often catalogued as being proxies of mortality risk measures. Previous studies suggest that there are sensible differences according to the source of risks considered (Costa, 1999). Therefore, we argue that a measure of mortality risk should be employed instead of a disease-related measure in order to capture an overall measure of all possible risks that may be influencing individual risk perceptions. Risk perceptions may be sensitive to the use of alternative measures and the decision model may as well considerably change due to this feature.

¹ in this papers, lung cancer risks perceptions are claimed to be a proxy of overall risks of smoking.

Risk perceptions are often estimated by means of probability-based measures to empirically examine information decision models. Viscusi (1990) tested weather the bayesian decision model fits to explain the smoking decision using a subjective probability measure. This measure was elicited asking in a survey the following question “ for every 100 people of a given population how many of they would acquire lung cancer due to smoking. Although, it shows some theoretical advantages² respect to alternative qualitative measures, this measure does not capture the concept of “attributable risk" typically employed in the epidemiological literature. Todate, there is still no evidence of any risk perceptions measure that is claimed to be consistent with epidemiological studies.

In this paper we report empirical evidence on the relationship between risk perceptions and smoking behaviour focusing on a probability based measure of overall mortality by means of an empirical application to Spain. Spain is a clear example where reduction of smoking prevalence constitutes a major health policy goal as its stands as the

European Union country with the highest share of smoking population. According to the existent epidemiological evidence, attributable mortality risk in Spain is of 14% from 1978 to 1992. Premature mortality represents about a third of total mortality from what it follows a loss of between 10 and 20 years of life. In this study we elicit smoking risk perceptions using a subjective relative risk perception measure of life years lost due to smoking. Moreover, we develop a discrete empirical decision model to test the role of mortality risk perception in the smoking decision.

² It permitted to compute a measure of the perceived probability of a particular hazard and then test the applicability of the bayesian information updating approach.

2. DATA AND EMPIRICAL MODEL

The data used in this study is a representative survey of the Spanish population³. The original sample was made up of 2551 questionnaires using professional sampling according to region, province and municipality. Over a half of the sample were males and 49% were middle and low class. The 35% are current smokers and the 96% have been smoking for more than a year, the remainder are non smokers (65%) from which the about 26% are ex - smokers.

The survey contains information on perceived years of life lost due to smoking ( and therefore relative mortality risk perceptions), smoking behaviour and socio-economic characteristics. The interview began by asking brief questions to focus the respondents attention on the general topic of smoking related hazards. For instance, respondents were asked whether they believe that smoking is bad for health and whether smoking was related with a several health diseases. Respondents then were asked as well to elicit the perceived life expectancy of a smoker and immediately the perceived life expectancy of a non-smoker. From this results relative mortality risks where easily estimated⁴. The reason for using this elicitation procedure is based on Slovic et al (1982). Individuals easily understand lifetime lost rather than annual risk of relatively low-probability events. Moreover, a control variable was considered to test whether respondents were well informed⁵.

MR=[E-11100 LEⁿs MR = LEs

LES

³ Data was collected through a national computer based survey conducted by a private firm (INNER Research) during June 1997. From the best of our knowledge there is no other specific survey available involving risks perceptions in Spain. Those who took part were 18 and older, both male and female.

⁴ The mortality risk (MR) would according to the following formula :

here ns LE refers to life expectancy if not smoker and ^s LE refers to life expectancy if the individual is a smoker.

The nature of the relationship between smoking and its associated risks can be contemplated as endogenous. As risk perceptions may be influenced by the smoking status, we test the existence of simultaneity using a Hausman test (Hausman ,1978). The empirical model is the following.

Assume there are two states of the world premature death or not. A rational individual would smoke if the benefits obtained from smoking (first term of the following equation) exceed the expected loses (second term of the equation) as follows:

\[(U (s m o k e) - U (d o n ^ {\prime} t)) + \pi (V - U (s m o k e)) > 0\tag{1}\]

Equation (1) can be parameterised using a linear model (2):

\[S M O K E _ {i} ^ {*} = \beta_ {0} + \beta_ {1} P M R _ {i} + \beta_ {2} X _ {I} + \mu\tag{2}\]

5 They were asked if the smokers had a higher probability of acquiring diabetes (an unrelated smoking disease). Results suggest that a 79% of the sample asked correctly this question.

where MRP (mortality risk) reflects the overall smoking loss, X is the vector of variables involving the determinants of the smoking decision, refers to coefficient vectors and is the random error.

However, as far as is not directly observable, we define a corresponding indicator variable SMOKE such that takes that value 1 if and the value 0 otherwise. Assuming a normal distribution for the random term then a probit model can be estimated.

Risk perceptions of premature mortality may be affected be the individual and socioeconomic , is represented by (3) :

\[M R P = f (\alpha , X) = \alpha_ {0} + \alpha_ {i} X _ {i} \varepsilon_ {i}\tag{3}\]

The survey database employed contains some information on socio-economic variables and lifestyles. We hypothesise that smokers would show lower risks perceptions than non-smokers and socio-economic variables such as education, income and regional variables may influence the perceptions of risk. The variables included were AGE as a set of dummy variables, HOUSEHOLD referring to the household size, GENDER, EDUCATION as a dummy variable denoting an increasing education levels. We include two regional variables BARCELONA and MADRID as may influence the prevalence of smoking and finally social class (CLASS) as a proxy of income.

3. ESTIMATION RESULTS

Table 1 shows the distribution of perceived life expectancy for smokers and nonsmokers. The average values suggest that there is approximately a perceived loss of 5 years of life due to smoking and the average relative mortality risk perceived is 0.179. However, as expected, smokers show a reduced risk perception (0.135) compared with non-smokers and ex - smokers (0.20 and 1.19 respectively). This result may be indicative that risk perceptions may be an explanatory variable explaining the decision to smoke. Comparing this estimates with the actual attributable risk (0.14), we find that smokers show a slightly underestimation of the relative mortality risk whereas non-smokers and ex -smokers overestimate the relative mortality risk. Moreover, males show a reduced risk perception than female and risk perceptions seem to increase at an advanced age.

Table 2, shows the distribution of the relative mortality risk perceptions. Results are indicative of a large heterogeneity in the perception of risks between individuals. 42% of smokers and 36 and 39% of non-smokers and ex - smokers respectively underestimate the mortality risk. Furthermore, 20% of smokers and 34 and 28% respectively of non-smokers and ex -smokers overestimate the attributable risk. This results show in line with previous work (Viscusi, 1990, 1992) that the role of risk information campaigns may tend to overestimate the risks of non-smokers as they have no experience with smoking. However, age differences show a different pattern compared to previous research and the heterogeneity of our findings do not confirm previous the evidence achieved in previous studies suggesting that smokers show an overestimated risk perception. Moreover, differences with previous research confirm that risk perceptions are sensible to the elicitation procedure employed.

Results from the risk perceptions and smoking probability equations are presented in table 3. The first two equations refer to the risk perceptions determinants whereas the third and forth refer to the smoking decision. A first source of results confirms previous research. In particular, males show lower risk assessment and a higher probability to smoke. Smokers perceived risk is 5,2% lower than non-smokers risk perception. Being in a middle age (30-44 years old) is associated with a higher probability of smoking. The existence of endogeneity of risk perceptions is rejected using the Hausman test computing the predicted smoking probability in the perceived risk equation. Nevertheless, a second source of results, show partially opposite results with prior research ( Viscusi, 1990, 1991, 1992 and Liu and Hsiech, 1995). According to our data, we cannot confirm the Viscusi´s approach that argues that public information tends to overestimate the overall risk perceptions. Risk perception estimates are only sensible to advanced ages (>60). Education also shows an opposite pattern with prior research. The more educated individuals are the less sensible to the warnings related to relative mortality (i.e "Tobacco kills" ) and extreme education levels are associated with higher smoking probabilities. Finally, living in a large city as Madrid or Barcelona is associated with a lower risk perception and a higher smoking probability (the later significant only for Madrid).

4. CONCLUSIONS

This paper has provided evidence on the heterogeneity of risk perception estimates using a relative risk mortality measure. From our results, it appears to be clear the significant role of risk perceptions as influencing the smoking decision. This result confirms the applicability of the Viscusi´s (1992) classification defining as : "cognitive with rational limitations". That is risk perceptions play a relevant role whenever the individual decision is mislead by informational constraints. Even though some share of the population may overestimate mortality risk perceptions, it's also true that a large part of the population shows underestimates risk perceptions. Therefore, public information under certain circumstances may be able to shift information towards more accurate mortality risk perceptions.

References

1. Costa, J (1999). Conducta Individual, assegurament i elecció col·lectiva: una aplicació a la política sanitària. PhD dissertation. Universitat de Barcelona.
2. Hausman, JA (1978). Specification tests in econometrics. Econometrica, 46: 1251-1271.
3. Hsieh, C-R, Yen, L-L and Liu , J-T (1996). Smoking, health knowledge, and antismoking campaigns: an empirical study. Journal of Health Economics, 15:87-104.
4. Hu, TW, Ren,QF, Keeler, TH.E and Bartlett, J(1995). The demand for cigarettes in California and behavioural risk factors . Health Economics, 4,7-14.
5. Lichtenstein, Slovic and Fischoff et al (1978). Judged Frequency of Events. Journal of Experimental Psychology: Human Learning and Memory, 4 : 6,551-578.
6. Liu, JT and Hseih CR (1995). Risk Perception and smoking behaviour : empirical evidence from Taiwan. Journal of Risk and Uncertainty,11 :139-157.
7. Slovic, P; Fischoff, B and Lichtenstein, S (1982). Facts versus fears : understanding perceived risk. In D. Kanheman, P. Slovic and A. Tversky (eds). Judgement Under Uncertainty : Heuristics and Biases, Cambridge: Cambdridge University Press.
8. Viscusi, WK and O’connor Ch (1984). Adaptive responses to Chemical Labeling: Are workers Bayesian decision makers?. American Economic Review,74:942-956.
9. Viscusi, WK (1990) Do smokers underestimate risks? Journal of Political Economy, 98 1253-1269.
10. Viscusi, WK (1991). Age variations in risk perceptions and smoking decisions. The Review of Economics and Statistics, LXXIII (4).
11. Viscusi, WK (1992). Smoking: making the risky decision. Oxford University Press.

Tables

Table 1. Perceived life expectancy and perceived mortality relative risk

Table 2. Distribution of the relative mortality risk perceptions

Table 3 Mortality risk perceptions and smoking probability equations