Beliefs about harms of cigarette smoking among Norwegian adults born from 1899 to 1969. Do variations across education, smoking status and sex mirror the decline in smoking?

BACKGROUND AND AIM: Smoking is one of the most important causes of socioeconomic disparities in morbidity and mortality. The aim of this study was to examine if beliefs about harms of smoking differed across gender, smoking status and education among Norwegian adults born between 1899 and 1969.
METHODS: Using data from a nationally representative survey of smoking habits and a multinomial logit/negative binomial two-stage hurdle model design, we examined (first hurdle) the associations between birth cohort, gender, education and smoking status and four beliefs about cigarette smoking: i) smoking is not harmful, ii) do not know if smoking is harmful, iii) any number of cigarettes per day (CPD) is harmful and iv) smoking more than a given nonzero number of CPD is harmful, and (second hurdle) the predicted number of CPD that could be smoked without causing harm (from outcome iv).
RESULTS: The probability of believing that smoking was not harmful was close to zero, regardless of birth cohort, sex, education and smoking status. The probability of not knowing if smoking was harmful decreased from around 0.7 to almost zero across cohorts. The probability of believing that smoking more than zero CPD was harmful increased from less than 0.1 to around 0.7, while the probability of believing that there is some safe level of smoking increased with cohorts born from 1900 to 1930 before declining. Respondents with primary/secondary education consistently believed smoking to be less harmful compared to respondents with tertiary education, but cohort trajectories were similar. DISCUSSION: The similar birth cohort trajectories in beliefs about the harms of smoking do not support the idea that Norwegian adults with lower education has had qualitatively different beliefs about the harmfulness of smoking compared to those with higher education. The persistent and large socioeconomic gradient is likely a result of other factors.

Introduction

In most developed countries, cigarette smoking has declined substantially over the last decades, primarily due to lower smoking prevalence in successive birth cohorts [1-6]. This decline has been markedly different among men and women, and among groups with different socio-economic position (SEP) [2, 4–7]. Moreover, at the time when smoking prevalence began to decline, the rate of decline was faster among higher educated men compared to other groups [6-8]. In Norway, the prevalence of daily smoking decreased from around 20 to five percent among men and women with tertiary education in the period 1996 to 2016. Among men/women with primary education, daily smoking prevalence decreased from 46/42 to 25/22 percent in the same period [9]. This is in line with numbers from most other Northern European countries and the United States [10-13].

The persistent and large differences in smoking between SEP-groups is one of the most important causes of socioeconomic differences in morbidity and mortality today [14-16] and projections of smoking patterns and health outcomes for the year 2040 predict that smoking will continue to be a leading cause of mortality, socioeconomic differences in mortality [17].

The unequal rate of increase, and later decrease, in cigarette smoking between men and women, and groups with different SEP in the 20th century has been explained as a process of diffusion whereby resourceful groups adopt new practices or products (innovations) more quickly than others [18-20]. As argued by de Walque (2010), one such innovation could be the emergence of new information about the detrimental health effects from smoking from the 1960s and onwards, presented in scientific reports such as the U.S. Surgeon General’s report on smoking and health from 1964 [21], but also in articles published in the general press, for example “Cancer by the Carton” published in readers Digest as early as 1952 [22]. Accordingly, health information is a central element of tobacco control policies [23] and the object of several articles (4, 11, 12 and 20) in the WHO Framework Convention on Tobacco Control [24].

Theoretical models of health behaviour, such as the theory of planned behaviour [25] and the Health Belief Model [26], argue that beliefs about risks play a key role for people’s health related choices, including smoking [27]. Health information may affect behaviour in different ways, for example by altering attitudes or beliefs about smoking [28]. Moreover, models that aim at explaining how intentions and beliefs affect behaviour, such as the Health Belief Model [26] and Cognitive Dissonance theory [29] assume that the relationship between intentions, beliefs and behaviour is, to some degree, rational [30].

People with higher education are generally better informed about the harms of cigarette smoking [31-34], and several studies argue that there is a causal association between education and most health behaviours, including smoking [35-37], although some studies argue that the effect of education on smoking is not a result of being better informed, but partly or wholly a result of other “third” variables, for example delay discounting [38-41]. The high smoking prevalence among medical doctors in the 1940s and 1950s [42, 43] and the nonconstant differences in smoking prevalence between men and women with similar levels of education [6, 44] suggest that health information alone cannot explain the persistent educational gradient in smoking.

The aims of this paper are therefore to examine A) how the following four beliefs about harms of cigarette smoking varied across gender, education, smoking status among Norwegian adults born between 1899 and 1969: i) not believing smoking to be harmful, ii) not knowing if smoking is harmful, iii) believing that smoking any number of CPD is harmful and iv) believing that smoking more than a given number of CPD is harmful, and B) the predicted harmless number of CPD from outcome iv).

This enquiry serves two purposes: First, it examines whether beliefs about harms of cigarette smoking have been stable across cohorts. Second, it sheds light on the association between beliefs about harms and smoking, in a period when smoking declined substantially, and at different rates, among men and women with different levels of education. This is of interest because informing the public about the risks from smoking has been regarded as crucial for reducing the popularity of smoking [10], even though few, if any, studies have examined the long-term associations between risk perceptions, smoking prevalence and socio demographic characteristics.

Material and methods

Data

Data stem from Statistics Norway’s nationally representative survey of smoking habits among adults (16–74 years) conducted each year from 1973 to 2020 in accordance with the Act relating to Official Statistics. Questions about beliefs about harms of cigarette smoking was included from 1973 to 1995, with the exception of 1974, 1975, 1991 and 1993. Responses from 1995 were omitted because the question differed from previous years. The mean annual number of respondents in the included surveys was 1 972. Response rates were above 85 percent in the period 1973–1990 [45] and around 70 percent in the period 1992 to 1994.

Respondents below 25 years of age were excluded to ensure that respondents were old enough to have been able to complete at least three years of tertiary education. The total number of respondents included, aged 25–74 years, was 35 487.

Measures

Beliefs about harms of cigarette smoking: Respondents were asked: How many cigarettes do you think you could smoke per day without harming your health? Additional response options were Do not know and I do not think smoking is harmful (see Table 1 for the number of respondents and distributions on demographic and risk perception variables).

Table 1

Descriptive statistics and variable description for hurdle model 1 and 2, men and women from 25 to 74 years, 1973 to 1994.

	Hurdle model 1	Hurdle model 2
	(N = 31357)	(N = 10614)
	Mean (sd) or percent	Mean (sd) or percent
Age	47.4 (14.2)	46.5 (13.8)
Period/real price of cigarettes in NOK	58.3 (2.3)	58.4 (2.4)
Birth cohort	1936.1 (15.6)	1936.9 (15.1)
Risk perception Hurdle model 1
Do not believe smoking is harmful	1.4%	-
Do not know if smoking is harmful	28.8%	-
Believe the harmless number of CPD = 0	35.9%	-
Believe the harmless number of CPD > 0	33.8%	-
Risk perception Hurdle model 2
Number of CPD that can be smoked without causing harm	-	6.0 (5.7)
Sex
Men	47.7%	52.0%
Women	52.3%	48.0%
Smoking status
Daily/occasionally	53.1%	39.8%
Do not smoke	46.9%	60.2%
Education
Primary/secondary	84.0%	83.9%
Tertiary	16.0%	16.1%
Region
Oslo	21.1%	23.3%
Eastern Norway excl. Oslo and surrounding areas	29.5%	29.2%
Southern and Western Norway, and Trøndelag	38.4%	37.6%
Northern Norway	11.0%	9.8%

Smoking status was determined by asking respondents: Do you ever smoke? Respondents who answered yes were then asked if they smoked daily or occasionally. Daily and occasional smokers were combined into current smokers.

Education: Respondents were asked whether they had completed nine years of compulsory education (primary), additional three years of college education (secondary) or at least three years of university education (tertiary). Respondents with primary or secondary education were combined.

Respondents were asked to name their geographic region of residence. Four regions could be identified across all survey years: The capital (Oslo) and surround areas, eastern Norway, southern/western Norway and northern Norway.

Hurdle model analysis

The structure and wording of the dependent variable (beliefs about harms of smoking) provided a methodological challenge. Respondents were presented with three equally valid response options, two of which a required binary response (do not believe smoking is harmful and do not know the number of cigarettes that can be smoked per day without causing harm) and one which required a specification of frequency (number of cigarettes that can be smoked per day without causing harm). Respondents were asked to respond to only one of the three questions.

One solution to this response structure was to use a hurdle model [46] in which the first hurdle consisted of calculating the likelihood of responding to the three different outcomes i) not harmful, ii) do not know, and iii) number of cigarettes per day (CPD) is harmful, using a multinomial logistic regression model, and in which the second hurdle consisted of estimating the predicted number of CPD reported in the third outcome using a zero inflated count model.

However, we would argue that the responses provided in the third outcome represent two qualitatively different phenomena. An answer of zero CPD indicated that the respondents believed that any level of cigarette use was harmful, while providing a response of one or more CPD indicated that the respondent believed there was a risk threshold where smoking more than a given number of cigarettes was harmful. We therefore believed that a better approach was to use a hurdle model in which the first hurdle (H1) consisted of estimating the likelihood of the four discrete outcomes: i) not harmful, ii) do not know, iii) any number of CPD is harmful (harmless number = 0) and iv) more than a given number of CPD is harmful (harmless number > 0) using a multinomial logistic regression model, and in which the second hurdle (H2) consisted of estimating the predicted number of harmful CPD from outcome (4) using a zero-truncated negative binomial regression model. Negative binomial regression was preferred to poisson regression because of overdispersion. For both H1 and H2, independent variables were education, smoking status, geographic region, sex, birth cohort, age and real price of cigarettes as a proxy for period (see below).

To examine if higher educated male cohorts regarded smoking as more harmful at an earlier point in time compared to other groups, we constructed a set of additional models for both H1 and H2 with the following interactions: 2) cohort X sex, 3) cohort X education, 4) cohort X smoking status, 5) cohort X sex X education, 6) cohort X sex X smoking status and 7) cohort X sex X education X smoking status (see S1 and S2 Tables for beta coefficients and model fit). For both H1 and H2, AIC and BIC (separately and combined) favoured the model that included a three-way interaction between smoking status, birth cohort and sex (model H1_6 and H2_6 in S1 and S2 Tables). Goodness of fit was tested with a series of likelihood-ratio tests. The tests indicated that adding variables increased model fit for all nested models in H1 (p ≤ 0.01, S1 Table), except for the interaction between education and birth cohort in Model H1_3 (p = 0.33). In H2, adding variables increased model fit for all nested models (p ≤ 0.02, S2 Table), except for adding education to the three-way interaction between birth cohort, sex and smoking status in Model H2_6 (p = 0.17).

From model H1_6 (first hurdle), we calculated adjusted predicted probabilities and 95 percent confidence intervals of responding to the four possible outcomes: 1) not harmful, 2) do not know, 3) harmless number of CPD = 0, and 4) harmless number of CPD > 0) for every 10th birth cohort (1899, …, 1969) at all values of sex, education and smoking status using the margins-command in Stata 15 [47]. From model H2_6 (second hurdle), we calculated the adjusted predicted number and 95 percent confidence intervals of cigarettes that respondents believed could be smoked per day without causing harm for every 10th birth cohort at all values of sex, education and smoking status. Associations were tested by calculating the marginal effects (dy/dx) of birth cohort for every 10th birth cohort at all values of sex, education and smoking status for both H1 and H2 (S1 and S2 Figs).

Age, period and cohort (APC) considerations

To examine the association between birth cohort membership and beliefs about harms of cigarette smoking, it was necessary to account for the two other temporal dimensions survey year and respondents’ age. However, since each temporal dimension is a perfect linear function of the two others (period (survey year) = cohort (birth year) + age), simultaneous estimation of all three effects is not possible without breaking the linear relationship, most often by imposing one or more restriction(s) [48]. This is referred to as the APC identification problem [49].

Several workarounds to this problem have been proposed [50-53]. Due to the relatively short series of surveys available (1973 to 1990) we employed a variant of an Age Period Cohort Characteristic (APCC) model [52] where age and birth cohort were entered as continuous variables and period was represented by a linear trend of the real price of cigarettes. The rationale behind using real price was that the period variable should represent short-term exogenous shocks and that long-term variations in beliefs should be a result of inter cohort ageing (age effect) and intra cohort change (cohort effect) in line with Ryders argument that social change is primarily the result of variations between cohorts and life cycle variations within cohorts [54]. A similar model was employed by Farkas to examine age, period and cohort effects upon female employment in the US [55]. Choice of APC model and alternatives are discussed below.

Results

In total 31 357 respondents had complete information on all relevant variables (Table 1). Of these, around 1 percent (n = 452) believed that smoking did not cause harm, 29 percent (n = 9 028) did not know if smoking was harmful, 36 percent (n = 11 263) believed that zero cigarettes could be smoked per day without causing harm, and 34 percent (n = 10 614) believed that one or more cigarette(s) could be smoked per day without harm, all survey years combined (Table 1). Furthermore, 52 percent of the respondents were women, 53 percent were current (daily or occasional) smokers and 16 percent had completed a bachelor’s or a master’s degree (tertiary education) or equivalent. The mean real price of a pack of 20 cigarettes was 58.3 Norwegian crowns (NOK) (range 55.4–66.3). After a drop from 66.3 NOK in 1973 (around 10 USD at that time) to 57.2 NOK in 1977, the real price was stable around 58 NOK (around 8.0 USD in 1994).

Of all respondents in the first hurdle model, 10 614 (34 percent) had provided a non-zero estimate of the number of cigarettes they believed could be smoked per day without causing harm (mean 6.0 CPD, range 1–60) and were consequently included in the second hurdle (H2). There was a higher fraction of men and smokers in the H2 sample compared to the full H1 sample.

Estimates from first hurdle (H1)

The birth cohort specific probabilities of the four outcomes from H1 (model H1_6) are presented in Fig 1. In general, the probability (pr) of believing that smoking does not cause harm (Fig 1, first row) was below 0.10 for all birth cohorts, and close to zero for cohorts born after 1930.

Fig 1

Adjusted predicted probabilities of i) believing that smoking is not harmful, ii) not knowing the safe number of CPD, iii) believing that the harmless number of CPD is zero and iv) believing that the harmless number of CPD is above zero. Male and female smokers and non-smokers with primary/secondary or tertiary education born in the period 1899 to 1969 (Model H6_1).

The probability of answering that they did not know how many CPD caused harm (Fig 1, second row) decreased strongly across cohorts from around 0.7 to near zero inn all groups. Female non-smokers with primary/secondary education had the highest initial probability of replying that they did not know (pr = 0.79, 95% CI: 0.76–0.83), while male non-smokers (pr = 0.53, 95% CI: 0.47–0.59) and male smokers (pr = 0.53, 95% CI: 0.46–0.60) with tertiary education, and female smokers with tertiary education (pr = 0.55, 95% CI: 0.48–0.62), had the lowest initial probabilities. Both among smokers and non-smokers born up until the 1950s, respondents with primary/secondary education had a higher probability of answering that they did not know if smoking was harmful, compared to corresponding groups with tertiary education.

In contrast, the probability of believing that the harmless number of CPD was zero (Fig 1, third row) increased rapidly across cohorts. However, the probabilities were initially higher among non-smokers compared to smokers. Among the latest cohort, born in 1969, the probability of believing that any number of CPD was harmful was highest among female non-smokers with tertiary education (pr = 0.79, 95% CI: 0.76–0.82), and lowest among female smokers with primary/secondary education (pr = 0.68, 95% CI: 0.63–0.72) and male non-smokers and smokers with primary/secondary education (pr = 0.69, 95% CI: 0.65–0.73 in both cases). Group differences were largest among cohorts born in the period 1929 to 1939. For example, among women born in 1929, the probability of believing that smoking more than zero CPD was harmful varied from 0.17 (95% CI: 0.16–0.18) among male smokers with primary/secondary education and 0.20 (95% CI: 0.18–0.21) among female smokers with primary/secondary education, to 0.41 (95% CI: 0.39–0.43) among male and 0.45 (95% CI: 0.43–0.47) among female non-smokers with tertiary education.

The probability of believing that smoking one or more CPD was harmless (Fig 1, fourth row) increased with birth cohort up until the 1930s, before declining in all groups. Among those born in 1929, men and women with tertiary education who smoked had the highest probabilities of believing that smoking a non-zero number of CPD was harmless (pr = 0.48, 95% CI: 0.46–0.51 for men and pr = 0.49, 95% CI: 0.46–0.51 for women). Among female non-smokers with primary/secondary education born in the same year, the probability was 0.21 (95% CI: 0.20–0.22). Among the latest cohort, the probability was highest among female smokers with primary/secondary education (pr = 0.28, 95% CI: 0.24–0.32) and lowest among female non-smokers with tertiary education (pr = 0.17, 95% CI: 0.14–0.20).

Estimates from second hurdle (H2)

Among those who believed they could smoke one or more CPD without being harmed, the predicted number (pr_n) of cigarettes from model H6_2 decreased from 8.1 (95% CI: 6.9–9.4) among those born in 1899 to 3.7 (95% CI: 3.2–4.1) among those born in 1969, all groups combined.

As shown in Fig 2, the highest predicted number of harmless CPD was reported by male smokers with primary/secondary education born in 1899 (pr_n = 11.3, 95% CI: 9.5–13.2). In comparison, the predicted number among female non-smokers from the same cohort was around three to four CPD (pr_n = 3.1, 95% CI: 2.5–3.6 for tertiary education and pr_n = 3.9, 95% CI: 3.2–4.6 for primary/secondary education).

Fig 2

Predicted number of cigarettes that can be smoked per day without causing harm.

Male and female smokers and non-smokers with primary/secondary or tertiary education born in the period 1899 to 1969 (Model H6_2).

Among the latest cohort, born in 1969, those with highest estimate of harmless CPD was still male smokers with primary secondary education (pr_n = 4.7, 95% CI: 4.0–5.4), but the differences between groups were small. The lowest estimates were found among male non-smokers and women with tertiary education (pr_n = 2.3, 95% CI: 1.9–2.6 for men and pr_n = 2.6, 95% CI: 2.2–3.0 for women).

Among female non-smokers, the predicted harmless number of CPD remained low and stable across all cohorts, regardless of education. If we compare those born in 1899 and 1969, the predicted number decreased by 0.4 points among female non-smokers with tertiary education (from pr_n = 3.1, 95% CI: 2.5–3.6 to pr_n = 2.6, 95% CI: 2.2–3.0) and 0.6 points among female non-smokers with primary/secondary education (from pr_n = 3.9, 95% CI: 3.2–4.6 to pr_n = 3.3, 95% CI: 2.8–3.9). When comparing those born in 1899 and 1969, the largest absolute decrease in harmless CPD was found among male smokers with primary/secondary education (6.7 points, from pr_n = 11.3, 95% CI: 9.5–13.2 to pr_n = 4.7, 95% CI: 4.0–5.4), followed by male smokers with tertiary education (5.2 points, from pr_n = 8.9, 95% CI: 7.3–10.4 to pr_n = 3.7, 95% CI: 3.1–4.2) and women with primary/secondary education (5.1 points, from pr_n = 8.9, 95% CI: 7.4–10.5 to pr_n = 3.8, 95% CI: 3.2–4.4).

Discussion

This study demonstrated that the probability of believing that smoking was not harmful was close to zero, regardless of sex, education and smoking status, among Norwegian adults born from 1899 to 1969, while the probability of not knowing if smoking was harmful decreased from around 0.7 to almost zero across birth cohorts. In contrast, the probability of believing that any amount of smoking was harmful increased from less than 0.1 to around 0.7 across birth cohorts, while the probability of believing that there was some safe level of smoking (above zero CPD) increased with cohorts born from 1900 to 1930 before declining.

The best fitting model (no interaction between education and cohort) did not support the argument that people with lower education had qualitatively different cohort trajectories for beliefs about the harms of cigarette smoking compared to those with higher education. However, compared to respondents with tertiary education, respondents with primary/secondary education were more likely to answer that they did not know how harmful smoking was (hurdle 1, outcome 2), less likely to believe that more than zero CPD was harmful (hurdle 1, outcome 3) and had a higher threshold value for how many CPD they believed could be smoked without harm (hurdle 2). However, among the latest cohorts, there were little or no educational differences.

The observed convergence in beliefs about harms of smoking among people with different levels of education has not been accompanied by a similar convergence in smoking behaviour. As shown by Vedøy [2], when comparing respondents born in the 1960s and 1970s in Norway, the smoking prevalence among men and women with primary or secondary education was more than 20 percentage points higher compared to those with tertiary education. Results from the US show similar discrepancies [4]. This is in line with the idea that educational differences in health behaviours are a result of unmeasured “third” variables [38, 56]. It should be noted that the pattern was the same even when we included education in the interaction term (models H1_5 and H1_7), which indicate that the similar trajectories were not a result of model specification.

Instead, the overall impression is that while education was associated with knowing that smoking was harmful (outcome 2), differences in assessing the number of cigarettes per day believed to be harmful (outcomes 3 and 4) were primarily due to smoking status. The increase in probability of believing that there is no harmless level of smoking (hurdle 1, outcome 3) took place at an earlier point in time among non-smokers compared to smokers, smokers were more likely to believe that a given number of CPD could be smoked without causing harm (hurdle 1, outcome 4) and, as shown by the results from hurdle model 2, the number of CPD they believed could be smoked without being harmed was higher. This is likely related to unrealistic optimism or self-exempting beliefs among smokers [57]. The finding that men generally believed they could smoke more CPD without being harmed compared to women (hurdle model 2) is in accordance with other studies of smoking related risk perceptions and sex [58, 59].

Differences between education groups, men and women, and smokers and non-smokers were less pronounced among respondents in late compared to early birth cohorts. However, in contrast to the discrepancy between beliefs about harms from smoking and smoking prevalence in the case of education, the increasing and converging probabilities of believing smoking to be harmful among men and women mirror the decrease and gender convergence in smoking prevalence observed in most European countries [2, 60].

From a risk communication perspective, the high over all probability (pr) of believing that smoking any number of CPD is harmful among smokers born in 1969 (pr = 0.69, 95% CI: 0.65–0.73), the low probability of answering that they do not know how harmful smoking is (pr = 0.05, 95% CI: 0.04–0.05), and the even lower probability of believing that smoking is not harmful (pr < 0.01) suggest that smokers have adopted beliefs about the health risks from smoking that are in line with health authorities and health professionals.

Similarly, the results do not support the idea that people overestimate the absolute risks of smoking to a large degree, as discussed by Krosnick et al., [61]. According Bjartveit and Tverdal, smoking one to four CPD is significantly associated with dying from ischaemic heart disease and all-causes [62]. Results from our study showed that, among all respondents born in 1969, the probability of believing that smoking a given nonzero number of cigarettes was harmless (outcome 4) was 0.23 (95% CI: 0.20–0.26), and among these, the overall predicted number of harmless CPD was 3.7 (95% CI: 3.2–4.1). This indicate that about a quarter of the respondents born in the late 1960s underestimated the risk from cigarette smoking in this study, but not by much.

In light of the substantial increase in beliefs about harms of smoking across birth cohorts shown in this study, the relatively large fraction with lower education who continue to smoke raises questions. If smokers believe they can smoke fewer and fewer CPD without being harmed, but maintain their smoking behaviour, this should lead to increased levels of cognitive dissonance [63, 64], which again could lead to fatalism [65] and an increased fraction of “hard core” smokers.

Several studies have shown that smokers with lower education have lower intentions to quit, more often believe that they will continue to smoke, are less likely to succeed in quitting and more likely to initiate smoking [2, 4, 5, 66, 67]. Moreover, a study from England found that smokers with low social grade were equally likely to try to quit smoking compared to high social grade, but that the likelihood of success among the latter group was twice as large compared to the former group [68].

This supports the idea that smokers with lower education or low socioeconomic acknowledge the risks of smoking, but that other aspects of smoking, such as stress relief, outweigh the risks. Having lower education is strongly associated with having lower disposable income and working in jobs that are associated with high levels of job strain and lower levels of job control [69]. Several studies support the idea that social and economic hardship are crucial in explain why low skilled workers and groups with lower education smoke more, and are less likely to quit, compared to groups with higher education [70, 71]. Accordingly, a study from Ireland found that education in itself only accounted for 13 percent of the total class inequality in smoking, while economic and social difficulties accounted for almost 40 percent [72], which suggests that social and economic strain are major factors for socio-economic differences in smoking, even after education is accounted for.

Limitations

The main limitation of this study was that the results were sensitive to how we defined period. We used the real price of cigarettes as a proxy for short-term changes that likely affect smoking behaviour, but not primarily beliefs about harms. There are certainly other models that could have been considered, for example hierarchical age–period–cohort (mixed effects) models [51, 73] and models using the intrinsic estimator [74], both which have previously been used in tobacco research [75, 76]. However, there has been much debate about the underlying assumptions of these models [77-81]. We therefore employed a theory driven and statistically simpler model that explicitly limited the period effects to be short-term variations in price, thereby forcing temporal variations to be a result of inter-cohort (ageing) and intra-cohort change, in line with Ryders theoretical discussion of social change [54].

However, to examine how sensitive our choice of period variable was, we calculated a set of cohort profiles from a series of models (plotted in S3 Fig) where, in the left panel, age and cohort were continuous and the period was: i) a continuous measure of working-time equivalents in minutes for purchasing a pack of 20 cigarettes among industrial workers, ii) a continuous measure of the consumer price index (CPI) for tobacco, iii) omitted, and, in the right panel, age and cohort were dummy variables and: iv) period was a continuous measure of real price, v/vi) period were dummy variables, and the first two/last two values of period was constrained to 1973/1988 respectively (“classic” constrained coefficient models), and lastly, vii) period was recoded to be orthogonal to a time trend (Deaton normalization with fixed effects) [82]. Estimates from the main model (H1_6) was included in the left panel for comparison.

Alternatives i), iii), iv) and vii) were to a large degree similar to the best fitting model (H1_6), while ii) and v/vi) were not. In the case of ii) the discrepancies were not surprising given that CPI increases steadily over the survey period, from 47 to 66, and was therefore in conflict with our assumption that period should model short-term variation with no discernible trend. In the case of v/vi), the inconsistent estimates and large variations underscore the problem of arbitrarily constraining two adjacent values in an APC model.

A second limitation is that although we had data for a long series of birth cohorts, the number of survey years and age groups included were limited (see S3 Table). We do not know if risk perceptions among smokers and non-smokers in the period after 1994 were similar to what we found in the study period. However, the survey years included covered a period with rapid changes in smoking behaviour and in which the risks from smoking received much attention.

A third limitation is that long-term variations in cigarette smoking is a product of several factors besides beliefs about harms or risk perceptions, including, but not limited to, restrictions on where to smoke, price, purchasing power and social norms. Moreover, some of these factors interact. For example, restrictions will likely affect both beliefs about harms and social norms. Some of these changes will likely be captured by variations in real price, but the endogenous nature of beliefs and risk perceptions means that we cannot claim to have isolated any causal effect of education or smoking status on beliefs about harms of smoking. Nevertheless, if differences in beliefs were an important factor for long-term educational differences in smoking, we would expect that beliefs would vary systematically with education. The results did not support this expectation.

Conclusions

Given the emphasis on informing the public to reduce cigarette smoking, and the large differences in smoking between groups with lower and higher education, beliefs about harms of cigarette smoking should vary systematically with education. This study found that beliefs about harms of cigarette smoking were more pronounced among those with higher education, but the differences were small and did not mirror the observed educational differences in smoking prevalence across birth cohorts. From this, we argue that educational differences in information about the harms of smoking cannot be the main reason for the large socioeconomic differences in smoking observed over time.

Marginal effects of birth cohort for every 10th birth cohort at all values of sex, education and smoking status in the first hurdle (H1).

(PDF)

Click here for additional data file.

Marginal effects of birth cohort for every 10th birth cohort at all values of sex, education and smoking status in the second hurdle (H2).

(PDF)

Click here for additional data file.

Cohort trajectories with different period specifications.

(PDF)

Click here for additional data file.

Regression coefficients, standard errors and p-values for models included in the first hurdle (H1).

(PDF)

Click here for additional data file.

Regression coefficients, standard errors and p-values for models included in the second hurdle (H2).

(PDF)

Click here for additional data file.

Number of respondents by birth cohort and survey year.

(PDF)

Click here for additional data file.

3 May 2022

15 Jun 2022

We would like to thank the reviewer and the editor for being given the possibility of submitting a minor revision of our manuscript entitled “Beliefs about harms of cigarette smoking among Norwegian adults born from 1899 to 1969. Do variations across education, smoking status and sex mirror the decline in smoking?”

We have addressed the issues regarding data availability and style requirements, and the reviewer's comments below.

Journal Requirements:

1. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at

https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_body.pdf and

https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf

Our reply: We have read through the PLOS ONE's style requirements and believe that our manuscript meet these requirements.

2. We note that you have indicated that data from this study are available upon request. PLOS only allows data to be available upon request if there are legal or ethical restrictions on sharing data publicly. For more information on unacceptable data access restrictions, please see http://journals.plos.org/plosone/s/data-availability#loc-unacceptable-data-access-restrictions.

In your revised cover letter, please address the following prompts:

a) If there are ethical or legal restrictions on sharing a de-identified data set, please explain them in detail (e.g., data contain potentially sensitive information, data are owned by a third-party organization, etc.) and who has imposed them (e.g., an ethics committee). Please also provide contact information for a data access committee, ethics committee, or other institutional body to which data requests may be sent.

b) If there are no restrictions, please upload the minimal anonymized data set necessary to replicate your study findings as either Supporting Information files or to a stable, public repository and provide us with the relevant URLs, DOIs, or accession numbers. For a list of acceptable repositories, please see http://journals.plos.org/plosone/s/data-availability#loc-recommended-repositories.

We will update your Data Availability statement on your behalf to reflect the information you provide.

Our reply: With regards to data availability, there are legal restrictions on sharing the de-identified data set. The data set is owned by a third party (Statistics Norway) and cannot be shared by us directly. However, the data is freely available to all researchers from the Norwegian Centre for Research Data and can be ordered free of charge at https://search.nsd.no/en/study/39e1c69e-aec1-44f5-963c-edaeb340c668. Creating a user profile and logging in is required.

3. Your abstract cannot contain citations. Please only include citations in the body text of the manuscript, and ensure that they remain in ascending numerical order on first mention.

Our reply: There are no citations in the abstract

4. Please review your reference list to ensure that it is complete and correct. If you have cited papers that have been retracted, please include the rationale for doing so in the manuscript text, or remove these references and replace them with relevant current references. Any changes to the reference list should be mentioned in the rebuttal letter that accompanies your revised manuscript. If you need to cite a retracted article, indicate the article’s retracted status in the References list and also include a citation and full reference for the retraction notice.

Our reply: The believe the reference list is correct and there are no changes

Review Comments to the Author

Reviewer #1: 1 The manuscript is written with a measurable and focused objectives using the optimum use of available data using a theoretically sound and rationalized modelling procedure. They have given the reasons for adopting hurdle model analysis and not choosing other competitive techniques (mixed effect and hierarchical models with visualization) explicitly.

Our reply: Thank you for your positive review of our manuscript. All changes in the manuscript are shown in red.

In addition to the changes made following your comments, we have made three changes that should be mentioned specifically.

First, we now denote the calculated probabilities from model H1_6 as “adjusted predicted probabilities” instead of “marginal mean probabilities”. Although both descriptions are used in the Stata manual, “adjusted predicted probabilities” better describe what we have calculated.

Second, in the first manuscript, the adjusted predicted probabilities calculated from model H1_6 were calculated using the over option (margins, at(cohort=(1899(10)1969)) over(smoking_status education sex)). This option divides the sample into subgroups according to smoking_status, education and sex and then calculated the probabilities of the four outcomes. After a discussion with colleagues and a close reading of the stata manual, it seems more correct to use the at option instead (margins, at(cohort=(1899(10)1969) smoking_status=(1 2) education=(1 2) sex=(1 2)). In this case, Stata uses all the data and calculates the probabilities as if all respondents were men, women, smokers, non-smokers etc. This avoids the problem of there being unobserved group differences that we were not able to control for. In practice, this change had no substantial impact on the estimates, and differences were most often around 0.01 units.

Third, we have now specifically described the higher probability of responding “Do not know” among respondents with primary/secondary education. This was briefly mentioned in the discussion, but not stated in the results-section. We have included the following in lines 241-44:

“Both among smokers and non-smokers born up until the 1950s, respondents with primary/secondary education had a higher probability of answering that they did not know if smoking was harmful, compared to corresponding groups with tertiary education.”

2 Table -1 where the description of the variables are shown might be represented in accordance with nature of the variables (nominal to ratio) and not by mean and sd in all cases as it does not offer much information.

Our reply: We agree with this comment and have now made a more compact Table 1 where ratio variables are described by means and standard deviations and where nominal variables are described by percentages. Ranges for the ratio variables are now only mentioned in the text.

3 The claim of the manuscript may be toned down little bit in terms of representing the Beliefs about harms of cigarette smoking among Norwegians born from 1899 to 1969. As they are giving a countrywide estimates using Norway’s nationally representative survey of smoking habits from 1973 to 1995 in which several exclusions are made as per age group and non response rate. It would be helpful if they offer the exact number of cohort for each year (from 1899-1969) in all 22 years data set as supplementary table.

Our reply: We have moderated the claims in the title by including adults in the title: “Beliefs about harms of cigarette smoking among Norwegian adults born from 1899 to 1969…”, in the abstract (Background): “The aim of this study was to examine if beliefs about harms of smoking differed across gender, smoking status and education among Norwegian adults born between 1899 and 1969” and (Discussion): “The lack of substantial educational differences in beliefs about the harms of smoking do not support the idea that Norwegian adults…” and in the first sentence of the discussion: This study demonstrated that the probability of believing that smoking was not harmful was close to zero, regardless of sex, education and smoking status, among Norwegian adults born from 1899 to 1969…

We also included the following section when discussing the limitations of the study:

“A second limitation is that although we had data for a long series of birth cohorts, the number of survey years and age groups included were limited (see S3 Table). We do not know if risk perceptions among smokers and non-smokers in the period after 1994 were similar to what we found in the study period. However, the survey years included covered a period with rapid changes in smoking behaviour and in which the risks from smoking received much attention.”

A complete table of respondents by birth cohort and survey year have been included as S3 Table.

4 some syntax errors might be corrected like in line 70/pg4 whether do they really mean a 'longer' education or a 'higher' education? Similarly line 202/pg13 the syntax ("Of the 31 357, 10 614 (34 %)") of the sentence is little cognitively overloaded sentence which may be rewritten.

Our reply: We now use higher/lower education in all cases.

The sentence on page 13 now reads:

“Of all respondents in the first hurdle model, 10 614 (34 %) had provided a non-zero estimate of the number of cigarettes they believed could be smoked per day without causing harm (mean 6.0 CPD, range 1-60) and were consequently included in the second hurdle (H2).”

5 I am particularly impressed with the limitation section where logical rationales are offered for obligation to choose one method over another.

Our reply: Thank you for your comment.

6 A section on effect size of selected model may be added with the p values in result section. As p value is a single number which sometimes may ignore an asymmetrical effect size. The parameters showing the explanatory powers of the models with increasing complexity of models may be added in supplementary file for the interested readers.

Our reply: Interpreting coefficients from multinomial logistic regression is difficult, even when they are exponentiated (relative risk ratios). We calculated relative risk ratios for all models, but given the small beta coefficients in some cases, some of the RRRs, especially the constants, became very large and not substantially meaningful. We therefore calculated predicted probabilities, which are much easier to interpret and take into account the other variables in the model. For the resubmitted manuscript, we also included two supplementary figures (S1 Fig and S2 Fig) which show the effect size (dy/dx) for the same groups reported in Fig1 and Fig2 at every 10th birth cohort. At the end of the “Hurdle model analysis” section we have now included the following sentence:

“Associations were tested by calculating the marginal effects (dy/dx) of birth cohort for every 10th birth cohort at all values of sex, education and smoking status for both H1 and H2 (S1 and S2 Fig).”

We agree that measuring effect sizes in multinomial logistic regression is complicated. With regards to the explanatory powers of the models we have now included a table showing goodness of fit in S1 Table and S2 Table and included the following section describing goodness of fit of the models under Material and Methods:

“Goodness of fit was tested with a series of likelihood-ratio tests. The tests indicated that adding variables increased model fit for all nested models in H1 (p<0.01), except for including the interaction between education and birth cohort in Model H1_3 (p=0.33, S1 Table). In H2, adding variables increased model fit for all nested models (p<0.02), except for adding education to the three-way interaction between birth cohort, sex and smoking status in Model H2_6 (p=0.17, S2 Table).”

Increased gof was also observed for the less complex nested models not shown in the supplementary files (S1 Table and S2 Table), where all independent variables were entered stepwise.

We hope these changes address your concerns.

Sincerely,

Tord Finne Vedøy and Karl Erik Lund

Submitted filename: Response to Reviewers.docx

Click here for additional data file.

6 Jul 2022

Beliefs about harms of cigarette smoking among Norwegian adults born from 1899 to 1969. Do variations across education, smoking status and sex mirror the decline in smoking?

PONE-D-21-31872R1

Dear Dr. Vedoy,

We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements.

Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication.

An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at http://www.editorialmanager.com/pone/, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to-date. If you have any billing related questions, please contact our Author Billing department directly at authorbilling@plos.org.

If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org.

Kind regards,

Rajnish Joshi

Academic Editor

PLOS ONE

25 Jul 2022

PONE-D-21-31872R1

Beliefs about harms of cigarette smoking among Norwegian adults born from 1899 to 1969. Do variations across education, smoking status and sex mirror the decline in smoking?

Dear Dr. Vedoy:

I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department.

If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact onepress@plos.org.

If we can help with anything else, please email us at plosone@plos.org.

Thank you for submitting your work to PLOS ONE and supporting open access.

Kind regards,

PLOS ONE Editorial Office Staff

on behalf of

Dr. Rajnish Joshi

Academic Editor

PLOS ONE