Introduction There is ample evidence in the medical research literature to show that the harmful effects of smoking are real. As a result, most governmental agencies around the world attempt to discourage smoking among smokers and nonsmokers alike by requiring cigarette companies to print health-warning messages on their product packages as a means to deter smoking. Little, however, is known about the credibility of these cigarette-warning messages perceived by both smokers and non-smokers alike on the effectiveness of these messages. Other studies have also examined socio-psychological factors other than health warning messages such as peer pressure, bad breadth, et al., (Pechmann, et al, 2003) which can also influence consumer desires whether or not to smoke. No study to our knowledge has incorporated this factor other than just health warning messages in the same model to study this phenomenon. Furthermore, most research on this topic has focused on consumers in western industrialized countries, even though the issues related to smoking can be just as severe in developing countries and warrants more research attention. In this research, we propose to study the credibility of cigarette warning messages on the effectiveness of these messages on deterring smoking behavior and the mediating effects of the perceived health and social risks in an integrated structural equation model (SEM) framework in an emerging country- Turkey. The study should provide insights into how these factors may influence smokers and nonsmokers differently and enable public policy makers to institute programs or strategies targeting the two different segments of consumers. Literature Background and Hypotheses Copious research has reported on the efficacy of Health Warning Message (HWM) in the Health Communications and Marketing literature (for reviews see Hammond, 2011; Noar, et al, 2015; 2016). In the USA, current rules only require health warnings be printed with tiny fonts on cigarette packages. The effectiveness of this format to deter smoking has come into serious questions. Other countries including countries in the EU have long required pictorial depiction of health hazards resulting from smoking be printed on cigarette packages that must occupy a substantial portion (50% or more) of the surface area on the cigarette package. The assumption is that “a picture is worth a thousand words”. However, the results are equivocal at best (Sabbane, et al., 2009). The real impact of HWM continues to be a point for debate as there are other tobacco consumption reduction methods such as increased taxes and smoking bans (Ruiter & Kok, 2005) various governments can and have imposed. Previous research has shown that when consumers believe that smoking behavior leads to unhealthy consequences and is unwise and unattractive, they generally would refrain from engaging in this behavior (Azjen & Fishbein, 1980). However, consumers who think that they are invincible and that these health hazards will never affect them but only other people, may have engaged in the process of rationalization and may actually discount the credibility of these claims. If so, the HWM may become less effective (Kees, et al., 2006). We therefore formulate the following hypothesis: H1: Perceived credibility of HWMs increases the effectiveness of HWMs. On the other hand, the extent to which consumers may perceive that cigarette HWMs to be credible can better persuade them about the health and social risks such as yellow teeth, bad breath, bad smell, et al., leading to social rejection associated with smoking (Pechmann, et al., 2003). Hammond’s review article (2011) provided some evidence that size and the presentation format can enhance the believability of HWMs. We should expect then that a greater perceived credibility will lead to better persuasion. Similarly, Kowitt, et al. (2016) also show in their study that the believability of HWMs vary among the adolescents regarding cigar smoking. Again, the assumption is that if they perceive the messages to be more credible, they will be more effective in persuading them about the harmful effects of smoking. We thus formulate the following two hypotheses: H2: Perceived credibility of HWMs increases negative social risk perception of smoking H3: Perceived credibility of HWMs increase negative health risk perception of smoking Past research has revealed that knowledge about the negative consequences of a behavior can lead to behavior change even though this may be contingent upon how difficult it is to quit a behavior such as smoking (Nourjah, et al., 1994). Consumer behavior literature has confirmed that due to various factors, attitude may not correlate with behavior even in the context of smoking (Ruiter &Kok, 2005; Solomon, 2018). Extant literature suggests that HWMs may be more successful in influencing attitude (Cameron & Williams, 2015). This research focuses on the health risk perception’s potential impact on intention to quit smoking rather than the act of actually quitting smoking. Furthermore, as mentioned earlier, other factors such as smoking ban and price increases, et al. may cause smoking behavior change. Hence we present the following hypothesis: H4: Perceived health risks associated with smoking influence the effectiveness of HWMs. Smoking has become less socially acceptable in western industrialized countries in recent years. Many nations and states have now banned smoking in public places such as colleges, restaurants and airports including second hand smokes. This creates a social stigma for many smokers. Smokers, as a result, has to bear greater social risks these days. This may also cause them to develop a negative attitude toward smoking as well as reducing their intention to smoke (Chuang & Huang, 2012). As Turkey is a country whose citizens care more about social acceptance, we postulate the following hypothesis: H5: Perceived social risks of smoking affect the effectiveness of HWMs. As mentioned earlier, the effectiveness of HWMs may vary according to subjects’ responses to such messages. They may discount the message veracity by a process of rationalization or even by the extent to which they may have been addicted to the habit. This is so particularly for smokers who may have engaged in such behavior for a long time. Hence, one should expect these messages to be more persuasive and thus more effective in dissuading non-smokers from starting the habit. Smokers, on the other hand, are likely to remain unconvinced and view these messages as exaggerations and therefore less affected by these messages (Hammond, 2009; Hahn & Renner, 1998; Pechmann & Knight, 2002). Based on this discussion, we postulate that the impacts of HWMs’ credibility on perceived health and social risks, as well as the messages’ effectiveness will vary depending on the smoking status of the subjects. We expect greater impacts for non-smokers. Therefore, we tested the SEM for smokers and the same model for nonsmokers in this study. The model is shown in Figure 1 in the Appendix. MethodologyWe designed a four parts questionnaire to measure the constructs in the model we used in this study. In the first part of the questionnaire, we first presented the 14 HWMs currently required by law in Turkey to be displayed on cigarette packages. Furthermore, each of these messages is to be rotated on the packages. We then used the semantic differential scale items comprising of five bipolar adjectives derived from previous literature (Beltramini, 1988; Renee & Cameron, 2015) to measure the perceived credibility of HWMs: (1) believable- unbelievable, (2) convincing- unconvincing, (3) reasonable- unreasonable, (4) easy to understand- hard to understand, (5) informativeuninformative. In the second part of the questionnaire, we operationalized the effectiveness construct by asking the respondents to indicate the extent to which they perceived these HWMs as effective on a 1 item semantic differential scale: “HWMs motivate me to quit or not quit smoking” for smokers and “HWMs reduce my curiosity and desire to try smoking” for nonsmokers. In the third part of the questionnaire, we measured health and social risks perceptions by using 4 statements. Respondents were asked to indicate their agreement/disagreement with each of the 5- point Likert scales culled from Kim (2006): Smoking makes me “contract various diseases and become addicted (health risks); Smoking makes me look “stupid, filthy (social risks). The last part of the questionnaire asked respondents about their smoking status and demographics including age, gender and the level of education. We collected the data by posting a copy of the questionnaire online on a website dedicated to discussing higher education issues in Turkey (www.akademikpersonel.org). We obtained an effective sample size of 432 after removing those with incomplete answers. Male and female respondents constituted 50 % of the sample each. The largest percentage of respondents was in the age group of 17 to 22 (35 %). Respondents (249) who have achieved the level of a college undergraduate degree or equivalent on education constituted 58 % of the sample. We conducted Chi-sqaure analyses to test significance of demographic differences between the two groups of smokers and nonsmokers. The results revealed no significant differences on age (χ2/3df=4.83; p=0.185), gender (χ2/1df=2.70; p=0.123) and the level of education (χ2/2df= 0.873; p= 0.646) between the smoking and nonsmoking subsamples. ResultsWe used WarpPLS 5.0 to conduct the structural model analyses as this analysis tool does not require the data set to be normally distributed (Kock & Lynn, 2012). We first tested the measurement scales with respect to their convergent and discriminant validities through standardized factor loadings. We conducted an Exploratory Factor Analysis (EFA). We eliminated two items on the credibility scale, which did not reach at least 0.5 on factor loadings for both subsamples. The two remaining items are “believable” and “convincing”. We then evaluated reliabilities of the scales for both subsamples, results for both subsamples showed that both credibility and health risk scales achieved Cronbach’s α scores higher than 0.7. Even though the Cronbach α score for the two items scale to measure social risk only reached a value of 0.6 for each of the subsample, this is considered adequate due to the smaller number of items used (Field, 2013). We assessed the assumptions of PLS-SEM. Variance inflation factors (VIFs) are well below the recommended 3.30. Hence, both vertical and lateral collinearity assumptions were met, thus deemed to be free from the common method bias (Kock & Lynn, 2012). We then evaluated convergent and discriminant validities of the latent construct. Discriminant validity was achieved as the square roots of AVE (average variance extracted) scores were higher than and correlation of the factor with another measure (Fornell & Larcker, 1981). For each subsample, composite reliabilities of all factors were higher than 0.8 and indicators’ loadings higher than 0.5 on the latent constructs. Thus, all scales used in the model achieved convergent validities for each subsample. We then proceeded to test the structural model for each of the subsamples. The results showed that for nonsmokers, the model explained 12% of total variance in the effectiveness of HWMs and for smokers 33%. All path coefficients were significant for nonsmokers except the health risks → HWMs effectiveness (β= 0.05, p= 0.23). Other path coefficients all showed significance as follows: perceived credibility of HWMs → perceived health risks (β= 0.154, p= 0.01), perceived credibility → perceived social risks (β= 0.148, p= 0.014), perceived credibility of HWMs → perceived effectiveness (β= 0.285, p < 0.01) and perceived social risk → perceived effectiveness (β= 0.148, p < 0.014). For the smokers’ subsample, the path coefficient between the perceived health risks and effectiveness was also shown to be non-significant (β= 0.06, p= 0.19). All other path coefficients all showed significance as follows: perceived credibility of HWMs → perceived health risks (β= 0.197, p < 0.01), perceived credibility → perceived social risks (β= 0.247, p< 0.001), perceived credibility of HWMs → perceived effectiveness (β= 0.503, p < 0.001) and perceived social risk → perceived effectiveness (β= 0.213, p < 0.001). These results confirmed H1, H2, H3 and H5 but not H4 for both smokers and nonsmokers. Discussion and Conclusion This study reveals some useful insights about how HWMs on cigarette packages may influence smokers and nonsmokers differently in an emerging country- Turkey, which can be used by public policy makers to formulate effective anti-smoking campaigns to target the two groups. The effect sizes appeared to be greater for smokers than nonsmokers. This suggests that if these HWMs can be crafted to be perceived as more credible, the smokers can be more persuaded to quit smoking than for nonsmokers to start smoking. These results appear to be consistent across the board whether we are considering the direct effect or indirect effects through the mediating variables. This is perhaps somewhat consistent with what we would expect based on our earlier discussions. Smokers may be more resistant to HWMs due to the fact they are already engaged in such behavior and therefore more likely to discount the messages as well as the associated health and social risk perceptions. For nonsmokers, on the other hand, it may not take as much to convince them about the harmful effects of smoking and therefore greater perceived credibility of HWMs may be less critical in persuading them not to start smoking. It is important then when designing such HWMs to consider carefully whether these messages are perceived to be credible. Some of the US cigarette warning messages printed in tiny scripts are ambiguous such as smoking can be harmful to your health clearly can not be expected to be highly credible. Research has also shown that pictorial messages can be more persuasive. However, research has also shown that the results are far from being conclusive (Sabbane, et al., 2009). More research is warranted, especially in emerging countries. It is also interesting to note that social risks perceptions are more persuasive than health risks perceptions. Future communications strategies should place a greater emphasis on this factor and not just on health risks. Limitations and Suggestions for Further Research Although our study is able to provide some useful insights into how perceived credibility of HWMs can influence the effectiveness in deterring smoking among nonsmokers and smokers alike either directly or indirectly through perceived health and social risks, we would like to point out several limitations of this study. First, we collected our data online using a web portal in Turkey that focuses on discussing higher education issues in general and therefore the respondents’ level of education is higher than the general population as well as the age group being younger. Future study should collect data from other age groups and education levels to produce results that are more generalizable. We conducted our study in one emerging country- Turkey. In future studies, more countries should be included. As the results revealed the prominence of social risks over health risks in influencing the effectiveness of anti-smoking campaigns, the issue about how to promote the social wellbeing of not smoking (promotional) versus the health risks aspects of the campaign warrants further research.