The efficacy-desire model (EDM) proposes that the intention to quit smoking (Iq) is a function of the difference between the motivation to smoke (Ms) and the motivation to quit (Mq):(1)

The focus on these competing motivations is not novel; other models of health behavior change, such as PMT [17] and the HBM [18], suggest that intending to quit an unhealthy behavior is a function of influences on these competing motivations. Indeed, any theory of rational choice suggests that all the smoker needs in order to quit is more desire to do so than to continue smoking [27].

The distinction between the motivations stems from the reward system’s powerful influence on goal seeking [21] and its circuits specialized for detecting both harmful (negative) and beneficial (positive) environments [28]. These circuits produce corresponding forms of negative and positive affect that, respectively, underlie desires to avoid or approach such objects as cigarettes [29]. Although these desires are often reciprocally related, they are independent sources of motivation that can have unique influences and effects.

In addition to the essential role of desires, the EDM recognizes that motivation is also determined by the perceived efficacy to satisfy desires. Self-efficacy is a familiar concept that has long been featured in models of behavior change [30]. In regard to smoking, even if smokers desire to quit the habit, they are unlikely to try unless they believe that they can implement the behavior. Neuroscience models of addiction also focus on self-efficacy by emphasizing the important role of the brain’s control system in undermining the ability to quit an addiction. Theories of behavior change, such as the TRA (15), treat desires (e.g., attitudes) and efficacy as additive influences on intentions. However, because both efficacy and desire are needed to motivate behavior, the EDM treats these expectancies and desires as multiplicative determiners of motivation, a common assumption in psychological models of motivation [31]. Thus, inserting the respective efficacies (Eq, Es) and desires (Dq, Ds) for quitting and smoking into eq. (1) produces:(2)

For an addiction such as smoking, both components of Ms are likely to be high. Indeed, the more one practices a behavior, the greater the skill and sense of efficacy for controlling it [32]. The same is unfortunately not the case for quitting. Even if the smoker wants to quit (Dq), no motivation and hence intention will be formed unless the smoker’s sense of efficacy (Eq) is also high. As is often the case for addictive habits [33], the smoker may strongly desire to quit but not believe that it is possible to do so. However, in deriving predictions from this model, it is important to consider individual differences in the various components of the model and the ways in which they are related to each other.

Figure 1 shows the relations between the components of the model at the individual level. Although desire and efficacy to engage in a behavior are likely to be positively related, we show no relation between Ds and Es under the assumption that Es has reached asymptote in most smokers, leaving little room for any relation with Ds. However, for quitting an addictive habit, such as smoking, the relation between efficacy and desire to quit is likely to be negative. Efficacy for quitting the behavior is at its peak in the early stages of acquiring the habit, usually in adolescence, when the young smoker believes he or she will not have much difficulty stopping [34]. However, as the habit progresses, the smoker finds it increasingly difficult to stop even if the desire to do so increases. This process creates an inverse relation between Eq and Dq.

Neuroscience models of addiction specifically predict that greater frequency of smoking (represented in Figure 1 by parameter β) reduces Eq while it simultaneously increases Ds. It is also likely that frequency of smoking increases Dq, given what we know about smokers’ wishes to quit. But even leaving this out of the model, neuroscience models of addiction predict that the more one smokes, the lower one’s quit-efficacy (hence lower Mq) and the greater one’s desire to continue smoking (hence higher Ms). This disparity between Mq and Ms makes it difficult for the smoker to quit and shows why smokers are so conflicted by their addiction, wishing to quit but nevertheless continuing the habit.

The model makes interesting predictions regarding the effects of a warning, which, based on what we know about their effects [9]–[11], should increase Dq and reduce Ds. Figure 1 shows such a warning (whose intensity is indicated by parameter α) directly affecting Dq. Because Dq and Eq are inversely related, the effect on Dq is:(3)

Replacing Dq in eq. (2) with eq. (3) shows that the model makes the novel prediction that for an addictive habit, Mq is an inverted U-shape function of Eq:

That is, assuming that individual differences in Eq range from negative to positive valence, Mq rises as Eq increases, but at an intermediate point, begins to decline. Although theories of behavior change predict that efforts to change behavior increase as efficacy increases, the EDM suggests that, for an addictive habit, this effect only holds up to a point, after which the motivation to do so declines. Furthermore, whether the habit is addictive or not, the effect of α depends on Eq, with an enhanced effect for positive values of Eq and a depressed effect for negative values of Eq.

The path linking Ds and Dq in Figure 1 suggests that these desires should be inversely related. However, consistent with a bivalent model of affect [29], we assume that these desires are somewhat independent. Hence, the effect of the warning on Ds is:(4)

Eq. (4) expresses the intuitive result that Ds is positively related to the heaviness of the habit (β) and inversely related to the strength of the warning (α). Inserting eqs. (3) and (4) into eq. (2) yields the following overall relationship between Iq and the respective efficacies for quitting and smoking:(5)

We show examples of the hypothetical relation between Iq and Eq for different values of α in Figure 2A, assuming that Es and β are constant and that Es is higher on average than Eq (.5 vs. 0). The inverted U-shape relation is especially apparent when α  = 0. This shows that persons who smoke will have equally weak intentions to quit smoking not only when their efficacy is low but also when it is high. Indeed, absent any health warnings, smokers will have the greatest intentions to quit when their efficacy is at a moderate level. The prediction that low efficacy produces low intentions is not surprising since most theories expect this result. However, the model also predicts that those who think they can quit easily will not be motivated to do so either. This is a critical prediction of the model that will be tested for the first time in the present research.

A second critical prediction of the model is the effect of the warning. As seen in Figure 2A, the effect of α primarily increases the intention to quit among those with That is the point in the relation where all three curves in Figure 2A converge. Indeed, those with weaker Eq than −Es actually begin to exhibit a reduction in quit intention. Thus, the model makes the counterintuitive prediction that those with the strongest desires to quit (i.e., those who have smoked the longest) will be least motivated to respond rationally to warnings about the hazards of their habit, and this will be the case despite the fact that their response to the warning (created by an increase in α) is just as strong as the response among those with weaker desires to quit. This may explain the paradoxical effects of warnings observed in previous research. The quit-enhancing effects of increases in α will primarily be observed among those whose efficacy for quitting exceeds their efficacy for smoking. Indeed, warnings for those with weak efficacy for quitting will actually result in weaker intentions to quit.

We tested the major predictions of the model in an experimental context in which smokers were randomly assigned to see one example of a pack of cigarettes with a warning that was varied systematically in intensity across experimental conditions. This provided the opportunity to observe the effects of a warning in the context of individual differences in both the efficacy and desire components of the model. In addition to the predicted U-shape function shown in Figure 2A, we tested the prediction that increases in the intensity of warnings (represented by α) produced by adding an emotionally charged picture will lead to divergent effects on Iq depending on Eq. In addition, the EDM predicts that the greater the amount smoked (β), the lower the intention to quit. However, variation in β should only shift the curve up or down (it should be independent of Eq and α), and it should only interact with Es, which we assume is at a high and relatively fixed level for all smokers. In addition, as shown in Figure 1, frequency of smoking should be inversely related to Eq but positively related to Ds. Finally, in support of the expected inverse relation between Eq and Dq, length of time smoking (using age of the smoker as a proxy) should be positively related to Dq but negatively related to Eq.

The research was reviewed and approved by the Institutional Review Board (IRB) of the University of Pennsylvania, which adheres to the principles of the Belmont Report. As the survey was conducted over the Internet, was completed anonymously, and posed minimal risk, the IRB waived the requirement for written consent. However, participants were informed that the survey involved research, that their responses were anonymous, and that their participation was entirely voluntary. Thus, proceeding to take the survey was considered documentation of consent.

We tested the model’s predictions using warnings that were tested for use in the U.S. by the FDA [13]. We used the same adult Internet panel used by the FDA (Research Now [35]) and similar warning labels that FDA evaluated. However, smokers who had participated in the earlier FDA test were excluded from the study.

Panel members were included in the study if they reported having smoked at least 100 cigarettes in their lives and if they currently smoked cigarettes “every day” or “some days” (N = 3297). Most of the sample reported smoking every day (62%), a lower proportion than the nearly 80% observed in recent national surveys of U. S. smokers [36]. In this test, approximately 160 smokers (56.5% female) in each of two age groups (18–24 and 25+ years) were randomly exposed to one of 10 computer screen images of a cigarette pack containing a warning about the hazards of smoking (see examples in Figure 3). The mean ages of the two age groups were 22.1 (SD  = 1.60) and 44.3 (SD  = 13.91). As expected, the older smokers were more likely to smoke every day compared to the younger group (76.5% vs. 45.7%), X²(1)  = 329, p<.001.

At the lowest level of intensity (α  = 0), the smoker saw a hypothetical pack of cigarettes on its side with one of 3 text warnings recently mandated by the U. S. Congress (Figure 3A): Cigarettes cause cancer, Cigarettes are addictive, or Smoking during pregnancy can harm your baby. These statements are factually correct but do not convey the importance and emotional impact imparted by pictorial warnings [9], [10]. In the middle level of intensity (α  = .5), the smoker saw the front of a similarly designed pack but with both the text and a picture that covered the top half of the pack (Figure 3B). In the third condition (α  = 1), the smoker saw a frontal view of the pack with a picture, the base text, and in addition, explanatory text that elaborated on the basis for the warning (Figure 3C). Elaborated warnings have been used in Canada since they were introduced in 2000 [9]. We expected the additional text to enhance the warning, resulting in the highest level of α. There were two versions of the elaborated text for the addiction and pregnancy warnings and one for the cancer warning, but we did not include a picture plus base text version for the cancer message. Thus, there were 10 different conditions with respondents nested within each condition in the experiment.

Respondents were permitted to view the warning image for as long as they wished; however, they were not permitted to return to it after leaving the screen. They then answered a series of questions about their reaction to the warning. We assessed quit intentions (Iq) with the following question: “How likely do you think it is that you will try to quit smoking within the next 30 days?” This question format is commonly used to determine intent because it captures both the desire and ability to engage in the behavior within a specified time period [15]. This question was answered using a scale from (1) very unlikely to (4) very likely. Those with no opinion (3.2%) were assigned the score of 2.5.

We assessed efficacy beliefs for quitting (Eq) by averaging agreement with two moderately correlated items (r  = .33, p<.001): “It is hard for a smoker to quit smoking” (reversed scored) and “I do not need help from anyone to quit smoking.” Both items were rated on a scale from (1) strongly disagree to (5) strongly agree. We also assessed previous quit attempts with the question: “During the past 12 months, have you stopped smoking for one day or longer because you were trying to quit smoking?” (Yes/No). This item was assessed to provide a behavorial measure of quitting that should also exhibit the inverted U-shape relation with Eq prior to exposure to the warning.

We assessed the direct effects of the warnings on measures of Dq and Ds as checks on the success of the manipulation of warning intensity (α). Our measure of desire to quit smoking (Dq) was response to: “How much do you want to quit smoking?” with answers ranging from “not at all” (1) to “a lot” (4). To assess Ds, we asked two questions that probed emotional reactions to smoking and desire to smoke a cigarette: “Imagine you are smoking right now. How good or bad would you feel smoking a cigarette right now?” with responses ranging from “very good” (1) to “very bad” (4). The other item asked for agreement with: “I want a cigarette right now” (reversed scored). The items were correlated (r  = .33, p<.001) and were averaged to define a measure of aversion to smoking, the inverse of Ds. As expected by the model in Figure 1, our measures of Dq and -Ds were correlated (r  = .25, p<.001). Finally, answers to the question about frequency of smoking, “Do you smoke, every day (1), some days (0) or never?” were used to assess β.

We used linear regression to test eq. (5) using self-efficacy to quit as a measure of Eq and three levels of α (0,.5, 1) as values representing the experimental warning conditions. A test of the effect of β was conducted with a second model in which smoking frequency was added as a predictor. Predicted scores from these models were plotted to provide a visual comparison of model predictions with those in Figure 2A. A test of the U-shape relation between Eq and prior quitting behavior was also tested using logistic regression and quit attempts in the 12 months prior to the experiment. We conducted regression analyses to test the predicted effects of age, α and β on all of the observed mediators in Figure 1 (Eq, Dq, −Ds). We did not have a measure of Es, which we assumed was relatively high for all smokers and which should be positively related to β in any case. A test of the critical hypothesis that efficacy for quitting is inversely related to desire to quit was conducted by regressing Dq on Eq. As predicted by eq. (3), those with the weakest efficacy should have the strongest desire to quit.

The finding that smokers with weaker self-efficacy for quitting actually became less intent on quitting following exposure to warnings is not a new phenomenon. This adverse effect of health information has been observed in studies of smoking and alcohol use [17]. Indeed, the classical study by Janis and Feshbach [39] found that, as the fear arousing character of messages increased, intention to follow through with the recommended health practice declined. This finding led to the prediction that fear arousing messages will exhibit an inverted U-shape relation to behavior change, whereby increases in fear initially lead to greater change but to decreases after an intermediate level has been reached. Contrary to this hypothesis, however, subsequent research failed to find the inverted U-shaped relation. Indeed, most research finds a weak but positive relation between fear arousal and message acceptance [40].

The EDM predicts a positive relation between the intensity of the message (i.e., α) and intentions/behavior when the efficacy for the recommended behavior exceeds that of the pre-existing unhealthy behavior. However, the model predicts decreases in behavioral intentions for those with relatively weak self-efficacy. It is quite likely that this was the case in Janis and Feshbach, which tested the effects of complex recommendations for repeated tooth brushing in early adolescents. Thus, the EDM could predict the effect observed by Janis and Feshbach and others that messages that increase fear regarding the recommended behavior will nevertheless backfire for those who have weak efficacy to change. The finding that such boomerang effects occur more often for addictive behaviors [17] is also consistent with the model. Furthermore, the EDM provides more precise predictions for the conditions under which one can expect such boomerang effects.

The tendency to reject messages as they increase in fear-arousing capacity has been ascribed to “defensive processing” in which the recipient of the message argues against the message and thus rejects its recommendation [40]–[42]. However, the EDM and the present results suggest that defensive processing is not necessary to explain message rejection. Smokers at all levels of self-efficacy reported increased unpleasant thoughts about smoking and felt disinclined to smoke a cigarette shortly following exposure to the warning. These reactions suggest that they did not reject the message outright. Instead, the results are more consistent with the prediction from the EDM that the multiplicative relation between desire and self-efficacy to quit leads to less motivation to adopt the recommended behavior among those with low Eq. As their desire to quit increased, their negative level of self-efficacy reduced rather than increased their motivation to quit, leaving them less intent on quitting.

Aside from effects of warnings on quit intentions, the EDM predicts that addictive behaviors will exhibit an inverted U-shape function in relation to quit efficacy. This novel prediction was supported both for prior attempts to quit as well as immediate reactions to warnings. Thus, it was not just intentions that reflected this pattern but reports of behavior. Although this relationship has not to our knowledge been identified before, it is consistent with the observation that those who engage in behaviors that are difficult to quit often lose the motivation to desist from the unhealthy habit even if they succeed in reducing the behavior [33]. The EDM’s explanation for this phenomenon is that as the efficacy for quitting increases, it is matched by reduced desire to quit. As a result, the motivation to quit declines. However, increasing the desire to quit, as was done in the present experiment, should allow the beneficial effects of high quit efficacy to emerge.

The model also shows why smokers find it so difficult to quit. If smokers succeed in reducing their smoking habit as represented by the influence of β (perhaps by ingesting nicotine through a patch), they will experience less Ds, which will reduce Ms. Indeed, the model indicates that a reduction in β will raise Iq independent of α and Eq. However, as proposed in Figure 1, a reduction in β increases Eq. This moves them closer to the right side of the inverted-U relation with Iq. Thus, any favorable effects mediated by Ds will be offset and potentially outweighed by an opposite effect on Eq. These opposing forces could actually lead to a decrease in quit intention. As a result, once an addictive habit is created, the forces that motivate it can entrap the addict in a difficult to break cycle of continued dependence on the habit.

Despite the favorable effects of warnings suggested by the EDM, the model also predicts that the quit intentions of those with weak efficacy beliefs will be reduced as a result of exposure to warnings. This suggests that an effective public health strategy will require a two-pronged approach. One priority should be to encourage all smokers to reduce their habit as a transitory goal to eventually quitting. Although most smokers may not be able to quit, they may be able to reduce the strength of their habit. Indeed, recent surveys suggest that U.S. smokers have been doing exactly that [36]. The model suggests that recent efforts to reduce Es, through price increases and restrictions on smoking in public, may have been responsible for these reductions. That is, as long as is greater than zero, decreasing Es should increase Iq. Since this condition is likely to be the case for heavy smokers, efforts to reduce Es should increase Iq. However, because reductions in smoking can also sap the motivation to quit, it will also be necessary to counteract this tendency with repeated exposure to health warnings that are periodically refreshed so that their effect does not wear off. This could enable persons who smoke to maintain stronger quit intentions and to break free of the habit.

This study has limitations that should be examined in subsequent research. The findings are based on only a single exposure and the effects of warnings may intensify as smokers are exposed to warnings over time. The model predicts that the lightest smokers with the strongest self-efficacy for quitting will be most successful in quitting. However, we have not tested this hypothesis here. We also did not have a very sensitive measure of β, which the model suggests is a major factor in intentions to quit. Finally, we did not assess self-efficacy for smoking, which the model predicts will have an influence on quitting. We assumed that it was relatively high compared to self-efficacy for quitting. However, future research should look at variation in this form of self-efficacy as well. This parameter could be assessed by asking about barriers to smoking that the smoker experiences, such as restrictions on places to smoke, increases in prices for cigarettes, and social disapproval for smoking.

In conclusion, the EDM sheds light on why models that assume a rational response to warnings do not account for the behavior of persons addicted to a behavior such as smoking. The model translates insights from current neurobiological models of addiction [19], [22]–[25] into concepts that have been employed in models of behavior change [15], [17], [18]: namely, attitudes toward addictive habits (desire) and the lack of control over their cessation (self-efficacy). The model is also consistent with neuroscience theories that postulate separate approach and avoidance motives that underlie affective experience and that can produce extreme conflict in persons with a serious addiction such as smoking. Thus, the model shows how the barriers to quitting an addictive habit such as smoking are so imposing that they can trap the person who is addicted in a continuing cycle of dependence and frustrate efforts to reduce this life-threatening habit.