Results

All purchase attempts were completed during July and August 2000. Out of 320 purchase attempts, 77% were made by females. Overall, 12.2% of purchase attempts were successful (table 1). Smokers were more successful than non-smokers (McNemar's χ² for paired data: 11.13, p < 0.001). Bivariate analysis of the 320 purchase attempts revealed that sales rates increased with the age of the youth (table 1), but were not affected by the youth's sex, the clerk's sex, the adult supervisor, or the order of purchase. The average estimated age for the clerks who completed sales was 27.2 years compared to 36.4 years for clerks who refused sales (p < 0.001). Clerks asked the age of the youth in 5.3% of purchase attempts and for ID in 89.1% of attempts. Illegal sales were much more common when ID was not requested (table 1). When ID was requested, smokers completed purchases more often than non-smokers (table 1). Nineteen sales were made to smokers with no request for age or ID compared to nine sales to non-smokers under these conditions.

Regression modelling was utilised to assess the impact of the two protocols while controlling for other variables. The regression model controlled for the sales clerk's sex and estimated age, order of purchase attempt, adult supervisor, whether the teen was asked for his or her age and, independently, whether ID was requested. The models did not include as predictor variables, being able to lie, present ID, tell a story, or plead, as these strategies were only allowed for the smokers. Protocol, the clerk's age, and being asked for ID were the variables that comprised the final regression model. The “smoker” protocol increased the rate of sale nearly sixfold over that for the non-smokers (odds ratio (OR) 5.7, table 2). Completed sales were significantly more likely to occur when the estimated age of the clerk was less than 21 years and when ID was not requested (OR 7.6, table 2). The regression model was adjusted for the community in which the sale attempt occurred. The pair of belonging among the matched pairs of smokers and non-smokers was not included in the final model as this variable has a colinear relation with the community each pair operated in.

POST-HOC ANALYSIS

Post-hoc analysis was performed to gain insight into what might account for the greater likelihood of a sale for the “smoker” protocol given that these teens were allowed flexibility in the strategies used to attempt a purchase. Out of 160 purchase attempts by smokers only, teens lied about their age or claimed to have left their ID elsewhere 86% of the time, presented their own ID 8% of the time, and resorted to pleading 6% of the time. Only one smoker possessed ID. Considering only smokers, a sale was more likely when ID was presented (41.7%, 5/12) than when it was not (15.5%, 23/148, p < 0.05). Pleading (n = 9) was never successful. Paradoxically, smokers were sold tobacco 83% (19/23) of the time when they did not lie and 7% (9/137) of the time when they did lie (p < 0.001). The youths did not have to lie if the clerk responded to their initial request. The logistic regression analyses revealed that the adult supervisor accompanying the teen, the order of purchase, the clerk's sex, and whether the teen was asked his or her age were not significant predictors of successful attempts among the smokers. Sample size limitations precluded assessing the protocol variable of “pleading” in the model development, and multicolinearity between telling a story and lying about one's age precluded inclusion of these two variables in the final model. The buyer's sex and age were not included in these regression models because the buyers were too homogeneous on these variables (80% female, 60% age 16 years). Age of the clerk (< 21 years), being asked for ID, and the presentation of ID were the variables composing the final regression model. Smokers were significantly more likely to purchase cigarettes successfully if the clerk was less than 21 years of age, if they were not asked for ID, or if they produced ID (table 3). This final model was adjusted for the community where the purchase attempt occurred though there was found to be no significant effect on the model after controlling for the individual youths (likely due to the colinearity of youth and community).

Discussion

In unfamiliar stores, smokers were nearly six times more successful at purchasing tobacco than were non-smoking youths following a standard compliance check protocol. While the “smoker” protocol was far more realistic than the standard compliance check protocol, it was unrealistic in one important aspect. Young smokers usually buy tobacco in their own communities and indicate that knowing the person behind the counter is a crucial factor in determining success (JR DiFranza, unpublished data, 2000). The violation rates we obtained using smokers strange to the store personnel are very likely to underestimate the difficulty faced by young smokers living in these same communities. The standard compliance check protocol using non-smoking youths who are strangers to store personnel is not a valid measure of the ability of local youths to purchase tobacco in a setting of strong law enforcement.

The standard protocol differs in several aspects from the natural behaviour of young smokers. It employs non-smoking youths who may not be experienced at purchasing tobacco and who cannot lie about their age or present ID. Smokers were twice as likely to be sold tobacco with no questions asked. This might reflect their greater past experience at purchasing, or greater self confidence. Lack of experience on the part of the non-smokers may have contributed as much to their lack of success as the constraints of the protocol. This is an important issue since some states purposely limit the number of checks youths may do during the annual statewide Synar compliance survey.30State officials in South Carolina allow youths to perform a maximum of nine checks, but in practice, they average only 4.2 checks per youth, almost assuring that all checks are performed by inexperienced youths.31 If the goal is to measure merchant compliance accurately, it would seem prudent to ensure that the buyers are practised and confident.

When our smokers lied about their age or ID in communities with high compliance rates, it had little effect on the measured violation rate because clerks wanted to see an ID. In a small pilot study involving only one 15 year old girl, Hyland and colleagues reported that lying had no impact on her ability to purchase tobacco.32 Since the overall violation rate was 41% in that study, it appears that lying has little impact on measured violation rates whether the law is well enforced or not.32 However, youths from communities with aggressive enforcement indicated that if they were able to buy tobacco once by lying, they could return to the same store day after day to buy from the same clerk (JR DiFranza, unpublished data, 2000). Thus, lying may have little impact on measured compliance rates but it could have a big impact on the long term availability of tobacco to individual youths and their friends.

Radecki has reported that presenting ID with an underage birth date increases the rate of completed sales.33 In this study, only one smoker had an ID. Although the ID proved that she was underage, showing it resulted in a dramatic increase in her ability to purchase over that of the other smokers (OR 27, 95% confidence interval (CI) 3.4 to 212). It is possible that other aspects of this youth's appearance or behaviour could have contributed to this large effect. Perhaps having the ID allowed her to project confidence.

Some authors have reported a positive impact from enforcement on the prevalence of smoking among minors, while others have not.2-6 34-38 Studies in three sites all reported violation rates under 10%, but impacts on the prevalence of smoking were seen in only two.3 5 34-36 38 While measured compliance rates may have been comparable, the ability of neighbourhood youths to purchase tobacco may have been quite different. A crucial question is, what do merchants do when approached by neighbourhood youths who they know are not associated with enforcement? Since merchants can learn to evade being caught in the act, law enforcement might only be effective when it is accompanied by the voluntary cooperation of merchants. It is well established that the voluntary cooperation of merchants in the absence of enforcement is never enough,16 39-41 but voluntary cooperation (that is, merchant buy-in) may be necessary in order for enforcement to impact on smoking prevalence rates.

Two changes to the standard protocol might ensure that the compliance test would be a reasonably valid approximation of what happens when underage smokers attempt to purchase tobacco in stores where they are not known. Youths should produce ID showing that they are underage. Fictitious names and addresses can safeguard confidentiality. If non-smokers are used, they must be experienced and confident. The rate of successful purchases for non-smokers continues to increase even after 60 attempts.24

The fact that the violation rate recorded by our “standard” protocol was within the range reported by local enforcement officials (under 10%) suggests that the protocol we used is equivalent to that used in the enforcement programmes, lending support to the validity of this study. Further research is needed to determine if these results are generalisable to other settings.

The high rate of illegal sales by clerks under 21 years of age provides a strong argument for prohibiting such individuals from selling tobacco.

Since we wanted the smokers to act naturally, we did not dictate the strategies they would use. This resulted in small numbers and poor distribution for a number of predictive variables, limiting the analyses. More research is needed to determine which aspects of the standard protocol contribute to its low sensitivity. By providing the non-smokers with a financial incentive for each pack of cigarettes purchased, we may have biased our study toward a null finding since the standard compliance check does not employ such an incentive. It is typical to train youths before employing them to conduct compliance checks and the fact that our non-smokers did not rehearse may have biased the results to show a greater difference between the two protocols. A logical goal for future research would be to vary different aspects of the protocol one at a time to determine which are important.

The standard compliance check underestimates the willingness of clerks to sell tobacco to strange youths. This may be due to the confidence displayed by underage smokers, their experience, and their ability to produce an ID, even if the ID shows them to be underage. The compliance check does not measure the willingness of clerks to sell to familiar youths. This limits our ability to measure the effectiveness of enforcement efforts, the crucial factor in evaluating the impact of enforcement on the prevalence of tobacco use. More accurate methods for measuring youth access to tobacco from commercial sources are needed. A measure of merchant “buy-in” may help to fill the gap between the standard compliance check and the real world of underage smokers.