This study indicates the influence of country of residence on adolescent smoking intentions was only partially mediated through the ASE determinants. Country also modified the influences of similar adolescent cognitions on smoking intentions. Contrary to the predictions of the ASE model, country appeared to have a large direct influence on adolescent smoking intentions. In our within-country investigation of the same UK participants, we found the external influences of ethnicity (African-Caribbean, Indian, Pakistani and white) gender and socio-economic disadvantage on adolescent smoking intentions were almost entirely mediated by the ASE determinants . Being a white boy had a small direct influence on intention. Otherwise ethnicity, gender and socio-economic disadvantage had no direct effects on smoking intentions. Additionally, the external influences of ethnicity, gender and socio-economic disadvantage did not modify the predictive effects of equivalent ASE determinant scores on intention. Thus, the findings of the between-country study reported here are very different to the findings of our within-country study . The findings of the between-country study reported here may have implications for theoretical development and the future development of pan-European adolescent smoking prevention interventions.
Implications of the findings for theoretical development
Why might the effects of equivalent social influences (social norms, perceived pressure and modelling) scores on adolescent smoking intentions have varied by country? First, it is conceivable that some role models such as pop singers were excluded from the salient role set used to assess modelling and this would affect the relative influence of modelling on smoking intentions in each country. However, it seems unlikely that we omitted salient actors from the role sets used to assess social norms and perceived pressure. Second, if equal weighting of all sources of social influence was broadly appropriate in one country, but not in the other, this would create the apparent differences between countries regarding the influence of social norms, modelling and perceived pressure on intention to smoke. Third, one influence, for example, mothers, may have the greatest influence on adolescents in both countries. However, equal weighting of all the sources of the social influence indices may mask the influence of mothers. Thus, both countries may have a similar modelling index but one has a high prevalence of maternal smoking, the social influence with the greatest influence on adolescent smoking intentions, while the other has a low prevalence of maternal smoking.
One of the most striking findings was the large difference between countries in self-efficacy. A typical Spanish adolescent not only has greater self-efficacy than a typical English adolescent but self-efficacy also has a greater influence on Spanish adolescents' smoking intentions. This could have arisen if the meanings of the social situations used to measure self-efficacy in this study vary between countries which may influence beliefs about how much individual control may be exerted by respondents .
Differential weighting of the ASE determinants does not threaten the validity of the ASE model. However, if the ASE model allows that the relative importance of the ASE determinants varies with respect to adolescent smoking intentions, then this implies the relative importance is determined by some organizing construct specific to the overarching culture of individual countries which is currently missing from the model. We included country × age and country × gender interactions, but these were not significant. Additionally, previous analysis of the UK data indicated that participants' ethnicity, age socio-economic status and gender did not moderate the influence of the predictive effects of ASE determinants on smoking intentions . These findings suggest that the organizing construct is specific to individual countries and understood similarly by all adolescent sub-groups within countries based on ethnicity, age, socio-economic status, and gender.
Country of residence had a large direct influence on intention (1.72 points on a 7 point scale). Previous analysis of the UK data indicted that participants' socio-economic status and ethnicity did not independently directly influence smoking intentions . This suggests that between-country variation in both the socio-economic status and ethnicity of the participants does not underpin the additional explained variance (independent effect) of country on adolescent smoking intentions. Two possible explanations for the direct influence of country on adolescent smoking intentions are related to affective and moral beliefs which are not included in the model. First, Connor and Armitage (1998)  distinguish between the influences of instrumental, affective and moral beliefs on attitudes. Our study only measured attitudes that focus on instrumental beliefs. Second, Ajzen proposed that moral norms could operate alongside other determinants of intention and directly influence intention when decisions have ethical or moral dimensions . The concept of moral norms has been extended to cover personal norms, where the use of a moral framework is problematic though not entirely redundant, such as with adolescent smoking [23–25]. That is, individuals have an inter-related set of values, closely allied to self-identity. Not only are some value systems/self-identities more compatible with smoking than others, but these value systems/identities and their associations with smoking or non-smoking may vary by country. If this is the case, whatever the nature of these value systems/identities, they appear to be shared by most English adolescents in our sample regardless of ethnicity but are distinct from those shared by Spanish adolescents in our sample. Understanding of the relationships between value systems/identities and adolescents' smoking intentions is relatively underdeveloped. However, moral beliefs have been shown to be a major cause of differences in smoking prevalence amongst UK Bangladeshi female and male adolescents .
Viewing cigarettes as a social handicap or social facilitator did not predict intention in either country. If confirmed, social acceptance could be omitted from the model.
Implications of the findings for European trans-national adolescent smoking prevention interventions
Adolescent smoking prevention interventions that are driven by the ASE model aim to change the cognitions underpinning the ASE determinants in order to alter behavioural intention and, thus, future smoking. Differential weighting of ASE determinants between countries has two implications for the development of effective trans-national adolescent smoking prevention programmes. First, the usefulness of each ASE determinant as a predictor may vary according to country. A pan-European intervention to resist peer pressure, for example, would be predicted a priori to be effective. However, our results suggest that although this type of intervention could potentially influence UK adolescent smoking intentions, the same type of intervention would have little effect on Spanish adolescent smoking intentions. Interventions that aimed to promote self-efficacy, on the other hand, would be more likely to have a greater influence on Spanish adolescent smoking intentions than on UK adolescent smoking intentions. Second, given the large variations between countries in the predictive effects of particular ASE determinants, understanding the possible country specific organizing constructs highlighted above may be important for the effective implementation of trans-national adolescent smoking prevention programmes. Without this additional understanding, trans-national adolescent smoking prevention initiatives may fail, but the overall results may hide important successes within some countries. Additionally, we have argued that value systems/identities and some attitudes that are underpinned by salient affective and moral beliefs may be important influences on adolescents' smoking intentions, but understanding of these is poorly developed. Given the large influence these variables may potentially have, we suggest their examination could be an important component of future trans-national teenage smoking prevention programmes.
European trans-national adolescent smoking prevention initiatives are rare and have had variable results [2, 3]. A life-skills smoking intervention for German speaking pupils in Austria, Denmark, Germany and Luxembourg had little effect on current smoking . The ESFA adolescent smoking prevention intervention study was conducted in six countries (Denmark, Finland, Portugal, Spain, the Netherlands and the UK) and was underpinned by the ASE model . The interventions had an apparent moderate effect on regular smoking in some countries after twenty four and thirty months. However, this study had methodological problems including the non-randomized nature of some data, potential biases in data collection and analysis, very high attrition rates, relatively high exclusion rates and adjustment for covariates in the analyses in order to increase power. The interventions used in the ESFA project were tailored to the cultural circumstances of each participating European country. However, this tailoring was based on practical and logistical concerns rather than theoretical considerations and appropriateness. Spain and the UK participated in the ESFA project but it is unclear how the ASE model informed the development of the interventions used in these countries. The ESFA project results may have arisen because cultural factors were not adequately accounted for . Given our results, it is perhaps unsurprising the ESFA project found the English and Spanish interventions had very different effects on adolescent smoking outcomes.
In this study, the ASE model explained 37% of the variance of intention in Spain and 29% in the UK, which Sutton describes as explaining a medium-high percentage of the variance . However, our study was cross-sectional so we can only say that the ASE determinants are associated with adolescent smoking intentions. Prospective studies have however, confirmed the predictive ability of the ASE model . Nonetheless, any conclusions concerning the ability of the ASE model to predict adolescent smoking intentions based on our findings should be regarded as tentative.
There are two possible biases that could have caused spurious differences in smoking intentions between countries and thus, the large unmediated effect of country. The first bias relates to the translation of the questionnaire from the original Dutch questionnaire. The intended meaning of the questions may not have been adequately reflected in the British and Spanish translated versions. This bias is an unlikely explanation for the differences in intention between countries, however, because the questionnaires showed good reliability in each country. The second bias could arise if participants' interpretations of the questions vary between individuals or between groups based on age, gender, or country. Thus, even if the translations were completely accurate, participants' interpretations could depend on the social context of their lives. However, bias is an unlikely cause of the direct unmediated effect of country of residence on smoking intentions because the differences between the predicted maximum and minimum intention scores arising from each ASE determinant (Figure 2) were less than the unexplained difference in intention between countries.
Behavioural intention was assessed using one question, which is commonly the case . However, two meta-analyses concluded that behavioural intention is a fairly robust construct and the type of measure of behavioural intention does not greatly influence the predictive ability of behavioural intention [28, 29].
The ASE model and the closely related TPB are currently still being used by researchers and health promoters as a basis for the development of adolescent smoking prevention interventions in many countries. Both the ASE model and the TPB have similar underpinning assumptions and these assumptions are currently commonly considered to be valid. The aim of the investigation was to test the validity of the underpinning assumptions of the ASE model. We used data that were collected in 1997 and are thus, relatively old. Nevertheless the ASE model should not, in theory, be dependent on the age of the data. In other words there is no reason to assume that a theoretical model such as the ASE model should lose its applicability over a ten year period. The ASE model and its underpinning assumptions are valid for each country . However, this investigation indicates that the importance of the ASE determinants varies according to country and the ASE model may fail to capture important cultural factors.