This is one of the few studies examining concordance between adolescent proxy reports and mothers’ self-reported smoking status. Similar to Barnett et al. (1997)
, we found good concordance between mothers’ self-reports and adolescent proxy reports on smoking, although adolescents reported a higher prevalence of smoking among their mothers than the mothers self-reported. Based on the first query, we observed a 96% agreement rate among adolescent–mother pairs and a 94% agreement rate at the second query. In addition, concordance was higher between adolescent reports and mothers’ first query than mothers’ second query. The concordance between the mothers’ first query and second query was also good, indicating good consistency in mother reports on smoking.
Although concordance between mothers’ first query and second query was good, one factor that could contribute to the discordant responses between the mothers and their children is the fact that some of the mothers’ smoking status changed between two queries. Therefore, the accuracy of the child’s report could depend on when the children provided data in relation to when their mothers provided data. Overall, 4.5% (46 women) reported a change in smoking status between the two queries; 19 women reported that they started smoking and 27 women reported that they quit smoking, suggesting that this impact was limited.
To be consistent with previous research (Barnett et al., 1997
), we compared reports of current smoking with reports of never and quitting. However, when we compared reports of current with reports of never smoking only, the kappa values at the second query increased, Κ = 0.62 (95% CI
= 0.51–0.72), as did the Κ value based on mother–mother reports, Κ = 0.70 (95% CI
= 0.59–0.80; data not shown).
We identified two factors associated with discordant smoking reports. In contrast to Barnett et al. (1997)
, we found that younger adolescents, not older ones, were more likely to provide concordant responses with their mothers. However, the adolescents in Barnett et al. ranged in age from 9 to 13 years, compared with 11–13 years in our study. A closer inspection of concordance rates observed by Barnett et al. reveals that, consistent with our results, 11-year-olds had the highest concordance rates.
We also found that experimenters were more likely to misclassify their mothers’ smoking status than nonexperimenters. Of those adolescents who disagreed, 63.3% misclassified their mothers as smokers. Because maternal smoking places mothers’ adolescent children at increased risk of smoking (Wilkinson et al., 2008
), it is possible that the adolescents’ own smoking behavior influenced their decision to classify their mothers as smokers. In addition, because age and experimentation were moderately correlated (r
= 0.22, p
< .01), older adolescents may have interpreted some parental smoking behaviors, like experimentation, as a manifestation of earlier smoking and therefore misclassified their mothers (Pomerleau et al., 2005
Overall, the kappa values in our study were moderate but the percent agreements were high. It has been suggested that kappa values are influenced by the bias and prevalence of the outcome (Banerjee & Fielding, 1997
), such that a high prevalence index (when the prevalence of a positive rating is very high or very low) results in lower kappa values (Sim & Wright, 2005
). In our study, the prevalence of nonsmoking was very high, and the bias, or the extent to which raters disagree, was also low, leading to the moderate kappa values.
Our study had several limitations. First, the mothers’ self-reported smoking status was not biochemically validated. Second, smoking data from fathers were limited, so we could not reliably compare father and adolescent reports. Third, the mother and adolescent data were not collected concurrently. However, maternal data were available from two timepoints, one before and another after the adolescents’ reports on their mothers’ smoking behavior. Fourth, unfortunately while we have detailed data from the mothers about their smoking behavior (e.g., self-reported level of nicotine dependence, number of cigarettes smoked per day, when first cigarette is smoked during the day), we did not have similar information from the children regarding the mothers’ smoking behavior. Therefore, we are unable to address the important issue of whether adolescents can provide more detailed information about their parents’ smoking behaviors, other than whether or not they do/did smoke.
In conclusion, the results of this study indicate that adolescent proxy reports on maternal smoking are accurate, suggesting that adolescent reports on mothers’ smoking behavior can be used as a proxy to obtain data if the mothers’ self-report data are not available. Our results further suggest that when reports are not collected concurrently, it may be more reliable to use self-report data that were obtained from the mothers prior to the proxy report obtained from their adolescents, rather than the other way around. In addition, in future studies, if the adolescent’s proxy report is used in lieu of the mother’s report, whenever possible, analyses should control for the adolescent’s age and experimenter status. Finally, the results suggest that latent variable modeling is warranted.