ABSTRACT
We combined insights from various theories and models of media learning, and advanced an indirect model accounting for the mechanisms underlying the media influences on knowledge acquisition. Our model was largely supported by the data from a two-wave longitudinal panel survey with a nationwide sample of Korean adults. It was found that both personal cancer history and cancer worry were positively associated with exposure to stomach cancer information from the media. In turn, exposure to media information was positively related to reflective integration of that information, which ultimately leads to stomach cancer knowledge only among people with high levels of social capital. These findings suggest that media uses and effects are not only an individual but also a contextually dependent experience.
Acknowledgment
We appreciate the comments from two anonymous reviewers.
Notes
1 According to Callegaro and DiSogra (Citation2008), online panel studies are divided into two categories: probability-based and volunteer opt-in panels. When the opt-in panel is used, it is impossible to calculate response rates. Instead, some possible metrics, such as completion rate and attrition rate, should be reported.
2 Kim and Yoo (Citation2013) showed that SIOPS was very strongly related to five scales of Korean occupational prestige, concluding that SIOPS is a valid indicator of social capital in Korea. Many researchers have also adopted SIOPS to measure social capital in the Korean context (e.g., Choi, Yoo, Sohn, Oh, & Yeo, Citation2012; Shim & Seol, Citation2010).
3 We included only control variables significantly related to the endogenous variables because the results of the trimmed model may be more reliable than those of the full model. For instance, Little (Citation2013, p. 197) recommended that a SEM model should “include covariates (control variables) selectively and prune the nonsignificant effects.” Similarly, Seaman and Weber (Citation2015, p. 227) wrote that “covariates should be carefully selected and included in the model; throwing in anything and everything that may be important should definitely be discouraged.” Because any control variable can “soak up variance that may in fact be random perturbations” (Little, Citation2013, p. 195), including nonsignificant control variables in the model may lead associations among theoretical variables to be statistically significant by chance.
4 We did not consider different causal orderings of cancer knowledge and the other variables because cancer knowledge was from the follow-up data while the others were from the baseline data. Also, because the reflective integration of the media information can occur only after exposure to that information, the reverse causality between reflective integration and media use might not be the case. In addition, it makes more sense that personal cancer history and family cancer history come before other key variables in our study than vice versa.