Contextual Antecedents and Political Consequences of Adolescent Political Discussion, Discussion Elaboration, and Network Diversity

Abstract

Understanding how adolescents come to be informed participants in a democracy is a key concern in political socialization scholarship. However, our understanding of this process is hampered by limited research on the antecedents of a sufficiently wide array of communication behaviors and cognitions, in addition to a limited repertoire of knowledge outcomes in adolescent research. This study seeks to further the literature by addressing how discussion frequency, elaboration, and network diversity are related to factual and structural knowledge among adolescents. In addition, we utilize multilevel modeling to assess both school and parental effects on the various elements of political discussion. Results suggest that frequency of discussion is related to both factual and structural knowledge, whereas discussion elaboration is related only to structural knowledge. The multilevel models suggest that aspects of both schools and families are related to discussion frequency, elaboration, and network diversity.

Keywords:

• knowledge structure density
• heterogeneity
• parental news mediation
• family communication patterns
• cross-cutting

In addition to primary funding from the Center for Information and Research on Civic Learning and Engagement (CIRCLE), this research effort also received financial support from the Columbus Foundation through Kids Voting Central Ohio and the School of Communication and the College of Social and Behavioral Sciences at the Ohio State University. The authors would like to thank Suzanne Helmick (of Kids Voting Central Ohio); Dwight Groce, Keith Bossard, Saundra Brennan, and Susan Martin (of the Columbus Public School District); and Tiffany Thomson, Lindsay Hoffman, and Jessica Flanders (former graduate students in the School of Communication at the Ohio State University) for assistance in the gathering of the data.

Notes

1. When students were contacted by phone in 2006, interviewers first asked to speak with the parent to assess the age of the student. If the student was 18 or older, the student was then approached for consent. For students younger than 18, parental consent was obtained before the student was approached for assent.

2. We examined the racial makeup of each school to see if our differences were based on over- or undersampling in a given school. However, it appears that our sample is fairly representative of the actual racial makeup in each of the schools. The ethnicity that was the majority in each school according to district reports was also the ethnicity of the majority of the respondents for that school in our sample. Thus, any bias comes from some schools being relatively underrepresented in our sample.

3. The political figures that students had to identify were the new chief justice of the U.S. Supreme Court, the secretary of defense, the vice president of the United States, a CIA agent at the center of a recent scandal, the governor of Florida, the secretary of state, the congressman indicted for violation of campaign finance laws, the speaker of the House of Representatives, the nominee for a seat on Supreme Court, the House minority leader, a senator and former presidential candidate, the governor of California, and the U.S. representative who called for pullout of troops from Iraq.

4. Experimental evidence from several studies has demonstrated that manipulating the number of interconnections within the structure of mediated information to which individuals are exposed can subsequently increase the density of their knowledge structure as measured here (e.g., Eveland, Cortese, Park, & Dunwoody, 2004; Eveland et al., 2004). Other research has found predictable patterns of association between this measure and self-reported communication behaviors (Eveland & Hively, in press), providing some evidence for the validity of this measure (see also Eveland & Shen, 2008).

5. Unlike most measures in the political communication literature emphasizing heterogeneity, disagreement, or cross-pressures, this measure of diversity ignores the characteristics of the ego and focuses on diversity among alters (see Eveland & Hively, in press), which we believe is closer to the concept advanced in theories of deliberative democracy.

6. If students indicated that they did not discuss politics at all in a week during the campaign, they were considered to not have a discussion network. Therefore, they would not have a diversity score and subsequently were excluded from the analysis. This approach is consistent with previous studies examining discussion networks (see Marsden, 1987).

7. FCPs were assessed both in 2005 and 2006. While it would have been ideal to utilize the 2005 measures, this decreased sample size considerably because we had a much smaller student N in 2005 than 2006, and even fewer student panel respondents with data from both waves. However, among students for whom we have both 2005 and 2006 FCP data, the scores were moderately to strongly correlated (r = .43 for concept and .33 for socio). Therefore, 2006 measures were utilized as a substitute for 2005 measures.

8. While the reliability scores of the FCP measures are lower than we would prefer, they are consistent with what is typically found in studies of adolescents using FCPs (see, for example, Meadowcroft, 1986; Pingree et al., 2000).

9. While this is more variance than is typically desired, multilevel modeling (MLM) is generally considered to be robust in regard to problems due to heterogeneity of variance (Raudenbush & Bryk, 2002).

10. News use was operationalized by averaging the z-scores of each media form's frequency and corresponding attention item based on the evidence presented by Eveland, Hively, and Shen (in press).

11. These two models were utilized in order for us to be able to better examine the effect of frequency of discussion on the two indicators of political knowledge. The first model allows us to assess the impact of discussion, including those who do not discuss. The second model does not include individuals who don't discuss (i.e., discussion frequency = 0) because those individuals have a missing value for diversity (as noted in the Method section) and thus are excluded from the analysis. Not assigning a diversity score to those who do not discuss also limits the sample size more generally. These analyses were rerun assigning those who did not discuss politics a diversity score of zero (rather than setting them to missing) and the results effectively remained the same.

12. MLM (see Park, Eveland, & Cudeck, 2008; Raudenbush & Bryk, 2002) is a technique that can be used when OLS assumptions regarding independence of errors can no longer be made due to nesting effects. Given that these data were collected based on course affiliation within a specific school district, it seems apparent that this assumption of independence is no longer valid. Moreover, MLM permits simultaneous examination of variables at different levels of analysis—in this case at the individual level (e.g., child age or gender) as well as at the school (ethnic diversity) or classroom (e.g., communication aspects of the curriculum) level.

13. The incorporation of independent variables gathered from our 2005 parent and teacher surveys limited our sample to the 210 respondents with data from 2006 for whom we had obtained parental and teacher data in 2005. The sample size for each analysis is reported in the tables, and it varies from 210 because when there are too few cases in a nesting group all subjects in this group are excluded. To ensure that these respondents were not substantively different from respondents for whom 2005 parental data were unavailable, we compared students for whom we did and did not have 2005 parental data and found that there were no significant differences in our outcome variables.

14. Predicted values based on the “average” respondent as described above were pluralistic, 3.84; protective, 2.86; consensual, 2.88; and laissez-faire, 2.52.

15. Predicted values based on the “average” respondent as described above were pluralistic, 7.60; protective, 6.75; consensual, 6.87; and laissez-faire, 6.45.

16. Predicted values based on the “average” respondent as described above were pluralistic, .36; protective, .36; consensual, .19; and laissez-faire, .24.