• No results found

The dual impact of sophistication on the breadth and depth of an individual’s

considerations and the homogeneity of those considerations helps explain some of these results. As my results show, increasing sophistication produces an increase in the complexity of a

respondent’s mental networks and associations. At the same time, Zaller (1992, 121) argues that increasing sophistication will also strengthen respondent resistance to the messages they receive, implying that the homogeneity of the considerations a respondent has will increase. Analysis of the correlation between party identification and ideology supports this assertion. Among those who correctly identified two or fewer individuals (Levels 1 through 3), the multiple imputation estimate of the correlation between party identification and ideology is .47 (p < .001); among those with three or more correct identifications (Level 4), the multiple imputation estimate of the correlation between party identification and ideology jumps to .67 (p < .001).47

This increase in breadth, depth and homogeneity of considerations are, in fact,

contradictory to each other and could produce the curvilinear model complexity in the results. If one presumes that the homogeneity of a respondent’s considerations is negatively associated with the number of predictors necessary to explain that respondent’s policy preferences, then, at some point, the increase in the number of core values that relate to a particular policy preference will be overwhelmed by the decrease in the number of core values necessary to explain that core preference. Under such conditions, a relatively complex model would collapse into a fairly simplistic model.

The fall-off in the BIC` measure of the model’s predictive power is more problematic because there is no suitable measure to judge its severity. Raftery (1995) suggests that a

difference of 10 points between two models is “very strong” evidence of a significant difference in model fit. Since the Adjusted R2 for each stratum does increase, the BIC` measure shows that

the model does not fit as well as it could,given the small number of individuals in the strata. Whether this means that the fit of the model in Level 4 (three or four correct identifications) is poorer than one should expect remains in question.

In conclusion, this research began with the premise that the respondent’s level of political sophistication would radically impact the way in which that respondent thought about gun

control. The results of a series of estimations of the contextual model on the sample stratified by political sophistication reveals that this premise is largely untrue. Of course, the model does change across levels of political sophistication; however, these changes appear only in the marginal predictors. The core drivers of gun control policy opinions, ideology, party identification, and gender-based differences, remain consistent across almost every stratum.

For elites, the results of this chapter paint something of grim picture. While elite messages might be able to sway some respondents to change their opinions, the ways in which respondents think about their gun control policy preferences are largely set. It will take a massive investment of energy or a major social catastrophe to jar the mental networks of

respondents into seeing the gun control issue in a new light. Evidence of this assertion abounds. Even after attempted and successful presidential assassinations, gun-related massacres in schools and the rise of tension due to fear of terrorist incidents, the partisan divide on gun control

remains about the same. Proponents and opponents of gun control have made statements that were intended to teach the public how to think about gun control, and the public has learned the lesson well.

END NOTES

41 Functionally, I expect that the decreased capability of the unsophisticated to simultaneously consider several core

values when determining a gun control policy preference will lead to a situation where some individuals consider one or a set of related core values to the exclusion of another. This condition will tend to attenuate the estimate of the relationship between each core value and gun control policy preferences because, for those individuals not

that I am essentially positing an uncontrolled conditional relationship between core values and policy preferences among unsophisticated individuals. I would further assert that this effect (i.e., selectively considering the various core values attached to the issue of gun control) is largely driven by factors occurring in close temporal proximity to (or contemporaneously with) the survey itself. If this were the case, these conditioning effects would be largely unmeasurable outside of a controlled, experimental survey atmosphere.

42 Of course, correlation between political sophistication and certain predictors in the model could produce the

appearance of a relationship between sophistication and gun control policy preferences. I would argue that such relationships are most likely spurious and would disappear in a multivariate environment.

43 An interaction is statistical device that allows the magnitude of a coefficient to vary with change in the value of a

second variable. When a researcher stratifies a sample on a particular variable and conducts separate estimates on every stratum, she is explicitly permitting each coefficient’s magnitude to vary between strata. Therefore, stratifying a sample and conducting estimates on every stratum effectively creates an interaction between the stratifying variable and every variable in the model.

44 The exception to this pattern is the highest levels of sophistication. At this level, the model again collapses to a

very simple four-factor solution. As I will discuss later in this chapter, this result is not as counter-intuitive as it may seem. The dual-track impact of increasing sophistication (increasing “breadth and depth” of considerations and increasing homogeneity among those considerations implies that one should expect non-monotoneity in model complexity (Delli Carpini and Keeter 1996, 237).

45 In the case of Adj. R2, the sample size figures into the calculation in both the numerator and denominator of the

function.

46 In this equation, m stands for the model under consideration and k stands for the number of independent variables

in the model.

Related documents