Testing Zaller’s Reception and Acceptance Model in an Intense Election Campaign

Zaller’s book and related works (e.g., Feldman and Zaller 1992), have been heavily cited in the literature. By way of comparison, he is among the recipients of the “Philip E. Converse Book Award” given annually to the author of an outstanding book at the American Political Science Association Meeting. The other recipients are Downs (1957), Campbell et al. (1960), Fiorina (1981), Schumann and Presser (1981), Iyengar and Kinder (1987), Stimson (1991) and Page and Shapiro (1992). According to the scholarly journal archive JSTOR, Zaller (1992) is cited in 28 papers published in the American Political Science Review between 1992 and 2002 (see http://​www.​jstor.​org/​). This is less than Downs (1957) and Fiorina (1981) but more than all the other “outstanding” books.

See JSTOR http://​www.​jstor.​org/​.

From the beginning of campaign, a majority of voters named free trade as the most important issue and its importance grew up as the campaign progressed. The analysis of the televised media coverage indicated that news coverage was dominated by FTA (Johnston et al. 1992, pp. 115–116).

Party positions on FTA were taken before the campaign started and did not change during the course of the campaign. It thus makes sense to consider party messages as exogenous to public opinion; see Doern and Tomlin (1991) and Fraser (1989). Our approach is entirely consistent with Zaller’s assumption that elites come first. Of course this does not rule out the possibility that these positions were partly in response to voters’ anticipated reactions. In his later work, Zaller (1998) acknowledges a possible reciprocal relationship. See also Gabel and Scheve (2007).

All hypotheses are stated in static terms but later we offer dynamic versions.

It could be suggested that by selecting on reception we are also selecting on levels of political awareness. We have three responses. First, logically party messages should have an impact only among those who have received these messages. Second, even though there is a relatively strong correlation between awareness and reception (see Table 1) the correlation is not overwhelming. There remains substantial variation in political awareness even among those who received party messages (in our sample almost 15% of those who scored 0 on political awareness knew the position of each party and leader and 16% of those very highly aware who scored 1 did not know; see the Appendix A for a description of variables). Third, we have performed analyses of acceptance including all respondents (see Appendix B), and the findings are similar.

With regard to those who do not have predispositions, the most aware should not behave differently from the least aware individuals. They should receive and accept both messages, and as a result they should be ambivalent.

For many people this was not a familiar issue. Only 52% knew where each of the parties and leaders stood on the issue of Canada/US ties and knowledge of party positions was strongly correlated with political awareness (see Table 1).

The variable is construed as a scale going from 0 to 1. An alternative dichotomous operationalization of that variable produced very similar results.

This is very similar to Zaller’s index of factual information which counts the number of items the respondent has correctly answered.

Zaller’s model assumes that predispositions do not affect reception. The author argues that ‘this specification of the model would be inappropriate if … individuals engaged in “selective reception” of political information’ whereas the empirical evidence in the extant literature shows that ‘selective reception does not occur on a sufficiently broad scale to affect the diffusion of major political ideas, and hence poses little danger to [his] formulation of the reception-acceptance model’ (Zaller 1992, pp. 139–140).

We also performed analyses with partisan (party identification) predispositions as well as with a combination of partisan and ideological orientations, and we obtained similar results.

Graphical evidence and formal statistical tests revealed the presence of heteroscedasticity (the variance of the error terms was not constant). Consequently, we used robust standard errors to estimate our OLS regressions.

It could be argued that multinomial logit is equivalent to binary logit and that we should use the simpler technique. However, even if statistically both models estimate essentially the same coefficients, multinomial logit should be preferred to binomial logit to avoid the loss of efficiency due to discarded information (given that binomial logit ignores the presence of a third choice). See Alvarez and Nagler (1998).

As indicated above (see note 6 and Appendix B) the results are similar if we include all respondents in our model.

Some might argue that our indicator of predispositions is too close to our dependent variable (opinion on FTA). In fact, the correlation between position on Canada/US ties and the dependent variable is .58, a strong correlation (as it should be) but not overly strong, which means that the two variables are not the same. The first is a general attitude and the second an opinion on a specific issue. And here our approach is similar to that of Zaller, who relates opinion on the Vietnam War to general attitudes about war (hawk-dove predispositions) and Goren, who relates opinions on specific welfare programs to general views about government intervention. An alternative strategy would be to measure underlying economic interests (human capital or sector/industry exposure) in the passage of the FTA. However, the correlation between education and opinion on FTA is very weak (.05). As for sector/industry, we do not have the appropriate data. We believe that our operationalization adequately reflects underlying predispositions.

To guard against the misspecification of the model we used robust standard errors.

Note that this pattern does not emerge when we consider all respondents. See Appendix B.

We thank one of the referees who suggested that interpretation. Note that this explanation is exogenous to Zaller’s model.

In the case of low awareness we put every individual at 0 on the awareness scale, in the case of high awareness every individual was put at 1, and for middle awareness every individual was assigned a score of .5. As for the moment of campaign, “beginning” refers to the first 17 days of the campaign, “end” to the last 15 days, and “middle” to the 15 days between beginning and end. We used Clarify software to calculate these predictions (Tomz et al. 2001; King et al. 2000).

The probabilities were computed based on N = 603 for those with positive predispositions and N = 832 for those with negative ones. The standard errors range respectively between .04 and .13 for positive predispositions group, and between .02 and .09 for negative predispositions group.