Ideological group influence: central role of message meaning

1. This model is not a formal path diagram but a conceptual model that represents our key theoretical predictions. In the interest of clarity, it omits lower-order paths required to estimate the model as well as other statistical information (e.g., variances, disturbance terms, etc., Hayes & Preacher, Citation2013). The full model for Experiment 1 is depicted in the online supplement.

2. These researchers demonstrated their proposed effects by pairing ambiguous messages about a telephone answering machine with credible or less credible sources.

3. Additionally, although we employ ambiguous messages in our studies, we note that ideological interpretations seem pervasive, even in scenarios in which message content appears unambiguous (cf. Cohen, Citation2003). For example, Crystal O’Conner’s statements left little ambiguity regarding her own interpretation of her service policy. Nonetheless, religious sympathizers and secular critics interpreted the policy along ideological lines, with the latter group imposing their own views.

4. See online supplement for details concerning calculation of these values, given the different scaling of our ideology items.

5. It is also important to note that the conditional indirect effects were not significant among individuals with moderate political identification who did not identify with either political group. These nonsignificant results reflect that ideological influence processes occurred only for individuals who identified with a liberal or conservative ideology, and not for individuals who lacked this social identity.

6. Careful readers will notice that we do not include the conditional indirect effects in our model results for this Experiment. The reason for this omission is twofold. First, because the mediator (social meaning) in Experiment 2 was operationalized as a binary variable, the conditional indirect effect does not have its usual interpretation (since the a path in this model now represents the effect of source on the log-odds of the mediator, rather than on the value of the mediator, itself). Second, because our primary goal in Experiment 2 was to assess the effect of manipulating the mediator (alteration condition), and because the interpretation of mediation models with binary M is not well-known, we opted to avoid these issues altogether and simply report the regression results from our two-step model.

7. We included two statements in order to enhance the generalizability of our effects and make the case that these results are not tied to one particular item or interpretation. We had identical predictions for both statements.

8. Careful readers may notice the inclusion of geographical information (Massachusetts vs. Texas) along with our source information in both studies. This was added to increase the realism of the manipulation by paring each senator with a ‘Blue’ or ‘Red’ state. However, we have obtained these same patterns of effects (source influencing meaning, meaning interacting with ideology to influence attitudes) in a variety of subsequent studies that did not include this information. This information was also absent in past research on social meaning and attitudes (Cohen, Citation2003; Wood et al., Citation1996). Thus, we feel confident that this ancillary geographical information is not the root cause of the effects reported here.

9. Additionally, we replicated this analysis in a mixed model regression predicting agreement as a function of time (time 1 = 0 vs. time 2 = 1) and our other model variables. In this case, the critical interaction manifested as a four-way recipient ideology x interpretations x alteration condition x time interaction. To avoid the complexities of decomposing a four-way interaction, we chose to report the simpler change score analysis here in the main body of the paper. For interested and ambitious readers, however, we have included this analysis in the online supplemental material.

10. Note that because statement did not alter the results of this model, we collapsed across statement conditions (see online supplement). Furthermore, in addition to this parsimonious theoretical model, we replicated our key result in a model containing all possible interactions among our key variables, which may also be found in the online supplement.

11. Of course, it is possible that the real influence of imagining a different ideological source was somewhat muted by social desirability concerns – perhaps it is difficult to admit the influence a source cue might have on one’s judgment when directly questioned. If so, there may be additional variation in attitudes accounted for by (presumably peripheral) evaluations of source cues. However, the absence of a source cueing effect is not a necessary precondition for our model. The claim we are making is that ideological meanings exert a direct, causal influence on attitudes – that ideological meanings matter. Whether source cues also matter in certain contexts (or mattered more than recipients were willing to self-reported) does not negate this point.

12. Although beyond the scope of the present article, initial pilot data support this notion, finding that participants make the same inferences about the social meaning of statements regardless of whether they are placed under cognitive load.