Act-r.psy.cmu.edu

- The Dynamics of Political Candidate Evaluation Sung-youn Kim ([email protected]) Department of Political Science, Stony Brook University Milton Lodge ([email protected]) Department of Political Science, Stony Brook University Charles Taber ([email protected]) Department of Political Science, Stony Brook University John Q. Public, a computational model of political mind on the spot influence the evaluations of ob- cognition which incorporates both cognitive and affective The accessibility of situational factors, to- mechanisms, is employed as a voter facing political cam- gether with the content and structure of prior beliefs paign information. A series of hypothetical, computa- determine what considerations and feelings come to tional experiments show that the model successfully re- produces a set of well-known empirical phenomena found – (Attitude Construction and Colorization) An atti- in electoral research and research on political cognition.
tude toward an object is constructed and/or up- Specifically, in response to issue and candidate infor- dated continuously in real time. That is, it is colored mation, the model reproduces 1) practice, recency, and by those thoughts and feelings that come to mind spreading activation effects on recall, 2) cognitive and at the time of information processing.
attitude priming effects, 3) question order and word- – (On-line Processing) An affective summary evalua- ing effects in survey research, and 4) both on-line and tion (valence) is linked to every object in memory that has been evaluated in the past and is updated Given the space limit, only some of axiomatic char- continuously on every exposure to new information, acterizations of the model and some of the simulation thereby reflecting the weighted influence from all results were presented here. (For formal presentation momentarily accessible information. That is, the and full discussion of the model and simulation results, valence of the new information at the time of up- see Kim, Lodge, & Taber, MPSA 2004 Conference Pro- dating is colored by the respective valence of those thoughts and feelings elicited at the time of expo-sure.
The model is built using ACT-R. Together with the cog- nitive mechanisms embedded in ACT-R, an affective, at- The computational experiments consisted of three exper- titudinal mechanism was incorporated into the model iments that were identical except for the model’s initial based on a set of axioms, some of which are the follow- ideological beliefs - typical conservative, moderate, and liberal beliefs. The initial ideological beliefs were derivedlargely from the NES (National Election Survey) 2000.
• (Hot Cognition) Most social concepts in memory are Within each experiment, an identical set of 24 pieces of affectively charged. The affective evaluation linked to campaign information were presented to the model, with a concept in long term memory can be positive or neg- varying order and wording and with three intervals. Im- ative or close to zero, indicating either a non-attitude portant parameter values used for the experiments were: 1) anchoring and adjusting (ρ = 0.7); 2) the parametergoverning the effect of affective strength of chunks on • (Attitude Priming) The information in memory that their decay rates (α = 0.1); 3) affective congruence ef- is affectively congruent to the information being pro- fect parameters. All other parameters were left to their ’common’ values (e.g. d = 0.5). In the below, some ofthe results were briefly presented.
• (Primacy of Affect) Affect can not only be triggered The Figure 1 shows the changes in attitudes towards automatically without conscious appraisal of an at- two fictitious candidates, James and Edwards, and two titude object, but also is primary in the sense that political parties, Republican and Democratic Parties, as it comes into working memory before other conscious the set of campaign information is processed (with mod- thoughts and appraisals enter into the judgment pro- erate initial belief). As expected, the valence of attitudes towards and the preference over the objects change over • (On-line and Memory-based Processing) time as the campaign information is integrated into theattitudes towards the objects. In the end, the model – (Memory-based Processing) Different, often con- was slightly positive toward Democratic Party, slightly flicting considerations and feelings that come to negative toward Republican Party, but positive towards Changes in Attitudes towards selected Nodes in Memory with moderate initial beliefs
Changes in Full Activations of Concepts (Cognitive and Attitudinal Priming)
Figure 3: Activation Levels of Chunks with ”Republi- Figure 1: Changes in Candidate Evaluations Over Time later. When the information were presented later, the both candidates, though slightly preferring the Demo- model’s attitudes toward both James and Edward in- creased mainly due to the changes in influences of the The Figure 2 shows the changes in full activation lev- information on its attitudes towards candidates.
els of two chunks, ”James” and ”Democrat”, over time,which is the sum of base-level activation, spreading of Changes of Attitudes in Survey Response
Changes of Attitudes in Survey Response
activation, and the attitude priming effect.
Changes in Full Activations of Nodes James and Democrat in Memory
Figure 4: Changes of Attitudes due to Order Effect Figure 2: Changes in Full Activations over Time This paper is drawn from Kim, Lodge, & Taber (MPSA2004 Conference Proceeding).
The Figure 3 shows cognitive and attitude priming effects with liberal initial belief. It shows the full acti- vation levels of four chunks in memory when the model Kim, Lodge, & Taber (2004). A Computational Model processes the word ”Republican” (i.e., ”Republican” is of Political Cognition: The Dynamics of Candidate a prime). At the moment of processing, ”rich” was both Evaluation. Proceedings of Midwest Political Science semantically associated with and attitudinally congruent to the prime. ”dumb” was only attitudinally congruent to the prime. ”Edward” was not related to the prime in Campaign Information and the Dynamics of Candi- any way (a base case). ”intelligent” was only attitudi- date Evaluation. APSR 89. (pp. 309–326).
nally incongruent with the prime. Consistent with cog-nitive and attitude priming effects, the order of chunks Lodge & Taber (2002). The Primacy of Affect for Politi- in terms of their activation levels, from the highest to cal Candidates, Groups, and Issues: An Experimental the lowest, were ”rich”, ”dumb”, ”Edward”, and ”intel- Test of the Hot Cognition Hypothesis.
The Figure 4 shows a question order effect in survey re- Zaller & Feldman (1992). A Simple Theory of the Sur- sponse. It compares the expressions of attitudes toward vey Response: Answering Questions versus Revealing the fictitious candidates with those when the informa- Preferences. APSR 36. (pp. 579–616).
tion about candidates’ party affiliations were presented

Source: http://act-r.psy.cmu.edu/wordpress/wp-content/uploads/2012/12/549KIM.PDF