People encounter evidence of varying quality every day—in courtrooms, medical consultations, and political discourse. How do they integrate that evidence into a judgment? Two dominant frameworks have shaped our understanding of this process. The first, motivated reasoning, holds that individuals selectively discount or ignore information that conflicts with their prior beliefs or desired conclusions (Kunda, 1990; Lord et al., 1979; Molden & Higgins, 2005). The second, Bayesian updating, holds that rational individuals weigh each piece of evidence by its perceived diagnostic strength (Tversky & Kahneman, 1974). Neither framework, however, adequately accounts for a well-documented empirical pattern: the dilution effect, in which adding weak-but-valid evidence to a set of strong evidence actually reduces the overall persuasive impact (Birnbaum, 1974).

We propose an alternative account grounded in information integration theory (Anderson, 1981; Kaplan, 1992): the Averaging Model of Information Integration. Rather than selectively weighting or ignoring evidence, individuals may compute a weighted psychological average across all available information. This averaging process predicts the dilution effect naturally—weak evidence pulls the average toward the midpoint, attenuating the impact of strong evidence. Critically, we further predict that political motivation moderates this averaging process: when evidence supports a preferred conclusion, individuals will engage in more inclusive averaging (allowing weak evidence to “rescue” or dilute strong opposing arguments), whereas in opposition-aligned conditions, individuals will more readily discard weak arguments and engage in selective processing consistent with motivated reasoning.

Understanding how citizens integrate evidence is foundational to addressing two of the most pressing problems in democratic discourse: political polarization and the persistence of misinformation. If individuals average rather than filter information, then the common strategy of providing a factual correction alongside weaker supporting arguments may inadvertently dilute the corrective effect—a counterintuitive and practically consequential implication. In polarized environments, where citizens are routinely exposed to evidence bundles of mixed quality on politically salient issues, the averaging process may help explain why high-quality corrections so often fail and why partisan divides persist even when accurate information is widely available (Simon, 1990; Gigerenzer & Gaissmaier, 2011). By specifying the mathematical and cognitive mechanism underlying this process, the proposed research provides empirical grounding for designing more effective communication, legal, and policy interventions.

To test this theory, we will conduct a multi-arm survey experiment (N = 3,000) manipulating two orthogonal factors: (1) evidence bundle composition—strong discriminative arguments only vs. strong plus weak arguments—and (2) political alignment—evidence that supports vs. opposes participants’ preferred political outcomes. Participants will evaluate a hypothetical but politically salient court case, rendering judgments of defendant guilt, confidence, and downstream policy support. By systematically varying the ratio of strong to weak arguments, we can cleanly identify whether weak evidence dilutes the impact of strong evidence—a signature prediction of averaging models not shared by traditional additive models. We predict a main effect of evidence quality and, critically, a significant Alignment × Evidence Composition interaction, driven by the asymmetric engagement of averaging under motivationally favorable versus unfavorable conditions.

The C-SPAM Seed Grant will support the following activities over a 12-month period: (1) Stimulus development and piloting: creation of realistic court case vignettes, evidence arguments (strong and weak variants), and political framing manipulations; (2) Survey design and IRB approval; (3) Data collection via online survey platform (Qualtrics) with quota sampling to ensure political diversity; (4) Statistical analysis in R, including robust linear modeling and sensitivity analyses; (5) Manuscript preparation and submission to a top-tier political science or psychology journal; (6) Conference presentation at APSA or SPSP to disseminate findings to the broader research community.

These activities will position the research team for federal funding (NSF, NIMH) by providing robust preliminary evidence of the averaging model in political contexts. Preliminary data demonstrating the hypothesized interaction effect will be essential for a competitive NIH or NSF proposal, where funding agencies prioritize “proof of concept” from seed grants.