Table of Contents
Fetching ...

When Truth Misleads -- Phase-Aware Coherence Detection for Misinformation Correction Across Epistemic Communities

Heimo Müller, Andreas Holzinger

Abstract

Truth can mislead not because it is false but because delivering it through the wrong channel or authority to an audience with a different epistemic frame can harden misbelief rather than reduce it. Conventional fact checking assumes a shared epistemology that corrections from credible institutional sources will be received constructively across audiences. Evidence suggests instead that effectiveness declines and can reverse into backfire as the distance between source positioning and recipient epistemic orientation grows. We introduce Phase Aware Coherence Detection PACD which operationalises epistemic orientation from pre intervention assessments across institutional trust scientific epistemology conspiracy or anti elite openness and experiential or alternative epistemology. In a study of 45 participants we stratified individuals into three epistemic clusters and randomly assigned them to traditional fact checking phase aware coherence analysis or control before evaluating three claims 5G health effects urban trees and air quality and mRNA vaccine mechanism. Traditional fact checking shifted beliefs toward truth on average but produced substantial backfire and sharply reduced effectiveness among more sceptical participants. Phase aware coherence analysis produced little average change overall yet remained stable across epistemic clusters and substantially reduced backfire. On identity adjacent claims traditional fact checking increased confidence while reducing accuracy whereas phase aware framing avoided this harm. These findings motivate a shift from content centric to epistemology centric correction when truth misleads the problem is often a mismatch between audience orientation and the reference frame of the intervention.

When Truth Misleads -- Phase-Aware Coherence Detection for Misinformation Correction Across Epistemic Communities

Abstract

Truth can mislead not because it is false but because delivering it through the wrong channel or authority to an audience with a different epistemic frame can harden misbelief rather than reduce it. Conventional fact checking assumes a shared epistemology that corrections from credible institutional sources will be received constructively across audiences. Evidence suggests instead that effectiveness declines and can reverse into backfire as the distance between source positioning and recipient epistemic orientation grows. We introduce Phase Aware Coherence Detection PACD which operationalises epistemic orientation from pre intervention assessments across institutional trust scientific epistemology conspiracy or anti elite openness and experiential or alternative epistemology. In a study of 45 participants we stratified individuals into three epistemic clusters and randomly assigned them to traditional fact checking phase aware coherence analysis or control before evaluating three claims 5G health effects urban trees and air quality and mRNA vaccine mechanism. Traditional fact checking shifted beliefs toward truth on average but produced substantial backfire and sharply reduced effectiveness among more sceptical participants. Phase aware coherence analysis produced little average change overall yet remained stable across epistemic clusters and substantially reduced backfire. On identity adjacent claims traditional fact checking increased confidence while reducing accuracy whereas phase aware framing avoided this harm. These findings motivate a shift from content centric to epistemology centric correction when truth misleads the problem is often a mismatch between audience orientation and the reference frame of the intervention.
Paper Structure (44 sections, 3 equations, 7 figures, 11 tables)

This paper contains 44 sections, 3 equations, 7 figures, 11 tables.

Figures (7)

  • Figure 1: Main results: Group $\times$ Epistemic Phase interaction.(A) Mean belief change toward truth (95 % CI) as a function of experimental group and epistemic phase cluster. Group A (traditional fact-checking, red) exhibits a monotonically declining slope ($-0.79$) from anti-institutional C1 to pro-institutional C3; Group B (phase-aware coherence analysis, teal) maintains a near-flat profile (slope $+0.08$) across all clusters. The dashed line marks zero change. (B) Backfire rates (proportion of observations where belief moved away from truth) by group and phase cluster; bold percentages indicate each group's overall rate. (C) Mean belief change toward truth by claim type and group (error bars $= 95\,\%$ CI), illustrating the complementarity of the two interventions across epistemic complexity and identity-load.
  • Figure 2: Left: Histogram of $\varphi$ ($N=45$), colour-coded by cluster. Dashed lines mark buffer-zone boundaries; the dashed curve is a kernel-density estimate. C1 (red, $n=5$) clusters near 0 (anti-institutional); C3 (blue, $n=19$) clusters near $\pi$ (pro-institutional). Right: Violin plots with jittered individual data points.
  • Figure 3: Left: Dimension profiles ($I$, $S$, $C$, $E$) by cluster (mean $\pm$ 95 % CI). $I$ and $S$ increase C1$\to$C3; $C$ and $E$ decrease. Right: Heatmap of all 15 belief items by cluster (green=agree, red=disagree; brackets = $\varphi$ contribution direction).
  • Figure 4: Pearson correlation matrix of phase-defining items and $\varphi$ ($N=45$). Within-block correlations are positive; cross-block $\{I,S\}$ vs. $\{C,E\}$ correlations are negative, confirming the two construct poles are genuinely opposed.
  • Figure 5: Mean source trust (0=no trust, 1=complete trust; mean $\pm$ 95 % CI) by cluster. Institutional and scientific sources increase C1$\to$C3; alternative health and anecdotal sources show the opposite gradient, cross-validating the dimension profiles from Part A of the survey.
  • ...and 2 more figures