Split signals of credibility: the impact of regime type and polarization on audience costs

Abstract

International threats vary: some deter, others provoke, with consequences for escalation, bargaining leverage, and leaders’ survival. Audience-cost research links credibility to domestic punishment yet typically collapses regime diversity into a democracy–autocracy binary and sidelines polarization. This thesis addresses both gaps by comparing democracies, anocracies, and autocracies and testing how affective polarization reshapes credibility within each. Theory: regime type sets a baseline for audience costs—high in democracies, low in autocracies, and intermediate in anocracies—while polarization diffuses punishability in democracies (weakening credibility), concentrates it in anocracies (strengthening credibility), and leaves autocracies largely unchanged due to insulation. Design merges MID with COW covariates and V-Dem regime and polarization measures to build a dyadic panel of 1,491 dyad–years (1816–2014) with standard controls (capabilities, contiguity, alliances, trade, GDP, target regime). Logistic regressions evaluate whether targets reciprocate (low perceived credibility) or comply (high perceived credibility) as a proxy for audience-cost likelihood; hypotheses address baseline regime differences and polarization’s moderating effect within regime families. Results in the main models are statistically significant: democracies display higher inferred audience-cost potential than autocracies, anocracies fall between; polarization erodes credibility in democracies, amplifies it in anocracies, and shows no systematic effect in autocracies. Scope conditions include proxy measurement of audience costs, regime coding choices, and polarization indices; these constraints qualify but do not undermine the contribution linking domestic punishability to international threat credibility. Definitions, coding thresholds, and data limitations are stated to maintain transparency and interpretive caution across cases and time.