AbstractCross-national comparisons in political science routinely assume independent observations, risking overconfident inferences when shared historical lineages drive apparent correlations. We introduce a Bayesian phylogenetic survival framework to confront Galton's problem in the study of regime longevity. Using the LIED dataset (1789-2022), we transform 29 empirically observed regime-type "taxa" into spell-based survival data and construct a neighbor-joining phylogeny from seven binary institutional traits (e.g., suffrage rules, competitive elections) via Hamming distances. We then fit two multilevel survival models in brms/Stan: (i) a "dynamic" model treating spells as independent, and (ii) a "phylogenetic" model embedding regime-type relatedness through a random-effect with covariance matrix A derived from the tree. Comparing models via leave-one-out cross-validation, Bayes R², and credible-interval widths, we find that initial wealth (log₁₀ GDP per capita) strongly prolongs 19th-century regime spells, attenuates in the mid-20th century, and partially rebounds after 1980. Incorporating phylogenetic covariance modestly "shrinks" effect estimates toward the grand mean and widens uncertainty, yielding a small but non-negligible improvement in model calibration (ΔELPD ≈-1.8, SE = 1.8). Sensitivity checks-comparing complete-case versus multiple-imputation treatments and alternative GDP series-confirm the robustness of substantive conclusions. Our approach offers a principled remedy to non-independence in comparative politics by uniting evolutionary phylogenetics with Bayesian survival analysis.