How do the structural drivers of civil conflict relate to one another empirically, and do those relationships hold across peace and war regimes? This paper presents a statistical validation of the causal architecture underlying a Bayesian network model of conflict, drawing on data spanning climate stress, resource scarcity, political instability, and economic development across a global panel from 1980 to 2025.
We construct a multi-source dataset linking heat stress, labor and food shortages, energy access gaps, military expenditure, and political instability indices to binary conflict outcomes from UCDP-PRIO. Spearman correlations validate individual links in the proposed Bayesian network, confirming that state fragility, energy deprivation, and low development cluster together as the dominant structural axis of conflict risk, while climate stress carries independent signal not reducible to poverty or institutional weakness.
We then conduct principal component analysis across pooled and regime-stratified samples to examine whether the dependency structure shifts between peace and war. Four components together explain nearly all variance in the data. Two of them, the fragility-development axis and the climate stress axis, are stable across regimes, justifying pooled treatment in the Bayesian network. The remaining two are regime-dependent: fractionalization loads onto the state power dimension only during war, and the role of income shifts across regimes. This means the network cannot treat all nodes symmetrically. Relationships that appear structural in aggregate may be conflict-specific, and those that appear conflict-specific may reflect deeper economic reorganization under war conditions.
These results provide empirical grounding for a Bayesian network that encodes regime-sensitive structure, with pooled pathways where the evidence supports stability and differentiated conditional probabilities where it does not.