Abstract
Motorcyclists account for ~83.8% of Thailand’s road-traffic deaths, yet fatalities per crash—the metric most relevant to intervention—have rarely been modelled for motorcycles using frameworks that handle both underdispersion and zero-fatality inflation. Building on prior aggregate-level zero-inflated Conway–Maxwell–Poisson (ZICMP) research on Thai road deaths, we compared Poisson, CMP, Zero-inflated Poisson (ZIP), and ZICMP models, each incorporating location-by-collision-type interactions, using seven years of motorcycle crash data (2019-2025; n = 23,335)—to our knowledge the first interaction-based ZICMP analysis of this outcome. The ZICMP-interaction model fit best (AIC = 32,940.71; BIC = 33,367.76; ν ̂ = 2.54, confirming underdispersion). Head-on and fixed-object collisions raised expected fatalities (IRR = 2.77 and 1.69; p < 0.001). Nighttime crashes proved far more lethal than daytime ones (39.6% vs. 25.3% fatal; OR = 0.23, p < 0.001), while April (Songkran) showed the highest non-fatal odds (OR = 1.36, p = 0.002), suggesting its toll stems from crash volume, not severity. Interactions also cut lethality at intersections (head-on, ~37%; IRR = 0.49) and access points (fixed-object, ~70%; IRR = 0.21)—patterns additive models miss. Nighttime open roads, not festivals, emerge as the priority target for motorcycle-safety interventions.