To enhance highway bus traffic safety and reduce accident rates, data from 2, 320 samples are collected, including bus accident frequency, traffic volume, road characteristics, weather conditions, and holiday proportions, from three highways in eastern China．Accounting for accident heterogeneity, Poisson-lognormal（PLN）, random parameters Poisson-lognormal（RP-PLN）, and random effects Poisson-lognormal（RE-PLN）models are established．The Bayesian method and Markov chain Monte Carlo simulation estimate the posterior distribution of model parameters．The modeling results quantitatively assess the impact of various factors on highway bus accident frequency．Findings indicate that the RP-PLN model exhibits superior goodness-of-fit compared to the PLN and RE-PLN models．Road segment length, traffic volume, and holiday proportion show positive correlations with accident frequency, while temperature, relative humidity, and visibility demonstrate negative correlations．Holding other factors constant, a 1% increase in road segment length and traffic volume increases accident frequency by 0.90% and 0.62%, respectively．A one-unit increase in holiday proportion increases accident frequency by 8.30%．Conversely, a one-unit increase in temperature, relative humidity, and visibility reduces accident frequency by 5.72%, 8.82%, and 6.55%, respectively．