Text, graphic, or both? Modeling pedestrian preferences for eHMI formats with trust and risk propensity as latent factors

External human–machine interfaces (eHMIs) can support communication between autonomous vehicles (AVs) and pedestrians, but preferred information formats may vary across traffic contexts and individuals. This study examined pedestrian preferences for Textual, Graphic, and Textual-and-Graphic eHMIs using a stated preference survey. Five scenario attributes—roadway median type, AV type, lane position, traffic volume, and weather—were varied across 16 scenarios. Five hundred respondents each completed four choice tasks, yielding 2,000 observations. Multinomial logit and integrated choice and latent variable models were used to examine associations between format choice, scenario attributes, sociodemographic characteristics, walking habits, three dimensions of unsafe-crossing tendency, and trust in AVs. Textual-and-Graphic eHMI was selected most frequently (47.7%), followed by Graphic eHMI (33.4%) and Textual eHMI (18.8%). Model estimates indicated that format preferences varied with roadway median type, AV type, lane position, traffic volume, weather, gender, and age. Signal-disregard tendency was associated with lower utility for Textual eHMI, whereas the reported statistics did not support an independent association between trust and format choice. The findings provide design-oriented evidence about stated eHMI preferences but require validation using behavioral experiments before safety conclusions can be drawn.