This　study　investigates　the　differences　in　the　factors　affecting　the　injury　severity　of　speeding-related　crashes　occurring　in　the　daytime　and　nighttime.　Two　log-likelihood　ratio　tests　are　conducted　to　validate　whether　speeding-related　crashes　classified　by　daytime　and　nighttime　should　be　modeled　separately.　The　result　proves　that　separate　modeling　is　necessary.　Two　correlated　random　parameter　order　probit　models　with　heterogeneity　in　means　are　conducted　using　the　data　collected　from　2018　to　2020　in　the　United　States.　Model　estimation　results　show　that　urban　areas,　speed　limits,　and　young　and　older　drivers　are　temporal　instability.　Angle　crashes,　head-on　crashes,　intersections,　downhill,　exceeding　the　speed　limit,　drunk　driving,　and　motorcycles　are　statistically　significant　in　both　models　with　an　increased　crash　severity.　Interaction　and　heterogeneity　effects　between　random　parameters　are　also　reported.　For　instance,　large　trucks　driving　above　the　speed　limit　are　more　likely　to　increase　the　probability　of　severe　injury.