I am a traffic engineer that deals with crash reports quite often. One thing we include at the beginning of every crash analysis we do is a disclaimer that the contents of the crash report are subjective and solely the work of the responding officer. This means that junction/non-junction or crossing/non-crossing can vary a lot. The data are very messy with inaccuracies in location and direction of travel. (Side note, it is super frustrating when the road winds a lot and the eastbound direction is traveling westerly, for example.) The data that are reported to FHWA and NHTSA are even more generalized that those kept by the individual states. Additionally, pedestrian and bicyclist crashes tend to be underreported unless there is a severe injury. This can greatly skew the data to look more rosey than reality.

The cited data [0] is presented in a super confusing way - the categories are "At Intersection" (18%), "Not At Intersection" (73%), and "Other" (9%). Apparently "other locations such as roadsides/shoulders, parking lanes/zones, bicycle lanes, sidewalks, medians/crossing islands, driveway accesses, shared-use paths/trails, non-traffic way areas, and other sites," which make up the "Other" category, are neither intersections nor non-intersections.

I'm gonna guess the "Not At Intersection" category refers solely to jaywalkers.

[0] https://crashstats.nhtsa.dot.gov/Api/Public/ViewPublication/...