Thanks for sharing your analysis and map, looks like you're off to a great start! My understanding of the NYPD data is that the time of day and even exact date of crashes are only loosely accurate, and I have been cautioned against doing analysis at the day and hour intervals for that reason.
I recently worked on a similar project for a non-profit in Hell's Kitchen called Chekpeds. They are currently in the process of acquiring additional funding for another phase of work on the web app, primarily to incorporate non-map charts similar the ones in your IPython Notebook, but more focused on comparing trends by NYC administrative areas such as City Council and Community Board so that local activists can use the tool to advocate for traffic safety solutions.
I know that Noel has been advocating for better quality of the crash data with city officials, but am not sure where that process is currently at. As I'm sure you've noticed, something like 20% of all crashes aren't geocoded, and the most common "contributing factor" category is "unspecified". From my understanding the data quality seems to vary between police precinct as well.
Anyway, it's likely that we'll start on phase 2 of the Crashmapper project at some point this summer. If you're interested in potentially working together let me know.