I’ve always been proud of our Salisbury University GIS students, and love to push them as far as their little minds can handle it. You may recall that last Spring I had my Advanced GIS students perform independent GIS projects and present those projects as posters at an Undergraduate Research Conference. Well, this Fall I am teaching GIS Programming, and have 7 awesome students (pictures and bios to follow). We started the year off learning spatial SQL with Postgres and PostGIS. We have now moved into Python, which includes Arcpy as well as other Python packages.
The semester was going so well, and the students were so responsive to anything I asked, I said what the heck, let’s try something crazy. So, I showed the students how to use two Python geocoding packages (geocoder and censusgeocode) and then said:
why don’t we conduct a research project over the next week to test the match rates and positional accuracies of the Google API and the United Census Bureau API.
So yeah, I gave them a week to put this together: design, code, analyze, and write. And, like most of my students at this level, they didn’t disappoint me. This meant they had to integrate a lot of what they have learned over the years (programming, GIS, statistics, etc.).
I just uploaded their work onto researchgate:
I was surprised by how little there is out there in terms of quantitative assessment of geocoding accuracies. I hope you have a chance to click on the link and check out the working paper (we will submit it to a journal sometime soon). Also, I included a short abstract below so that you can see the results of our work (note: our paper includes the original data and the source code for performing the geocoding):
Undergraduate Research in Action: Evaluating the positional differences between the Google Maps and the United States Census Bureau geocoding APIs
Abstract: As part of a class assignment in GIS Programming at Salisbury University, students evaluated 106 geographically known addresses to determine the match rate and positional accuracy obtained using the Google and the United States Census Bureau geocoding application programming interface (API)s. The results showed that 96.2% of the addresses supplied by the Google API were successfully geocoded, while 84% of the addresses supplied by the Census Bureau API were successfully geocoded. Further, the Google API matched 90% of the addresses with a ROOFTOP designation. The average positional accuracy of the Google derived addresses were 80m overall, and 65m for those geocoded with the ROOFTOP designation while the Census Bureau positional accuracy was 271.09m.
So yeah, this is what you can do with 7 GIS undergraduates at Salisbury University: they work hard, fast, and are a very creative bunch.