I happened across an article on the North Carolina State University research blog. One of their statistics graduate students has taken data from a decade-old Harvard University linguistics survey and turned it into graphical magic.
The survey is Dr. Bert Vaux’s Dialect Survey, which was conducted in 2002 and included over 30,000 Americans from all 50 states (even though the maps only show the contiguous 48 states).
What NCSU student Joshua Katz did was take the geolocated data from that survey and apply a “k-nearest neighbor” smoothing algorithm to estimate the likelihood at every point in the U.S. of using a particular dialect or word choice.
Continue Reading