Starting with the point-referenced data from Bert Vaux’s online survey of English dialects, we used a k-nearest neighbor smoothing algorithm to estimate the probability of seeing a particular answer–eg, whether a person would say soda, pop, or coke–at every point in the continental US.
The composite map gives a picture of the overall distribution, coloring each cell according to whichever answer is estimated to be most likely at that location. The more clearly one answer dominates, the darker the color. Individual maps show estimated probability of each particular answer at a given location, with larger probabilities shown in red and smaller probabilities shown in blue. At the moment, only the four most popular answers for each survey question are displayed.