Oooh, what a big, random sample! Let's generalize this over the entire population of horse owners/riders, worldwide!!!
If the sample were truly random, that sample size is good enough to give a margin of error of only +/- 8% at a 95% confidence level. That is, the sample matches the population of the area the samples were drawn from, within +/- 8% 19 times out of 20. You’re quite right in saying that the results cannot be generalized beyond that area — and certainly not worldwide.
Of course, there could be a great many other problems with the methodology. However, the results are credible, because we know that the general population is getting bigger faster than the average horse.