Actually, it's worse than that. It sounds like the researchers looked at the FCC's census-tract numbers and multiplied each by the number of households in each

Actually, it's worse than that. It sounds like the researchers looked at the FCC's census-tract numbers and multiplied each by the number of households in each census tract. That inherits a big error baked into the census-tract numbers. In the FCC's numbers, if you have a census tract wherein one household has access to AT&T, one other household has access to Comcast, and none of the other households have any access to any ISP at all, that's a census tract with two providers. I suspect there aren't any census tracts where the situation is quite that extreme, but I suspect many "competitive" census tracts have at least some households that actually only have one option. That said, at least census-tract aggregation is better than the zip-code level aggregation that preceded it...