The article could be wrong as bucketing by %tile excludes the un-served. It could be that the 16 modem can serve ppl that are otherwise out of range for 16e, ju

The article could be wrong as bucketing by %tile excludes the un-served. It could be that the 16 modem can serve ppl that are otherwise out of range for 16e, just at super low speeds. If a lot of 16 users are in places the 16e doesn't go, that skews the 16's bottom 10%tile's mbps lower, but also means more users are served and therefore it has a better modem. Given this data is from Ookla's Speedtest, I wonder if a lot of people in super super low reception areas wind up running Speedtest to see why their connection is slow, and are thus overrepresented in the data. If that were the case, the conclusion is even more statistically off.