Depends on what you're measuring.
If you're measuring speed per connection, then including people without connections is wrong.
Speed per person: technically, yes, scribbling in "0" for those without connections is arguably correct. How meaningful is this number? If you have a town full of Luddites who refuse to connect regardless of availability, it's going to lower your average; but piping 10-gigabit fiber to every home in that town won't change anything except the level of money in your pocket.
"Entitled to answer 0 if they haven't hooked up your water" is not a very good analogy and is probably a red herring. Where I live, you can't have a legal residence without having active water, sewage, and electric service; but phone and internet service aren't considered to be necessary for it to be habitable. So if I'm not hooked up to water, "0" is a meaningless number because either there's a temporary service interruption (which is counted in a different way), it's not a legal residence, or it's time to call a lawyer. Sorry, but "there is no water" isn't the same as "water pressure is zero".
And it's hardly fair to count somebody's internet connection as 0 if they don't want one. It's like saying my phone service is crap because I haven't asked for a phone line. Your local grocery store doesn't carry food because you don't shop there.
Average or mean or median or whatever connection speed is one metric, but "per person" seems like a weird way to go about it. Let's see -- posit two homes, side-by-side. Both have identical service. One has 10 people living there, the other has 2. Are you going to divide the connection speed by the number of people living there? What if three of them are hanging out next door?
OK, you didn't say "per person", let's say you mean "per household." If a household is comprised of people who get all their Internet via phone service, one phone per person, how are you counting them? Do you add all the phones' rates and come up with a total? No, can't do that -- you've already said you didn't want a per-connection rate. But obviously 10 people armed with 100Mb connections (let's pretend they're actually getting that) will total 1Gb of available data. What happens if one household has two wired connections? Because I've done that myself. Two ADSL connections; one for me, one that everybody else shared. Per household was the sum of the two, but it seems to me that you have to count that as two connections; it required two phone lines, two modems, two routers, and twice the cost per month.
Hey, I could have brought in a third line. So if we're counting things your way, I should count that as a third line at "0" because I chose not to.
What about an abandoned house, or one at least with nobody in residence? What does that count as?
I have three phones. (Please don't ask why, I'll start to whimper.) What is my available speed? I can't meaningfully aggregate the connections. OTOH I could, say, watch Youtube on one while downloading movies on another. But is the aggregate speed really useful? Most of the time I'll be using one at a time, with the other two effectively idle. (Actually, the oldest phone sits in my car and plays music while I'm driving. It's more convenient than playing silly buggers with the connections every time I climb in.) And yet, they are three separate connections. I COULD perform multiple actions if I needed to. Or I could lend one to a friend who left his charger at home. Or... whatever.
If you don't understand why people are disagreeing with you, that's why. Any speed metric that isn't "per connection" doesn't make sense, because you end up with a metric that isn't really useful. Running around trying to count _potential_ connections as part of a data rate statistic? That way lies madness.
Mind you, there ARE things you can do with what you're saying. If there's a place where there are homes and businesses that can't get Internet, then that's not the same as saying they're getting 0bps of service; but there IS a metric that applies: percentage of homes or businesses that have service available. If you want to argue that that's at least as important a statistic as speed, then a lot of people will agree with you. It's just that you're trying to lump it into the speed metric, which makes both metrics invalid.
And then there's the listed-vs-actual bandwidth. Or the burst-vs-steady bandwidth. I'm on gigabit cable, which means that at 6pm I'm probably sharing my bandwidth with my neighbors; but at 3am, probably not so much. So there are arguments about all of that, and they're probably all valid things to measure and report, and TFA's quoted report is probably over-simplifying by ignoring all that.
Please don't fall into the same trap.