Apologies, I should not have indicated that it "proves" that this is so. Rather, it does paint a convincing picture that something is there. It is akin to a smell test, if you will.
So, yes, lets debate the confounding factors. If you can name some factors, they should guide us in how we would build tests to explore them. I didn't claim it was easily testable, but it is certainly testable.
Things are "testable" when you can control all variables except one (in this case, MHL). This is rarely possible in real life, except when a natural experiment presents itself:
So in this case, it might be "the same city, before and after MHL was enacted." Then we could graph the number of cyclists and the number of injuries, and you'd have something.
I would argue that things are "easily" testable when you can control all of the variables. For most of history, we have rarely ever controlled all variables outside of the easy cases. Typically, we define away much of the extra stuff. That has not stopped us testing what we can, just be sure to disclose everything else. Obviously, things are more confidently testable if you have controlled all variables, but that would be a crippling condition.
Basically, following the link in wikipedia will get you to https://en.wikipedia.org/wiki/Scientific_control which allows for "A scientific control is an experiment or observation designed to minimize the effects of variables other than the independent variable (i.e. confounding variables)." Which is much less strongly stated than what you have. The idea is still there, of course.
ok. It's not really a "crippling condition" when natural experiments do crop up from time to time, though.
For example: if Amsterdam suddenly had an MHL, that would be one. In fact, any city that instituted MHL would be a natural experiment. The number of riders, the number of accidents, the % of accidents that involve injuries -- all those things would change, while the city design and transit situation (other than bikes) presumably would not.
So, yes, lets debate the confounding factors. If you can name some factors, they should guide us in how we would build tests to explore them. I didn't claim it was easily testable, but it is certainly testable.