Do you know what empirical means?
Do you? Please go and find a definition.

Oh, heck, I know you won’t because you don’t want to put that much effort into trolling.

“Empirical research is a way of gaining knowledge by means of direct observation or experience. It is used to answer empirical questions, which must be precisely defined and answerable with data.”

This study uses pure numbers. No anecdotes, no ‘psychological studies’, (which, although they may be scientific, don’t get me wrong, aren’t empirical), nothing qualitative, just numbers.

Now, what part of that claims reasonable sources of error? In the scientific community, we publish and take on preliminary studies all the time. The data we want can be hard to get, (read: expensive, in time/money/both), and the best way to convince someone to give us what we need to do a proper study is to find/perform a preliminary study that suggests the world doesn’t work quite the way we think it does.

That said, I can see why you might trash the study, because it does seem to have plenty of sources of error that you can see. And I agree, it’s definitely not conclusive, but, like correlation, it does waggle its eyebrows suggestively while mouthing “look over here”.

And besides all that, the point of the study was “There does not exist”. When you consider sources of error, you should also consider the influence those errors might have on the data. The types of sources that Mike, (and you), mention, would reduce the impact, not eliminate it, (Less drivers on the road, passengers talking instead of drivers, etc.).
The study shows no significant increase.

Now, if you’re capable of thinking, or if you have a strong, unhelpful bias, you might realize that doesn’t mean the study is scot-free.
What about un-reported accidents? Maybe people on cellphones get into a lot more of those, but don’t report it because there’s no one around, or their insurance would freak and they’re afraid the insurance might realize they’re on the phone while driving. So there COULD be an increase, but it would be invisible.
Maybe, there’s nearly no one who drives & uses a cellphone, (As opposed to Mike’s point, where the ratio of drivers to passengers using cellphones would reduce the estimated # of cellphone drivers to 1/3 or 1/4).

And THEN, we can consider another thing that’s kinda been mentioned. We do, of course, have a lot of studies that people who use cellphones while driving get in a lot more accidents. That people who do, are a lot more distracted, pay less attention to road, don’t notice drivers around them, etc.
All those should be a pretty good reason to ignore this study, right?

Wrong. These studies don’t necessarily contradict each other.
All those studies above indicate, but do not prove, that cellphones cause traffic accidents. What they prove is what they say they prove and no more: Statistically speaking, the population of drivers who talk on cellphones are more likely to get in an accident.
Correlation.

As DCX2 said:
“I also agree with the point that bad drivers are still bad even without a cell phone to talk on.”
So bad drivers might use cellphones more often while driving, but the population of drivers are not significantly affected by using a cellphone.

TL;DR: Yes, the study doesn’t mean much. No, it is not wrong, no, you should not throw it out.
What it means, is that closer research should be done into establishing or disproving a causal link between using a cellphone while driving & driving accidents.

Alternatively, we could play it safe and just ban cellphone use . . . nope, wait, we already have well established data that says when we do that, accidents increase because people try to hide their law-breaking rather than stopping potentially unsafe practices.