If tests aren't accurately predicting usage, should we not develop more accurate tests?

It doesn't get more accurate. We should just scrap the combined "MPGe" (and EU equivalent) and stick with "electric range" and "MPG". Both of those can be fairly accurately predicted as separate values. How they combine is entirely up to the individual.

I agree with everything you're saying, but this part. As you stated before, those are intended to allow for an apples to apples comparison and make it easier for the consumer to judge a car's funel economy, without having to do their own math (which - lets face it - most people suck at).

If the underlying usage pattern doesn't reflect a typical average use, that's an issue, that can be adressed. And when studies show that they don't why not take that as a call to improve upon the methodolooy?

There's always going to be the caveat that one's one usage pattern might deviate greatly from the standard, and absolutely it's a must, that the individual values are indicated, so people CAN do their own math. But having a standard combined measure is still a useful tool.

Addendum. I have to admit to really only having read the article just now.

The scientists have called on EU regulators to adjust their measurements to fit the real-world findings, urging stricter controls as to how fuel consumption of plug-in hybrids is measured. According to the findings, the CO2 limits a car manufacturer’s fleet of vehicles is allowed to emit on average needs to be urgently adjusted.

Contacted by SWR, the EU Commission, which is responsible for CO2 limits, declined to comment on the findings.

The German Association of the Automotive Industry said it believed that existing measures to determine fuel consumption and CO2 emissions were reliable.

This is the real underlying issue here. It's the EUs regulation on CO2 emissions reduction, that car manufacturers are abusing here. They are designing their car in such a way, that they look good on paper and can pass the requirements of the regulation, while their real-world emissions are much higher. (And in that regard, it's not too unlike Dieselgate. Minus the intentional technical manipulation ofc.)

While your claim, that it's the individuals responsibilty how they use their car is obviously true for an individual car's fuel consumption, that realization is also utterly useless as a basis for effective policy. There needs to be a standard, and that standard better reflect an empirical assessement of reality.

The issue isn't really that the estimate is wrong, it's that it's wrong by an enormous amount -- and one that's been increasing every year. I don't think that the study is trying to say that these vehicles are inefficient as some kind of absolute judgment, but that they're less efficient than estimated (although there are big differences based on vehicle make and model).

I don't think the problem really lies with manufacturers, it's that the current tests aren't accurate enough to predict real-world usage closely enough. Although, driver input is mediated by computer systems and if on-board systems are being too aggressive in switching over to ICE, I suppose that's a manufacturers problem.

Really, they've been doing these very large studies for a long time. The sample size is large enough to capture the full diversity of driving styles and it cannot be a few outliers skewing results. Since 2012, the disparity between estimated and observed fuel usage has grown every year. Why? Why is it changing and why is it always changing in the same direction?