<I don't know why some like Mark have such blind faith in "computer models" and object to criticism on that point, but those of us in the programming field see this result more often than not. Yet another flawed EPA study.

This is like saying we shouldn't do manned space flight because sometime o-rings get too cold.

A mountain of software goes into these things. Bugs will happen every time. That doesn't mean that we should give up on this whole whacky data processing concept.

The fact that they found the problem and announced it should make everyone happy, especially you. By the model you use (yes Phil, even you are a modeler) to predict modeler's behavior this should not even occur: since the error "supported their beliefs", it should have made them happy and they should have left it in and kept quiet about it. No?

What speaks VOLUMES is this quote...
"Our deadline for getting this out is the end of 2003 and this software issue may impact our ability to meet that deadline," Martyak said. "We still believe strongly in the tie between the particulate matter and the health effects that it creates."



Here we see the study's sponsor admiting to holding strong beliefs. Good for him/her.



Holy smoke, Phil! Didn't you ever hear of black lung/white lung diseases? Asbestos-caused lung cancer? How could any sane person disagree with that statement? Believe me, it is well proven that inhaling particulate matter affects your health.

But now tell me that the results of "computer models" do not "somehow" deliver what the sponsor expects. If not, like here, hey just re-work them a bit.

This is not typically how it works. Just like in any other scientific endeavor, the results of the prediction are compared to real-world data. When you see the discrepancies is when you "re-work things a bit". Not to fit your beliefs, but to make the model better predict what the world actually does.

If a model is constantly contradicted by real data, then it will be discredited and no one will use it.

This predilection is why double blind studies are good science and why computer models aren't. They too easily get co-opted by the modelers subconscious preferences even without trying.
They cannot be trusted.

I think you're missing the point about computer models. They are not used for proving anything or predicting the future with certainty. Of course they can't be trusted: they are just our best guess about how something is going to work. After a model has been around for a long time and its results have shown to be sufficiently accurate, sometimes people start to accept it as being a good model and stop tinkering with it so much. Still, it's just a model.

We still use the Newtonian model of the universe. Even though it's guaranteed to give you incorrect results if your measurements are sufficiently precise, it is still a good enough approximation to be quite useful.