David Allison

I control for something, and I think I'm doing a good thing by that, but in fact, I've created an inadvertent association.

I could go on and on with statistical obscura.

I think it's the confounding, particularly but not only by culture and socioeconomic status and social class education.

Second would be the measurement error, in particular, again, the non-random measurement error.

And then I would probably say some selection biases that are a little hard to specify.

Miguel Hernan at Harvard

is perhaps one of the most thoughtful people on talking about the other ways these biases can creep in, in terms of when people start the study, who gets in the study, and when we consider their exposure as occurring.

He does some ways to try to correct that.

He thinks that's more important than confounding.

So I think those are probably the three biggest ones.

Those are the intrinsic issues.

I think the other one, which is very big, but not intrinsic, is I'm going to say honesty, and maybe that's a little too strong, but I think it's the honesty or maybe the sincerity, perhaps is a better way of saying it, the sincerity of reporting by the investigators.

I don't think many investigators are lying in an explicit sense, but I do think there are both intentional and unintentional

efforts at distorting.

That is, people want to tell a story and they emphasize some things and de-emphasize others.

They hide some things and don't hide others.

So it's not an explicit lie, but there's a manipulation of the information.

For the majority of editors, it's lack of ability, lack of resources, and lack of courage.

For a minority,

of editors, it's limitations on the ability to really go in and suss it all out.