[gumby]

> Even in cases where people obviously should use controls, such as clinical trials, they frequently don’t.

This is not true in my experience (I have designed and run pharmaceutical clinical trials in humans). Can you give some examples?

Some treatment trials do compare the existing standard of care vs the new one rather than placebo against proposed treatment, but those are certainly controlled studies too.

I do know of cases where a control is impossible, such as some surgical procedures, though even then sometimes sham surgery is performed (this is controversial). Those are singly-blinded controls as the surgeon knows.

[redavni]

It is endemic in statistical studies based on medical data. They reach a conclusion that is statistically supported but never do the work to verify it.

[gumby]

A statistical study is not a clinical trial, which was the term used by the GP poster.

[PaulHoule]

Definitely you want to use controls if you want regulators to take your work seriously. A lot of stuff gets found on pubmed that isn't meant for regulators and it doesn't.