"doing it" is is inherently different from the naturally occurring action. It might be useless

A common way to improve things is to use what we know is good, and do it intentionally.

if we know that married people are happier, - a hypothetical example - we might want people to get married.

but the example exemplifies the problem. People that get married without our advice, are probably more compatible and more in love than those to whom we advice that "get married, it is good for you"


The same principle applies for anything that is found to be good and we try to do it on purpose.

Analysis:

1) The most common problem is that the artificially created situation is different from the naturally occurring one. I am not against artificial things. They are just different. The guy getting married "because it makes people happy" is having a different marriage than the more natural marriage.

2) Another error is mistaking trait and intervention. 

if optimistic people are richer happier etc. - then, goes the nativity, lets teach people to be optimistic. 

But teaching optimism might not make people optimistic. Even if it does, the nature of this optimism will be different from the naturally occurring one. 

3) Less central is mixing of correlation and causation. But this is a well known problem. This too, might cause the above intervention error. if something is correlated with a positive outcome, a native observer will try to induce the correlate, even though the correlate does not have any causative relation.

Example for error 2: Mindfulness trait is strongly correlated to many positive psychological measures (Brown 2003,2007), but the theory that practicing mindfulness (meditation) works better than placebo (the effect of "doing something you believe will help")  is not yet proven. All meditation studies do not contain an active placebo control. There is one study I know of coauthored by Richard Davidson, which does not show meditation to be superior.