Finding lambda for Box-Cox transformations for non-statisticians

I've learned that statisticians are not to be trifled with, because every time I read anything about selecting lambda with the slopes of log-likelihoods and whatnot, it feels like I'm trying to grasp oil. That's on fire.

That's why I fudge it and use the lambda where the resulting skew of the transformed set is the closest to zero (and less than twice the standard error of skew, I'm not a barbarian).

My question is this: does my shameless workaround of my stupidity only make the pure hearts of statisticians harumph, or is it actually methodologically dangerous somehow?
The Box-Cox transformation is solely to make the data more normal, as you mentioned. Sadly, there is no method that is guaranteed to find the “correct” or “best” value of lambda. What most people do is try several types of lambda and pick the one with the lowest standard deviation as that may be the most normal.

Edit: I should add that another option is, like you mention, to use the lambda which results in the lowest skew.


Ambassador to the humans
To be honest I'd be more interested in why you're doing a box cox transformation in the first place. For the most part I believe that the majority of times people use a box cox transform it's fairly unnecessary.
I don't think I ever checked back on this thread, thank you all.
I am doing a Box-Cox so I can view and rapidly compare skewed data (hospital costs and length of stay) using only averages, which is what SQL comes shipped with.