ASA Connect

Hi All,

I have been working on a project that explores increasing the control group and decreasing the treatment group with relation to power in an experiment. Here the treatment group is hard to come by hence its small.

I have found that sometimes this method does give more power and then after a certain increase yields nothing more in terms of a power increase.

Has anyone got any views on this as I found it really counter intuitive that this would be the case. I am thinking that one would only spot large to medium effects due to a small treatment group but at this point I could be wrong:)

A rule of thumb that I've heard says that the gain from adding controls per case greatly diminishes after 3. You are seeing what one should expect. Here's an interesting approach to assessing the value of additional (or fewer) controls per case: Optimal case/control ratio in a case-control study

Marie,

If you calculate the power as though you were comparing the treatment to a parametric expectation this will give a limit on the power that one can get by increasing the size of the control group.  That is, an infinitely large control group is the same as a parametric expectation.  I looked at a special case of this in an old paper of mine: 

Ginevan, M.E. 1981. A Poisson trials approach to interpopulation comparisons of cause of death data. Environmental Research 25:147-159.

Best Wishes,

Michael