Statistics for everyone: ch.32 Calculation of power

easier R than SPSS with Rcmdr : Contents

ch.32 Calculation of power

So let’s calculate the power when comparing the mean of two groups.

If the expected mean difference and standard deviation between the two populations are 8 and 15, respectively, as before, let’s say that recent trends suggest that we can recruit about 100 patients a year. In that case, I wonder if calculating power with an estimate of a period of 1 year or so would yield enough power?

The expected power is 85%, so if you are satisfied at this level, you can estimate the duration of the study by taking into account about 1 year for the patient’s recruitment and other preparations and follow-up.

After all, when the 2 pieces of information, Difference in Means and Standard deviation, are determined, the graph will look the same. It just depends on whether you look at Numbers or the Power. In other words, in the same graph, it is a matter of ‘give a value of x and get a value of y, or a value of y and a value of x’, so in the end, ‘calculate the sample size’ and ‘calculate the power’ are the same.

The 5 menus that calculate the power in this way are essentially the same as calculating the sample size each, so I will omit them.

On the other hand, after completing the research, sometimes reviewers asks writers to calculate the power. This is called ‘post hoc power calculation’, which is actually of no use. This is not to calculate whether the test strength is high or low after the study is over. It makes no sense to calculate ‘power’ because it is itself a calculation of the probability of what will happen in the future.

This is only meaningful if you use the data obtained from this study to estimate how many samples you will need for your next study.

easier R than SPSS with Rcmdr : Contents

=================================================