I measured cell size of human muscle and I wanted to examine the effect of age on the parameter. I graphed my data as a scatter plot and calculated Pearson's correlation coefficient but then also split the sample into 2 groups; one above and one below the mean age of the group and compared those with a t-test.
Is there any literature that would validate this method of splitting up my groups, i.e., using the mean as a cut off?
You will almost always lose useful information if you take a continuous variable and split it into groups. Your scatter plot directly displays the information you have collected, and is the best way to show others what you found.
My sense from your question is that there may be a non-linear relation between your parameter and age, so that the old/young t-test seems more significant than the linear correlation. The extensive discussion of binning continuous variables goes into useful detail on the dangers of binning and better ways to proceed with non-linear relations among variables.
