The distribution of a random process will usually be symmetrically distributed about the mean. Sometimes early results may be skewed. In theory results should approach the expected distribution over time as a more complete sampling is taken. This underlies the need to have an adequate sample size, to have appropriate controls and not to draw conclusion based on early results.
Significance: An intervention aimed at a group or characteristic that is very different from the average may initially appear to be more or less successful than it really is (Morton and Torgerson).
Parameters:
(1) number of results available
(2) randomness and bias of the study
(3) stability in the underlying system
(4) within-subject variability or measurement error
You would expect a regression to the mean if:
(1) The initial results involve a small sample.
(2) Subjects and results are random and unbiased.
(3) The system is the same as when the distribution was determined.
(4) There is large within-subject variability or measurement error (Linden et al).
You would expect results to continue to be skewed when the study is complete if:
(1) Almost all of the results have been collected.
(2) Data is biased or nonrandom.
(3) Something fundamental has changed.
(4) There is a small within-subject variability or measurement error.
The magnitude of the regression to the mean effect can be calculated using statistical formulae.