How Much of the Placebo ‘effect’ Is Really Statistical Regression?

but in this case this formula cannot be used to calculate the values in Table 11. The second point concerns the authors’ observation that ‘by using the average of enough pretreatment measures one can reduce the size of regression to any predetermined level. ‘This is based upon an optimistic view of the correlation structure for repeated observations. It assumes that a sample of repeated measurements taken in a short period before treatment can be regarded as a simple random sample from a period covering the initial assessment and the post treatment assessment. A similar criticism is the substance of my third point. According to McDonald, Mazucca and McCabe, ‘the regression effect can be eliminated entirely by two pretreatment observations of the outcome variable. In this case, the first observation is used to select for treatment and the second to measure change for treatment.’ The model which justifies this observation by the authors is as follows. We have X,.as ‘selection’ measurement, X, as ‘baseline’ measurement and X3 as ‘evaluation’ measurement. Then,