Paired Continuous Data, Analysis with Help of Correlation Coefficients

Abstract

The t-value obtained from an unpaired analysis of paired data produces biased results. This is, because the level of correlation between unpaired data is assumed to be zero, and this may not be true for paired observations. Particularly, repeated measurements in one subject produces usually results more similar than those from single measurements in separate subjects. Repeated measurements, thus, tends to produce a positive correlation. However, this is not always true. Negative correlations will be observed, if completely different treatment effects are examined in one subject. This is, because the responders to one treatment are more at risk of being non-responder to the other treatment and vice versa. Indeed, correlations is a very basic phenomenon in statistical analyses, and it almost entirely determines the results of regression analyses.