The effect of autocorrelation on the results of visually analyzing data from single-subject designs.

Abstract

Single-subject research designs are used to conduct clinical research and outcome evaluation in occupational therapy. Confusion exists regarding the best method to analyze and interpret single-subject data. One hundred graphs displaying the results of published single-subject research were examined to determine the influence of autocorrelation on the visual inferences made by the original investigators. The graphs were selected from 20 articles published over 10 years in seven rehabilitation journals. The data were extrapolated and lag 1 autocorrelation coefficients computed for both the baseline and treatment phases. Data analysis focused on two issues: (a) whether a relationship existed between the amount of autocorrelation present in a graph and the conclusion on the basis of visual analysis and (b) whether the amount of autocorrelation varied across different phases of the single-subject graphs. When a significant degree of autocorrelation was present, researchers using visual analysis were more likely to conclude that there was no clinically significant change in performance. Autocorrelation values were significantly higher in the treatment phases of the single-subject designs. Additional research is needed to establish a set of decision rules to assist clinicians in using visual analysis to evaluate the results of single-subject research.