Optimal number of baseline sessions before changing phases within single-case experimental designs

• Published in: Behavioural processes Vol. 191; p. 104461
• Main Authors: Barnard-Brak, Lucy, Watkins, Laci, Richman, David
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.10.2021
• Subjects: Baseline; biomedical research; experimental design; phase transition; Sessions; Single case; Single subject; standard deviation; Single case; Baseline; Single subject; Sessions
• ISSN: 0376-6357; 1872-8308; 1872-8308
• DOI: 10.1016/j.beproc.2021.104461

•We examined the optimal number of baseline sessions that produced minimal bias.•As the number of baseline sessions increased, the bias in effect size estimates decreased.•Second, we examined what would be the minimum number of baseline sessions associated with varying levels of bias.•As the number of baseline sessions increases, the standard deviation for the phase decreased.•When considering five or ten percent bias, the optimal level of standard deviation was 0.59 or less. Recommendations vary considerably for the minimum or optimal number of baseline sessions to conduct within single-case experimental design clinical analyses or research studies. We examined the optimal number of baseline sessions that produced minimal bias. First, we examined the relation between the number of baseline sessions and the degree of bias in calculating estimates of treatment effect size. As the number of baseline sessions increased, the bias in effect size estimates decreased, r = -0.36, p < 0.001. s, we examined what would be the minimum number of baseline sessions associated with varying levels of bias. Bias of approximately ten percent was associated with four to five baseline sessions. Bias of about five percent was associated with six to seven baseline sessions. Third, we examined the relation between standard deviation and varying levels of bias. As the number of baseline sessions increases, the standard deviation for the phase decreased, r = -0.89, p < 0.001. Fourth, we examined what value of standard deviation in the baseline phase was associated with equal to or more than five versus ten percent bias. When considering five or ten percent bias, the optimal level of standard deviation was 0.59 or less.