Randomization tests for multiple-baseline designs: An extension of the SCRT-R package

Abstract

Multiple-baseline designs are variants of single-case designs well suited to behavioral research. In this article, we want to bring these designs to the attention of experimental psychologists and social and behavioral researchers in general, discuss such designs' advantages and limitations for valid inference in behavioral research, and suggest a statistical data-analytic technique to complement visual inspection, together with software to conduct those analyses.A multiple-baseline design consists of a series of replicated single-case designs, in which the replications are carried out at the same time. They extend the basic singlecase AB phase design by implementing several of those AB designs simultaneously to different persons, behaviors, or settings (Ferron & Scott, 2005;Onghena & Edgington, 2005). For convenience, these separate persons, behaviors, or settings will henceforth be called units.

Making treatment effect inferences from multiple-baseline data: The utility of multilevel modeling approaches

“…The degrees of freedom for an inference about the average Although the utility of multiple-baseline designs is well established, there is no consensus on how to analyze and present the resulting data. Among the methods traditionally employed are visual analyses (Parsonson & Baer, 1992), randomization tests (Bulté & Onghena, 2009;Koehler & Levin, 1998;, ordinary least squares regression (Huitema & McKean, 1998), first-order autoregressive models (McKnight, McKean, & Huitema, 2000), and more general time-series models (Velicer & Fava, 2003). In recent years, multilevel models (also called hierarchical linear models, or mixed linear models) have also been suggested as a method for combining single-case data within and across studies (Nugent, 1996;Shadish & Rindskopf, 2007;Van den Noortgate & Onghena, 2003a, 2003b.…”

“…A multiple-baseline First, the multilevel model assumes that the time series are independent of each other. In some multiple-baseline applications, however, one would expect the units to be interdependent (Bulté & Onghena, 2009;. A multiple-baseline design across behaviors will often produce series that are interdependent, because the treatment of one behavior may affect other behaviors in the individual and because an unmeasured variable that contributes to the error may have an impact on multiple behaviors.…”

A comparison of mean phase difference and generalized least squares for analyzing single-case data

“…Randomization tests have been suggested for a variety of single-case designs including ABAB (Onghena, 1992) and alternating treatments designs (Onghena & Edgington, 1994), whereas recent adaptations have focused on more complex situations (Lall & Levin, 2004;Levin, Lall, & Kratochwill, 2011). Open-source software has been made available for some of these tests (Bulté & Onghena, 2009) enhancing their applicability. The requirement for introducing randomization in the design can be seen as a gain in terms of scientific credibility , but it also makes the design less practical.…”

Probability and Visual Aids for Assessing Intervention Effectiveness in Single-Case Designs

“…After the two-phase comparison results are obtained, the researcher still has to (decide how to) combine the information in order to have an estimate for the whole (e.g., multiplebaseline,ABAB) design. This decision is inherent to SCED data analysis, although procedures like randomization tests (Heyvaert & Onghena, 2014; using the software described in Bulté & Onghena, 2008;2009), the d-statistics (Hedges et al, 2012 or multilevel models (Moeyaert, Ugille, Ferron, Beretvas, & Van den Noortgate, 2014), are more directly applicable to designs involving within-or across-subjects replication. …”