Exploring Equivalence Testing with the Updated TOSTER R Package

Abstract

Equivalence testing is arguably under utilized by experimental researchers. Due to limited software support for such analyses, and little education on the topic in graduate programs, the utilization of equivalence testings still appares to be low. One option for equivalence testing is the use of two one-sided tests (TOST). The TOSTER R package and jamovi module, originally developed by Daniel Lakens in 2017, was created to make TOST more accessible to the average researcher. In the past two years, I have made significant changes to the TOSTER package in order to increase its accessibility and provide more robust analysis options for researchers. In this paper, I will detail the changes to the package and highlight new analysis options that will make TOST easier for the average quantitative researcher.

The reality of virtual reality

“…To provide statistical evidence for the equality of measures, we used the Wilcoxon two one-sided test (TOST) (1) as a robust equivalence test for all non-significant comparisons ( Caldwell, 2022 ). The TOST procedure allows us to provide support for the null hypothesis, i.e., the absence of effects large enough to be considered worthwhile.…”

Advancing the Shift-of-Strategy approach: Shifting suspects' strategies in extended interviews.

“…Table 4 displays comparisons of each interview condition using Welch’s t tests, and if these tests were nonsignificant, an equivalence test is reported, with bounds set at d = 0.20. Equivalence tests were performed using the TOSTER package (Caldwell, 2022; Lakens, 2017) for R. Participants in the Direct condition rated their performance as significantly better than those in the SoS-Standard condition and SoS-Reinforcement condition. There was no significant difference between participants’ self-assessments in the SoS-Standard condition and the SoS-Reinforcement condition nor was there evidence of equivalence between these conditions.…”

Validity of Apple Watch, Garmin Forerunner® 935 and GENEActiv for estimating energy expenditure during close quarter battle training in Special Forces soldiers

“…Bias corrected and accelerated 95% CIs for the least products regression estimates were calculated using bootstrap resampling (Canty et al., 2021). Equivalence tests were performed using the TOSTER package (version 0.8.0) (Caldwell, 2022; Lakens, 2017). Non‐parametric limits of agreement were calculated in a custom Excel spreadsheet (Microsoft Corp).…”