Statistical approach

To evaluate the potential ways in which memory issues could influence proprioceptive accuracy, two sets of analyses were conducted using a Bayesian model selection approach (McElreath, 2016).

In the analysis, we considered the influence of amplitude, delay, and sensory and environmental conditions on the accuracy of self-turn performance. This allowed us to investigate the research goal: whether integration of vision and proprioception is crucial only for encoding the movement or also for the storage of the movement.

Starting from a full model that included all the interactions and variables of interest, predictors were removed until the most plausible model was obtained according to information criteria (Wagenmakers & Farrell, 2004; Yamashita, Yamashita, & Kamimura, 2007). The Watanabe-Akaike Information Criterion (WAIC, Gelman et al., 2014; Vehtari et al., 2017) was used as information criteria to evaluate the models. WAIC is the corresponding Bayesian version of the commonly used Akaike information criterion (AIC, Akaike 1973) and lower WAIC values indicate a better model. The selected models were interpreted by means of estimated parameters, graphical representations, and planned comparisons.

Bayesian generalized mixed-effects models were used as they allow for the complex structure of this data (Gelman et al., 2013). Specifically, data are characterized by: (1) a continuous non-normally distributed dependent variable (i.e., self-turn error); (2) within-subject factors (i.e., Perception and Environment conditions); (3) a quantitative independent variable (i.e. Amplitude of the passive rotation); (4) and a within-subject factor (i.e., Delay). Random intercepts were included to account for participants’ interpersonal variability, while the other variables were considered as fixed effects. Gamma distribution, with logarithmic link function, was specified as the family distribution of the generalized mixed-models to account for the distribution of the data.

Analyses were performed using the R software version 3.6.1 (Team & et al. 2018). Models were estimated using the R package brms (Bürkner & et al. 2017). All models used default prior specification of the R package brms. Detailed prior specifications are reported in the supplemental online material.