Our perception is subject to a process of adaptation that changes its operating properties (Clifford et al., 2007). This process manifests itself in a plethora of perceptual illusions and so-called aftereffects. It also plays an important role during the perception of multistable stimuli, such as a Necker cube (Fig. 1A). These are compatible with several, typically two, comparably likely perceptual interpretations. In the case of the Necker cube, one can perceive it as “upwards” or “downwards,” and during continuous viewing the perception switches between these alternatives (Fig. 1B).

Typically, time series for such multistable stimuli are fitted using the Gamma distribution (Fig. 1C). This assumes that individual dominance phase durations are exchangeable, i.e., the order in which they are drawn is unimportant. However, this ignores the effect of prior perception through adaptation (Ee, 2009) that makes individual dominance phases serially dependent on prior ones, violating assumptions about independent and identically distributed samples. In other words, each dominance phase is drawn from its specific distribution, whose parameters are also determined by the prior perceptual history. The bistablehistory package solves this problem by accounting for the slow accumulation process via a homogeneous first-order process (Pastukhov & Braun, 2011), providing tools for fitting time series using various distributions. It also allows for fitting while accounting for additional random or fixed factors. In addition, it provides a tool for extracting the estimated accumulated adaptation or computing it directly for further usage. The package aims to streamline the time series analysis for perceptual multistability and experiments on continuous perception in general. The package is built using the Stan probabilistic programming language (Carpenter et al., 2017). Thus, it provides posterior distributions, the ability to compare models via information criteria (Vehtari et al., 2017), etc. In addition, the package provides Stan code for performing the estimation and an example that explains how to implement a custom Stan model that relies on it.