TEfits: TEfits: Time-evolving model fits
In akcochrane/TEfits: Time-evolving model fits

TEfits

R Documentation

TEfits: Time-evolving model fits

Description

Data is described, interpreted, and tested using indices such as d prime, mean, or psychometric function threshold. This package serves to allow the same questions to be asked about time-evolving aspects of these indices, namely, the starting level, the amount of time that the index takes to change, and the asymptotic level of that index. Nonlinear regression applied to time-evolving functions is made as intuitive and painless as is feasible, with many extensions if desired.

Details

There are three broad approaches to parametric model fitting in TEfits (i.e., fitting a curve to change over time, such as a learning curve). The first, recommended, approach uses the nonlinear functionality of the brms package for Bayesian modeling with Stan. The second approach uses base R likelihood-optimization, and minimizes dependencies. The third approach uses a stepwise approach and the lme4 framework.

There is a last approach that, rather than fitting a parametric curve to change over time, uses basis functions over the dimension of time (e.g., trial number) to approximate arbitrary fluctuations. This can be a powerful way of controlling for unknown variations over time, as well as recovering those timecourses. The implementation is a straightforward augmentation of standard mixed-effects models; see time_basisFun_mem and vignette('mem_basis_vignette',package='TEfits').

Bayesian time-evolving fits

The TEbrm function is the recommended user-oriented function of the TEfits package. This function, and its associated helper functions, automates the parameterization of nonlinear time-evolving models. Apart from this, it is largely simply a wrapper for brm. Each nonlinear parameter of change is defined within its own [possibly generalized or mixed-effects] linear model. See TEbrm for examples.

Maximum likelihood time-evolving fits

The TEfit function is the associated user-oriented function. It allows for nonlinear fitting of time-related change in an outcome variable by estimating that outcome variable's value at each timepoint. See TEfit, TEfitAll for fitting a TEfit model to subsets of data (e.g., individual participants), and vignette('TEfits_tutorial') for an introduction to the framework.

Nonlinear regressors in [generalized] linear models

While TEfit and TEbrm are intended to interrogate time-evolving trends themselves within data, TEfits also includes several extensions to common regression functions that allow for seamless incorporation of a nonlinear [exponentially saturating] variable of time. These functions approach time-evolving dynamics from a common perspective in behavioral research: Behavior "of interest" occurs after transient initial bias in performance (e.g., initial task learning needs to occur before the target behavior can be effectively measured). The following functions use a stepwise method to first estimate the rate of change in the outcome variable, then use these rates to transform the variable of time into a saturating interpolation between 1 [starting offset] and 0 [asymptotic level], and finally include the time variable in the corresponding [g]lm[er] model: