Computing Bayes factors for evidence-accumulation models using Warp-III bridge sampling

Gronau, Quentin Frederik; Heathcote, Andrew; Matzke, Dora

doi:10.3758/s13428-019-01290-6

Cited by 25 publications

(32 citation statements)

References 76 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…As practical advice, we recommend to keep the following four points in mind when using the bridgesampling package (see also Gronau et al 2020aGronau et al , 2019. First, one should always check the posterior samples carefully.…”

Section: Discussionmentioning

confidence: 99%

See 1 more Smart Citation

bridgesampling: An R Package for Estimating Normalizing Constants

Gronau¹,

Singmann²,

Wagenmakers³

2020

J. Stat. Soft.

Self Cite

161

112

View full text Add to dashboard Cite

Statistical procedures such as Bayes factor model selection and Bayesian model averaging require the computation of normalizing constants (e.g., marginal likelihoods). These normalizing constants are notoriously difficult to obtain, as they usually involve highdimensional integrals that cannot be solved analytically. Here we introduce an R package that uses bridge sampling (Meng and Wong 1996; Meng and Schilling 2002) to estimate normalizing constants in a generic and easy-to-use fashion. For models implemented in Stan, the estimation procedure is automatic. We illustrate the functionality of the package with three examples.

show abstract

Section: Discussionmentioning

confidence: 99%

“…The black solid line shows the standard normal proposal distribution and the gray histogram shows the posterior samples. Figure also available at https://tinyurl.com/y7owvsz3; see alsoGronau et al 2020aGronau et al , 2019.…”

mentioning

confidence: 99%

bridgesampling: An R Package for Estimating Normalizing Constants

Gronau¹,

Singmann²,

Wagenmakers³

2020

J. Stat. Soft.

Self Cite

161

112

View full text Add to dashboard Cite

show abstract

“…nitive modelsAnnis et al, 2019;Gronau, Heathcote, & Matzke, 2019). The choice of Bayesian model selection was based on their property of statistical consistency(Gronau & Wagenmakers, 2019), whereas other commonly used methods that assess out of sample predictive accuracy show biases towards detecting effects…”

mentioning

confidence: 99%

Think fast! The implications of emphasizing urgency in decision-making

Evans¹

2020

Preprint

View full text Add to dashboard Cite

Evidence accumulation models (EAMs) have become the dominant explanation of how the decision-making process operates, proposing that decisions are the result of a process of evidence accumulation. The primary use of EAMs has been as "measurement tools" of the underlying decision-making process, where researchers apply EAMs to empirical data to estimate participants' task ability (i.e., the "drift rate"), response caution (i.e., the "decision threshold"), and the time taken for other processes (i.e., the "non-decision time"), making EAMs a powerful tool for discriminating between competing psychological theories. Recent studies have brought into question the mapping between the latent parameters of EAMs and the theoretical constructs that they are thought to represent, showing that emphasizing urgent responding -- which intuitively should selectively influence decision threshold -- may also influence drift rate and/or non-decision time. However, these findings have been mixed, leading to differences in opinion between experts in the field. The current study aims to provide a more conclusive answer to the implications of emphasizing urgent responding, providing a re-analyse of 6 data sets from previous studies using two different EAMs -- the diffusion model and the linear ballistic accumulator (LBA) -- with state-of-the-art methods for model selection based inference. The findings display clear evidence for a difference in conclusions between the two models, with the diffusion model suggesting that decision threshold and non-decision time decrease when urgency is emphasized, and the LBA suggesting that decision threshold and drift rate decrease when urgency is emphasized. Furthermore, although these models disagree regarding whether non-decision time or drift rate decrease under urgency emphasis, both show clear evidence that emphasizing urgency does not selectively influence decision threshold. These findings suggest that researchers should revise their assumptions about certain experimental manipulations, the specification of certain EAMs, or perhaps both.

show abstract

“…Model 3 had condition effects on both starting point β and drift rate δ (see Supplementary Figure S3). We computed log marginal likelihoods for each model with warp-III bridge sampling (Gronau, Heathcote, et al, 2020;Meng & Schilling, 2002;Meng & Wong, 1996), using the bridgesampling R package (Gronau, Singmann, et al, 2020), with 6 repetitions and a maximum of 20,000 iterations. Based on the marginal likelihoods we then calculated Bayes factors to determine the most plausible model.…”

Section: Exploratory Analysesmentioning

confidence: 99%

Prior information use and response caution in perceptual decision-making: No evidence for a relationship with autistic-like traits

Retzler¹,

Boehm²,

Cai³

et al. 2020

Preprint

View full text Add to dashboard Cite

Interpreting the world around us requires integrating incoming sensory signals with prior information. Autistic individuals have been proposed to rely less on prior information and make more cautious responses than non-autistic individuals. Here we investigated whether these purported features of autistic perception vary as a function of autistic-like traits in the general population. We used a diffusion model framework, whereby decisions are modelled as noisy evidence accumulation processes towards one of two bounds. Within this framework, prior information can bias the starting point of the evidence accumulation process. Our pre-registered hypotheses were that higher autistic-like traits would relate to reduced starting point bias caused by prior information and increased response caution (wider boundary separation). 222 participants discriminated the direction of coherent motion stimuli as quickly and accurately as possible. Stimuli were preceded with a neutral cue (square) or a directional cue (arrow). 80% of the directional cues validly predicted the upcoming motion direction. We modelled accuracy and response time data using a hierarchical Bayesian model in which starting point varied with cue condition. We found no evidence for our hypotheses, with starting point bias and response caution seemingly unrelated to AQ scores. Alongside future research applying this paradigm to autistic individuals, our findings will help refine theories regarding the role of prior information and altered decision-making strategies in autistic perception. Our study also has implications for models of bias in perceptual decision-making, as the most plausible model was one that incorporated bias in both decision-making and sensory processing.

show abstract

Computing Bayes factors for evidence-accumulation models using Warp-III bridge sampling

Cited by 25 publications

References 76 publications

bridgesampling: An R Package for Estimating Normalizing Constants

bridgesampling: An R Package for Estimating Normalizing Constants

Think fast! The implications of emphasizing urgency in decision-making

Prior information use and response caution in perceptual decision-making: No evidence for a relationship with autistic-like traits

Contact Info

Product

Resources

About