Confronting Quasi-Separation in Logistic Mixed Effects for Linguistic Data: A Bayesian Approach

Kimball, Amelia E.; Shantz, Kailen; Eager, Christopher; Roy, Joseph

doi:10.1080/09296174.2018.1499457

Cited by 16 publications

(12 citation statements)

References 41 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This may actually be tremendously useful for people using hierarchical linear models. When using appropriate random effect structures (see Barr et al, 2013;Bates et al, 2015), these models are known to run into convergence issues (e.g., Kimball, Shantz, Eager, & Roy, 2018). To remedy such convergence issues, a common strategy is to drop complex random effect terms incrementally (Matuschek, Kliegl, Vasishth, Baayen, & Bates, 2017).…”

Section: Preregistrations and Registered Reportsmentioning

confidence: 99%

Researcher degrees of freedom in phonetic research

Roettger

2019

Laboratory Phonology: Journal of the Association for Laboratory Phonology

View full text Add to dashboard Cite

The results of published research critically depend on methodological decisions that have been made during data analysis. These so-called 'researcher degrees of freedom' (Simmons, Nelson, & Simonsohn, 2011) can affect the results and the conclusions researchers draw from it. It is argued that phonetic research faces a large number of researcher degrees of freedom due to its scientific object-speech-being inherently multidimensional and exhibiting complex interactions between multiple covariates. A Type-I error simulation is presented that demonstrates the severe inflation of false positives when exploring researcher degrees of freedom. It is argued that combined with common cognitive fallacies, exploitation of researcher degrees of freedom introduces strong bias and poses a serious challenge to quantitative phonetics as an empirical science. This paper discusses potential remedies for this problem including adjusting the threshold for significance; drawing a clear line between confirmatory and exploratory analyses via preregistration; open, honest, and transparent practices in communicating data analytical decisions; and direct replications.

show abstract

Section: Preregistrations and Registered Reportsmentioning

confidence: 99%

Researcher degrees of freedom in phonetic research

Roettger

2019

Laboratory Phonology: Journal of the Association for Laboratory Phonology

View full text Add to dashboard Cite

show abstract

“…The first analysis asked whether the three participant groups showed differential effects of Cognate Status, Noise Condition, or their interaction. Unfortunately, due to model convergence issues (perhaps the result of quasi-separation; Kimball et al, 2019 ), it was not possible to fit a sufficiently complex mixed effects logistic regression to the accuracy data. We therefore present descriptive statistics for both real words and nonwords in order to qualitatively evaluate response strategies across groups and conditions.…”

Section: Analysis and Resultsmentioning

confidence: 99%

Modulation of Cross-Language Activation During Bilingual Auditory Word Recognition: Effects of Language Experience but Not Competing Background Noise

Fricke

2022

Front. Psychol.

View full text Add to dashboard Cite

Previous research has shown that as the level of background noise increases, auditory word recognition performance drops off more rapidly for bilinguals than monolinguals. This disproportionate bilingual deficit has often been attributed to a presumed increase in cross-language activation in noise, although no studies have specifically tested for such an increase. We propose two distinct mechanisms by which background noise could cause an increase in cross-language activation: a phonetically based account and an executive function-based account. We explore the evidence for the phonetically based account by comparing cognate facilitation effects for three groups of native English listeners (monolinguals, late (L2) learners of Spanish, and heritage Spanish speakers) and four noise conditions (no noise, speech-shaped noise, English two-talker babble, and Spanish two-talker babble) during an auditory lexical decision task in English. By examining word recognition in the dominant language, the role of language control mechanisms is minimized, and by examining three different types of competing noise, the role of energetic vs. informational masking can be assessed. Contrary to predictions, we find no evidence that background noise modulates cross-language activation; cognate facilitation is constant across the four noise conditions. Instead, several indices of word recognition performance are found to correlate with aspects of linguistic experience: (1) The magnitude of the cognate facilitation effect is correlated with heritage listeners’ self-ratings of Spanish proficiency; (2) Overall noise deficits are marginally larger for heritage listeners with lower English vocabulary scores; (3) Heritage listeners’ Spanish self-ratings predict their magnitude of informational masking; (4) For all bilinguals, the degree of masking incurred in both English and Spanish two-talker babble is correlated with self-reported daily exposure to Spanish; and (5) The degree of masking incurred by Spanish babble is correlated with Spanish vocabulary knowledge. The results enrich our understanding of auditory word recognition in heritage speakers in particular and provide evidence that informational masking is most subject to modulation due to variation in linguistic experience. It remains to be seen whether cross-language activation is modulated by noise when the target language is the less dominant one.

show abstract

“…Another practical reason is that complex random effects structures that are often required for linguistic data analysis often do not, or not easily, converge with lme4. In contrast, complex random effects structures are more likely to converge when fitting Bayesian models (Eager & Roy, 2017; Kimball et al., 2019; Sorensen & Vasishth, 2015).…”

Section: A Hands‐on Poisson Regression Analysis Examplementioning

confidence: 99%

Poisson regression for linguists: A tutorial introduction to modelling count data with brms

Winter

Bürkner

2021

Language and Linguist. Compass

View full text Add to dashboard Cite

Count data is prevalent in many different areas of linguistics, such as when counting words, syntactic constructions, discourse particles, case markers, or speech errors. The Poisson distribution is the canonical distribution for characterising count data with no or unknown upper bound. Given the prevalence of count data in linguistics, Poisson regression has wide utility no matter what subfield of linguistics is considered. However, in contrast to logistic regression, Poisson regression is surprisingly little known. Here, we make a case for why linguists need to consider Poisson regression, and give recommendations for when Poisson regression is more appropriate compared to logistic regression. This tutorial introduces readers to foundational concepts needed to understand the basics of Poisson regression, followed by a hands‐on tutorial using the R package brms. We discuss a dataset where Catalan and Korean speakers change the frequency of their co‐speech gestures as a function of politeness contexts. This dataset also involves exposure variables (the incorporation of time to deal with unequal intervals) and overdispersion (excess variance). Altogether, we hope that more linguists will consider Poisson regression for the analysis of count data.

show abstract

Confronting Quasi-Separation in Logistic Mixed Effects for Linguistic Data: A Bayesian Approach

Cited by 16 publications

References 41 publications

Researcher degrees of freedom in phonetic research

Researcher degrees of freedom in phonetic research

Modulation of Cross-Language Activation During Bilingual Auditory Word Recognition: Effects of Language Experience but Not Competing Background Noise

Poisson regression for linguists: A tutorial introduction to modelling count data with brms

Contact Info

Product

Resources

About