Stratified distributional analysis-a novel perspective on RT distributions.

Area of Science:

• Cognitive Psychology
• Computational Linguistics
• Psycholinguistics

Background:

• Response times (RTs) and their distributions offer insights into cognitive processes.
• Understanding how word properties like frequency and length affect RT distributions is crucial for cognitive theories.
• Existing methods may not fully capture the complex relationship between word characteristics and RT variability.

Purpose of the Study:

• To introduce a novel statistical methodology, stratified distributional analysis (SDA), for quantitatively assessing determinants of response time distributions.
• To investigate the influence of word frequency and length on response time distributions using large-scale lexical decision data.
• To determine the best probability distribution and word occurrence measure for modeling response time distributions.

Main Methods:

• Applied stratified distributional analysis (SDA) to millions of lexical decision response times from the English and British Lexicon Projects.
• Analyzed response time distributions as a function of word frequency and word length.
• Compared lognormal, Wald, and gamma distributions, and evaluated word form frequencies, discourse contextual diversity, and user contextual diversity.

Main Results:

• Response time distributions were best described by a lognormal distribution when word occurrence was quantified using contextual diversity measures.
• SDA, utilizing a hierarchical Bayesian framework, provided participant-level posterior distributions for distributional parameters.
• The methodology enabled probabilistic predictions of response times based on word frequency and length.

Conclusions:

• Stratified distributional analysis (SDA) effectively elucidates the mechanisms governing response time generation by fitting probability distributions.
• The findings highlight the utility of contextual diversity measures in modeling word processing.
• SDA is a versatile tool applicable to various cognitive tasks beyond lexical decision, including word-naming and eye-tracking.