Behavior Research Methods最新文献

Latent class analysis (LCA) can help identify unobserved classes of individuals in a population based on collected categorical data. It is commonly used in psychology to test hypotheses about sources of heterogeneity and class characteristics. However, careful decision-making is required in the modeling process. Its flexibility may explain why it is becoming more commonly used in psychology; however, it also highlights that there are many decision points in the modeling process, thus warranting a systematic literature review to document the use of LCA in psychology, mapping both the prevalence and quality of LCA studies. This systematic review followed the PRISMA guidelines and involved a comprehensive search across multiple databases, yielding 7,580 records related to latent class analysis. After removing duplicates and selecting a representative subsample, 377 documents were assessed for eligibility. Of these, 251 publications (comprising 313 LCAs) met the inclusion and exclusion criteria and were reviewed for this study. Each study was meticulously coded to map how the authors performed and reported each step of the LCA. Our analysis of these studies, in comparison with published guidelines, revealed notable discrepancies in how LCA is applied and reported. To support researchers in enhancing the quality of future LCA applications, we summarize key recommendations in a final section that outlines best practices for future LCA applications. The findings indicate a growing use of LCA in psychology but also highlight the need for greater methodological rigor and transparency in its implementation.

Vigilance decrement refers to the decline in sustained attention over time during prolonged tasks, which often leads to increased errors and accidents. However, to date, there are no experimental tasks that simultaneously measure changes in vigilance, cognitive control, and mind-wandering (MW) across time-on-task. We adapted the Attentional Network Test for Interactions and Vigilance-executive and arousal components (ANTI-Vea) task to integrate mind-wandering measures along with assessments of vigilance and cognitive control. By inserting thought probes (TPs) at different frequencies per block, we aimed to identify the optimal TP rate to capture mind-wandering changes without interfering with the measurement of vigilance, thereby providing an integrative assessment of changes in mind-wandering, cognitive control, and vigilance across time. We conducted two experiments: one in the laboratory with 90 students from the National University of Córdoba, Argentina, and another online, as a replication, with 180 students from the University of Granada, Spain. Participants were divided into three groups (4, 8, 12 TPs per block) and completed the ANTI-Vea-TP task. The results revealed that the inclusion of TPs was effective in detecting changes in mind-wandering over time-on-task. Moreover, TP frequency did not have a significant effect on mind-wandering reports, vigilance, or cognitive control over time-on-task. We discuss the potential suitability of this tool for investigating the interaction between vigilance, cognitive control, and mind-wandering, in both laboratory and online environments, which is essential for evaluating different theories of vigilance decrement.

This paper describes a dataset consisting of manually annotated nouns from a corpus of longitudinal day-long audio and hour-long video recordings collected monthly from 44 babies from age 6 months to age 17 months. This dataset was created as part of a larger project, called SEEDLingS, that examines the development of infants' language comprehension before and after their first birthday, from earliest comprehension to the early days of word production. This paper provides an overview of the corpus, describes how and why the nouns from the corpus were annotated, and discusses considerations for the reuse of this dataset for future work. The described annotations and relevant metadata are publicly available alongside this manuscript.

Research into dyadic social interactions has expanded significantly, enabling a deeper understanding of the dynamic processes underlying interpersonal communication. As the use of larger datasets becomes increasingly common in this field, the need for scalable and efficient analytical tools has grown. Automated coding methods, such as those provided by OpenPose, an open-source software for detecting and tracking human motion, offer significant advantages for analysing the movement of two individuals during an interaction. However, the processing and analysis of large quantities of JSON output files generated by OpenPose remain a considerable challenge. To address this, we introduce duet, an R package designed to streamline the processing and analysis of OpenPose output data, particularly in the context of dyadic interactions. The package provides a suite of functions for data cleaning, interpolation, kinematic analysis, and visualisation, offering researchers a comprehensive and user-friendly workflow. By simplifying the handling of OpenPose data, duet aims to facilitate large-scale, automated analysis of dyadic social interactions, with minimal coding experience, thereby advancing methodological capabilities in social and behavioural sciences.

Optional stopping refers to the practice of repeatedly performing a statistical analysis on a dataset as new data are collected until a pre-specified decision criterion is reached. This procedure is often adopted because of its effectiveness in optimizing data collection. Discussions of optional stopping to date have primarily centred around statistical issues, with relatively little consideration of any methodological implications of this procedure. Building on recent work drawing attention to methodological biases arising from the use of optional stopping, we highlight experimenter awareness of the current evidence state during data collection (experimenter evidence unmasking) as a salient methodological confound of optional stopping. We argue that experimenter evidence unmasking has the potential to influence an experimenter to implicitly or explicitly modify their behaviour in ways that can reduce the internal validity of an experiment. We conclude by offering recommendations for circumventing this confound and for the transparent reporting of experimenter evidence masking procedures.

Process tracing and process modeling are the two primary behavioral approaches for uncovering human decision-making processes. However, both approaches face significant limitations: process tracing offers a large and oftentimes confusing number of measures, while process modeling relies on a minimal number of comparable trials for reliable model fitting. In our study, we explore how we can combine mouse cursor tracking and the drift diffusion model (DDM) in order to both reduce the number of cursor measures and circumvent the minimal trial amount requirements of DDM fitting. One hundred three participants completed 90 trials in a random dot kinematogram (RDK). A total of 18 cursor measures were taken from the mouse cursor tracking literature and used to predict drift rate, threshold separation, and non-decision time of the DDM via partial least squares regression. Four cursor measures contributed significantly to the prediction of the DDM parameters. When reducing the available trials, these cursor measures, in combination with response time and accuracy, performed better and remained more stable in the prediction of DDM parameters than model fitting. Our results lower the barrier for applying mouse cursor tracking for novice researchers by highlighting important cursor measures and their mapping to psychological constructs of decision-making, while also offering an approach for behavioral scientists to investigate DDM components in experimental setups with a restricted number of trials.

Megastudies and crowdsourcing studies are a rich source of information for word recognition research because they provide processing times for thousands of words. However, the high cost makes it impossible to include all words of interest and all relevant participant groups. This study explores the potential of fine-tuned large language models (LLMs) to generate lexical decision times (RTs) similar to those of humans. Building on recent findings that LLMs can accurately estimate word features, we fine-tuned GPT-4o mini with 3000 words from a megastudy. We then gave the model the task of generating RT estimates for the remaining words in the dataset. Our findings showed a high correlation between AI-generated and observed RTs. We discuss three applications: (1) estimating missing RT data, where AI can fill in gaps for words missing in some megastudies, (2) verifying results of virtual experiments, where AI-generated data can provide an additional layer of validation for results of virtual experiments, and (3) optimizing human data collection, as researchers can run simulations before conducting studies with humans. While AI-generated RTs are not a replacement for human data, they have the potential to increase the flexibility and efficiency of megastudy research.

Face stimuli used in face perception research often focus on between-model variability, underrepresenting within-model variability across conditions and time. However, exposure to within-model variability is crucial for developing stable representations of faces. Here, we introduce the Chinese Face and Body Dataset (CFBD), a publicly accessible resource that captures within-model variability to represent a broad spectrum of appearance and image variations in laboratory and natural settings. The CFBD comprises 2,195 images from 117 models, including both laboratory photos taken by researchers and personal photos donated by models. Each model is depicted in 10 to 31 photos, coded for attributes such as the time photos were taken, facial expressions, viewing angles, and environmental contexts. Independent participants also rated these photos based on facial attractiveness, trustworthiness, and distinctiveness. The results revealed that the CFBD captures a wide range of variations across appearances and image attributes, and the within-model variances in trait ratings are comparable to, if not greater than, the between-model variances. Moreover, within-model variances in the trait ratings differ by image type, with personal photos being rated as more attractive, distinctive, and trustworthy than their laboratory counterparts. By capturing a diverse range of appearances and images of Chinese individuals, the CFBD provides valuable resources that expand face datasets, potentially advancing our understanding of robust face representation.

Researcher degrees of freedom in data analysis present significant challenges in social sciences, where different analytical decisions can lead to varying conclusions. In this work, we propose an example of an exploratory multiverse simulation to empirically compare various decision pathways to identify an effect's sensitivity to different analytical choices. The approach is demonstrated on the congruency sequence effect (CSE), a well-studied phenomenon in cognitive control research. We reviewed existing literature to identify common non-theory-specific analytical decisions, such as outlier exclusion criteria and hypothesis testing methods, and incorporated these into our simulation framework. Using 20,000 simulated datasets, we compared the true positive rates (TPR) and false positive rates (FPR) across 50 different decision pathways, resulting in a total of 1 million analyses. Our results indicate substantial differences in power and type I error rates across the analytical pathways, with some posing a significant risk of producing high false positives. The findings underscore the importance of running extensive simulations to investigate different data handling and hypothesis testing approaches in certain research fields. This case study serves as an example for conducting similar simulation procedures in research fields characterized by high variability in analytical decisions when investigating an otherwise widely accepted effect.