Psychological methods最新文献

Intensive longitudinal data, increasingly common in social and behavioral sciences, often consist of multivariate time series from multiple individuals. Dynamic factor analysis, combining factor analysis and time series analysis, has been used to uncover individual-specific processes from single-individual time series. However, integrating these processes across individuals is challenging due to estimation errors in individual-specific parameter estimates. We propose a method that integrates individual-specific processes while accommodating the corresponding estimation error. This method is computationally efficient and robust against model specification errors and nonnormal data. We compare our method with a Naive approach that ignores estimation error using both empirical and simulated data. The two methods produced similar estimates for fixed effect parameters, but the proposed method produced more satisfactory estimates for random effects than the Naive method. The relative advantage of the proposed method was more substantial for short to moderately long time series (T = 56-200). (PsycInfo Database Record (c) 2025 APA, all rights reserved).

Causality is a fundamental part of the scientific endeavor to understand the world. Unfortunately, causality is still taboo in much of psychology and social science. Motivated by a growing number of recommendations for the importance of adopting causal approaches to research, we reformulate the typical approach to research in psychology to harmonize inevitably causal theories with the rest of the research pipeline. We present a new process which begins with the incorporation of techniques from the confluence of causal discovery and machine learning for the development, validation, and transparent formal specification of theories. We then present methods for reducing the complexity of the fully specified theoretical model into the fundamental submodel relevant to a given target hypothesis. From here, we establish whether or not the quantity of interest is estimable from the data, and if so, propose the use of semi-parametric machine learning methods for the estimation of causal effects. The overall goal is the presentation of a new research pipeline which can (a) facilitate scientific inquiry compatible with the desire to test causal theories (b) encourage transparent representation of our theories as unambiguous mathematical objects, (c) tie our statistical models to specific attributes of the theory, thus reducing under-specification problems frequently resulting from the theory-to-model gap, and (d) yield results and estimates which are causally meaningful and reproducible. The process is demonstrated through didactic examples with real-world data, and we conclude with a summary and discussion of limitations. (PsycInfo Database Record (c) 2025 APA, all rights reserved).

Intensive longitudinal data analysis, commonly used in psychological studies, often concerns outcomes that have strong floor effects, that is, a large percentage at its lowest value. Ignoring a strong floor effect, using regular analysis with modeling assumptions suitable for a continuous-normal outcome, is likely to give misleading results. This article suggests that two-part modeling may provide a solution. It can avoid potential biasing effects due to ignoring the floor effect. It can also provide a more detailed description of the relationships between the outcome and covariates allowing different covariate effects for being at the floor or not and the value above the floor. A smoking cessation example is analyzed to demonstrate available analysis techniques. (PsycInfo Database Record (c) 2025 APA, all rights reserved).

Experiments with fully crossed designs are often used in experimental psychology spanning several fields, from cognitive psychology to social cognition. These experiments consist in the presentation of stimuli representing super-ordinate categories, which have to be sorted into the correct category in two contrasting conditions. This tutorial presents a linear mixed-effects model approach for obtaining Rasch-like parameterizations of response times and accuracies of fully crossed design data. The modeling framework for the analysis of fully crossed design data is outlined along with a step-by-step guide of its application, which is further illustrated with two practical examples based on empirical data. The first example regards a cognitive psychology experiment and pertains to the evaluation of a spatial-numerical association of response codes effect. The second one is based on a social cognition experiment for the implicit evaluation of racial attitudes. A fully commented R script for reproducing the analyses illustrated in the examples is available in the online supplemental materials. (PsycInfo Database Record (c) 2024 APA, all rights reserved).

In psychology, we often want to know whether or not an effect exists. The traditional way of answering this question is to use frequentist statistics. However, a significance test against a null hypothesis of no effect cannot distinguish between two states of affairs: evidence of absence of an effect and the absence of evidence for or against an effect. Bayes factors can make this distinction; however, uptake of Bayes factors in psychology has so far been low for two reasons. First, they require researchers to specify the range of effect sizes their theory predicts. Researchers are often unsure about how to do this, leading to the use of inappropriate default values which may give misleading results. Second, many implementations of Bayes factors have a substantial technical learning curve. We present a case study and simulations demonstrating a simple method for generating a range of plausible effect sizes, that is, a model of Hypothesis 1, for treatment effects where there is a binary-dependent variable. We illustrate this using mainly the estimates from frequentist logistic mixed-effects models (because of their widespread adoption) but also using Bayesian model comparison with Bayesian hierarchical models (which have increased flexibility). Bayes factors calculated using these estimates provide intuitively reasonable results across a range of real effect sizes. (PsycInfo Database Record (c) 2024 APA, all rights reserved).

Many psychological dimensions seem bipolar (e.g., happy-sad, optimism-pessimism, and introversion-extraversion). However, seeming opposites frequently do not act the way researchers predict real opposites would: having correlations near -1, loading on the same factor, and having relations with external variables that are equal in magnitude and opposite in sign. We argue these predictions are often incorrect because the bipolar model has been misspecified or specified too narrowly. We therefore explicitly define a general bipolar model for ideal error-free data and then extend this model to empirical data influenced by random and systematic measurement error. Our model shows the predictions above are correct only under restrictive circumstances that are unlikely to apply in practice. Moreover, if a bipolar dimension is divided into two so that researchers can test bipolarity, our model shows that the correlation between the two can be far from -1; thus, strategies based upon Pearson product-moment correlations and their factor analyses do not test if variables are opposites. Moreover, the two parts need not be mutually exclusive; thus, measures of co-occurrence do not test if variables are opposites. We offer alternative strategies for testing if variables are opposites, strategies based upon censored data analysis. Our model and findings have implications not just for testing bipolarity, but also for associated theory and measurement, and they expose potential artifacts in correlational and dimensional analyses involving any type of negative relations. (PsycInfo Database Record (c) 2024 APA, all rights reserved).

Longitudinal randomized controlled trials (RCTs) have been commonly used in psychological studies to evaluate the effectiveness of treatment or intervention strategies. Outcomes in longitudinal RCTs may follow either straight-line or curvilinear change trajectories over time, and missing data are almost inevitable in such trials. The current study aims to investigate (a) whether the estimate of average treatment effect (ATE) would be biased if a straight-line growth (SLG) model is fit to longitudinal RCT data with quadratic growth and missing completely at random (MCAR) or missing at random (MAR) data, and (b) whether adding a quadratic term to an SLG model would improve the ATE estimation and inference. Four models were compared via a simulation study, including the SLG model, the quadratic growth model with arm-invariant and fixed quadratic effect (QG-AIF), the quadratic growth model with arm-specific and fixed quadratic effects (QG-ASF), and the quadratic growth model with arm-specific and random quadratic effects (QG-ASR). Results suggest that fitting an SLG model to quadratic growth data often yielded severe biases in ATE estimates, even if data were MCAR or MAR. Given four or more waves of longitudinal data, the QG-ASR model outperformed the other methods; for three-wave data, the QG-ASR model was not applicable and the QG-ASF model performed well. Applications of different models are also illustrated using an empirical data example. (PsycInfo Database Record (c) 2024 APA, all rights reserved).

It is unappreciated that there are four different approaches to power analysis for detecting misspecification by testing overall fit of structural equation models (SEMs) and, moreover, that common approaches can yield radically diverging results for SEMs with many items (high p). Here we newly relate these four approaches. Analytical power analysis methods using theoretical null and theoretical alternative distributions (Approach 1) have a long history, are widespread, and are often contrasted with "the" Monte Carlo method-which is an oversimplification. Actually, three Monte Carlo methods can be distinguished; all use an empirical alternative distribution but differ regarding whether the null distribution is theoretical (Approach 2), empirical (Approach 3), or-as we newly propose and demonstrate the need for-adjusted empirical (Approach 4). Because these four approaches can yield radically diverging power results under high p (as demonstrated here), researchers need to "match" their a priori SEM power analysis approach to their later SEM data analysis approach for testing overall fit, once data are collected. Disturbingly, the most common power analysis approach for a global test-of-fit is mismatched with the most common data analysis approach for a global test-of-fit in SEM. Because of this mismatch, researchers' anticipated versus actual/obtained power can differ substantially. We explain how/why to "match" across power-analysis and data-analysis phases of a study and provide software to facilitate doing so. As extensions, we explain how to relate and implement all four approaches to power analysis (a) for testing overall fit using χ² versus root-mean-square error of approximation and (b) for testing overall fit versus testing a target parameter/effect. (PsycInfo Database Record (c) 2024 APA, all rights reserved).

Moderated nonlinear factor analysis (MNLFA) has emerged as an important and flexible data analysis tool, particularly in integrative data analysis setting and psychometric studies of measurement invariance and differential item functioning. Substantive applications abound in the literature and span a broad range of disciplines. MNLFA unifies item response theory, multiple group, and multiple indicator multiple cause modeling traditions, and it extends these frameworks by conceptualizing latent variable heterogeneity as a source of differential item functioning. The purpose of this article was to illustrate a flexible Markov chain Monte Carlo-based approach to MNLFA that offers statistical and practical enhancements to likelihood-based estimation while remaining plug and play with established analytic practices. Among other things, these enhancements include (a) missing data handling functionality for incomplete moderators, (b) multiply imputed factor score estimates that integrate into existing multiple imputation inferential methods, (c) support for common data types, including normal/continuous, nonnormal/continuous, binary, ordinal, multicategorical nominal, count, and two-part constructions for floor and ceiling effects, (d) novel residual diagnostics for identifying potential sources of differential item function, (e) manifest-by-latent variable interaction effects that replace complex moderation function constraints, and (f) integration with familiar regression modeling strategies, including graphical diagnostics. A real data analysis example using the Blimp software application illustrates these features. (PsycInfo Database Record (c) 2024 APA, all rights reserved).