Psychological methods最新文献

Testing the presence of mediation effects is important in social science research. Recently, Bayesian hypothesis testing with Bayes factors (BFs) has become increasingly popular. However, the use of BFs for testing mediation effects is still under-studied, despite the growing literature on Bayesian mediation analysis. In this study, we systematically examine the performance of the BF for testing the presence versus absence of a mediation effect. Our results showed that the false and/or true positive rates of detecting mediation with the BF can be impacted by the prior specification, including the prior odds of the presence of each path (treatment-mediator path or mediator-outcome path) used in the design stage for data generation and in the analysis stage for calculating the BF of the mediation effect. Based on our examination, we developed an R function and a web application to determine sample sizes for testing mediation effects with the BF. Our study provides insights on the performance of the BF for testing mediation effects and adds to researchers' toolbox of sample size determination for mediation studies. (PsycInfo Database Record (c) 2024 APA, all rights reserved).

Latent change score (LCS) models within a continuous-time state-space modeling framework provide a convenient statistical approach for analyzing developmental data. In this study, we evaluate the robustness of such an approach in the context of accelerated longitudinal designs (ALDs). ALDs are especially interesting because they imply a very high rate of planned data missingness. Additionally, most longitudinal studies present unexpected participant attrition leading to unplanned missing data. Therefore, in ALDs, both sources of data missingness are combined. Previous research has shown that ALDs for developmental research allow recovering the population generating process. However, it is unknown how participant attrition impacts the model estimates. We have three goals: (a) to evaluate the robustness of the group-level parameter estimates in scenarios with empirically plausible unplanned data missingness; (b) to evaluate the performance of Kalman scores (KS) imputations for individual data points that were expected but unobserved; and (c) to evaluate the performance of KS imputations for individual data points that were outside the age ranged observed for each case (i.e., to estimate the individual trajectories for the complete age range under study). In general, results showed lack of bias in the simulated conditions. The variability of the estimates increased with lower sample sizes and higher missingness severity. Similarly, we found very accurate estimates of individual scores for both planned and unplanned missing data points. These results are very important for applied practitioners in terms of forecasting and making individual-level decisions. R code is provided to facilitate its implementation by applied researchers. (PsycInfo Database Record (c) 2024 APA, all rights reserved).

Bayesian model comparison (BMC) offers a principled approach to assessing the relative merits of competing computational models and propagating uncertainty into model selection decisions. However, BMC is often intractable for the popular class of hierarchical models due to their high-dimensional nested parameter structure. To address this intractability, we propose a deep learning method for performing BMC on any set of hierarchical models which can be instantiated as probabilistic programs. Since our method enables amortized inference, it allows efficient re-estimation of posterior model probabilities and fast performance validation prior to any real-data application. In a series of extensive validation studies, we benchmark the performance of our method against the state-of-the-art bridge sampling method and demonstrate excellent amortized inference across all BMC settings. We then showcase our method by comparing four hierarchical evidence accumulation models that have previously been deemed intractable for BMC due to partly implicit likelihoods. Additionally, we demonstrate how transfer learning can be leveraged to enhance training efficiency. We provide reproducible code for all analyses and an open-source implementation of our method. (PsycInfo Database Record (c) 2024 APA, all rights reserved).

There has been increased interest in practical methods for integrative analysis of data from multiple studies or samples, and using factor scores to represent constructs has become a popular and practical alternative to latent variable models with all individual items. Although researchers are aware that scores representing the same construct should be on a similar metric across samples-namely they should be measurement invariant-for integrative data analysis, the methodological literature is unclear whether factor scores would satisfy such a requirement. In this note, we show that even when researchers successfully calibrate the latent factors to the same metric across samples, factor scores-which are estimates of the latent factors but not the factors themselves-may not be measurement invariant. Specifically, we prove that factor scores computed based on the popular regression method are generally not measurement invariant. Surprisingly, such scores can be noninvariant even when the items are invariant. We also demonstrate that our conclusions generalize to similar shrinkage scores in item response models for discrete items, namely the expected a posteriori scores and the maximum a posteriori scores. Researchers should be cautious in directly using factor scores for cross-sample analyses, even when such scores are obtained from measurement models that account for noninvariance. (PsycInfo Database Record (c) 2024 APA, all rights reserved).

Careful consideration of the tradeoff between Type I and Type II error rates when designing experiments is critical for maximizing statistical decision accuracy. Typically, Type I error rates (e.g., .05) are significantly lower than Type II error rates (e.g., .20 for .80 power) in psychological science. Further, positive findings (true effects and Type I errors) are more likely to be the focus of replication. This conventional approach leads to very high rates of Type II error. Analyses show that increasing the Type I error rate to .10, thereby increasing power and decreasing the Type II error rate for each test, leads to higher overall rates of correct statistical decisions. This increase of Type I error rate is consistent with, and most beneficial in the context of, the replication and "New Statistics" movements in psychology. (PsycInfo Database Record (c) 2024 APA, all rights reserved).

Bayesian hypothesis testing procedures have gained increased acceptance in recent years. A key advantage that Bayesian tests have over classical testing procedures is their potential to quantify information in support of true null hypotheses. Ironically, default implementations of Bayesian tests prevent the accumulation of strong evidence in favor of true null hypotheses because associated default alternative hypotheses assign a high probability to data that are most consistent with a null effect. We propose the use of "nonlocal" alternative hypotheses to resolve this paradox. The resulting class of Bayesian hypothesis tests permits more rapid accumulation of evidence in favor of both true null hypotheses and alternative hypotheses that are compatible with standardized effect sizes of most interest in psychology. (PsycInfo Database Record (c) 2024 APA, all rights reserved).

Assessing measurement invariance is an important step in establishing a meaningful comparison of measurements of a latent construct across individuals or groups. Most recently, moderated nonlinear factor analysis (MNLFA) has been proposed as a method to assess measurement invariance. In MNLFA models, measurement invariance is examined in a single-group confirmatory factor analysis model by means of parameter moderation. The advantages of MNLFA over other methods is that it (a) accommodates the assessment of measurement invariance across multiple continuous and categorical background variables and (b) accounts for heteroskedasticity by allowing the factor and residual variances to differ as a function of the background variables. In this article, we aim to make MNLFA more accessible to researchers without access to commercial structural equation modeling software by demonstrating how this method can be applied with the open-source R package OpenMx. (PsycInfo Database Record (c) 2024 APA, all rights reserved).

A scale to measure a psychological construct is subject to measurement error. When meta-analyzing correlations obtained from scale scores, many researchers recommend correcting for measurement error. I considered three caveats when correcting for measurement error in meta-analysis of correlations: (a) the distribution of true scores can be non-normal, resulting in violation of the normality assumption for raw correlations and Fisher's z transformed correlations; (b) coefficient alpha is often used as the reliability, but correlations corrected for measurement error using alpha can be inaccurate when some assumptions of alpha (e.g., tau-equivalence) are violated; and (c) item scores are often ordinal, making the disattenuation formula potentially problematic. Via three simulation studies, I examined the performance of two meta-analysis approaches-with raw correlations and z scores. In terms of estimation accuracy and coverage probability of the mean correlation, results showed that (a) considering the true-score distribution alone, estimation of the mean correlation was slightly worse when true scores of the constructs were skewed rather than normal; (b) when the tau-equivalence assumption was violated and coefficient alpha was used for correcting measurement error, the mean correlation estimates can be biased and coverage probabilities can be low; and (c) discretization of continuous items can result in biased estimates and undercoverage of the mean correlations even when tau-equivalence was satisfied. With more categories and/or items on a scale, results can improve whether tau-equivalence was met or not. Based on these findings, I gave recommendations for conducting meta-analyses of correlations. (PsycInfo Database Record (c) 2024 APA, all rights reserved).

We introduce a new meta-analysis estimator, the weighted and iterated least squares (WILS), that greatly reduces publication selection bias (PSB) when selective reporting for statistical significance (SSS) is present. WILS is the simple weighted average that has smaller bias and rates of false positives than conventional meta-analysis estimators, the unrestricted weighted least squares (UWLS), and the weighted average of the adequately powered (WAAP) when there is SSS. As a simple weighted average, it is not vulnerable to violations in publication bias corrections models' assumptions too often seen in application. WILS is based on the novel idea of allowing excess statistical significance (ESS), which is a necessary condition of SSS, to identify when and how to reduce PSB. We show in comparisons with large-scale preregistered replications and in evidence-based simulations that the remaining bias is small. The routine application of WILS in the place of random effects would do much to reduce conventional meta-analysis's notable biases and high rates of false positives. (PsycInfo Database Record (c) 2024 APA, all rights reserved).

Individual differences are studied with a multitude of test instruments. Meta-analysis of tests is useful to understand whether individual differences in certain populations can be detected with the help of a class of tests. A method for the quantitative meta-analytical evaluation of test instruments with dichotomous items is introduced. The method assumes beta-binomially distributed test scores, an assumption that has been demonstrated to be plausible in many settings. With this assumption, the method only requires sample means and standard deviations of sum scores (or equivalently means and standard deviations of percent-correct scores), in contrast to methods that use estimates of reliability for a similar purpose. Two parameters are estimated for each sample: mean difficulty and an overdispersion parameter which can be interpreted as the test's ability to detect individual differences. The proposed bivariate meta-analytical approach (random or fixed effects) pools the two parameters simultaneously and allows to perform meta-regression. The bivariate pooling yields a between-sample correlation of mean difficulty parameters and overdispersion parameters. As a side product, reliability estimates are obtained which can be employed to disattenuate correlation coefficients for insufficient reliability when no other estimates are available. A worked example illustrates the method and R code is provided. (PsycInfo Database Record (c) 2024 APA, all rights reserved).