Subjects: Psychology >> Social Psychology submitted time 2023-03-28 Cooperative journals: 《心理科学进展》
Abstract： In the fields of psychology, education, and clinical science, researchers have devoted increasing attention to the intraindividual dynamics of behaviors, minds, and treatment effects over time, making personalized modeling a growing concern. Traditional cross-sectional and longitudinal studies only have a few measurement time points for each individual, which is not suitable for studying intraindividual dynamics. Intensive longitudinal design collects a set of measures from individuals at multiple time points with higher frequency over longer periods. With its strengths in more immediate, accurate, and authentic assessments, this design is more suitable to investigate the dynamics and mechanisms of intraindividual processes. With the development of mobile phones and other mobile devices, researchers can conveniently collect intensive longitudinal data for various aspects of psychology, including individual emotion, personality, cognition, and behavior patterns. The intensive longitudinal design has recently become one of the most prominent and promising approaches in psychological research, but most of these studies still relied on traditional analyzing methods. We first reviewed a conventional method of intensive longitudinal data analysis, the multilevel linear model (MLM), and discussed its limitations in analyzing intensive longitudinal data. We then introduced the principles, empirical applications, strengths, and weaknesses of two advanced modeling methods, dynamic structural equation model (DSEM) and group iterative multiple model estimation (GIMME). DSEM is a top-down approach of modeling intensive longitudinal data while GIMME is a bottom-up one, both being implemented in commonly used software. DSEM is one of the most promising methods for intensive longitudinal modeling and can be regarded as a multilevel extension of the dynamic factor model (DFM). It combines the strengths of various modeling approaches, including multilevel modeling, time-series modeling, structural equational model (SEM), and time-varying effects modeling (TVEM). GIMME is a dynamic network method initially proposed for functional magnetic resonance imaging (fMRI) data analysis and has recently been applied to intensive longitudinal data analysis. It combines individual- and group-level information to estimate network models at both levels, bridging nomothetic (population) and idiographic (individual) approaches to intensive longitudinal data analysis. By introducing these two advanced modeling methods, the current review can help deepen the understanding of the top-down approach and bottom-up approach and clarify their strengths and weaknesses in the intensive longitudinal data analysis. To help empirical researchers better understand the modeling of DSEM and GIMME and show the advantages of the two models compared with MLM, we provided a tutorial on how to analyze the intensive longitudinal data with the three models (i.e., MLM, DSEM, and GIMME), respectively. We presented the analyzing processes step by step and explained how to interpret the results of these models accordingly. By comparing the output results of the three models, the current review summarized the characteristics of each model. The corresponding Mplus and R codes were provided in the appendixes. Along with the three modeling methods mainly introduced in the current review, we also provided a general introduction of other common modeling methods in the intensive longitudinal data analysis. The current review summarized the popular models in the intensive longitudinal data analysis on their strengths and weaknesses and guided researchers to select suitable modeling methods in different situations. The current review contributes to the development and application of the advanced methods of intensive longitudinal data analysis and helps researchers better understand the dynamic process behind the intensive longitudinal data.
Subjects: Psychology >> Statistics in Psychology submitted time 2021-05-08
Abstract： In the fields of psychology, education, and clinical science, researchers have devoted increased attention to the dynamic changes and personalized modeling of individuals' behaviors, minds, and treatment effects over time. Intensive longitudinal data is a set of measures collected at multiple time points with higher frequency over shorter periods. Thus, it can be used in the analysis of the dynamics and mechanisms of within-person processes. In recent years, intensive longitudinal design has become one of the most prominent and promising approaches in psychological research. However, many of these researches still rely on traditional data analysis methods. Many models have been proposed to analyze intensive longitudinal data, including top-down approaches (e.g., dynamic structural equation model, DSEM) and bottom-up approaches (e.g., group iterative multiple model estimation, GIMME). Both of the methods can conveniently model autoregressive and cross-lagged effects in intensive longitudinal data.
Peer Review Status:
Awaiting Review
Subjects: Psychology >> Statistics in Psychology submitted time 2020-10-27
Abstract： In the research of psychology, education, and organizational behavior, researchers often encounter multilevel data with hierarchical structures (e.g., participants may cluster within communities, classes, or clinics). Ignoring the hierarchical structures of data may lead to a violation of the independence assumption of some models, resulting in biased parameter estimates. Therefore, researchers often need to conduct multilevel modeling to solve the statistical problems caused by non-independent observations. However, due to the limitation of objective conditions, in real studies, the sample sizes of level 1 and/or level 2 are often small in hierarchical data. Traditional frequentist-based maximum likelihood (ML) approach, which relies on large-sample theory, might lead to problems in parameter estimation and model convergence in multilevel modeling with small samples. In contrast, Bayesian approach is often more advantageous in small samples, but it is also more susceptible to the subjective specification of priors. To investigate the potentially detrimental effects of inaccurate prior information on Bayesian approaches and compare their performance to the traditional approaches, we conducted a series of simulations under the multilevel model framework with different dependent variable types (i.e., continuous normal, continuous non-normal, and binary dependent variables), sample sizes and intraclass correlation coefficients (ICCs). In sum, the results revealed the devastating impacts of inaccurate prior information on Bayesian estimation, especially in the cases of larger ICC, smaller level 2 sample size, and smaller prior variance. When the dependent variable was non-normal or binary, these negative effects were more obvious. The present study investigated the impacts of inaccurate prior information on Bayesian estimation and provided advice on the specification of priors. We hope that it could contribute to strengthening the theoretical and practical understanding of prior specifications.
Peer Review Status:Awaiting Review
Subjects: Psychology >> Statistics in Psychology submitted time 2020-05-14
Abstract： Psychological researches focus on describing, explaining and predicting behavior, and having a good understanding of the association between variables is an essential part of this process. Regression analysis, a method to evaluate the relationship between variables, is widely used in psychological studies. However, due to its highly focus on the interpretation of sample data, the traditional ordinary least squares regression has several drawbacks, such as over-fitting problem and limitation on dealing with multicollinearity, which may undermine the generalizability of the model. These drawbacks have an inevitable influence on the promotion and prediction of the model conclusion. With the rapid development of methodology, Least absolute shrinkage and selection operator (Lasso) regression has been emerged to better compensate for the limitations of traditional methods. By introducing a penalty term in the model and shrinking the regression coefficients to zero, Lasso regression can achieve a higher accuracy of model prediction and model generalizability with the cost of a certain estimation bias. Besides, Lasso regression can also effectively deal with the multicollinearity problem. Therefore, it has been widely used in medicine, economics, neuroscience and other fields. In psychology, due to the limitations of computer computing power, researchers used to mainly rely on hypothesis testing to understand the association among variables to verify theories. Now, with the rapid development of machine learning, a shift from focusing on interpretation of the regression coefficients to improving the prediction of the model has emerged and become more and more important. Therefore, based on fundamental theories and real data analysis, the aim of this paper is to introduce the principles, implementation steps and advantages of the Lasso regression. With the help of statistic science, it is promising that more and more applied researchers will be called upon to focus on the emerging statistical tools to promote the development of psychology.
Subjects: Psychology >> Statistics in Psychology submitted time 2019-03-01
Abstract： Measurement invariance (MI) is a pre-requisite for comparison between groups under the framework of latent variables. But in traditional multi-group factor analysis, scalar invariance is almost unachievable in practice. The Bayesian approximate MI proposed by Muthén and Asparouhov (2013) compensates for this limitation to some extent by providing a zero-mean, small-variance prior for the differences in measurement parameters. It allows for small differences between groups and avoids the problem caused by strict restrictions in classical method, such as poor model fitting, awkward model modifications and higher Type I error rate. These strengths make this new approach a more suitable method in the practical research. Throughout the paper we summarized the principles and advantages of the approximate MI approach, and a real data set was analyzed to demonstrate the validity and practical usefulness of this approach using Mplus. "
Peer Review Status:Awaiting Review
Subjects: Psychology >> Statistics in Psychology submitted time 2018-12-27
Abstract： Structural equation modeling (SEM) has been widely used in psychological researches to investigate the casual relationship among latent variables. Model estimation can be conducted under both the frequentist framework (e.g., maximum-likelihood approach) and the Bayesian framework. In recent years, with the prevalence of Bayesian statistics and its advantages in dealing with small samples, missing data and complex models in SEM, Bayesian structural equation modeling (BSEM) has developed rapidly. However, in China its application in the field of psychology is still insufficient. Therefore, this paper mainly focuses on presenting this new research method to applied researchers. We explain the theoretical and methodological basis of BSEM, as well as its advantages and disadvantages compared with the traditional frequentist approach. We also introduce several commonly used BSEM models and their applications. "
Hits
Hits
Downloads
Comment