Structural Equation Modeling is a powerful tool for analyzing causal relationships among multiple variables and has huge application potential in many disciplines. Our previously launched "Structural Equation Modeling Based on R Language" covers a series of topics such as introduction to the principles of structural equations, global and local estimation of structural equations, model construction and adjustment, latent variable analysis, composite variable analysis, and implementation of Bayesian methods for structural equations. With an introduction and a large number of case explanations, it systematically introduces the entire process of establishing, fitting, evaluating, screening and result display of structural equation models from the shallower to the deeper, which is widely recognized by students. After further communication and feedback after class, we found that the modeling process using structural equation models often encounters many 'special' situations: 1) there is a non-linear relationship between variables; 2) there is interaction between variables; 3) the data does not satisfy the normal distribution ;3) Variables are non-normal numerical variables, such as 0,1 data (in line with binomial distribution) and count data (in line with Poisson distribution), etc.; 4) Exogenous or endogenous variables are categorical variables, such as male and female , high, medium and low, different land types or forest types, etc. In "Structural Equation Modeling Based on R Language", we made a preliminary introduction to non-linear relationships between variables, non-normal variables and data analysis, but everyone still has great confusion when encountering these situations. These situations often require special handling. This time, we will provide a more in-depth explanation of the above issues so that everyone can face them calmly when encountering the above situations when using structural equation model modeling.
feature:
1. Explain the principles in a simple and easy-to-understand manner, emphasizing the importance of the principles;
2. Explain techniques and methods, provide complete supporting textbook data and provide long-term playback;
3. Combine with project cases to better connect with actual work applications;
4. Learn computer operations, independently complete case operation exercises, and track and analyze problems throughout the process;
5. The exclusive student assistance group assists in consolidating learning and communicating on practical work applications, and holds online Q&A sessions from time to time.
expert:
Teacher Zhang, from the Chinese Academy of Sciences, has long been engaged in research and teaching in structural equation modeling, community ecology, conservation biology, landscape ecology and ecological models. He has published many papers and has rich scientific research and practical experience.
Topic 1. Non-linear, non-normal, interaction and categorical variable analysis
Topic 2: Nested hierarchical data and data grouping analysis
Topic 3. Spatial autocorrelation data analysis technology
Topic 4. Practical Technology of Non-recursive Structural Equation Modeling
Topic 5. Technical practice of incorporating phylogenetic data into structural equation models
Topic 6. Time repeated measurement data analysis
Topic 7. Structural Equation Model Forecasting Issues - Direct Forecasting Implementation Approach
Topic 8: Paper writing, precautions and analysis of common problems and examples
Note: Each topic can be selected individually
Topic 1: Advanced Application of Structural Equation Model (SEM) and Analysis of Nonlinear, Non-Normal, Interaction and Categorical Variables
Modeling using structural equation models often encounters many 'special' situations: 1) there is a non-linear relationship between variables; 2) there is interaction between variables; 3) the data does not satisfy the normal distribution; 4) the variables are of non-normal type Numerical variables, such as 0,1 data (in line with binomial distribution) and count data (in line with Poisson distribution), etc.; 5) Exogenous or endogenous variables are categorical variables, such as men and women, high, middle and low, different land types or forest types wait. In the course "Structural Equation Modeling Based on R Language", we introduced the non-linear relationship between variables, non-normal variables and data analysis, but everyone still has great confusion when encountering these situations. These situations often require special handling. This time, we will provide a more in-depth explanation of the above issues so that everyone can face them calmly when encountering the above situations when using structural equation model modeling.
1: Nonlinear relationship and interaction analysis
1. Processing of nonlinear relationships among exogenous variables
2. Processing of nonlinear relationships among endogenous variables
3. Analysis of interaction relationships between variables
2: Non-normal data/variable analysis
1. Problems with non-normal data/variables
2. Non-normal data analysis
3. Variable analysis of non-normal variables
Three: Categorical variable analysis
1. Introduction to categorical variables
2. Analysis of exogenous variables as categorical variables
3. Analysis of endogenous variables as categorical variables
Topic 2: Advanced Application of Structural Equation Model (SEM) and Topic on Nested Hierarchical Data and Data Grouping Analysis
The data obtained in scientific research work often has the characteristics of nested/hierarchical/multi-level structure. This kind of data structure violates the assumption of data independence. The results obtained when directly using general regression (or generalized regression) and structural equation model analysis are inconsistent. Reliable, needs correction. In regression analysis, mixed effects models (nested models or multi-level models) need to be used for analysis to correct the impact of non-independence of data on the results. This article will first explore in detail the use of structural equation modeling to analyze nested/multi-level/stratified data. In addition, using structural equation modeling to group analysis of data is also an effective method for processing stratified data. The advantage of group analysis is that it can group data under a unified model framework, which is particularly effective for studies with small sample sizes. It also The significance of the differences between different grouping parameters can be tested for comparative analysis. Therefore, structural equation model data grouping analysis is included. We provide an in-depth introduction to data grouping analysis through several examples, giving everyone one more choice when encountering nested/hierarchical/multi-level data structures.
1: Basic principles of nested/hierarchical/multi-level data regression analysis
1. Overview of nested/multi-level/hierarchical data
2. Basic principles of mixed effects model analysis of nested/multi-level/stratified data
3. Basic principles of Bayesian method for analyzing nested/multi-level/stratified data
2: Structural equation model nested/hierarchical/multi-level data analysis
1. Implementation methods of nested/multi-level/hierarchical data structure equation model
2. Examples of balanced and unbalanced nested/multi-level/hierarchical data nested data knot equation models
3. Latent variable model nesting/multi-level/hierarchical data analysis
Three: Structural equation model data group analysis
1. The difference and connection between data grouping and nesting/stratification/multi-level and categorical variables
2. Structural equation model data group analysis
3. Latent variable model data group analysis
Topic 3: Advanced Application of Structural Equation Model (SEM) and Spatial Autocorrelation Data Analysis Technology
Sampling sample points usually contain spatial information. The similarity between two sample points that are close to each other is higher than the similarity between sample points that are far away. This is called spatial autocorrelation. Spatial autocorrelation causes samples to violate the independence assumption. Therefore, it is necessary to consider the impact of spatial autocorrelation on the results during the modeling process to eliminate the biased results caused by spatial autocorrelation on the model. Focusing on the spatial autocorrelation characteristics displayed by spatial data, we will discuss in detail the processing methods and processes of spatial autocorrelation data using the global estimation method, local estimation method and Bayesian method of structural equation model.
1: Spatial autocorrelation data regression model analysis
1. Overview of data space autocorrelation
2. Regression (mixed effects) model processes spatial autocorrelation data
3. Bayesian method for processing spatial data
2: Spatial autocorrelation data structure equation: local estimation method
1. The local estimation method incorporates the basic principle of spatial autocorrelation
2. Analysis of spatial autocorrelation data using local estimation methods (piecewiseSEM and brms)
Three: Under the spatial autocorrelation data structure equation: global estimation method
1. Basic principles of global estimation method (lavaan) for spatial autocorrelation data analysis
2. Example explanation of spatial autocorrelation analysis using global estimation method
Topic 4: Non-recursive Structural Equation Model Practical Technology
Structural Equation Modeling is a powerful tool for analyzing causal relationships among multiple variables and has huge application potential in many disciplines. The contents covered in the "Structural Equation Modeling Based on R Language" course we launched earlier are all recursive models. In fact, in the process of building a structural equation element model, through literature research, it will be found that there is an interaction between two variables (Reciprocal Interaction). For example, if A affects B, B in turn affects A; there will also be a cyclic interaction between three variables. Interaction Loop, for example, A affects B, B affects C, and C in turn affects A. These two situations are called non-recursive models in structural equation models. We will further explain the non-recursive structural equation type, and we will explain the modeling process of the non-recursive structural equation model in detail through several classic cases.
1. The difference between recursive models and non-recursive models 2. Precautions and implementation approaches for non-recursive model analysis
3. Explanation of classic cases of non-recursive models 4. Bayesian method (brms) to implement non-recursive models
Topic 5: Technical practice of incorporating phylogenetic data into structural equation modeling
For data containing species information, the distance of the species' kinship will affect the expression of the species' attributes. The attributes of species with close kinship will be more similar, and vice versa. Phylogenetic trees can quantify the distance of species relationships. This type of data also has the problem of non-independence, so it needs to be considered in structural equation modeling. We will share with you how to incorporate phylogenetic information into structural equation models to correct the biased results of the model.
1. Introduction to issues related to phylogeny
2. Ways to incorporate phylogeny-related data into structural equation models
3. The local estimation method (piecewiseSEM) realizes the integration of phylogeny-related data into structural equations
4. Bayesian method (brms) realizes the integration of phylogeny-related data into structural equations
Topic 6: Structural Equation Model (SEM) Time/Repeated Measures Data Analysis
Many studies require continuous observations at multiple time points (such as days/months/years), that is, repeated observation data or time data. When performing this type of data analysis, there are autocorrelation problems between samples at adjacent observation times that need to be corrected. In addition, the research purpose itself may be to explore the change of a certain observed variable of the research object over the observation time, that is, the growth curve model; it may also be to study the cross-interaction between two variables in the system, such as A and B. Two variables, the influence of A on B at time T1 is expressed at time T2, and the influence of B on A at time T1 is also expressed at time T2. This type of model is called a cross-lagged model (Autoregressive Cross-Lagged Model). Both the growth curve model and the cross-lagged model will have the problem of time autocorrelation. This course will focus on the above aspects and discuss in detail the global estimation method, local estimation method and Bayesian method of structural equation model analysis of time/repeated observation data.
1: Time/repeated measurement data regression model analysis
1. Introduction to the characteristics of time repeated measurement data
2. Regression model processing time/repeated measurement autocorrelation data
3. Bayesian method for time/repeated measurement data analysis
2: Time/Repeated Measurement Data Structural Equation Correction
1. Basic principles of local estimation method for processing time/repeated measurement data
2. Analysis of time autocorrelation data using local estimation methods (piecewiseSEM and brms)
Three: Cross-lagged model and growth curve model of time/repeated measurement data
1. Cross-lagged model of time/repeated measurement data (Autoregressive Cross-Lagged Model)
2. Growth Curve Model of time/repeated measurement data
Topic 7: Structural Equation Model (SEM) Prediction Issues
There are few literatures that describe in detail how to predict after the structural equation is established. Therefore, when displaying results, most literatures only show the regression relationship between variables in the model, that is, bivariate regression . The coefficient of this regression relationship is not consistent with the path coefficient given in the structural equation. In addition, there are structural equation software packages that borrow the concept of partial regression from linear regression to represent the "true" relationship between variables in the structural equation model. Unfortunately, the regression coefficients obtained by this method are different from those obtained by the structural equation. The path coefficients are also inconsistent. So how to express the relationship between variables based on the structural equation model coefficients? What is the meaning of bivariate regression between variables ? We will answer the above questions through examples and explore how to use structural equations for prediction.
1. Overview of prediction problems using structural equation models
2. Ways to realize direct prediction by structural equation model
3. Realization and expression of partial relationship between variables after structural equation modeling
Topic 8: Things to note when publishing Structural Equation Modeling (SEM) papers
Through actual case analysis, we will share the structural equation model modeling process, result display, and main precautions and common problems in paper writing to avoid problems with paper rejection.
