Composition and application of structural equation model;
The structural equation model consists of two parts, namely the measurement model and the structural model. This paper will mainly introduce the concepts and applications of the above two models.
1, measurement model
In practical research, not all concepts can be directly observed and measured.
For example, when we investigate the experience of sellers who love to buy, the seller experience here is actually an abstract concept and a comprehensive reflection of all measurable indicators of the platform. These indicators may include inquiries received by sellers through the platform, orders, satisfaction with major rights and interests, service speed and quality, and so on.
In SEM, if users experience these abstract concepts that cannot be directly measured, they are called "latent variables", while those variables that can be directly observed, such as query volume, are called "observed variables" or "explicit variables".
The more you know about the feedback of the seller on the effective observation variables of the platform, the more realistic and reliable the description of the seller's experience will be.
Based on the verification of the measurement model, we find that the seller's comprehensive experience of the platform can be interpreted as a collection of the seller's experience of the platform, the experience of rights and interests and the experience of service (satisfaction) to some extent.
It should be noted that the observed variables cannot fully explain the potential variables, and there are unexplained errors (also called residuals) in the overall measurement model. The influence of error size and distribution is also a part that needs to be considered in actual measurement.
2. Structural model
Different from the measurement model that tests the relationship between observed variables and latent variables, the structural model is mainly used to test the relationship between latent variables. If we only look at the structural model, it is the traditional path analysis, which aims to explain the causal or predictive relationship between variables.
With the deepening of research, we find that correlation analysis or univariate/multivariate regression analysis commonly used in past research is difficult to explain the causal relationship between variables. For example, when studying the buyer's willingness to renew, it is difficult to judge whether the experience affects the willingness to renew or the willingness to renew affects the experience only through correlation analysis.
However, only using multiple regression analysis, we can only find the independent influence of each experience dimension index on the willingness to renew the fee, while ignoring the interaction between each experience index.
Structural equation model has the following points to pay attention to:
1 and SEM are more used for confirmatory analysis.
Therefore, in practical research, it is necessary to combine business analysis, qualitative research, theoretical summary and other methods to set an initial theoretical model and then verify it.
In short, if the hypothesis is first, the influence path between concepts can be drawn first, and then transformed into a statistical model for correction;
2.SEM usually requires a relatively large sample size.
Because of the large number of variables processed by SEM, the relationship between variables is complex, and the samples
The scale will affect the stability and applicability of the overall analysis.
Generally speaking, the sample size needs to exceed 200. When there are many latent variables involved, the sample size can be set at 10 times of the topic size.
3. Use the original data directly when analyzing the data.
Because the mathematical statistics foundation of SEM is based on variance and covariance analysis, the covariance matrix of original data or samples should be directly used instead of standardized data or correlation matrix when using SEM to avoid wrong parameter estimation or error.