Instructions for analyzing the 2nd order PLS SEM model in SmartPLS 4

PLS SEM SECONDARY MODEL ANALYSIS GUIDE

USING SmartPLS SOFTWARE 4

A higher-order model is a type of structural analysis model in which the independent and dependent variables are classified into groups, and each group is modeled as a hierarchical structural model. (first-order structural model). This group was then used as the independent variable in a higher-order structural model.

A higher-order model can be viewed as an expanded first-order structural analysis model, with particular independent and dependent variables grouped and modeled as a first-order structural model. individual. These first-order models are called “lower-order models”, while the higher-order models containing these lower-order models are called “higher-order models”. ).

Higher-order models are used to examine relationships between variables at a more abstract and complex level, providing a more general view of the relationships between variables. It is often used in research in psychology, education, economics, and other social science fields to examine relationships between abstract concepts.

However, the use of higher-order models requires an understanding of structural analysis methods and practical experience to identify appropriate independent and dependent variables to group, and to establish relationships between variable groups.

First, to go deeper into the technique of analyzing high-level structural models, we need to talk briefly about the types of multidimensional measurement models.

A multidimensional measurement model is a type of model in measurement analysis in which the quantities being measured have not just one dimension but multiple dimensions, also known as multiple attributes. A simple example of a multidimensional measurement model is in the case of measuring a country’s development index. Instead of just measuring one dimension like GDP, the multi-dimensional measurement model will measure many dimensions such as GDP, education quality, life expectancy, health assessment… In the multi-directional measurement model, the quantities measured are defined as predictor variables and modeled using SEM (Structural Equation Modeling) or PLS (Partial Least Squares) models. This allows analyzing relationships between attributes and finding hidden relationships between attributes.

Following Edwards and Bagozzi (2000), we have two basic patterns representing the association between lower-order latent variables and higher-order latent variables.

Reflective constructs are a type of variable in which the measured variable is said to reflect a latent variable.

In the reflective model, latent variables (also called factors) are said to reflect aspects and dimensions of a common latent variable. Observed variables are considered to reflect or measure these latent variables. It is important that observed variables must have a strong correlation with the latent variables they measure. Observed variables are not considered to influence each other or contribute to the determination of latent variables.

Different from reflective latent variables (reflective constructs), structural latent variables (formative constructs) are a type of variable in SEM (Structural Equation Modeling) analysis, in which measured variables (observed variables) are considered formative ( form) a latent variable.

In contrast to the reflective model, in the formative model, observed variables are said to create or form latent variables. This means that observed variables influence latent variables and contribute to their determination. Therefore, the researcher cannot eliminate any observed variable of this type of latent variable. In the formative model, observed variables are not expected to be strongly correlated with each other or with latent variables.

In models 1 and 2, we change the direction of the arrows at the indicators of the first-order latent variable to add two new models, but these two types are rarely used, so we will not consider them here. consider.

Thus, before analyzing PLS SEM, you need to establish a research model based on the theoretical foundations mentioned above combined with research objectives and references from previous studies. . In addition, the choice between reflective and formative models depends on the theoretical significance and database of the research. It is important to ensure that the choice is appropriate to the research objectives and the nature of the variables being studied. Note that defining a reflective or formative model is not always straightforward, and may require careful consideration and knowledge of the specific field of study.

We will have 2 ways to implement the high-order PLS SEM model, both ways are the same in stage 2, only different in stage 1. In stage 1, we will find indicators for the latent variable high level.

The first way is that we create a second-order variable at stage one by including all the indicators of the first-order variable as shown.

Evaluating the model in this first stage requires assessing the criteria of reliability level, convergence accuracy level, and discrimination accuracy level. We have mentioned the evaluation method in previous articles so we will not present it again here.

In case the model in stage one does not meet the requirements, we need to reprocess the data in the following ways:

Remove observations (Observed variables) that do not ensure the above conditions and run again from the beginning.
Re-study the questionnaire and conduct the survey again.
Contact Le Minh Data Service for quick support.

In addition, the issue of removing observed variables from the scale requires careful consideration before deciding. We recommend that you consult your Science Instructor for further implementation.

If the model in stage one meets the requirements, we will find indicators for the second-order latent variables by clicking on Open report selecting Latent variables and copying the data to an excel file. In this excel file, we will have indicators for quadratic variables, in this example AS, AW and QP.

At this point, the quadratic model is rebuilt as follows, this stage is considered phase 2. We should note that we should save the data into another file and insert it into SmartPLS to use for the quadratic model.

For the structural model, to change the arrow, right-click on the quadratic variable (Second-Order) and select Edit settings, in the measurement model, select formative.

Results of quadratic model analysis.

The study will conclude the research hypothesis based on the results of the model in phase two. Bootstrapping analysis to test p value.

Here both research hypotheses are accepted. The remaining issue is to discuss the research results and provide recommended solutions, if any. Note, when discussing results to make recommendations, it is necessary to consider combining with the results the indicators of latent variables in stage one.

You can refer to SmartPLS documents in these two books by Professor. Nguyen Minh Ha (Ho Chi Minh City Open University). In both of these books, the professor only practices on SmartPLS 3. As for the current SmartPLS 4 version, there is only one English book by author Chua Yan Plaw (this book you can buy on amazone, the price is around 2 million or so). what).