4.1.4. Co-Citation Analysis

Co-citation analysis of cited references was performed as well. By definition, the reference can be a co-citation if the two documents are cited together by another document [63]. As shown in Figure 5, each point represents the cited author, and the color of the points is according to the number of co-citations. A total of 88,731 cited authors were detected, and only 800 authors met the threshold in which the minimum citation of an author is 800. As seen in Figure 5, 800 authors formed 7 different clusters that provide information related to the co-citation of this study. Overall, most of the co-citations are related to statistical methods such as the Latent growth curve model, evaluation in structural equation modeling, evaluation in PLS-SEM, and procedure in the simulation study. The highest total link strength in co-citation analysis is Muthen and McArdle, and the article is related to the simulation study and latent curve analysis.

**Figure 4.** Network visualization map of citations based on the documents.

**Figure 5.** Network visualization map of co-citation of cited authors.

#### 4.1.5. Keyword Co-Occurrence Analysis

The co-occurrence analysis focuses on the examination of the actual content of the publications based on the words derived from the author's keyword. This analysis can determine the trend of research topics in recent years. Figure 6 shows the network visualization map of the co-occurrence of keywords related to the panel survey data. Based on this analysis, 3875 keywords were retrieved from 1725 articles. However, there were only 49 keywords that met the minimum threshold of occurrences number of at least 10. As seen in Figure 6, 49 keywords formed 9 different clusters that provided information about the related topic of this study. The largest cluster was the red and green clusters which consisted of 11 keywords for each cluster. In addition, Table 6 shows the list of keywords as well as their co-occurrence frequencies in each cluster. In the context of the research topic, longitudinal study and adolescence showed the highest co-occurrences in this study with 292 and 169 repeated keywords, respectively. While in the context of statistical method, the Latent growth curve model was the most used in the analysis, with 294 co-occurrences keywords, followed by structural equation modeling, PLS-SEM, and partial least squares with 22, 14, and 11 respectively. Besides, the keywords in the same cluster shared a similar topic. Generally, most of the research topic for each cluster is related to mental health, psychology, child and adolescent development, and lifestyle.

**Figure 6.** The network visualization map of co-occurrences of keywords.

To examine the trend of research topics in the recent year, an overlay visualization map was produced. Figure 7 shows the co-occurrences of the keywords according to the time (in years). Based on the overlay visualization map, there were a few research topics in recent years, such as mental health, life satisfaction, cognition, aggression, effortful control, children, social media, and COVID-19. In the context of statistical methods, PLS-SEM has been used in recent years to analyze the panel survey data.


**Table 6.** Co-occurrence of author keyword in panel survey data.

**Figure 7.** The overlay visualization map of keywords according to year.

Overall, bibliometric analysis has fulfilled the first research question related to the trend of the Latent growth curve model (LGCM) and PLS-SEM in panel survey data. The result shows that the distribution, trend, and application of LGCM are more outstanding than PLS-SEM in analyzing the panel survey data. To answer the next two research questions, content analysis was employed. Content analysis is focused on exploring the PLS-SEM in analyzing the panel survey data. The reason behind exploring the PLS-SEM in the panel survey data is due to the bibliometric analysis that shows the development of this method is not well developed, even though this method has good potential in handling panel survey data.

#### *4.2. Themes Generation*

To answer the other two research questions related to the limitations of PLS-SEM in panel survey data and the procedure of the existing approach of PLS-SEM in panel survey data, content analysis was employed. This section explains the details of content analysis on the PLS-SEM approach for panel survey data based on the selected top 100 most cited papers. The explanation of the content analysis is divided into two different themes: (i) identification of PLS-SEM approaches and their limitations, and (ii) procedures of the method.

#### 4.2.1. Identification of PLS-SEM Approaches and Their Limitations

The exploration of the PLS-SEM approach is explained according to the evaluation of the measurement, and the structural model. Based on the reviewed articles, most of the researchers used the standard procedure to evaluate the measurement model as suggested by [64]. The measurement model involved the evaluation of indicator reliability, internal consistency reliability, convergent validity, and discriminant validity.

However, when evaluating the structural model, most of the top researchers use different approaches and procedures. As a summary, five approaches and procedures are identified to be used by researchers. The first approach is suitable for two periods of time and uses the different latent constructs at the pre-evaluation and post-evaluation named *pre- and post-approaches with different constructs*. The main purpose of this approach is to evaluate relationships between the exogenous variable at pre-evaluation, and the endogenous variable or outcome at post-evaluation. The relationship was evaluated using partial least squares (PLS) in the structural model. The second approach is known as the *Path Comparison approach,* and it is suitable for two periods of time. This approach uses the same latent construct at the initial (t0) and end (t1) of the evolution. This approach can measure the relationship of the latent construct using path analysis and the impact of time on the PLS model using a t-test. The third approach is named the *Cross-Lagged Panel Method (CLPM),* and the main purpose is to measure the direction, strength, and cause-effect relationship among latent constructs over time. This method is also suitable for two periods of time and uses the same construct at time 1 and time 2. The fourth approach also involves two periods of time and measures the same latent constructs at pre-evaluation and post-evaluation named *pre- and post-approaches with the same constructs.* The difference between this method compared to the other three is regarding the latent score used to develop the PLS model. This method uses the differences in scores from time 1 and time 2 to develop the PLS model. The fifth approach is the *evaluation approach* which involves more than two waves of time and uses the same constructs for evaluating participants over time. The main purpose is to evaluate the direct and indirect effects over time. This approach uses path analysis to evaluate the direct effects between latent constructs and indirect effects over time using a bias-corrected confidence interval.

Though there are five different approaches of PLS-SEM in analyzing the panel survey data, these approaches still have limitations and spaces for improvement. The obvious limitation for models 1, 2, 3, and 4 is related to the number of waves for the study, since these approaches are only suitable for two periods of time. In addition, these four models focus on the pre- and post-evaluations and do not measure the evaluations of effects over time. Besides that, model 5 has the limitation in evaluating the growth of trajectory even though this method is capable of handling studies with more than two periods of time. This model does not have one structural model to represent the whole changes in the repeated measure. In addition, this model cannot capture an individual trajectory, the mean of the trajectory of the sample or entire group, the evaluation of individual differences in trajectories, and assess the potential incorporation of predictors of individual differences in trajectories. Furthermore, this model is not flexible to handle the latent constructs simultaneously as independent and dependent in the same model, allowing for complex representations of growth and correlations of change. Table 7 shows the summary of the five different approaches practiced by researchers in analyzing panel survey data.


**Table 7.** Summary of approaches practiced in analyzing a structural model.

4.2.2. Procedure of the Approaches

This section explains the details of the procedure for several PLS-SEM approaches practiced by the researchers in analyzing the panel survey data. The procedure of the approach is explained according to the data collection phase and analysis phase. Table 8 shows the summary of the procedure for five approaches in PLS-SEM to analyze the panel survey data.

**Table 8.** The procedure of the PLS-SEM approach in analyzing panel survey data.

#### **Table 8.** *Cont.*
