Education Research Quantitative Analysis for Little Respondents: Comparing of Lisrel, Tetrad, GSCA, Amos, SmartPLS, WarpPLS, and SPSS

Many researchers are confused about which software to use because there is no research on software comparisons for quantitative research data analysis.The purpose of this study is to compare the results of quantitative research data processing in the field of education management using Lisrel, Tetrad, GSCA, Amos, SmartPLS, WarpPLS, and SPSS software for small samples or respondents. This research method is quantitative and research data analysis uses the four types of software to obtain a comparison of the results of the analysis. The analysis in this study focuses on the analysis of hypothesis testing and regression analysis. Regression analysis is used to measure how much influence the independent variable has on the dependent variable. The field of this research is education management and the research data uses quantitative data derived from questionnaire data for a small sample of 40 respondents with three research variables, namely the independent variable of transformational leadership and job satisfaction, while the dependent variable is teacher performance. Based on the results of the analysis using Lisrel, Tetrad, GSCA, Amos, SmartPLS, WarpPLS, and SPSS software, the results showed that for a small sample there was no significant difference in the significance value of p-value and t-value. There is also no significant difference in the determination value, and the correlation value in the resulting structural equation also has no significant difference in results, while for CB-SEM represented by Lisrel, Tetrad cannot process data with a Little respondents size. The novelty of this research is the result of comparative analysis of Lisrel, Tetrad, GSCA,


Introduction
Many different statistical software are available, and each offers a little something different for the user. What one chooses will depend on a variety of factors, including research questions, statistical knowledge, and coding experience. These factors may mean the cutting edge of data analysis, but as with any research, the quality of the data obtained depends on the quality of the conduct of the study. It is therefore important to remember that while we may have advanced statistical software and the knowledge to use it available, the results will not mean much if the data are not collected in a valid way. Data analysis is one of the research processes that is carried out after all the data needed to solve the problems studied have been obtained in full. Sharpness and accuracy in the use of analytical tools will determine the accuracy of conclusions, therefore data analysis activities are activities that cannot be ignored in the research process. Errors in determining the analytical tools can be fatal to the conclusions drawn and this will have an even worse impact on the use and application of the research results. Thus, knowledge and understanding of various analytical techniques is absolutely necessary for a researcher so that his research results are able to make a meaningful contribution to problem solving as well as the results can be scientifically justified. Some researches in the field of education management use statistical software tools such as Lisrel, Tetrad, GSCA, Amos, SmartPLS, WarpPLS, and SPSS. Many researchers in the field of education management are still hesitant in choosing which software to use.
Smart PLS or Smart Partial Least Square is statistical software that has the same purpose as Lisrel and AMOS, namely to test the relationship between variables, both among latent variables and with indicator variables, or manifest.The use of Smart PLS is highly recommended when we have a limited number of samples while the model being built is complex. this cannot be done when we use the two software above. they need sufficient samples. There are many researches in the field of education that use SmartPLS, such as those conducted by Budi Santoso, P., Asbari, M., . examines the role of job satisfaction and organizational citizenship behavior on performance. Putra, F., Asbari, M., Purwanto, A., Novitasari, D., & Santoso, P. B. (2021) investigated linking social support and performance in college. Johan, M. (2021) examines the effect of knowledge sharing and interpersonal trust on innovation. Purwanto, A., Santoso, PB, Siswanto, E., Hartuti, H., Setiana, YN, Sudargini, Y., & Fahmi, K. (2021) the effect of hard skills, soft skills, organizational learning and innovation capability on performance lecturer. Novitasari, D., Asbari, M., Purwanto, A., Fahmalatif, F., Sudargini, Y., Hidayati, L. H., & Wiratama, J. (2021). examined the effect of social support factors on the performance of elementary school teachers. Anas Ahmadi, E., Herwidyaningtyas, F.B., & Fatimah, S. (2020) examined the influence of organizational culture, work motivation, and job satisfaction on lecturer management performance. Jayus, J.A. (2021). Examining the effect of distributive justice, procedural justice and interactional justice on teacher involvement and teacher performance. . examined the relationship between teacher support, academic efficacy, academic resilience, and student engagement in Bahrain.  examined the relationship between teacher support, academic efficacy, academic resilience, and student engagement in Bahrain. Purwanto, A., J.T. Purba, I. Bernardo, and R. Sijabat. (2021) examines the effect of management innovation, transformational leadership and knowledge sharing on performance.  examined the role of organizational citizenship behavior (OCB), transformational and digital leadership on performance through mediating organizational commitment.
AMOS is a statistical software developed by IBM. Amos software is specifically designed to help test the hypothesis of the relationship between variables. Through this software, we can determine the level of strength of the relationship between variables, both between latent variables and manifest variables. How significant is the relationship between variables, and how fit the hypothetical model is compared to the real data in the field. The advantage of Amos is that we don't need a complex syntax or programming language to operate this software. For beginners, or those who are unfamiliar with programming languages, of course this is a distinct advantage. Through amos, simply describe the latent variable and the manifest variable, then relate them via the arrows provided. There have been many studies in the field of education using AMOS, such as  researching structural equation modeling using AMOS: confirmatory factor analysis for the workload of teachers in special education integration programs. Fathema, N., Shannon, D., & Ross, M. (2015) extending the technology acceptance model (TAM) to examine the use of faculty learning management systems (LMSs) in higher education institutions. Standage, M., & Treasure, D. C. (2002) the relationship between achievement goal orientation and multidimensional situational motivation in physical education. Kulinna, P. H., & Cothran, D. J. (2003) self-reported physical education teacher use and perceptions of various teaching styles. Skaalvik, E. M., & Skaalvik, S. (2010) teacher self-efficacy. Iordanoglou, D. (2007) relationship between emotional intelligence and leadership effectiveness, commitment, and satisfaction. Kokkinos, C. M., Charalambous, K., & Davazoglou, A. (2009)  SPSS (Statistical Package for the Social Sciences) is an application used to perform advanced statistical analysis, data analysis with machine learning algorithms, string analysis, and big data analysis that can be integrated to build a data analysis platform. SPSS is one of the most widely used and frequently used software for quantitative research. To produce a comprehensive data analysis, a researcher must know the use of this program. SPSS was born as a goal to make it easier for researchers to organize and produce the right data according to the established method. As of 2019, SPSS already has its 25th version.There are many researches in the field of education that use SPSS, such as those conducted by ebjan, U., & Tominc, P. Generalized Structured Component Analysis (GSCA) is a method that has been developed to complete the shortcomings that exist in Partial Least Square, namely in the overall goodest of fit. In this study, the GSCA method was used to model the structural factors that affect the nutritional status of children under five, which is a unidimensional structural equation. This study aims to examine and apply the SEM-GSCA approach with a unidimensional structural equation model for a case study of determining the factors that affect nutritional status in children under five in Genteng-Banyuwangi. The GSCA method in estimating its parameters uses the Alternating Least Square (ALS) method and in estimating the standard error of parameter estimation using the bootstrap method. To evaluate the GSCA model, three stages are carried out, namely, the first is an evaluation of the measurement model by looking at convergent validity, discriminant validity, composite reliability and average variance extracted (AVE). There have been many educational studies using the GSCA, such as those conducted by Sturman, E. D., & Zappala-Piemme, K. (2017)  School. Lisrel was developed by Karl Jöreskog and Dag Sörbom. Lisrel is the most widely used statistical software among researchers and practitioners. The advantage of lisrel software is its ability to identify relationships between complex variables. How to operate it, which consists of various options, both with syntax and with simple programs, making it more widely used by various groups. Syntax will certainly be preferred for users who are familiar with programming languages. While Simplis or simple lisrel is an alternative for those who are unfamiliar with programming languages.Many educational studies use Lisrel such as Dorman, J. P. (2003) relationship between school and classroom environment and teacher burnout: A LISREL analysis. Yuen

Method
This research method is quantitative, research data analysis uses Lisrel, Tetrad, GSCA, Amos, SmartPLS, WarpPLS, and SPSS software to obtain a comparison of the results of the analysis. The analysis in this study focuses on the analysis of hypothesis testing and regression analysis. Regression analysis is used to measure how much influence the independent variable has on the dependent variable. The data from this study used quantitative data derived from questionnaire data with a small sample of 40 respondents. In the data there are 3 variables, namely three research variables, namely the independent variables of transformational leadership and job satisfaction, while the dependent variable is the performance developed from Purwanto et al. (2020); Asbari et al. (2021) and Novitasari et al (2020) with the following research model: X is Transformational Leadership, Y1 is Job Satisfaction and Y2 is Teacher Performance. The relationship models to be analyzed are as follows: 1. The relationship between Transformational Leadership (X) and Teacher Performance (Y2). 2. The relationship between Transformational Leadership (X) and Job Satisfaction (Y1). 3. The relationship between job satisfaction (Y1) and teacher performance (Y2). 4. The relationship between Transformational Leadership (X) and Teacher Performance (Y2) through Job Satisfaction (Y1).

Results & Discussion
Testing the Significance of t-Value The first stage of data analysis is testing the significance of the relationship between the independent variable transformational leadership (X), job satisfaction (Y1) with the dependent variable teacher performance (Y2) by looking for t-Value using Lisrel, Tetrad, GSCA, Amos, SmartPLS, WarpPLS, and SPSS, the decision criteria if the t-Value value is greater than 1.96 or > 1.96 then the relationship is significant, if less than 1.96 or < 1.96 then the relationship is not significant. For WarpPLS does not produce a t-statistic value, the significance test can be seen on the p-value, so that the t-statistic value will be obtained.
The test results with 4 software for a direct relationship can be seen in Table 1 below:

Relationship between transformational leadership (X) and job satisfaction (Y1)
Based on the results of the software analysis, the results of the t-Value using Amos data cannot be processed. The t-Value using SmartPLS is 89,509, which is greater than 1.96, so it can be concluded that the relationship between X and Y1 is significant. The result of t-Value using SPSS is 21.424 which is greater than 1.96 so that it can be concluded that the relationship is significant so that it can be concluded that SmartPLS and SPSS give the same results.

Relationship between transformational leadership (X) and performance (Y2)
Based on the results of the software analysis, the results of the t-Value using Amos using Amos data cannot be processed. The t-Value using SmartPLS is 1.960, which is greater than 1.96, so it can be concluded that the relationship between X and Y2 is significant. The results of the t-Value using SPSS of 2.125 are greater than 1.96 so that it can be concluded that the relationship between X and Y2 is significant, so it can be concluded that SmartPLS and SPSS give the same results.

The relationship between transformational leadership (X) and performance (Y2) through job satisfaction (Y1)
Based on the results of the software analysis, the results of the t-Value using Amos data cannot be processed. The result of t-Value using SmartPLS is 0.822 which is smaller than 1.96 so that it can be concluded that the relationship between X and Y2 through Y1 is not significant. The result of t-Value using SPSS of 1.051 is smaller than 1.96, so it can be concluded that the relationship between X and Y2 through Y1 is not significant, so it can be concluded that SmartPLS and SPSS give the same results.

Testing the Significance of p-Value
The second stage is data analysis, namely testing the significance of the relationship between the independent variable transformational leadership (X), job satisfaction (Y1) with the dependent variable teacher performance (Y2) by looking for p-value using SPSS, Amos, SmartPLS, WarpPLS and SPSS software. The decision is that if the p-value is less than 0.050 or <0.050 then the relationship is significant, if it is more than 0.050 or >0.050 then the relationship is not significant.
The test results with 4 software for direct connection are as follows:

Relationship between transformational leadership (X) and job satisfaction (Y1)
Based on the results of the software analysis, the p-value results using Amos data cannot be processed. The p-value using SmartPLS is 0.000 less than 0.050 so it can be concluded that the relationship between X and Y1 is significant. The p-value using WarpPLS is 0.000 less than 0.050 so it can be concluded that the relationship is significant. The p-value using SPSS is 0.000 less than 0.050, so it can be concluded that the relationship between X1 and Y1 is significant, so it can be concluded that WarpPLS, SmartPLS and SPSS give the same results.

The relationship between transformational leadership (X1) and performance (Y2)
Based on the results of the software analysis, the p-value results using Amos data cannot be processed. The p-value using SmartPLS is 0.046 which is smaller than 0.050, so it can be concluded that the relationship between X and Y2 is significant. The result of p-value using WarpPLS is 0.410 less than 0.050 so that it can be concluded that the relationship between X and Y2 is significant. The result of the p-value using SPSS is 0.042 less than 0.050 so it can be concluded that the relationship between X and Y2 is significant t so that it can be concluded that WarpPLS. SmartPLS and SPSS give the same result.

The relationship between transformational leadership (X1) and performance (Y2) through job satisfaction (Y1)
Based on the results of software analysis, the p-value using SmartPLS was 0.411 more than 0.050, so it was concluded that the relationship between X and Y2 through Y1 was not significant. The result of p-value using WarpPLS is 0.450 which is greater than 0.050 so that it can be concluded that the relationship between X and Y2 through Y is not significant. The results of the p-value using SPSS of 0.302 is greater than 0.050 so that it can be concluded that the relationship between X and Y2 through Y is not significant.

Coefficient of Determination Test
Testing the coefficient of determination to calculate the influence of the independent variable on the dependent variable. In this study, the R Square termination coefficient was calculated for the independent variables of transformational leadership (X), Job Satisfaction (Y1) and Performance (Y2). The results of the R Square test using Amos, SmartPLS, WarpPLS and SPSS are as follows: Based on the results in Table 4, the R Square value for Job Satisfaction (Y1) using Amos cannot be run. The value of R Square for Job Satisfaction (Y1) using SmartPLS is 0.941 or 94.1%, meaning that the Job Satisfaction variable (Y1) is influenced by the transformational leadership variable (X) of 94.1% while the remaining 5.9% is influenced by other variables that are not discussed in this study. The R Square value for Job Satisfaction (Y1) using WarpPLS is 0.95 or 95%, meaning that the Job Satisfaction variable (Y1) is influenced by the transformational leadership variable (X) by 95% while the remaining 5% is influenced by other variables not discussed in this section. this research. The R Square value for Job Satisfaction (Y1) using SPSS is 0.939 or 93.9%, meaning that the Job Satisfaction variable (Y1) is influenced by the transformational leadership variable (X) by 93.9% while the remaining 6.1% is influenced by other variables not discussed in this study. . Table 4, the R Square value for Performance (Y2) using Amos cannot be run. The value of R Square for Performance (Y2) using SmartPLS is 0.852 or 85.2%, meaning that the performance variable (Y2) is influenced by transformational leadership variables (X) and job satisfaction (Y1) is 85.2% while the remaining 14.8% is influenced by other variables not discussed in this study. The R Square value for Performance (Y2) using SmartPLS is 0.85 or 85%, meaning that the performance variable (Y2) is influenced by transformational leadership variables (X) and job satisfaction (Y1) by 85% while the remaining 15% is influenced by other variables. which were not discussed in this study. The value of R Square for Performance (Y2) using SmartPLS is 0.844 or 84.4%, meaning that the performance variable (Y2) is influenced by transformational leadership variables (X) and job satisfaction (Y1) is 84.4% while the remaining 15.6% is influenced by other variables not discussed in this study.

Correlation Coefficient Test
The correlation coefficient shows the strength of the linear relationship and the direction of the relationship between variables. If the correlation coefficient is positive, then the two variables have a unidirectional relationship (Purwanto et al, 2020). This means that if the value of the variable X is high, then the value of the variable Y will be high as well. Conversely, if the correlation coefficient is negative, then the two variables have an inverse relationship. This means that if the value of the variable X is high, then the value of the variable Y will be low and vice versa. According to Hair et al (2017) to make it easier to interpret the strength of the relationship between two variables, the following criteria are provided: • 0 means There is no correlation between two variables • >0.00 -0.25 means the correlation is very weak • > 0.25 -0.50 means enough correlation • >0.50 -0.75 means strong correlation • > 0.75 -0.99 means the correlation is very strong • 1.00 means perfect correlation The results of testing the correlation coefficient for structural equations using Amos, SmartPLS, WarpPLS and SPSS software are as follows: The results of structural equations using Lisrel, Tetrad and Amos software cannot be run. The results of the structural equation using SmartPLS software obtained the equation is Y2 = 0.642X1+0.287X2+e, meaning that the correlation coefficient value of the influence of transformational leadership variable (X) on performance (Y2) is 0.642, meaning that there is a strong correlation and indicates that if the value of transformational leadership ( X) increases by 1 unit, while the value of job satisfaction (X2) remains, the performance value (Y2) will increase by 0.642 units. This means that the partial effect of transformational leadership on performance is 64.2%. The correlation coefficient value of the influence of job satisfaction variable (Y1) on performance (Y2) is 0.287, meaning that there is a sufficient correlation and shows that if the value of job satisfaction (Y1) increases by 1 unit, while the value of transformational leadership (X) remains, the performance value (Y2) will increase by 0.287 units. This means that the effect of job satisfaction (Y1) on performance partially is 28.7%.
The results of the structural equation using WarpPLS software obtained the equation is Y2 = 0.658X1+0.271X2+e, meaning that the correlation coefficient value of the influence of transformational leadership variable (X) on performance (Y2) is 0.658, meaning that there is a strong correlation and shows that if the value of transformational leadership ( X) increases by 1 unit, while the value of job satisfaction (X2) remains, the performance value (Y2) will increase by 0.658 units. This means that the partial effect of transformational leadership on performance is 65.8%. The correlation coefficient value of the effect of job satisfaction variable (Y1) on performance (Y2) is 0.271, meaning that there is a sufficient correlation and indicates that if the value of job satisfaction (Y1) increases by 1 unit, while the value of transformational leadership (X) remains, the performance value (Y2) will increase by 0.271 units. This means that the effect of job satisfaction (Y1) on performance partially is 27.1%.
The results of the structural equation using SPSS software obtained the equation is Y2=0.219+ 0.642X1 + 0.324Y1 + e, meaning that the correlation coefficient value of the influence of transformational leadership variable (X) on performance (Y2) is 0.642, meaning that there is a strong correlation and indicates that if the value of leadership transformational (X) increases by 1 unit, while the value of job satisfaction (X2) remains, the performance value (Y2) will increase by 0.642 units plus a constant of 0.219 units. This means that the partial effect of transformational leadership on performance is 64.2%. The correlation coefficient value of the influence of job satisfaction variable (Y1) on performance (Y2) is 0.324, meaning that there is a sufficient correlation and indicates that if the value of job satisfaction (Y1) increases by 1 unit, while the value of transformational leadership (X) remains, the performance value (Y2) will increase by 0.324 units plus the constant 0.219 units. This means that the effect of job satisfaction (Y1) on performance partially is 32.4%.
The results of the structural equation using the GSCA software obtained the equation is Y2=a+ 0.703X+ 0.229Y1 + e, meaning that the correlation coefficient value of the influence of transformational leadership variable (X) on performance (Y2) is 0.704, meaning that there is a strong correlation and shows that if the value of transformational leadership ( X) increases by 1 unit, while the value of job satisfaction (X2) remains, the performance value (Y2) will increase by 0.703 units n. This means that the partial effect of transformational leadership on performance is 70.3%. The correlation coefficient value of the influence of job satisfaction variable (Y1) on performance (Y2) is 0.229 meaning that there is a sufficient correlation and indicates that if the value of job satisfaction (Y1) increases by 1 unit, while the value of transformational leadership (X) remains, the performance value (Y2) will increase by 0.229 units plus the constant 0.229 units. This means that the effect of job satisfaction (Y1) on performance partially is 22.9%.
PLS-SEM is a causal modeling approach that aims to maximize the explained variance of the dependent latent construct. This contradicts the goal of CB-SEM to reproduce a theoretical covariance matrix, without focusing on the variance described. PLS-SEM becomes important when we deal with empirical research challenges such as smaller sample sizes. The estimation of the CB-SEM model requires a series of assumptions that must be met, including the normality of multivariate data, minimum sample size, and so on. But if the assumption of CB-SEM cannot be met, or the aim of the study is prediction rather than confirmation of structural relationships, then variance-based PLS-SEM is the appropriate method. Compared to CB-SEM results, which can be very imprecise when assumptions are violated, PLS-SEM often provides stronger estimates than structural models.
Another philosophical difference between CB-SEM and PLS-SEM: if the aim of the research is the testing and confirmation of theory, then the appropriate method is CB-SEM. On the other hand, if the research objective is prediction and theory development, then the appropriate method is PLS-SEM.The main objective is to maximize the differences described in the dependent construct but also to evaluate the quality of the data based on the characteristics of the measurement model. Given the ability of PLS-SEM to work efficiently with a much wider range of sample sizes and increased model complexity, and less stringent assumptions about the data, PLS-SEM can address a wider range of problems than CB-SEM.
In addition, because construct measurement is less constrained with PLS-SEM, constructs with fewer items (e.g., one or two) can be used than CB-SEM requires. Overall, when the measurement or nature of the model limits the use of CB-SEM or when the emphasis is on exploration rather than confirmation, PLS-SEM is an attractive and often more appropriate alternative.PLS-SEM can be applied to a wider variety of situations, researchers should always be aware of differences in the interpretation of the results, especially with regard to the nature of construct measurement. For example, can PLS-SEM be applied appropriately when measurement theory fails to meet the criteria of confirmatory factor analysis, including tests of convergent validity and tests of discriminant validity.
PLS-SEM estimates variable indicator loadings for exogenous constructs based on their predictions of endogenous constructs, not the joint variance among indicator variables in the same construct. Loadings in PLS-SEM by means of their contribution to the path coefficient.While CB-SEM gives poor results for the measurement model but significant structural model relationships, PLS-SEM offers acceptable results for the measurement model whereas structural model relationships are not significant. This suggests that the difference in the results is mainly a matter of the quality of the measurement model. Using "good" measures and data, these two approaches yield practically the same results. To properly apply CB-SEM and PLS-SEM, researchers must understand the objectives of developing each approach and apply them. Structural equation models with good measurement properties generally achieve comparable results with both approaches, especially when the CB-SEM model specifications are properly set up. Moreover, both approaches still have to consider issues such as the proper use and interpretation of formative versus reflective measures. These situations are often situations in which the nature of the measurement is questioned and the results may differ, thus requiring the researcher to make a reasonable judgment as to which approach is most appropriate.

Conclusion
PLS-SEM has several advantages and disadvantages relative to CB-SEM. The advantage lies in its robustness, meaning that it will provide a solution even if there are problems that could prevent a solution in SEM. First, poor measurement is one of the main barriers to obtaining an SEM solution. For example, when researchers are trying to test a structural model with a single item size or a combination of multiple one-and two-item measures, PLS may be an option because of identification problems that may occur in SEM. As we have noted, all recursive models were identified (showing no statistical identification problem) in PLS, even with single item sizes. Thus, while validating one-and two-item measures in the context of measurement theory does not mean much with SEM, PLS is not constrained by the problem. Based on the results of the analysis using GSCA, SPSS, SmartPLS and WarpPLS software, the results showed that for a small sample there was no significant difference in the significance value of p-value and t-value. There is also no significant difference in the determination value produced, and the correlation value in the resulting structural equation also has no significant difference in results, while for CB-SEM represented by Lisrel , tetrad and Amos cannot process data with a small sample size. Many researchers have used PLS for only this reason, given the perceived difficulty in the specification of formative models in SEM. PLS can also be a useful way to quickly explore a large number of variables to identify sets of principal component variables that can predict multiple outcome variables