Goodness Of Fit Test In Structural Equation Modeling with Unweighted Least Square (ULS) Estimation Method

Abstrak. Model persamaan struktural (MPS) adalah metode analisis multivariat yang


INTRODUCTION
SEM is a method that was formed because of the problem of measuring a variable where there is a variable that cannot be measured directly (Pagadala et al., 2023).Variables that cannot be measured are called latent variables which require a manifest variable as an indicator or measuring tool for the latent variable.There are several estimation methods in the SEM method including Maximum Likelihood (ML), Weighted Least Squares (WLS), Generalized Least Squares (GLS), Unweighted Least Squares (ULS) (Ghanbar & Rezvani, 2023).Each method has specificity in its use according to the distribution, sample size and scale of the data, as well as the underlying assumptions.According to Mokhtar et al. (2023) and Tarko (2023) some parameter estimation methods require strict assumptions related to the shape of the distribution and sample size.The ML and GLS methods require the assumption of normality in the data (Du & Bentler, 2022).While the WLS and ULS methods do not require the assumption of normality (Kyriazos & Poga-Kyriazou, 2023;Zulkifli et al., 2023).
ULS is a method that does not require special assumptions from the distribution of observed variables as long as the parameters are identified (Foldnes & Grønneberg, 2022;Hikmah et al., 2023).The ULS method has consistent and unbiased estimator properties and the process is relatively fast due to the simplicity of the method, but the ULS estimator is not an efficient estimator for large data (Chen et al., 2023;Dhaene & Rosseel, 2023).Therefore, this study will discuss the use of theoretical concepts or constructs that cannot be observed directly.In previous research by Mahmoodi et al. (2022), the Structural Equation Modeling method was tested for the analysis of factors affecting the quality of library services with the LISREL 8.80 program.In this study, the fit index test will be carried out with the ULS method using the survey data from the University of Lampung Library Service Quality questionnaire.

METHOD
This study uses primary data with the research population, namely visitors to the University of Lampung library class of 2015 and 2016.From this population, a sample of 100 library visitor respondents will be taken with a purposive sampling technique.The technique is done by taking samples based on certain percentages and considerations.Data collection using a Likert scale questionnaire with 5 categories used, namely strongly disagree (STS), disagree (TS), moderately agree (CS), agree (S) and strongly agree (SS) ranging from 1 to 5. The variables used are 3 latent variables and 12 indicator variables.
The steps in this research are (Robitzsch, 2022)

Measurement Model Specification
The measurement model is a model that describes the relationship between latent variables and observed variables.In this study, 12 observed variables were used which consists of 7 observed variables X and 5 observed variables Y is the path diagram: The model in matrix form becomes:

Structural Model Specifications
The structural model is a model that describes the relationship between latent variables.In this study, 3 latent variables were used, consisting of 2 exogenous latent variables and 1 endogenous latent variable.The following is a diagram the path.

Figure 2. Structural Model Path Diagram
Since is a scalar, that form is equal to its transpose.
(7) To obtain the value of the estimator so that the sum of the residual squares is as small as possible as possible, then we differentiate h'f against  ̂ and k'f against  ̂, then we get the following equation: and by taking and then an equation can be obtaineds follows: and Based on the parameter estimation results above, the parameter estimation in the structural equation model cannot be done by manual calculation because the structural equation model is analyzed simultaneously so that the population covariance (∑(θ)) is close to or equal to the sample covariance (S).In addition, the structural equation model also involves many equations and involves many parameters so it will be difficult to obtain the parameter estimation value.Therefore, the structural equation model analysis was carried out with the help of LISREL software.Based on Figure 3, the estimated value for all parameters at size 100, which  11 is 0.68.This means that the value of the relationship between employee competence and service quality is 0.68, while for  21 is 0.33, which means that the value of the relationship between library facilities and service quality is 0.33.
From the parameter estimation results obtained, the equation can be arranged for the structural model and measurement model.1, it can be seen that that the chi-square value above is 67,35 and has a p-value of 0,062 (p-value>0,05) which means the fit is good.For the RMSEA value value shows a good fit as well with a result of 0,057.Likewise, the NFI and PNFI results show a good fit model with each result obtained of 0,99 and 0,05.with each result obtained of 0,99 and 0,77.So that it can be concluded that the overall fit of the model is good.

CONCLUSION
The results of the three fit tests, namely the overall model fit test, the structural model fit test and the measurement model fit test with the Unweighted Least Square estimation method at a sample size of 100 have shown good results to explain the fit between the model and the data.Service quality variables at Lampung University Library ( 1 ) can be explained by employee competence variables ( 1 ) and library facilities ( 2 ) by 85%, while the remaining 15% is explained by other factors outside the model.

REFERENCE
: 1. Conducting a survey of visitors to the Universitas Lampung library 2. Inputting questionnaire data with a sample size of 100 3. Testing the validity and reliability of the questionnaire data using SPSS 23 4. Designing Model Specifications: Structural Model and Measurement Model 5. Path diagram construction Constructing a path diagram is to build a relationship between latent variables, namely  1 ,  2 ,  1 6.Model Estimation Parameter estimation using the Unweighted Least Square method by minimize half the sum of the squares of the elements of the leftover matrix.The estimates were obtained with the help of the Lisrel 8.80 program.7. Evaluating the fit test In general, the test of the overall fit of the model can be seen from the results of Goodness Of Fit (GOF) results, such as Chi Square, RMSEA, AGFI and PGFI (Dash & Paul, 2021; Hidayat & Wulandari, 2022).
Chi Square and RMSEA.While in the incremental fit test incremental fit test is the NFI and in the parsimony fit test is the PNFI.The following are the results of testing the overall suitability of the model can be seen in Table1.
-Structural Model -Measurement Model 5. Overall Model Fit Test The overall model fit test is used to evaluate the general degree of fit by looking at the goodness of fit results.Overall There are several criteria for testing a model, namely the Absolute Fit Test, Incremental Fit Test and Parsimony Fit Test.Absolute Fit Test, Incremental Fit Test and Parsimony Fit Test.In the absolute fit test The absolute fit test consists of