PLSSEM Guideline and Compliance Summary
A study of Diffusion of Legal Software Use in a Global Campus: Action Research at a Global Campus in China.
by Jeonghwan (Jerry) Choi, Aug. 2017
Partial Least Square Structural Equation Modeling (PLSSEM)
PLSSEM method gets a special attention from business and social science researchers for its robustness in psychometric model analysis. This critical advantage makes the PLSSEM as an alternative technique for SEM and it has become a key research method in recent years (J.F. Hair & Hult, 2016). For example, PLSSEM gives a more robust structural equation model convergence over CBSEM in many situations especially when a research model has many indicators, paths, and relationships among key variables and constructs (Chin, 2010; J.F. Hair & Hult, 2016; Henseler et al., 2014). For this benefit of PLSSEM, it gets popular in management disciplines such as marketing management, strategic management, human resources management, and particularly management information system field of studies (J.F. Hair & Hult, 2016; J.F. Hair, Ringle, & Sarstedt, 2011; Williams et al., 2015).
Considering the rules of thumb for selecting CBSEM or PLSSEM (J.F. Hair et al., 2011), we judged that our study was appropriate for applying the PLSSEM technique because this study’s main goal was to explore new knowledge and to extend an existing structural theory of the complex UTAUT. Prior to conducting the PLSSEM, we summarized prerequisite or necessary conditions and recommendations through synthesizing a few foundational literature (Chin, 2010; J.F. Hair & Hult, 2016; Latan & Ramli, 2013; Richter, Sinkovics, Ringle, & Schlaegel, 2016). By using our data, we validated these seven categories of PLSSEM guidelines:
· Data and sampling characteristics
· PLSSEM algorithm
· Outer model evaluation: reflective (mode A)
· Out model evaluation: formative (mode B)
· Inner model evaluation: recursive model
· Model fit
· Multigroup analysis
Results of validation of PLSSEM guideline and compliance are described in the following Table.
Overall, our data and research model showed good compliances to the proposed PLSSEM guidelines. It should be noted that we use the Smart PLS [ SmartPLS (v. 3.2.6). Ringle, Wende, and Becker, 2015. "SmartPLS 3." Boenningstedt: SmartPLS GmbH, http://www.smartpls.com] as the PLSSEM analysis tool.
Table
PLSSEM guideline and compliance
Characteristics  Guideline  Recommendation  Compliance 
Data and sampling characteristics  
Sample size  Ten times rule: the minimum sample size should be equal to the larger of 10x largest number of 1) formative indicator and structural paths directed at a particular latent construct (Wong, 2013)
Alternatively, Cohen’s sample size recommendation of statistical power and effect sizes takes into accounts. Cohen’s sample size recommendation (In this study, 5% significance level, 0.10 minimum R2, and maximum number of arrows pointing at a construct (5) = 205  · 10x formative indicator (N/A) · 10x structural paths directed a a latent construct (10 x 5 = 50)
· Cohen’s sample size recommendation = 205  Samples size n = 215 > 205 
Holdout  30% of original sample (Hair Jr & Hult, 2016)  > 30%  215/218 = 98.6% 
Missing data  Less than 5% of missing or screening out data  < 5%  3 / 218 data were screened out = 1.38% 
Distribution  Robust when applied to highly skewed data, but skewness and − kurtosis should be reported (Richter, Sinkovics, Ringle, & Schlaegel, 2016) 


PLSSEM Algorithm  
Weighting scheme  In general, the path weighting scheme is strongly recommended because it provides the highest R^{2} value for endogenous latent variable (Vinzi, Chin, Henseler, & Wang, 2010)  path weighting  path weighting 
Data metric  The standardized value setting  Mean 0, Var 1  Mean 0, Var 1 
Total maximum iteration  The standard maximum iteration is 300  300  300 
Abort criterion  The recommended number is 1.0E5  1.0E5  1.0E7 
Starting value  Initial outer weight can be set as 1.0  1.0  1.0 
Algorithm to handle missing data  Missing value treatment options are mean replacement, EM (expectationmaximization algorithm), and nearest neighbor (Hair, Ringle, & Sarstedt, 2013)  Mean replacement  Mean replacement 
Bootstrap subsample size  The number of bootstrap samples should be high but must be at least equal to the number of valid observations. As a rule, 5,000 bootstrap samples is recommended (Hair Jr & Hult, 2016, p. 132)  2000  4000  5000 
Bootstrap sign change  No sigh change option is recommendable because it results in the most conservative outcome (Hair Jr & Hult, 2016, p. 135)  No sign change  No sign change 
Significance level  Generally, 5% significance level is widely used in social science field of study  5%, Twotailed  5%, Twotailed 
Predictive relevance  In the process of blindfolding, omission distance (D) can be set between 5 and 10 (Hair Jr & Hult, 2016, p. 179)  5 £ D £ 10  D = 7 
Software feature  SmartPLS (v. 3.2.6). Ringle, C. M., Wende, S., and Becker, J.M. 2015. "SmartPLS 3." Boenningstedt: SmartPLS GmbH, http://www.smartpls.com.     
Outer model evaluation: Reflective (mode A)  
Indicator reliability  Recommended > 0.6 for exploratory research and > 0.7 for confirmatory research (Chin, 2010)  > 0.7  All indicators (factor loadings) are higher than 0.7 [0.737 ~ 0.939] 
Internal consistency reliability  The cutoff value for composite reliability is > 0.6 for exploratory research and > 0.7 for confirmatory research. The Cronbach’s alpha is not suggested for distinguishing  > 0.7  All composite reliabilities are higher than 0.7 [0.904 ~ 0.952] 
Convergent validity  The Average Variance Extracted (AVE) is > 0.5  AVE > 0.5  All AVE is higher than 0.5 [0.662 ~ 0.869] 
Discriminant validity  Fornell and Larcker (1981) criterion: Each construct’s AVE should be higher than its squared correlation with any other construct (Fornell & Larcker, 1981)  Square root AVE > Correlation  All square root AVE is larger than any other correlations with other constructs 
 Crossloading: Each indicator should load highest on the construct it is intended to measure (Chin, 2010)  Highest loading on the construct  Each indicator loaded highest on the intended construct 
 HeterotraitMonotrait Ratio (HTMT) should under 0.85 for each outer model indicators: (Henseler, Hubona, & Ray, 2016)  HTMT ratio < 0.85  All HTMT ratio is under 0.85 [0.358 ~ 0.824] except ‘Effort expectation and Facilitation’ [0.852] 
Item removal  If some item has been dropped to achieve a model fit, give additional information    No removed item 
Outer model evaluation: Formative  
 According to Confirmatory Tetrad Analysis (CTA), all constructs in the model are not formative constructs in 1% level of significance (Hair Jr & Hult, 2016, pp. 4647) 


Collinearity  The cutoff value for VIF should be smaller than 0.5. A stabilized estimation is suggested as ranging 2.5 ~ 3.3 (Hair et al., 2013)  VIF < 0.5  Most of outer VIF values are under 0.5 [1.000 ~ 4.014]. However, the VIF values of UP1 (6.070) and UP2 (6.425) are higher than the criterion. 
Construct removal  If a construct has been dropped due to collinearity, the problem should be reported. 
 No removed construct 
Inner model evaluation: Recursive model  
Path estimates  Reporting 1) Path coefficient 2) significance and confidence interval from bootstrapping  Bootstrapping is applied for the significant of the path coefficient with twotails of 5% = 1.96  Bootstrapping is applied for the significant of the path coefficient with twotails of 5% = 1.96 
R^{2} (Adjusted R^{2})  R^{2} acceptable level is contextdependent. (Hair Jr & Hult, 2016; Latan & Ramli, 2013).  0.25: Weak 0.50: Moderate 0.75: Strong  R^{2 }(Adj R^{2}) of Behavioral intention: 0.584 (0.578)
R^{2 }of Use Behavior 0.180 (0.172) 
Effect size f^{2}  Cohen’s statistical power analysis of effect size (Cohen, 1992)  0.02: Weak 0.15: Moderate 0.35: Strong  PE: 0.169 EE: 0.053 SI: 0.044 BI: 0.040 FA: 0.016 BI: 0.040 
Predictive relevance  The cross validated redundancy as a measure of Q^{2} is recommended because it includes the key element of the path model, the structural model, to predict eliminated data points (Chin, 2010; Hair Jr & Hult, 2016, pp. 183184).  Q^{2} > 0  Behavioral intention Q^{2} = 0.471 Use behavior Q^{2} = 0.151 
Model fit  
Standardized Root Mean Square Residual  SRMR  < .08  .059 
Squared Euclidean Distance  d_ULS  < .95  .803 
Geodesic Distance  d_G  < .95  .777 
Incremental fit measure  NFI  > 0.9  0.821 
rms Theta 

 0.177 




MultiGroup Analysis  
Bootstrapping for MGA  The number of bootstrap samples for Multi Group Analysis should be high but must be at least equal to the number of valid observations. As a rule, 5,000 bootstrap samples is recommended (Hair Jr & Hult, 2016, p. 132)  2000  4000  5000 
Path coefficient difference between groups  PLSMGA (Henseler’s MGA) with 5% of significance level.  tvalue > 1.96 pvalue < .05 or > .95  The path between Facilitation > Use behavior has a marginal difference (p = .055 




Note: It is important to note that these model fit assessment criteria often not be useful for PLS SEM and must be used with caution. These criteria are in their very early stage of research and not fully understood. However, these fit statistics give researchers to estimate the quality of the model when it is a reflective model (Hair Jr & Hult, 2016). In more detail, please see this Note of Caution (https://www.smartpls.com/documentation/functionalities/modelfit)
References
Chin, W. W. (2010). How to write up and report PLS analyses. Handbook of partial least squares, 655690.
Cohen, J. (1992). A power primer. Psychological bulletin, 112(1), 155.
Fornell, C., & Larcker, D. F. (1981). Evaluating structural equation models with unobservable variables and measurement error. Journal of Marketing Research, 3950.
Hair, J. F., & Hult, G. T. M. (2016). A primer on partial least squares structural equation modeling (PLSSEM): Sage Publications.
Hair, J. F., Ringle, C. M., & Sarstedt, M. (2011). PLSSEM: Indeed a silver bullet. Journal of Marketing theory and Practice, 19(2), 139152.
Hair, J. F., Ringle, C. M., & Sarstedt, M. (2013). Partial least squares structural equation modeling: Rigorous applications, better results and higher acceptance. Long Range Planning, 46, 112. doi:http://dx.doi.org/10.1016/j.lrp.2013.01.001
Henseler, J., Dijkstra, T. K., Sarstedt, M., Ringle, C. M., Diamantopoulos, A., Straub, D. W., . . . Calantone, R. J. (2014). Common beliefs and reality about PLS: Comments on Rönkkö and Evermann (2013). Organizational Research Methods, 17(2), 182209.
Henseler, J., Hubona, G., & Ray, P. A. (2016). Using PLS path modeling in new technology research: updated guidelines. Industrial Management & Data Systems, 116(1), 220.
Latan, H., & Ramli, N. A. (2013). The Results of Partial Least SquaresStructural Equation Modelling Analyses (PLSSEM). doi:10.2139/ssrn.2364191
Richter, N. F., Sinkovics, R. R., Ringle, C. M., & Schlaegel, C. (2016). A critical look at the use of SEM in international business research. International Marketing Review, 33(3), 376404.
Vinzi, V. E., Chin, W. W., Henseler, J., & Wang, H. (2010). Handbook of partial least squares: Concepts, methods and applications: Springer Science & Business Media.
Williams, M. D., Rana, N. P., & Dwivedi, Y. K. (2015). The Unified Theory of Acceptance and Use of Technology (UTAUT): A Literature Review. Journal of Enterprise Information Management, 28(3), 443488.
Wong, K. K.K. (2013). Partial least squares structural equation modeling (PLSSEM) techniques using SmartPLS. Marketing Bulletin, 24(1), 132.
'Choi's Research > 1.Progress in Research' 카테고리의 다른 글
[RESEARCH] KUSA/WKU AGILE MANAGEMENT RESEARCH GROUP (0)  2022.10.30 

Is the Psychological Safety Matter in the Manufacturing Industry? (0)  2018.07.21 
Good Questions for Case Study..!! (0)  2008.09.12 
Integral Innovation Leadership Research (0)  2008.07.25 
댓글0