A Critical Examination of Common Beliefs About PLSPM
 Hengkov
 PLS SuperExpert
 Posts: 1618
 Joined: Sun Apr 24, 2011 10:13 am
 Real name and title: Hengky Latan
 Location: AMQ, Indonesia
 Contact:
A Critical Examination of Common Beliefs About PLSPM
Hello all,
I already read six critical (myth) thinking by Ronkko and Evermann (RE, hereafter) in they article from Organizational Research Methods 2013. This post just for discussion about this issue and for clear all issue. I will begin with show some myth statement from RE compare with Wold original paper (May be some expert PLS will comment this issue future):
a) In Myth 6: PLS Is Most Appropriate for Exploratory or Early Stage Research
RE conclude PLS is not an appropriate choice for earlystage theory development and testing (p. 18). This statement contrast by Noonan and Wold research (see Noonan and Wold 1983). Noonan and Wold conclude PLS is not only a tool for systems analysis; it is also general tool for scientific development. Three ares can be mention: (1) exploratory factor analysis; (2) instrument development; and (3) theory development (p. 328). see also Wold (1975b, p. 350351).
b) In Myth 5: PLS Has Minimal Requirements on Sample Size
I think this myth clear and consistent by Wold state in original paper (1982 p. 25; 1985b p.231), PLS based on consitency at large with many indicators and sufficient sample size. Current article PLS recommend 10x path coefficient/predictors in model for require sample size (see Barclay et al. 1995, p. 292; Hair et al. 2011, p. 144).
c) Myth 4: PLS Can Be Used for Testing Null Hypotheses About Path Coefficients
RE conclude because NHST (null hypothesis significance testing) requires a test statistic with a known sampling distribution, the PLS path estimates cannot be used in NHST. This statement contrast by Noonan and Wold research (see Noonan and Wold 1983). Noonan and Wold state if the standard errors of the regression coefficients for the inner relations are computed from the diagonal elements of the inverse matrices of the predictor variables, the ttest or Ftest can be applied. Again, Wold (1975b, p. 353) also suggest for used confidence interval for hypothesis testing.
d) Myth 3: PLS Can Be Used to Validate Measurement Models
Because of these better alternatives, the measurement model should never be evaluated based on the composite loadings produced by PLS or any statistic derived from these. RE conclude that the idea that PLS results can be used to validate a measurement model is a myth (p. 14). For validate measurement models (Mode A), PLS used PCA and for Mode B PLS used canonical correlation (see Wold 1975, p.122; 1985a, p. 587). Principal Component versus Factor Analysis is typical for the different at issue between PLS and ML approach.
e) Myth 2: PLS Reduces the Effect of Measurement Error
For this issue I not have comment. I need testing again before response.
f) Myth 1: PLS Has Advantages Over Traditional Methods Because It Is an SEM Estimator
I think this myth clear explanation by Marcoulides et al. (2012, p. 721722).
Please reply another comment if anyone interest this issue.
Regards,
References:
Barclay, D., Higgins, C., and Thompson, R. 1995. “The Partial Least Squares (PLS) Approach to Causal Modeling: Personal Computer Adoption and Use as an Illustration,” Technology Studies (2:2), pp. 285309.
Marcoulides, G. A., Chin, W. W., and Saunders, C. 2012. "When imprecise statistical statements become problematic: A response to Goodhue, Lewis and Thompson," MIS Quarterly (36:3), pp. 717728.
Noonan, R., and Wold, H. 1983. “Evaluating School Systems Using Partial Least Squares,” Evaluation in Education, Oxford: Pergamon (7), pp. 219364.
Wold, H 1975a. “Soft Modeling by Latent Variables: The NIPALS approach,” In Gani, J. (Ed.), Perspectives in probability and statistics: Papers in honour of M.S. Bartlett on the occasion of his sixty fifth birthday pp. 117142.
Wold, H. 1975b. “Path Models with Latent Variables: The NIPALS Approach,” In Blalock et al. (Eds.) Quantitative Sociology. New York: Academic Press.
Wold, H. 1982. “Soft modeling: the basic design and some extensions,” In: Jöreskog K. G,. and Wold, H (Eds) Systems under indirect observation. Causality, structure, prediction, Vol II. NorthHolland, Amsterdam, pp 154.
Wold, H. 1985a. “Partial least squares,” In Kotz, S., and Johnson, N.L (Eds.), Encyclopedia of statistical sciences (Vol. 8, pp. 587599). New York: Wiley.
Wold, H. 1985b. “System Analysis by Partial Least Squares,” In Nijkamp, P., Leitner, G., and Wrigley, N (Eds.), Measuring the Unmeasurable. Martinus Nijhoff Publishers. Dordrecht, Boston, Lancaster, pp. 221251.
I already read six critical (myth) thinking by Ronkko and Evermann (RE, hereafter) in they article from Organizational Research Methods 2013. This post just for discussion about this issue and for clear all issue. I will begin with show some myth statement from RE compare with Wold original paper (May be some expert PLS will comment this issue future):
a) In Myth 6: PLS Is Most Appropriate for Exploratory or Early Stage Research
RE conclude PLS is not an appropriate choice for earlystage theory development and testing (p. 18). This statement contrast by Noonan and Wold research (see Noonan and Wold 1983). Noonan and Wold conclude PLS is not only a tool for systems analysis; it is also general tool for scientific development. Three ares can be mention: (1) exploratory factor analysis; (2) instrument development; and (3) theory development (p. 328). see also Wold (1975b, p. 350351).
b) In Myth 5: PLS Has Minimal Requirements on Sample Size
I think this myth clear and consistent by Wold state in original paper (1982 p. 25; 1985b p.231), PLS based on consitency at large with many indicators and sufficient sample size. Current article PLS recommend 10x path coefficient/predictors in model for require sample size (see Barclay et al. 1995, p. 292; Hair et al. 2011, p. 144).
c) Myth 4: PLS Can Be Used for Testing Null Hypotheses About Path Coefficients
RE conclude because NHST (null hypothesis significance testing) requires a test statistic with a known sampling distribution, the PLS path estimates cannot be used in NHST. This statement contrast by Noonan and Wold research (see Noonan and Wold 1983). Noonan and Wold state if the standard errors of the regression coefficients for the inner relations are computed from the diagonal elements of the inverse matrices of the predictor variables, the ttest or Ftest can be applied. Again, Wold (1975b, p. 353) also suggest for used confidence interval for hypothesis testing.
d) Myth 3: PLS Can Be Used to Validate Measurement Models
Because of these better alternatives, the measurement model should never be evaluated based on the composite loadings produced by PLS or any statistic derived from these. RE conclude that the idea that PLS results can be used to validate a measurement model is a myth (p. 14). For validate measurement models (Mode A), PLS used PCA and for Mode B PLS used canonical correlation (see Wold 1975, p.122; 1985a, p. 587). Principal Component versus Factor Analysis is typical for the different at issue between PLS and ML approach.
e) Myth 2: PLS Reduces the Effect of Measurement Error
For this issue I not have comment. I need testing again before response.
f) Myth 1: PLS Has Advantages Over Traditional Methods Because It Is an SEM Estimator
I think this myth clear explanation by Marcoulides et al. (2012, p. 721722).
Please reply another comment if anyone interest this issue.
Regards,
References:
Barclay, D., Higgins, C., and Thompson, R. 1995. “The Partial Least Squares (PLS) Approach to Causal Modeling: Personal Computer Adoption and Use as an Illustration,” Technology Studies (2:2), pp. 285309.
Marcoulides, G. A., Chin, W. W., and Saunders, C. 2012. "When imprecise statistical statements become problematic: A response to Goodhue, Lewis and Thompson," MIS Quarterly (36:3), pp. 717728.
Noonan, R., and Wold, H. 1983. “Evaluating School Systems Using Partial Least Squares,” Evaluation in Education, Oxford: Pergamon (7), pp. 219364.
Wold, H 1975a. “Soft Modeling by Latent Variables: The NIPALS approach,” In Gani, J. (Ed.), Perspectives in probability and statistics: Papers in honour of M.S. Bartlett on the occasion of his sixty fifth birthday pp. 117142.
Wold, H. 1975b. “Path Models with Latent Variables: The NIPALS Approach,” In Blalock et al. (Eds.) Quantitative Sociology. New York: Academic Press.
Wold, H. 1982. “Soft modeling: the basic design and some extensions,” In: Jöreskog K. G,. and Wold, H (Eds) Systems under indirect observation. Causality, structure, prediction, Vol II. NorthHolland, Amsterdam, pp 154.
Wold, H. 1985a. “Partial least squares,” In Kotz, S., and Johnson, N.L (Eds.), Encyclopedia of statistical sciences (Vol. 8, pp. 587599). New York: Wiley.
Wold, H. 1985b. “System Analysis by Partial Least Squares,” In Nijkamp, P., Leitner, G., and Wrigley, N (Eds.), Measuring the Unmeasurable. Martinus Nijhoff Publishers. Dordrecht, Boston, Lancaster, pp. 221251.

 PLS User
 Posts: 13
 Joined: Thu Apr 04, 2013 12:42 pm
 Real name and title:

 PLS Expert User
 Posts: 248
 Joined: Sat Jul 25, 2009 1:34 pm
 Real name and title:
Hi everybody,
As an answer I would recommend to read the following two articles:
Reinartz, W./Haenlein, M./Henseler, J. (2009): An empirical comparison of the efficacy of covariancebased and variancebased SEM, in:
International Journal of Research in Marketing, Vol. 26, No. 4, S. 332344.
Henseler, J./Chin, W. (2010): A Comparison of Approaches for the Analysis of Interaction Effects Between Latent Variables Using Partial Least Squares Path Modeling, in: Structural Equation Modeling: A Multidisciplinary Journal, Vol. 17, No. 1, S. 82109.
These articles seem for me much better simulated as the mentioned article above. At the end the results of PLSSEM and CBSEM are not so different after all especially when a model was built thoroughly. My personal belief is that PLSSEM offers much more freedom of acting in model building. Therefore a researcher needs much more knowledge about the statistic characteristics of PLSSEM. This as a general comment…
All the best,
Christian
As an answer I would recommend to read the following two articles:
Reinartz, W./Haenlein, M./Henseler, J. (2009): An empirical comparison of the efficacy of covariancebased and variancebased SEM, in:
International Journal of Research in Marketing, Vol. 26, No. 4, S. 332344.
Henseler, J./Chin, W. (2010): A Comparison of Approaches for the Analysis of Interaction Effects Between Latent Variables Using Partial Least Squares Path Modeling, in: Structural Equation Modeling: A Multidisciplinary Journal, Vol. 17, No. 1, S. 82109.
These articles seem for me much better simulated as the mentioned article above. At the end the results of PLSSEM and CBSEM are not so different after all especially when a model was built thoroughly. My personal belief is that PLSSEM offers much more freedom of acting in model building. Therefore a researcher needs much more knowledge about the statistic characteristics of PLSSEM. This as a general comment…
All the best,
Christian

 PLS User
 Posts: 13
 Joined: Thu Apr 04, 2013 12:42 pm
 Real name and title:
Regarding Myth 5: Sample Size:
I think a possible answer can be found in Chin, 2010, p.661ff:
PLS only performs a series of simple OLS regressions (i.e. correlations), thus, to determine the sample size you only have to consider the endogenous LV with the largest number of predictors, and not all LVs in the whole model.
anyone agree / disagree?
best
frederic
Chin, W. W. (2010). How to Write Up and Report PLS Analyses. In Handbook of Partial Least Squares: Concepts, Methods and Applications . V. Esposito Vinzi, V. E., Chin, W.W. Henseler, J., & Wang, H. (Eds.) Heidelberg, Dordrecht, London, New York: Springer, 655690.
I think a possible answer can be found in Chin, 2010, p.661ff:
PLS only performs a series of simple OLS regressions (i.e. correlations), thus, to determine the sample size you only have to consider the endogenous LV with the largest number of predictors, and not all LVs in the whole model.
anyone agree / disagree?
best
frederic
Chin, W. W. (2010). How to Write Up and Report PLS Analyses. In Handbook of Partial Least Squares: Concepts, Methods and Applications . V. Esposito Vinzi, V. E., Chin, W.W. Henseler, J., & Wang, H. (Eds.) Heidelberg, Dordrecht, London, New York: Springer, 655690.
 Hengkov
 PLS SuperExpert
 Posts: 1618
 Joined: Sun Apr 24, 2011 10:13 am
 Real name and title: Hengky Latan
 Location: AMQ, Indonesia
 Contact:
Hello Frederic and Nitzl,
Good comment.
Btw, I find another paper please check
http://ssrn.com/abstract=2195970 or http://dx.doi.org/10.2139/ssrn.2196120
Regrads,
Good comment.
Btw, I find another paper please check
http://ssrn.com/abstract=2195970 or http://dx.doi.org/10.2139/ssrn.2196120
Regrads,
Thanks Hengkov for this critical post :)
can you please check the second link, it is not working
http://dx.doi.org/10.2139/ssrn.2196120
can you please check the second link, it is not working
http://dx.doi.org/10.2139/ssrn.2196120
 Hengkov
 PLS SuperExpert
 Posts: 1618
 Joined: Sun Apr 24, 2011 10:13 am
 Real name and title: Hengky Latan
 Location: AMQ, Indonesia
 Contact:
Hi,
I find one book similar this topic about myths in statistical methodology.
In some chapter this book include some topic related with PLS such mediation, moderating, effect size and sample size. May be this book will make some researcher to next investigation. Please check
Editor = Lance, Charles and Vandenberg, Robert.
Title = Statistical and Methodological Myths and Urban Legends: Doctrine, Verity and Fable in the Organizational and Social Sciences
Year = 2009
Publisher = Taylor & Francis, New York.
Best Regards,
I find one book similar this topic about myths in statistical methodology.
In some chapter this book include some topic related with PLS such mediation, moderating, effect size and sample size. May be this book will make some researcher to next investigation. Please check
Editor = Lance, Charles and Vandenberg, Robert.
Title = Statistical and Methodological Myths and Urban Legends: Doctrine, Verity and Fable in the Organizational and Social Sciences
Year = 2009
Publisher = Taylor & Francis, New York.
Best Regards,
 Hengkov
 PLS SuperExpert
 Posts: 1618
 Joined: Sun Apr 24, 2011 10:13 am
 Real name and title: Hengky Latan
 Location: AMQ, Indonesia
 Contact:
Hi,
For clear Myth 4, I recommend to read one good book by Rex Kline (2004). Beyond significance testing: Reforming data analysis methods in behavioral research.
In chapter 1, 2 & 3 Kline (2004) very clear explain all about NHST (controversy, false, missunderstading etc).
Kline (2004, p. 86) suggest:
1. NHST suitable in exploratory research
2. If statistical test are used, information about power must be reported
3. Drop the word "significant" from our data analysis vocabulary
4. Used confidence interval
5. Replication is the best way to deal with sampling error
6. Compute effect size
Regards,
For clear Myth 4, I recommend to read one good book by Rex Kline (2004). Beyond significance testing: Reforming data analysis methods in behavioral research.
In chapter 1, 2 & 3 Kline (2004) very clear explain all about NHST (controversy, false, missunderstading etc).
Kline (2004, p. 86) suggest:
1. NHST suitable in exploratory research
2. If statistical test are used, information about power must be reported
3. Drop the word "significant" from our data analysis vocabulary
4. Used confidence interval
5. Replication is the best way to deal with sampling error
6. Compute effect size
Regards,
 Hengkov
 PLS SuperExpert
 Posts: 1618
 Joined: Sun Apr 24, 2011 10:13 am
 Real name and title: Hengky Latan
 Location: AMQ, Indonesia
 Contact:
Henseler et al. (2014) already answer R&E critical myths. I think not different with my comments past:
1. PLS is an SEM method that is designed for estimating composite factor models. Composite factor models
estimated by PLS provide unbiased implied covariances. Path coefficients appear biased if interpreted as
relationships between common factors. Model tests and fit measures can be applied.
2. PLS construct scores will be more reliable than sum scores if the indicators vary in quality (reliability or
impact) and if PLS has sufficient information (with regard to model complexity, strength of interconstruct
relationships, and to a lesser extent sample size) to estimate different weights.
3. PLS can help detect a wide spectrum of model misspecifications as long as not only the common factor model
but also the composite factor model is wrong. Analysts should particularly look at the exact fit (pvalue) as well
as approximate fit measures such as SRMR and RMStheta.
4. PLS in combination with bootstrapping can be used for null hypothesis significance testing (NHST) of path
coefficients. If PLS does not have sufficient information to estimate different weights (see Question 2) or if the
empirical bootstrap distribution is bimodal analysts should rely on BCa bootstrap confidence intervals.
5. PLS can still be applied when other methods do not converge or provide inadmissible solutions. PLS estimates
complex models in which the number of variables or parameters exceeds the number of observations.
6. PLS can be a valuable tool for exploratory research because it estimates a less restricted model (the composite
factor model), because it reliably provides estimates even in situations in which other methods fail, and because
as a limitedinformation approach it is less prone to consequences of misspecification in subparts of the model.
References:
Henseler, Jörg; Dijkstra, Theo K.; Sarstedt, Marko; Ringle, Christian M.; Diamantopoulos, Adamantios; Straub, Detmar W.; Ketchen, David J., Jr.; Hair, Joseph F.; Hult, G. Tomas M.; Calantone, Roger J. (2014), Common beliefs and reality about PLS: Comments on Rönkkö & Evermann (2013), Organizational Research Methods, 17 (2), in print.
1. PLS is an SEM method that is designed for estimating composite factor models. Composite factor models
estimated by PLS provide unbiased implied covariances. Path coefficients appear biased if interpreted as
relationships between common factors. Model tests and fit measures can be applied.
2. PLS construct scores will be more reliable than sum scores if the indicators vary in quality (reliability or
impact) and if PLS has sufficient information (with regard to model complexity, strength of interconstruct
relationships, and to a lesser extent sample size) to estimate different weights.
3. PLS can help detect a wide spectrum of model misspecifications as long as not only the common factor model
but also the composite factor model is wrong. Analysts should particularly look at the exact fit (pvalue) as well
as approximate fit measures such as SRMR and RMStheta.
4. PLS in combination with bootstrapping can be used for null hypothesis significance testing (NHST) of path
coefficients. If PLS does not have sufficient information to estimate different weights (see Question 2) or if the
empirical bootstrap distribution is bimodal analysts should rely on BCa bootstrap confidence intervals.
5. PLS can still be applied when other methods do not converge or provide inadmissible solutions. PLS estimates
complex models in which the number of variables or parameters exceeds the number of observations.
6. PLS can be a valuable tool for exploratory research because it estimates a less restricted model (the composite
factor model), because it reliably provides estimates even in situations in which other methods fail, and because
as a limitedinformation approach it is less prone to consequences of misspecification in subparts of the model.
References:
Henseler, Jörg; Dijkstra, Theo K.; Sarstedt, Marko; Ringle, Christian M.; Diamantopoulos, Adamantios; Straub, Detmar W.; Ketchen, David J., Jr.; Hair, Joseph F.; Hult, G. Tomas M.; Calantone, Roger J. (2014), Common beliefs and reality about PLS: Comments on Rönkkö & Evermann (2013), Organizational Research Methods, 17 (2), in print.