Science topic

Partial Least Squares - Science topic

Explore the latest questions and answers in Partial Least Squares, and find Partial Least Squares experts.
Questions related to Partial Least Squares
  • asked a question related to Partial Least Squares
Question
1 answer
I have tried to estimate with PLS-SEM (with the R package 'seminr') a model that, in its measurement part (or outer part), includes a bifactor structure (i.e., two specific dimensions and a common general dimension). The general factor is the one that is "causally" related to other latent variables (and all constructs are reflective). However, R does not estimate factor loadings for the specific factors, but only for the general factor. Is it possible to estimate the full bifactor structure with PLS-SEM? Is there a way to do so in 'seminr'?
(I assume that I cannot add causal relationships between the specific factors because the bifactor models assume orthogonality.)
Relevant answer
Answer
Follower
  • asked a question related to Partial Least Squares
Question
1 answer
When introducing new constructs or variables into a model and examining their effects in accordance with specific established theory, especially in a context where the theory has not been tested before. Is this theory development or theory testing and confirmation? Given that I need to justify my use of PLS, which is generally recommended for models where the emphasis may be more on theory development.
Relevant answer
Answer
Hi Ahmed! PLS-SEM is a method that generally used for prediction, while CB-SEM is the method better suited for theory confirmation. PLS-SEM is also used when the constructs are defined as composites, not as common factors as in the case of CB-SEM. It is a preferred method for theory development because of its more flexible in terms of data distribution and sample size, as well as because it allows to incorporate both reflective and formative constructs. Theory development studies work with theories that need to be validated in more contexts as well as further conceptual development, such as the incorporation of new variables. Depending on the level of the development of the theory, the purpose of the research in this case could be exploratory. In the example you give, as you introduce new variables it seems to be about theory development.
  • asked a question related to Partial Least Squares
Question
2 answers
How can I decide between Partial Least Squares (PLS-SEM) and Covariance-Based SEM (CB-SEM) for testing a conceptual model in management studies?
Relevant answer
Answer
Comparison of PLS-SEM and CB-SEM
PLS more useful for exploratory research and CB more suited for Confirmatory research
Focus
  • PLS-SEM: Prediction-oriented
  • CB-SEM: Theory testing
Model Type
  • PLS-SEM: Complex, formative, and reflective constructs
  • CB-SEM: Simple, primarily reflective constructs
Data Distribution
  • PLS-SEM: No normality required
  • CB-SEM: Normality required
Sample Size
  • PLS-SEM: Smaller
  • CB-SEM: Larger
Fit Evaluation
  • PLS-SEM: Variance explained (e.g., R², Q²)
  • CB-SEM: Goodness-of-fit indices (e.g., CFI, RMSEA)
send me you outline of research problem and design and Hope to give you specific suggestion. You can write at, cityju at rate of rediffmail dot com
  • asked a question related to Partial Least Squares
Question
1 answer
I have data from population based observation (not questionnaires but yearly observation from secondary database) and I already have a common model for each populations (6 groups, each has the same latent variables, observed variables, and the structural models are also the same shape). As the study is some kind of longitudinal basis (not independent to each other), am I still be able too use MGA (Multi Group Analysis)? My result does not pass the MICOM procedure, is doing a MICOM procedure an obligatory prior MGA in my specific case?
Relevant answer
Answer
They should pass MICOM. Otherwise, you may only qualitatively compare the results of group differences (but not based on a test).
  • asked a question related to Partial Least Squares
Question
1 answer
Dear Community,
I would like a question regarding the use of Partial Least Square Regression Analysis. Basically ,I am confused in units. For example, I have Year 1, Year 2, Year 3 land cover and Water Balance Components. The units of Water Balance Components are "mm", while the units of each landcover type for year 1, 2 and 3 are in square Km. I am confused how the different units will perform the PLSR test.
Either, I have to use the % difference in each Year or % of particular landcover type to the total area of the basin and similarly convert the water balance variables from "mm" to percentage.
Looking for a guidance. Please teach me.
Regards
Relevant answer
Answer
In my opinion, Normalizing each component will solve problem!
  • asked a question related to Partial Least Squares
Question
2 answers
What is partial least squared structural equation modelling?
Relevant answer
The partial least squares path modeling or partial least squares structural equation modeling (PLS-PM, PLS-SEM) is a method for structural equation modeling that allows estimation of complex cause-effect relationships in path models with latent variables.
  • asked a question related to Partial Least Squares
Question
6 answers
I am currently working with a study titled "Knowledge, Competency, Adoptability and Sustainability of Artificial Intelligence (AI) Technology Among Physician Entrepreneurs in GCC Countries". I am in dilemma that which SEM model AMOS/SmartPLS will suit for the study. The total responses received so far is 220 out of the population size of 400. The main objective is to find out knowledge , competency, adoptability and sustainability of Artificial intelligence among Physician entrepreneurs.
I will be much thankful to you, if you can extend your kind feedback on this.
Regards
Dr.Sharfras Navas
Relevant answer
Answer
With the latest SmartPLS release, you can run both PLS-SEM and CB-SEM (like Amos): https://www.smartpls.com/downloads
  • asked a question related to Partial Least Squares
Question
1 answer
Hi all,
May I know if you have any idea how to address issue related to limitation of non probability (convenience sampling) sampling in using PLS-SEM (partial least squares structural equation modeling)?
Thanks in advance.
  • asked a question related to Partial Least Squares
Question
10 answers
I know that some literature said that the minimum estimate for the path coefficient should be around 0.2, but is there any discretion or other opinion regarding this matter? Thank you for the attention.
Relevant answer
Answer
Very useful to follow. Good inquiry and answers. Thank you, professors.
Kind Regards,
  • asked a question related to Partial Least Squares
Question
6 answers
Dear All,
I want to build a higher order construct, based on literature, that has 5 components.
One of them has really high HTMT of 0.93.
If I want to perform the the two-stage approach, according to the literature, I have to check the lower order constructs, HTMT implicitly.
Is it fine to delete a dimension?
Thank you so much!
Relevant answer
Answer
You may want to review this contribution to get a clear idea of exactly what HTMT is all about and how an adverse outcome can be improved:
Henseler, J., Ringle, C. M., and Sarstedt, M. (2015). A New Criterion for Assessing Discriminant Validity in Variance-based Structural Equation Modeling., Journal of the Academy of Marketing Science, 43(1): 115-135.
  • asked a question related to Partial Least Squares
Question
8 answers
Can anybody share opensource Partial Least Squares (PLS) tools that is easy to handle, reliable and easy to learn analysis and interpretation?
Relevant answer
Answer
Dear Manojit,
You can find many PLS packages in R environment (https://www.r-project.org/) which is free.
Best regards,
Ludovic
  • asked a question related to Partial Least Squares
Question
10 answers
I am working on urban social sustainability and completed pilot study on 109 respondents. I want to know from the respected reasearcher that what will be next procedure for propose a model?
Relevant answer
Answer
Sultana Razia , EFA and CFA are used for scale development and validation.
If your goal is to develop a model to illustrate the relationships among different variables, you can proceed with SEM (or other statistical analyses) directly.
  • asked a question related to Partial Least Squares
Question
3 answers
I am a new learner to process the analysis in SMART PLS-SEM. there are many options to assess model fit. Could you please help me to provide a complete report of Model Fit as well as references?
Goodness of Fit Formula
GoF = squre root (squre R *AVE)
best regards
Relevant answer
Answer
Hi Baraaah
SmartPLS offers the following fit measures:
SRMR
NFI
Chi²
RMS_theta
For the approximate fit indices such as SRMR and NFI, you may directly look at the outcomes of a PLS or PLSc model estimation (i.e., the results report) and these criteria's values with a certain threshold (e.g., SRMR < 0.08 and NFI > 0.90).
For the exact fit measures you may consider the inference statistics for an assessment. Therefore, you need to run the bootstrap procedure and to use the “complete bootstrap” option in SmartPLS.
Please check out the following links
  • asked a question related to Partial Least Squares
Question
5 answers
I have extended the UTAUT2 model and I have already developed my model hypotheses. I read about various methods/techniques used such as (Factor Analysis, Partial Least Square (PLS), Structure Equation Model (SEM), Regression Analysis) and tools such as (SPSS, Smart-PLS, Mplus, R, PLS-graph, AMOS). But I am not sure which one to use.
Please I would like some advice about recommended techniques and tools. I am really looking to find out what you consider to be the most efficient method and your rationale as cost and time are limited factors.
Many thanks
Relevant answer
Answer
smart pls is do better than spss based my experiences
  • asked a question related to Partial Least Squares
Question
3 answers
If we have 4 variables and 3 -4 statements for each variable. How many global items we are suppose to use? What is min: and max: limit for global item ? What will be the effect if we ignore global items in the moderator and mediating variable?
thanks in advance!
Relevant answer
Answer
Agreed
  • asked a question related to Partial Least Squares
Question
4 answers
Dear All,
I'd like to use Smart PLS3 to run a PLS-MGA model, in which there are three variables of two values for grouping. I've come up with two questions;
1. First, do I have to report the results of model fit measures for each grouping variable separately or are the results for the complete model the one to go with?
2. Second, what if none of the fit measures is not satisfactory? What could be the reasons? What could be done to tackle the issue? I'd really appreciate any helpful reference in this regard.
An overview of the variables is as follows:
Five exogenous variables, with two of them being single-item, two three-item variables and one four-item variable.
One endogenous variable with 6 items.
Three grouping variables of two values.
The sample size is 160.
Any assistance would be welcomed.
Best Regards,
Saeed
Relevant answer
Answer
I absolutely agree with Cristian, you do not need to validate these measures.
  • asked a question related to Partial Least Squares
Question
3 answers
Kind colleagues,
I am running a PLS-SEM on SmartPls 3.0. I specified the model as with two higher orderd formative components (HOCs). The lower order construcst instead have reflective causality (it's a reflective-formative HCM). To estimate it I tried with the repeated indicators approach and calculate the internal, convergent and discriminant validity statistics for HOCs by hand. Everything works fine except the overall model fit (which I know to be a very debated topic) that gives me SRMR above 0.11 and an infinite Chi-square (which I interpret as too many degrees of freedom and too tiny sample - 260). By using the disjoint two-stage approach instead, my model works perfectly. Not only the path coefficients are all significant, but also the overall model fit is amazing (SRMR below 0.06). My question is: does it make sense to consider these overall fit measure in the two stages approach? I mean, is it even comperable to a specification estimated with the repeated indicators approach? How should I choose the best model?
I need the help of the community!
Thanks everyone
Relevant answer
Answer
Giacomo Buzzao, I am not using SmartPls, but I think, in general, estimating higher-order constructs using the two-stage approach seems reasonable due to complexity involved. What the two-stage approach does in your case is that it creates component scores for the first-order constructs using the PLS-SEM algorithm and then uses these component scores as the manifest variables of the formative constructs which themselves are also component scores. Another way of saying this is that your first-order constructs are predicting your second-order construct. PLS-SEM itself is a two-stage algorithm. Following this reasoning then, any fit measures or alike provided by software based on PLS-SEM algorithm should also be used for evaluating higher-order component models.
  • asked a question related to Partial Least Squares
Question
9 answers
The whole data was collected on Likert scale. I am using Prof. Gaston's plspm package in R for SEM modeling. I think different age groups must have differences. But, I can't test differences of more than than two groups at structural level. My question is, if I divide the data into various age based subgroups and prepare SEM models separately for each subgroup. Is it meaningful? How to justify the models are significantly different? I was not able to perform ANOVA test to check the difference among models. What should I use? Please guide me Thanks
Relevant answer
Answer
Smart PLS has excellent capabilities for multigroup PLS-SEM analysis. You can for instance access to statistical testing for path differences between groups. This software is not open source but you can freely download and use it for a free 30 day trial (maybe it could help)
  • asked a question related to Partial Least Squares
Question
5 answers
I am using Camo Unscrambler X for building an partial least square (PLS) regression model. 
I am confused which parameter should I consider for choosing optimal number of factors. Is it based on the lowest standard error of calibration (lowest SEC) and highest R2  ? Or lowest root mean square error of calibration (lowest RMSEC) and highest R?  
Relevant answer
Answer
Lowest rmsc
And high r2
  • asked a question related to Partial Least Squares
Question
6 answers
I want to conduct partial least squares structural equation modeling for my research in SmartPls software
. So, is there anyone who provides me the step by step guide of conducting SmartPls software? It may be video or written material. Thanks for your response.
Relevant answer
Answer
This latest article, https://authors.elsevier.com/c/1bs7Iz1m7PEMF can help you with the following:
- Partial least squares structural equation modelling (pls-sem) techniques using SmartPLS, and a step by step interpretation of SmartPLS results
- How to validate and report higher-order models or hierarchical component models (HCMs) or higher-order constructs
- How/when CB-SEM (i.e., AMOS) can be (should be) used for measurement model validation in a PLS-SEM (i.e., SmartPLS) study
- How to report multigroup analysis using SmartPLS (PLS-MGA)
- Global fit analysis using SmartPLS in studies with data from two different countries
- How to use and report confirmatory tetrad analysis in PLS-SEM (CTA-PLS in SmartPLS)
- How to report mediation tests using SmartPLS
Good luck!
  • asked a question related to Partial Least Squares
Question
6 answers
'Partial least squares path modeling: Time for some serious second thoughts'.
I would like to invite experts of the PLS-SEM to comment on this paper or please indicate any paper that has already made comment with respect to this paper.
Relevant answer
Answer
Yes, there have been many evaluations on methodology by reseachers in this area- PLS-SEM: Indeed a Silver Bullet, Joe F. Hair, Christian M. Ringle, and Marko Sarstedt. Others: Bagozzi, Richard P., Garson, G.David,
  • asked a question related to Partial Least Squares
Question
3 answers
Hi,
so i have a reflective-formative type models (as seen on the picture attached) and use the repeated indicator approach, and what i did on assessing the model based on what i understand was:
1. calculate the PLS algorithm and see the cronbach alpha, composite reliability and AVE criteria.
2. Since few of my variables on the LOC has low value of AVE composite reliability and cronbachs alpha, i had to eliminate some of the indicators so that i can meets the criteria. Ive read on some journals that if you eliminate an indicator from the LOC, u also have to remove the one in the HOC and vice versa.
The problem is, after i eliminate some of the indiators all of my LOC has already meet the AVE minimum value, but somehow my HOC still has a very low AVE value (0.284, as seen on the second pic attached).
So my question is do i have to keep eliminating the indicators until my HOC's AVE value acceptable (but then id lost quite a lot of indicators on LOC too)? Or am i doing it wrong? Is there any specific guides on this cause im kinda lost right now
Thank you
Relevant answer
Answer
In addition to the one above, you can try this:
Becker, J.-M., Klein, K. and Wetzels, M. (2012), “Hierarchical latent variable models in PLS-SEM: guidelines for using reflective-formative type models”, Long Range Planning, Vol. 45 Nos 5/6, pp. 359-394.
  • asked a question related to Partial Least Squares
Question
6 answers
How to analyze the questionnaire's results based on UTAUT model ( Venkatesh et al., 2003) via Partial Least Square (PLS)? How to measure satisfaction?
To clarify, I tend to invite a sample of users to use a web application and analyze their satisfaction by asking them to participate in the survey based on UTAUT model questionnaire. However, I am uncertain about the evaluation procedure.
Relevant answer
Nice Dear Djamel Toudert
  • asked a question related to Partial Least Squares
Question
9 answers
Dear Colleagues:
I am focusing a ratio/quotient construct/variables which is the product of the ratio of two (latent) constructs/variables and is a dependent variable and a mediator as well.
For this purpose, I measured the two variables at the seven-points Likert scale and then calculated the mean scores of each of the two variables. Finally, I calculated the ratio/quotient variables dividing the mean values of two variables.
After this computation, I tried to run my research model in SmartPLS, however, the results show very poor regression.
I have following questions in this context:
1. Is it right way to calculate the quotient/ratio construct? I followed the methodology of the following paper for the ratio calculation.
2. Can SmartPLS/Partial least squares be applied for dealing with such a quotient/ratio variable? The variable is my case is not following the assumption of ordinal data points/equal data intervals.
3. What software/data technique could be best for analyzing such a ration variable?
I highly appreciate your kind feedback in advance, thanks so much,
Best regards,
Muhammad Shujahat
Relevant answer
Answer
But see my comment. mean benefit / mean cost is not a variable. I'm going to stop following this question.
  • asked a question related to Partial Least Squares
Question
4 answers
(The example in R uses vectors to illustrate Y, not matrix.)
I understand with the R exemple using the yarn database, that when using plsr :
regpls <- plsr(Y ~X, ncomp = 5, data = mydata)
typeof(Y) = double  and  typeof(X) = double
typeof(mydata) = list...
BUT, as Y is a matrix in my case, I don't manage to turn it into a type recognised as "double", so far R recognise it as a list of doubles.
Could anybody help me with practical example, I guess my problem is more about how to qualify my data than on the use of plsr, at the moment. I haven't find any proper example so far neither in book or on internet, I hope the answer is pretty simple, and of course I'll put it there if I find it before any help comes :)
Relevant answer
Answer
Does anyone know why the pls depot package provides VIP scores for each principal component rather than for the model overall?
  • asked a question related to Partial Least Squares
Question
5 answers
Poring over the literature of Fault Diagnosis, I found out several conventional approaches, including Principal Component Analysis (PCA), Fisher discriminant analysis (FDA), Partial Least Squares(PLS), and Canonical variate analysis (CVA). Having said that, I am not assured of lastest or improved categories.
Relevant answer
Answer
Rahul Kumar I don't go along with that idea, there is a broad range of papers on this regard. Intensive research looking at recent developments and classification, however, is missing.
BTW, I appreciate your response and the recommended book.
  • asked a question related to Partial Least Squares
Question
3 answers
I am running a PLS-PM to evaluate Satisfaction "S".
I have defined a latent variable LV "D" and I concern how to define the manifest variables(MV).
I have:
  • 3 variables regarding satisfaction with "D", expressed on 7-likert scale
  • 1 variable regarding how many x you consume
  • 1 variable regarding how much you spend to buy x
My doubt:
I expect the satisfaction MVs to be positive correlated with LV and the last two MVs to be negative correlated. I have seen someone trasforming MVs into negativeMVs to have unidimensionality condition respected, but I would like to avoid this solution becouse it conceptually does not make much sense.
  1. can I consider my MVs to be reflective with LV since I have a MV about consuption and a MV about expences?
  2. in alternative, can I state my LV "D" to be explained by 3 reflective MVs and 2 formative MVs?
  3. suggestions to define my LV?
Thanks a lot for the attenction, any help will be highly appreciated.
Relevant answer
Answer
I think it would be better if before analyzing using PLS, testing the validity and reliability of the MV against LV is first performed. Later, invalid or unreliable MVs should not be included in PLS analysis. Usually a negative MV should have been indicated invalid.
  • asked a question related to Partial Least Squares
Question
2 answers
Hi everyone,
I want to analyse the correlation of phenolic compounds and bioactive activities (quantitative variables) of my samples, but I also want to include my sample information (qualitative variables) for the correlation analysis.
It seems that the common Pearson Correlation method can only used for quantitative variables. I have found that the Partial Least Squares Correlation method can be used for both quantitative and qualitative variables, but this method is not commonly used.
So my question is what is the difference between the Partial Least Squares Correlation and Pearson Correlation, which one is better ?
Thank you very much
Relevant answer
Answer
Hi Yun,
the difference between a Correlation and Partial Correlation is:
A correlation measures the linear relation between two variables.
A partial correlation measures the linear relation between two variables, after the variance of another variable has been partialed out.
You can imagine this by drawing three overlapping circles, which create a shared area altogether in the middle. Call them circle A, B and C. Now, if you only look at the area shared by A and B, while ignoring C, then ... this is a correlation between A and B. But if you erase the areas that A and C share, and B and C share, and only then look at the left over area that A and B share, you have a partial correlation.
(There exists also a semi-partial correlation, used in regression models, which only partials out the variance shared by the "C-variable" and the predicted variable.)
As you see, this has nothing to do with quantitative vs qualitative. Anyway, you always can calculate correlations (or partial correlations) that include qualitative variables, as long as they are binary (e.g., sample A vs. sample B), and as long as you do not interpret it as "linear" effect. For instance, if you calculate a correlation between a binary and a continuous variable, you can interpret it as a mean difference between the two samples. In this sense, you can partial out the "shared" variance (average difference) of the samples and your other variables, before you correlate these other variables. If your sample variable is not binary, but categorical, than you need a regression model, in which you predict one variable of interest, using the categorical one (defined as nominal factor) and the continuous one, with main effects and interactions. The test you need then is a (keyword) likelihood ratio test between the full and the reduced model. (You need to look this up :)). If this becomes significant. Then the relation between the continuous variables depends on the sample.
Hope this helps
Best, René
  • asked a question related to Partial Least Squares
Question
4 answers
Hi everyone!
I am working with small samples of national teams. Such small samples make it impossible to estimate the quality of the linear regression model, so I decided to apply partial least squares structural equation modeling. Though it seems to suitable for my purposes, I still have several questions regarding my models.
I developed three models. One includes both adolescents (M = 16) and adults (M ~ 23) (54 athletes), second - only adolescents (35 athletes, male and female), and third - only adults (19 athletes, female only). As can be expected, the first model is the best in terms of the significance, reliability of latent variables measures. However, I am not sure that it is the correct way to choose the model. It seems logical that the bigger sample provides "better" results. But is it ok to mix two different age groups?
Relevant answer
Answer
If previous research has suggested there are no differences between adults and adolescents on the characteristics you are studying, it should be OK. If there is any question of differences, you can account for it by adding an adult/adolescent dummy variable (0/1) to the model to determine if there is a difference.
  • asked a question related to Partial Least Squares
Question
8 answers
I got some data for my research. I run them on SPSS SEM but the R-square value is quite low. I heard that Smart PLS can offer higher R-square than SEM. Is that reliable?
Relevant answer
Answer
Hello Sann,
Higher R-squared for what? It would be helpful to know something about the research question, variable set, and proposed relationships to give you a more informative response to your query.
There is a substantial perspective distinction between covariance-based models and PLS models in SEM; that might be more germane to the choice than a claim that one method yields higher/lower values for some target statistic. See this link for some of the reasoning that PLS-fans offer for making one choice vs. another:
SPSS offers SEM modeling only via the add-on module, AMOS (or via its R extensions, in which you could run lavaan or sem). I assume you're referring to one of these options.
Good luck with your work.
  • asked a question related to Partial Least Squares
Question
6 answers
Urgent>> I am a master student doing a study about privacy behavior on Facebook.. my research model consists of two independent variables (cost and benefit), each independent variables consists of six items.. in addition to two dependent variables each of three items.. all measurement models are reflective.. the number of samples is 120.. all answers are within the range (1 to 5) or (strongly disagree to strongly agree).. I use the pls algorithm in smartpls software to analyse my model.. the problem is most of the outer loadings are below (o.7).. one variable has Cronbach's alpha below (0.7) while three varibles have AVE below (0.5). I try to delete the items with the least loadings but there is no recognized effect on the Cronbach's alpha and the AVE.. kindly I need to know how to fix this problem ??
Relevant answer
Answer
Is the number of items sufficient to give you a good loading? Any multi-collinearity among items or factors.
  • asked a question related to Partial Least Squares
Question
4 answers
Currently, variables are related to each other. Instead of performing a t-test for each variable between two groups, I was suggested to apply partial least square to calculate the p-value.
I read some publications about PLS and PLS-SEM. I still don't know how to use PLS to calculate group differences.
Relevant answer
Answer
Chao, a point to understand is that whether you are asked to compare mean differences for variables between two groups, for example, or the difference in relationships of variables due to two groups.
For the latter, as suggested by David and Alejandro, you can look at multi-group analysis in SmartPLS. It's not a complicated merhod and you can see YouTube videos to learn it.
However, if you are asked to compare mean values of variables using PLS, I am interested to know the answer as well.
  • asked a question related to Partial Least Squares
Question
9 answers
Hi,
Can anyone answer me the question about the steps to do in structural model in smart PLS?
I read in the book "Hair 2016 - A Primer on Partial Least Squares Structural Equation Modeling" they mentioned there are 6 steps to do in Structural Model Assessment procedure.
Step 1,2,3, are fine. But I wonder step 4,5,6 are always necessary to report? Or when we should report these steps?
Step 4: Assess f2 effect size
Step 5: Assess predictive relevance Q2
Step 6: Assess q2 effect size
Thank you
Relevant answer
Answer
q2=Q2included-Q2excluded/ (1-Q2included)
You can run blindfolding by removing one exogenous variable at a time and calculate using above formula.
  • asked a question related to Partial Least Squares
Question
4 answers
How partial least squares is related to Projection to Latent Structures method developed by H. Wold?
Is there any difference between partial least squares and Projection to Latent Structures method?
Relevant answer
Answer
Mei Peng Low thank you so much
  • asked a question related to Partial Least Squares
Question
3 answers
Hi. I'm using WEKA to build a QSAR model using PLS method with a training set included 102 instances and 206 attributes (chemical descriprtors). I want to select attributes that can:
- Build a model with correlation coefficient r2>0.5 and root mean square error RMSE<0.5.
- Cross validation (leave-one-out) r2>0.5, RMSE<0.5.
- Evaluate on external set: r2>0.5.
I'm totally newbie and have no idea to do that. I tried select attributes with CfsSubsetEval + BestFirst but the result RMSE is still >0.5. I continued to select some of attributes that could reduce the RMSE on both training set and cross validation but it's overfitting and the r2 on external set is only 0.2. Can you help me please? Thank you very much.
Relevant answer
Answer
In passing, you may try to use the genetic algorithm as an automatic selection method if your set of descriptors is large and you have no idea which feature is likely to be relevant for your target property prior to automatic selection.
PS: note that there are some physically meaningful features which are likely to be dominant in many QSARs:
- Size (represented by molecular weight or number of atoms or molecular volume, for example)
- Polarity (for example, represented by descriptors such as Total Polar Surface Area or, if you want to factor out the size, dividing Total Polar Surface Area by Total Molecular Surface Area)
- Hydrogen-bonding (could be represented by the number of hydrogen-bond donors, or by group contributions if you want to differentiate the different types of hydrogen bonding)
At least when intermolecular interactions come into play. When reactivity comes into play, you may be interested in the most relevant reactions to your target property and their associated descriptors like bond dissociation energies.
Best regards.
  • asked a question related to Partial Least Squares
Question
4 answers
What is the difference between formative and reflective partial least square structural equation models? And what tests need to be conducted for the aforementioned models?
Relevant answer
Answer
Yes, see my profile. If you send me a request I can provide it to you.
Best regards,
Florian
  • asked a question related to Partial Least Squares
Question
4 answers
In testing moderating effect of a moderating variable on relationship between IVs and DV (all the latent variables are with reflective indicators; and sample size over 200) using SmartPLS 3, firstly it obtained a significant moderating effect by using the product indicator approach. However, the test again using the two stage approach found no moderating effect.
As referred by Henseler & Chin (2010), the two-stage approach has the higher statistic power, i.e. it is the most likely approach to detect a significant interaction. But the results here showed the opposite!!
I would appreciate any suggestions regarding this.
Relevant answer
Answer
Examine your variables. If the moderator is metric or continuous in nature then you have to use product indicator technique. If it is non-metric then use multigroup analysis
  • asked a question related to Partial Least Squares
Question
7 answers
I would like to do analysis from SmartPLS. 
The objective  of my research is to establish relationship "X" to "Y" by the use of SmartPLS
X have have five domain and every domain have 12 items that mean 12x5=60 items
And "Y" have one domain with 6 items
Can I establish this relation from smartPLS? if yes than what type of test i have to with ?
kindly attached my analysis and tell me that i am going on right way or not ?
Relevant answer
Answer
Yes your model is OK, Please follow the recommendations of Hair et al. (2016) in analyzing the second order. Moreover, I see that your model is perfect and you just need to sketch the path for your statistical analysis that you could easily follow from the above-mentioned book.
Please feel free to contact me if you face any issues.
Best of luck
  • asked a question related to Partial Least Squares
Question
3 answers
One Week Workshop on Advanced Research Analysis Tools and Techniques Being Organised by Department of Commerce (SAP, DRS-III), University of Jammu, Jammu (J&K)
Research provides new domain of knowledge, new theories and laws, solution to a problem and open new areas of thought. Further, it has the potential to question and analyse the existing theories in accordance to the present circumstances. Research involves answering unanswered questions or exploring something which currently does not exist (Goddard & Melville, 2004). Hence, it is important for educational/ professional academicians and scholars to be acquainted higher order research techniques. In the recent past, many new dimensions of research have emerged catering to the vast changes taking place in the present day social situation. The quality of research depends on the understanding of the research problem, the attendant research methodology and the appropriate tools used for the analysis. This one week workshop will focus on qualitative research techniques, experimental research design and partial least squares structural equation modelling (PLS-SEM).
Relevant answer
Answer
I am late
  • asked a question related to Partial Least Squares
Question
12 answers
Hi everyone, 
 I am using smart PLS for one of my study. I have first order and second order constructs. In Smart PLS i got the AVE for first order constructs above 0.5, whereas the AVE for second order construct is less (0.30). is it acceptable ?if not do i need to report AVE for second order construct? support me with literature  ( My Model is Reflective- Reflective model. 
the other validity for both first order and second order are satisfying the cut off level ( say convergent, internal composite reliability etc.,)
Relevant answer
Answer
We can accept AVE < 0.5 if the composite reliability is > 0.6, because the convergent validity of the construct is still adequate (Fornell & Larcker, 1981)
  • asked a question related to Partial Least Squares
Question
103 answers
What is the free software that you have used for partial least square structural equation modeling?
Thanks in advance
  • asked a question related to Partial Least Squares
Question
18 answers
Is there any impact factor journal for PLS-SEM studies only? or any current special issue? 
Relevant answer
Answer
Behaviormetrika has a special issue on composite-based methods--not just PLS. Special issue editors are Marko Sarstedt (of SmartPLS) and Heungsun Hwang (creator of GSCA).
  • asked a question related to Partial Least Squares
Question
2 answers
I would like to know on how to calibrate for real-time time monitoring using FTIR probe, the reaction is a multi-components reaction for which Partial Least Square will be used.
Relevant answer
Answer
Thank you prof
  • asked a question related to Partial Least Squares
Question
5 answers
Am conducting a multgroup analysis in PLS-SEM using SmartPLS 3 and I came across an interesting situation.
There is a moderator variable which is having an effect on one group (lets say X1) and not having an effect on another (say X2), this is evident by the t-values and probability obtained. But the inter-group difference of t-value isnt significant. How to report this? Assume that the moderator has only two categorical levels.
What I inferred was that since there were no inter-group differences in terms of p-value, even though X1 has a significant p-value, it might be considered a sampling error and for the general population there is no moderating effect of X on the relationship. Thus, no heterogeneity was observed. 
Is this correct?
Relevant answer
Answer
Deaer Profssor Alshibly,
Sorry the delay in my answer. Howerver I'm very greatfull, for your fast and very rapid response e kind favor.
I will contact you again about a fundamental error and possible fatal error that I observed in aplications of PCA, in chemometric field. I'm sure about this fact. I wrote to Professor Iam Jollif (Rochester University, book author of "The PCA" - Springer Verlag)) retired now, and he agree with my observation. I invite you to analyse such fail and to write a chemometric article about this.Woul you mind in thing about this ?. I believ that this article will be a great contribuition to chemometrics. Basically the error iis a strong instability in PCA by SVD or eigenvalue diagonalization derivation. Can you you send me your e-mail to start our discussion?
Sincerely yours
Cesar
my private e-mail: mellocesar31@gmail.com
  • asked a question related to Partial Least Squares
Question
2 answers
Our plan is to use the SLAM-framework which includes a questionnaire with 65 questions. This framework is previously tested and validated. But for our case, we have 110 respondents witch is too little for the statistical analyses, such as factor analysis where you need 5 respondents per question. Do anyone have any suggestions on how to tackle this problem?
We want to use these questions, but do we have to reduce the number of questions before we send the questionnaire out?
Relevant answer
Answer
Dear Luvic,
If your questionary was valited, i think you should use the confirmatory factor analysis. Thus, the PLS (partial least square) need a small number of respondets.
Even though it was adopted by many in Exploratory Factor Analysis (EFA), the KMO test (general and by variable) was developed for the purpose of evaluating if the sample has sufficient size to use the EFA.
See:
Hair, J.F., Hult, G.T.M., Ringle, C. M., & Sarstedt, M. (2014). A Primer on Partial Least Squares Structural Equation Modeling (PLS-SEM). Thousand Oaks, CA: Sage.
Ringle, C. M., Silva, D., & Bido, D. (2014). Structural Equation Modeling with the Smartpls. Brazilian Journal Of Marketing, 13 (2), 56-73.
COHEN, J. (1988). Statistical Power Analysis for the Behavioral Sciences. 2nd ed. New York: Psychology Press.
If you need more details, please send more informations about your researsh.
hope this help
Besta regards
Dirceu
  • asked a question related to Partial Least Squares
Question
5 answers
Data set size = 283
IVs = 4
DV = 1
Factor = 1 (out of 4, one was selected)
Results
Press = 0.91
X explained 58%, Y explained 18%
Relevant answer
Answer
The acceptable PRESS value should be chosen according to the balance of model complexity and predictive error. Therefore, the PLS model with the least latent variables as well as preditvie error not significantly larger than the smallest one should be chosen. F-test can be used to test the significance. More details can be found in the following paper:
HAALAND D M, THOMAS E V, CHEM. A. Partial Least-Sqares Methods for Spectral Analyses. 1. Relation to Other Quantitative Calibration Methods and the Extraction of Quantitative Information [J]. Analytical Chemistry, 1988, 60(11): 1193-202.
  • asked a question related to Partial Least Squares
Question
4 answers
Most researchers have faced the problem disability of differentiate between A reflective analysis and conjunctive analysis in Structural Equation Modelling SEM. Kindly please I need any one assets in getting the explanation for this question.
Relevant answer
Answer
This is A very nice book Thank you Qais Almaamari
  • asked a question related to Partial Least Squares
Question
9 answers
Does Stata have the ability to perform a partial least squares analysis or another procedure which might help specify a model with low co-linearity among numerous predictors? This would be apart from a trial and error process and examining VIF for each try. Thank you in advance.
Relevant answer
  • asked a question related to Partial Least Squares
Question
8 answers
I am analysing soil microbial community data for classification and discrimination. PLS-DA seems more efficiency in seperating microbial groups. what is the difference between PCA and PLS-DA? When to use PLS-DA rather than PCA?
Relevant answer
Answer
PCA is totally unsupervised. With PLS-DA you do a regression between the group of classes - then you have already from the beginning defined your classes as a response variable, therefore more efficient separation, but then you need to know what classes each observation belongs to.
  • asked a question related to Partial Least Squares
Question
2 answers
i want to perform a test on biophysical parameters with different SAR polarizations 
Relevant answer
Answer
Hair et al adresses this questions in their book Hair et al (2017): A Primer on Partial Least Squares Structural Equation Modeling (Pls-Sem)
  • asked a question related to Partial Least Squares
Question
8 answers
I have three independent and one dependent variable in our model. I am using partial least square structural equation modeling to test the relationships. When tested separately, each independent variable has a significant relationship with the dependent variable. But when put together in the model, one of the independent variable becomes insignificant.
What does it mean? What could be the interpretation? 
Relevant answer
Answer
Farhan -
Because of the various ways in which the "independent" variables may interact, in addition to the misleading nature of "significance," it is probably best just to test your model's overall performance if you are looking for the 'best' predictions.   -----  However, you might first make sure you are looking for the best predictions, not for explanation, as discussed in the following, where the appendix example is instructive:
Galit Shmueli -
You might want to research the terms "model selection," "model validation," and "model testing." 
In general, a more complex model may have higher variance, but lower bias, and vice versa for less complex models, as noted in bias-variance tradeoff discussions in the area of "statistical learning."  However, I have found that if the less complex model does not use the 'best' regressor (i.e., predictor, independent variable), the estimate of sigma is inflated so that the higher estimated variance of the prediction error may actually be for the less complex case. 
Perhaps you could compare your model to other ones which would be consistent with subject matter theory.  If you compare them for fit, say graphically/visually, as I am about to suggest, you still need to save out some data for testing, to avoid overfitting.  -  You could, in many cases, and I am assuming it is practical in your case, though I don't know, compare different models on the same scatterplot by graphing y on the y-axis, and predicted-y on the x-axis.  That is included in the following as Figures 5 and 6:
I'm not sure how helpful the above may be for you, but I think it might be of some use. 
Cheers - Jim
  • asked a question related to Partial Least Squares
Question
13 answers
How can I detect common method bias in smartpls? Can you suggest some guidelines?
Relevant answer
Answer
 For PLS-SEM, common method bias (CMB) is detected through a full Collinearity assessment approach (Kock, 2015). VIF values should be lower than  the 3.3 threshold (Hair et al., 2017, Kock, 2015). This is indicative that the model is free from common method bias. Any value greater than 3.3 means the model is affected by CMB.
  • asked a question related to Partial Least Squares
Question
13 answers
Does anyone here know how to create three-way interaction effect using SmartPLS?
Relevant answer
Answer
 Hi, Chuah, Did you sucessfully use SmartPLS 2.0 to creat 3-way interaction? Can you tell how to do it? Thanks very much!
  • asked a question related to Partial Least Squares
Question
5 answers
Currently, we are working on a moderated mediation model, and we want to test this model using smart-pls is it appropriate?
  • asked a question related to Partial Least Squares
Question
5 answers
I want to calculate GoF = sqrt((average AVE) * (average R2))
My AVE generated by the SmartPLS are
0.7399
0.8225
0.7582
0.6398
0.6735
Is Average AVE means to take the average of these values
like (.7399+.8225+.7582+.6398+.6735 / 5)?
Furthermore, after running PLS algorithm, the value given in the output variable is Average R2?
Relevant answer
Answer
AVE is for validly and is about measurement model, not about structural model..
AVE for your constructs having four indicators can be calculated as:
AVE = (square loading of item1 + item2 + item3 + item4) / 4
  • asked a question related to Partial Least Squares
Question
5 answers
Dear PLS structural equation modeling professors and experts,
If I have 10 experts to judge about the indicators of a  project performance (As a first order construct) and I have 20 sample of completely independent projects whose  their indicators(observed variables) are to be judged by the10  experts, Could I say I have 20*10 means 200 samples and use it for parameters estimation in PLS(Partial least square), or I must get an average among 20 project and claim that I have only 10 sample or I may get an average of the expert judgments and claim that I have 20 samples?
Is it a problem of Pseudo-replication? If so will it lead to unreliable parameter estimation?
Sincerely Yours
Thanks in Advance
Relevant answer
Answer
Ali:
If I understand your problem correctly, you have a data matrix of p projects on q attributes made by N experts; all experts have rated exactly the same projects on exactly the same attributes. You are, quite reasonably, unwilling to assume that each individual expert's ratings (p x q in number) are independent of each other. In addition, you are interested in the possibility that there might be differences among experts in their models of project performance.
You can find relevant literature using the search term "policy capturing" (see Karelaia & Hogarth, 2008, for a review). A brief summary of the process for analyzing your data follows.
You can begin your examination of differences among experts by assessing their agreement in the ratings on each of the attributes (see LeBreton & Senter, 2007, for methods of assessing interrater agreement). If interrater agreement is low on some or all of the attributes, you can examine the ratings to see if there is any obvious clustering of the experts.
Next, you can see if the experts have similar mental models of project performance by calculating, for each expert, the matrix of intercorrelations among the attributes (across projects). If there seem to be substantial differences among the experts in their intercorrelation matrixes, you could conduct a cluster analysis to identify the different "points of view" (POV) among them.
If you find that the differences among the experts are not theoretically meaningful, then you can proceed to an analysis of the projects. In this analysis, as you and Shah have agreed, you would compute the mean rating (over experts) on each attribute of each project. The sample size for your analysis of projects (as distinct from your earlier analysis of experts) is p = 20.
If you find that there are theoretically meaningful differences among the experts, you can conduct a separate analysis for each POV. That is, you average the ratings only for the experts that are represented by that POV. If you found differences among the POVs with respect to their intercorrelation matrixes, you will almost certainly find differences among them in the structural models. In this case, you will have as many structural models as you have POVs.
It is not a foregone conclusion that you will have multicollinearity problems with your structural analysis. If you do find multicollinearity, it might be that this is inherent in the relationship between attributes (project cost, duration, and quality usually have negative correlations) or it might be an aspect of your experts' mental model that is unrelated to reality. If your raters really are experts, then it is more plausible that this is an aspect of reality rather than a misconception. Of course, if you have different POVs that differ in their intercorrelations among attributes, then one POV or the other (or perhaps both) are wrong (or perhaps basing their ratings on different sets of experiences). As Shah said, you should examine their educational qualifications, years and types of experience to see if you can explain the differences.
If you do get different POVs, you could identify the projects on which they differ the most and ask representatives from the different POVs to explain the basis for their judgments.
Good luck.
References
Karelaia, N., & Hogarth, R. (2008). Determinants of linear judgment: A meta-analysis
of lens model studies. Psychological Bulletin, 134, 404–426.
doi:10.1037/0033-2909.134.3.404.
LeBreton, J. M., & Senter, J. L. (2007). Answers to 20 questions about interrater reliability and interrater agreement. Organizational Research Methods, 11, 815-852.
  • asked a question related to Partial Least Squares
Question
10 answers
Hello,
I thought PLS was only interesting when the sample is not very important compared to the number of items, but I saw papers with roughly 200 responses and 20 items using PLS. Maybe it's a problem of formative vs. reflective variables? Thank you for your help.
Relevant answer
Answer
Current thought in the statistical literature is that PLS rarely offers any benefits over covariance-based SEM and/or OLS regression modeling. Some authors, especially Rönkkö et al. (2016) – see http://dx.doi.org/10.1016/j.jom.2016.05.002 – and Guide & Ketokivi (2015) – see http://dx.doi.org/10.1016/S0272-6963(15)00056-X – judge PLS very harshly. Given the widespread criticism, it typically makes sense to use covariance-based SEM or OLS regression modeling instead of PLS.
  • asked a question related to Partial Least Squares
Question
4 answers
Hello everyone,
i have two columns of data (x,y), which i want to fit with an explicit function:
 y=p1*(x-p2)^p3,
using non-linear least square algorithm. So as to calculate the Jacobian maxtix, i need to know the partial derivatives of the function with respect to three parameters. For p1 and p2, it is quite obvious that:
dp1=(x-p2)^p3;
dp2=-1*p1*p3*(x-p2)^(p3-1)
how about the third one ?  dp3=?
Normally, for the function 7^p3, dp3=7^p3*log(7). However, it is not  my case, since (x-p2) can be negative. Is this function non-differentiable w.r.t. p3?
Thanks in advance for your answer.
Liang
Relevant answer
Answer
Dear Liang, you are true. In the case p3 is a real variable z^p3 is only defined for z>0 because z^p3 = exp(p3*ln(z)).
Richard
  • asked a question related to Partial Least Squares
Question
5 answers
Please share any excel file if any body have
Relevant answer
Answer
Thank you so much... i will go through the problem... If got any problem i will again request you for help...
  • asked a question related to Partial Least Squares
Question
26 answers
I am more familiar with SEM tools than PLS. On the other hand, one student I am supervising wanted to use PLS on account that it has significantly smaller sample size requirements than SEM. However, I have my own misgiving with the lack of goodness of Fit statistics in PLS (to my knowledge). Is there any goodness of fit statistics (or its equivalent) in PLS?
I would like opinion from both PLS and SEM users on this matter.
Relevant answer
Answer
The answers above can be regarded as outdated. The linked paper explains how to assess the goodness-of-fit of PLS path models.
  • asked a question related to Partial Least Squares
Question
6 answers
My results from PLS-SEM analysis (in WarpPLS) shows weak effect size (f-squared). The path coefficient is 0.15 and p-value is significant. I am analyzing indirect relationship (mediation). How do I interpret such results? Also the VAF value shows full mediation. I am confused. 
Relevant answer
Answer
As the mediating effect is significant, then you can say that your mediating variable has the mediating effect (p value) and state how much the effect. Now, to judge the value if it is small or large, it referes to the type of relationships and variables being tested, as well as how these variables are compared to what is in the literature. . Good luck
  • asked a question related to Partial Least Squares
Question
3 answers
So my colleague and I tried to run PLSPM from a huge dataset which measures binary, ordinal, and numerical data.
Is there a way to do this in R using the plspm package? Tried using XLSTAT PLSPM package and it won't run the non-numerical data during the bootstrap.
We also like to know if there are ways to see whether the latent variable are meaningful to the model. If anyone can help us it'll be awesome! Thanks!!
Relevant answer
Answer
Right I've solved this, it was due to the old version of R which did not run the plugin properly :s
  • asked a question related to Partial Least Squares
Question
1 answer
Principal component analysis (PCA) to obtain the principal components (PCs) that best fits the shape of the face in order to relate the SNP data with that PCs by using the partial least squares regression (PLSR) method as it is explained in Claes et al. 2014? 
Don't understand very well the methodology there explained. 
Thanks in advance.
Relevant answer
Answer
PCA isn't method for regression but for data description. You can try PCA to find relation between SNP data and face shape data if you create one data matrix with both sets. Investigation loadings could tell something about relations.
Different approach to solve your task is to use some regression method for example PCR (principal component regression) or PLSR.
In my opinion there is no sense to calculate PLSR based on PCA data.
  • asked a question related to Partial Least Squares
Question
2 answers
When choosing a WAPLS model, apart from relying on performance (e.g. r2, r, max bias, RMSEP, scatter-plot etc..), the recommendation is to select the successive component if the model performance increases by >5% measured by the cross validated RMSEP (e.g. Braak & Juggins, 1993; Birks, 1998; Brooks & Birks, 2000; Barlow et al., 2013).
 Does this rule apply also to PLS technique?    
Relevant answer
Answer
Dear Gerhard,
Thanks for your response! My dearest Yvonne already answered my question via email. Keep up with the interesting work.
Best regards from Hollywood
Simona
  • asked a question related to Partial Least Squares
Question
8 answers
I found this table (as appears in the attached image)
in this book
Hair Jr, J. F., Hult, G. T. M., Ringle, C., & Sarstedt, M. (2013). A primer on partial least squares structural equation modeling (PLS-SEM). Sage Publications.
I need to understand how to use this table
Relevant answer
Answer
Basically, the table outlines that the sample size depends on three factor.
First, the number of variables or constructs that are intended to explain the dependent variable. Given you have a lot of variables; you need a higher sample size. This is why, if you look down the columns, the necessary number of participants increases, the more “arrows” point at the dependent variable.
Second, the estimated effect size. R square can be used to estimate the effect size (have a look at the effect size f). If you guess or prior research let you conclude that your independent variables will have substantial impact, in other words, a large effect size on the dependent construct, you need less participants. If you guess that the effect size is small, you need more participants to “find” the effect.
Third, the significance level. If you define the relationship as significant at a 10% alpha error, you need less participants compared to a defined alpha error of 1%. That is, the more you accept being susceptible to make an alpha error, the less participants you need.
Hope that helps.
Rafael
  • asked a question related to Partial Least Squares
Question
7 answers
Hello, 
I need to fit PLS regression model on my data. However, I could not find proper references on how to explain the results on SPSS. How to choose latent number and how to extract the model from the results for prediction. 
I really appreciate it if anyone can give me a reference or a tutorial regarding PLS on SPSS.
Thanks
Relevant answer
Answer
Here you have a whole playlist on PLS-R. This is only the first video. It will also talk about selecting model parameter. Enjoy!
  • asked a question related to Partial Least Squares
Question
3 answers
hi,
 I am a new user of SmartPLS. I would like to perform FIMIX-PLS, but I could not find any guideline to run the analysis in SmartPLS 3.
Relevant answer
Answer
hi gain
I found the answer myself
please follow the links below
  • asked a question related to Partial Least Squares
Question
8 answers
I have collected 84 responses through survey. However, I am not sure what analytical tools will be better to use. Any suggestions (step by step, if possible) for Quantitative data analysis would be of help.
Thanks. 
Relevant answer
Answer
Dear Abdul,
84 "samples" is a good number. I propose you to divide the dataset in two (64/20). Try to select representative samples in the two datasets. Develop a PLS calibration model with the 64 samples. Often we use cross-validation for this calibration. The 20 left out samples are finally used for prediction in order to estimate accuracy. If you need help, don't hesitate to contact me.
Ludovic.
  • asked a question related to Partial Least Squares
Question
6 answers
My request is primarily directed towards the optics to be a good user and not towards an expertise of the mathematical dimensions
Relevant answer
Answer
Dear James
Thank you very much for all these documents and for your help
best regards
  • asked a question related to Partial Least Squares
Question
1 answer
To overcome multicollinearity problem in multiple regression one uses either ridge regression, liu estimator, Partial least squares regression and etc.
Relevant answer
Answer
I would probably try simulation. An example is attached. Best wishes.
  • asked a question related to Partial Least Squares
Question
15 answers
want to use it for my some data analysis 
Relevant answer
Answer
There is some confusion in the discussion.
Gertrude asked about software for PLS path modeling but many of the tools that you have advised concerning PLS regression.
There are two free packages that you can use for PLS path modeling analysis:
- PLS-SEM toolbox (Matlab Language) 
- PLSPM package (R Language)
from CRAN repository
  • asked a question related to Partial Least Squares
Question
7 answers
I have a set of likert items all of which are constructed with scales of 7 categories, except for one likert item, which is constructed with a scale of 11 categories.
These likert items are to be used in a Structural Equation Model and analysed using Partial Least Squares.
My question is now:
- How, if at all, will this discrepancy influence the analysis?
Relevant answer
Answer
Hey Lars,
there exists a variety of ways to address the issue of categorical indicators in PLS-PM, e.g. Canatkuppi and Boari (2012) or Russolillo (2012). In a joint paper (under review), we developed a new consistent approach to deal with ordinal categorical indicators. In this context, we also run some simulations and we find out that for 7 and more categories the difference between PLSc and our approach is small. Hence, the scale of the indicators can be negected for more than 7 categories. For covariance-based estimators,similar studies exist, e.g. Rhemtulla et al. (2012).
I hope this helps. Best regards,
Florian
Rhemtulla, Mijke, Patricia É. Brosseau-Liard, and Victoria Savalei. "When can categorical variables be treated as continuous? A comparison of robust continuous and categorical SEM estimation methods under suboptimal conditions." Psychological methods 17.3 (2012): 354.
Boari, Giuseppe, and Gabriele Cantaluppi. "A PLS algorithm version working with ordinal variables." XLVI Riunione Scientifica della Società Italiana di Statistica. cleup, 2012.
Russolillo, Giorgio. "Non-metric partial least squares." Electronic Journal of Statistics 6 (2012): 1641-1669.
  • asked a question related to Partial Least Squares
Question
4 answers
multicoillinearity and outliers do exist in multivariate regression analysis. There are methods like ridge regression, principal component regression and etc.
Relevant answer
Answer
Hello,
There is a robust PLS variant called partial robust m-regression. For more details please see references (Matlab code available in the Tomcat toolbox):
(1) S. Serneels, C. Croux, P. Filzmoser, P. J. Van Espen, Partial Robust M-regression, Chemometrics and Intelligent Laboratory Systems 79 (2005) 55-64
(2) M. Daszykowski, S. Serneels, K. Kaczmarek, P. Van Espen, C. Croux, B. Walczak, TOMCAT: a MATLAB toolbox for multivariate calibration techniques, Chemometrics and Intelligent Laboratory Systems, 85 (2007) 269-277
In addition, you can simultaneously handle missing values and outliers:
(3) I. Stanimirova, S. Serneels, P.J. Van Espen, B. Walczak, How to construct a multiple regression model for data with missing elements and outlying objects, Analytica Chimica Acta, 581 (2007) 324-332
Best regards!
  • asked a question related to Partial Least Squares
Question
11 answers
Thaumastocoris peregrinus is  a major pest of eucalyptus in Brazil an we need to develop efficient methodologies to monitor this insect. 
Relevant answer
Answer
Hello,
Can you provide more information about your methods? Which sensors you are planning to use? On what type of platform (e.g. airborne, satellite)?
PLS-DA can be done with varius software from R to Matlab. If you think we can collaborate on long distance way we could collaborate.
  • asked a question related to Partial Least Squares
Question
13 answers
I would like to ask is there any rule to determine the optimal number of PLS components i.e. above which it will overfit and below  which it underfits. Is it like one third of the number of total samples?
Relevant answer
Answer
Tons of literature are available on that subject. References can be found on: http://www.chemometry.com/Site%20map.html
Mr. Faber, the owner of the website, is one of the specialists in this topic.
The simple answer to your question is: if your system has N components you will need N-1 variables. That is when there is no interaction between the components, Beer's law applies, your sample set is well designed, your measurements are nicely linear and so on.
In practice the optimal number is found by applying some cross validation (jack-knife, or other). The measurement set is divided in multiple sub-sets. The PLS models with increasing numbers of variables are developed on one set and tested on the other(s). This way an seemingly optimal number of variables is found in terms of prediction error.
What is crucial is validating the developed model on an independent dataset which was set aside before starting the model. The prediction on that set is the actual prediction error of your developed model. If it turns out that refinement is needed a new validation set must be measured/collected.
Good luck
  • asked a question related to Partial Least Squares
Question
3 answers
I got a bifactor model with 4 indicator each one and two groups. I understood that samples below 100 can´t provide stable results but otherwise I can´t add more cases. Is there a way to run measurement invariance with 40 cases each group or another similar procedure (perhaps partial least squares) or alternative (montecarlo simulation to detect sample size)?
Relevant answer
Answer
  • asked a question related to Partial Least Squares