Hi all,
I intend to model a binary choice decision (chosen=1, not chosen=0) using several latent constructs as predictors. Some of the constructs are (multi-item) reflective measures, the other is a multi-item formative measure (all items are Likert-scale based). I am looking for the best way to create the construct values required for my logistic regression.
Is it possible to use SmartPLS to operationalize the constructs albeit the binary nature of the dependent variable (DV)?
Specifically, can I use it to a) validate the psychometric properties of the respective constructs (formative: VIF, weights, etc.; reflective: CR, AVE, etc.) and b) "generate" latent variable scores (LVS) that I employ as independent variables in the logistic regression?
If so, are there any precedents in the literature that you are aware of?
To my knowledge, SmartPLS is/should typically not (be) employed for binary variables (e.g., Hair et al., 2012, p. 421) but am wondering if it might be feasible for the purpose above. I would not interpret the path coefficients from the constructs to the binary DV but merely "create" the LVS for the subsequent logistic regression.
Looking forward to your replies! In advance, thank you very much!
Benedikt Constantin
----
Quote: Hair, J.F./Sarstedt, M./Ringle, C. M./ Mena, J. A. (2012). An assessment of the use of partial least squares structural equation modeling in marketing research. Journal of the Academy of Marketing Science, 40. 414-433. (DOI 10.1007/s11747-011-0261-6)
Latent variable scores as input for logistic regression
-
- SmartPLS Developer
- Posts: 1284
- Joined: Tue Mar 28, 2006 11:09 am
- Real name and title: Dr. Jan-Michael Becker
Re: Latent variable scores as input for logistic regression
Two general thoughts:
1) You could use simple equal weighted averages of your indicators per construct. For a reflective construct this is usually not much worst (and in small samples even advantageous) then using more advanced PLS weights. For formative constructs, it is only advisable if the underlying formative weights are not very much different and you have a small sample.
2) You could include the binary dependent variable in your model and estimate a linear probability model. It has some disadvantageous to a logistic regression, but it can also be very similar. In addition, you could use that model only to estimate the LV scores and validate the measurement properties and then take the LV scores (as you have proposed) to do a logistic regression analysis.
1) You could use simple equal weighted averages of your indicators per construct. For a reflective construct this is usually not much worst (and in small samples even advantageous) then using more advanced PLS weights. For formative constructs, it is only advisable if the underlying formative weights are not very much different and you have a small sample.
2) You could include the binary dependent variable in your model and estimate a linear probability model. It has some disadvantageous to a logistic regression, but it can also be very similar. In addition, you could use that model only to estimate the LV scores and validate the measurement properties and then take the LV scores (as you have proposed) to do a logistic regression analysis.
Dr. Jan-Michael Becker, BI Norwegian Business School, SmartPLS Developer
Researchgate: https://www.researchgate.net/profile/Jan_Michael_Becker
GoogleScholar: http://scholar.google.de/citations?user ... AAAJ&hl=de
Researchgate: https://www.researchgate.net/profile/Jan_Michael_Becker
GoogleScholar: http://scholar.google.de/citations?user ... AAAJ&hl=de
-
- PLS Junior User
- Posts: 2
- Joined: Mon May 16, 2016 10:54 am
- Real name and title: Benedikt Constantin, PhD Student
Re: Latent variable scores as input for logistic regression
Thank you very much for your reply, Dr. Becker!
With regard to your second suggestion: How exactly would you use a linear probability model to estimate the required LV scores, that is what would be the independent/dependent variables? I might be missing something in your thought and am curious to better understand the idea.jmbecker wrote: 2) You could include the binary dependent variable in your model and estimate a linear probability model. It has some disadvantageous to a logistic regression, but it can also be very similar. In addition, you could use that model only to estimate the LV scores and validate the measurement properties and then take the LV scores (as you have proposed) to do a logistic regression analysis.
Re: Latent variable scores as input for logistic regression
Hi,
I am pretty newbie for SmartPLS and also have the same problem with binary variable as DV in model. I have read that weighted PLS (WPLS) is now available in SmartPLS 3 to test a model with binary outcome variable, but for further guide I haven't found any references. Could you explain more about that procedure?
Should I first put all LV and their indicators and DV as well to estimate the model fit, then using LV scores to calculate the logistic regression on DV?
Many thanks in advance.
I am pretty newbie for SmartPLS and also have the same problem with binary variable as DV in model. I have read that weighted PLS (WPLS) is now available in SmartPLS 3 to test a model with binary outcome variable, but for further guide I haven't found any references. Could you explain more about that procedure?
Should I first put all LV and their indicators and DV as well to estimate the model fit, then using LV scores to calculate the logistic regression on DV?
Many thanks in advance.
-
- SmartPLS Developer
- Posts: 1284
- Joined: Tue Mar 28, 2006 11:09 am
- Real name and title: Dr. Jan-Michael Becker
Re: Latent variable scores as input for logistic regression
I am not sure whether the weighted PLS will help with this issue.
The general procedure would be the following:
Build you model with your normal constructs as independent latent variables and the binary response as single item construct. Estimate the PLS model. The results would be similar to a linear probability model. This is not optimal but it should capture at least the correlation structure between the binary response and the independent variables approximately. Then, extract the latent variable scores and estimate a logistic regression outside of SmartPLS.
The general procedure would be the following:
Build you model with your normal constructs as independent latent variables and the binary response as single item construct. Estimate the PLS model. The results would be similar to a linear probability model. This is not optimal but it should capture at least the correlation structure between the binary response and the independent variables approximately. Then, extract the latent variable scores and estimate a logistic regression outside of SmartPLS.
Dr. Jan-Michael Becker, BI Norwegian Business School, SmartPLS Developer
Researchgate: https://www.researchgate.net/profile/Jan_Michael_Becker
GoogleScholar: http://scholar.google.de/citations?user ... AAAJ&hl=de
Researchgate: https://www.researchgate.net/profile/Jan_Michael_Becker
GoogleScholar: http://scholar.google.de/citations?user ... AAAJ&hl=de