Dependent Variable - Continuous versus categoric

[Mostafa.Bakr]

I am using SMART PLS in order to analyze the impact of several variables grouped into 4 main constructs on the dependent variable which is the directness of distribution . The directness of distribution refers to a given manufacturer's interest to contact the retailers either through direct sales channels using own captive sales force and trucks or through indirect channels using intermediaries.

My dependent variable is a binary dependent variable . when i introduced the results to SMART PLS , the constructs weakly explained the variation in the dependent variable i.e. directness of distribution nearly 38%.

So i had to make a work around which was to estimate the probability of employing direct channels of distribution using Logistic Regression (LR). Then i used this estimated probability (continuous variable) to be tested using SMART PLS , which lead to significant improvement in the coefficient of determination R2 value which increase to 80% . is this an acceptable solution?
Thanks
Mostafa

[jmbecker]

I would rather do it the other way round.

1) Estimate your model with the dichotomous DV in PLS. (The results would be similar to a linear probability model.)
2) Use the latent variable scores from this model and the dichotomous DV in a logistic regression to assess the effect of the latent variables on you DV.

You should not use the estimated probabilities in the PLS model because these contain already information on the predictors.

[Mostafa.Bakr]

Dear Dr. Jan-Michael Becker

I would like to thank you for your advice

I did follow the recommended steps and I integrated the latent variable scores produced from PLS into Logistic Regression (LR) to assess their effect on my DV.

The results are quite interesting and now I have an overall high level summary coming from the latent variables summarzizing each the effect of several variable on my DV. This enables me now to make conclusions in a more meaningfull sense rather than assessing the impact of 15 plus variables on the DV.

Just to confirm that I followed the correct steps, I did the following:

1) I referred to "Smart PLS Report"
2) Under "Final Results"
3) I referred to "Latent Variable"
4) I extracted the scores prevailing in the table and imported to SPSS

Just to confirm I followed the right approach and accessed the right data .

Thanks
Mostafa

[jmbecker]

Sounds correct.

[josemesquita]

Dear Professor Becker, which reference may I quote regarding the use of latent variables scores to perform a logistic regression?

Thanks in advance,

Jose Mesquita