Smart PLS versus Binary Regression (SPSS)

[Maximilian]

Hi,

I've been wondering about some contradicting outcomes, using SmartPLS versus SPSS (Binary Regression):

I'm aiming at validating a direct causal relation between an IV (expected value) and a DV (adoption), moderated by M (maintenance period). I therefore manipulated both, the expected value and the maintenance period.

1. When using Smart PLS, the direct relation between value and adoption is significant, so is the moderating effect of M, however, the direct relation between maintenance period and adoption is NOT significant.
2. When using SPSS (binary regression), ALL relations (both direct relations as well as the moderating effect) are significant.

(see the attached illustration 1)

Furthermore, when analyzing the same relation, but with supposed K.O. criteria in both manipulated stimuli, I get the following two soluation:

1. When using Smart PLS, all direct and moderating relations are significant
2: When using SPSS (binary regression), ONLY the moderating effect is significant (which, in my opinion, makes sense, since a long maintenance period is nut useful, when the expected value is too low, and vice versa)

(see attached illustration 2)

Since the findings tremendously differ regarding their interpretation, I was hoping to get some opinions about what to do.

Thanks a lot!

Max

[cringle]

When you use standardized data and the linear regression in SPSS, the results should match.

PLS requires metric data for the indicator. Where do you use binary data / why do you use binary regression?

Best regards
Christian

[Maximilian]

Hey Christian,

thanks so much for your answer! Well, since I'm using a binary DEPENDENT variable (0: adoption of product A, 1: adoption of product B), I thought using binary logistic regression seems to be a problem-adequate method. My indidcators are both metric (or at least metrically interpretable --> 7-point Likert scales). So - as far as I know - I meet the requirements of SmartPLS. However, the differences appear as illustrated above.

Any idea?

Thanks a lot,
Max

[cringle]

Well, PLS path modeling uses OLS regression - and not binary logistic regression. You need to extract the latent variable scores (maybe the unstandardized ones obtained from the IPMA) and then, run a binary logistic regression in SPSS (binary variable as a dependent; the latent variable scores as independent variables).

Best regards
Christian