spss and smartpls huge correlation difference

[mohd alsuwaidi]

Hi everyone, 

I used smartpls and spss to find out the correlations. The smartpls correlations are totally different than the spss. The spss result is more logical. For example, my model is like IV--> different mediators --> DV and one of the mediators is negative emotions which is negatively correlated to IV and DV and other mediators in Spss. However, that is not the case in smartpls. Negative emotions is positively related to most variables. The following is the correlation results of negative emotion with other variables in the model from spss and smartpls: 

Spss result 

Neg Emotion 

-.161* 

-.137* 

-.303** 

1 

-.245** 

-.220** 

0.062 

-.224** 

-.308** 

Smartpls result 

NEG emotion 

0.89 

.382** 

.372** 

.414** 

-.276** 

-0.033 

0.01 

0.007 

-0.067 

0.016 

.348** 

0.001 

Does anyone knows why there is a big different in spss and smartpls correlations? The change is huge from sig. negative relationship to sig. positive ones.

[CMR]

Hi

I check this issue many times and I never had differences in the correlations between SPSS and SmartPLS.

-> Estimate your PLS path model; use the settings in the PLS algorithm to obtain standardized results

-> SmartPLS generates the standardized LV scores using SmartPLS

-> SmartPLS also returns the correlations of these LV scores

-> If you copy and paste the LV scores to SPSS and compute the correlation of the LV scores, they exactly match the SmartPLS outcomes.

 Reasons for differences are various (SPSS settings and procedures, different data used etc.) but if you follow this procedure and do everything correctly, no differences should occur.

 Please also check the www.smartpls.de discussion forum.

 Best regards

Christian Ringle

http://www.sagepub.com/books/Book237345

[Ned Kock]

Are you referring to the path coefficients or the correlations among LVs? I assume you are referring to path coefficients, which are associated with the links among LVs.

 

I don’t know if this is the case here, but sometimes the signs of the weights end up reversed, leading to paths that are negative but that should be interpreted as positive.

 

This is more common when PLS Mode B is used (the so-called “formative” mode), and less common with PLS Mode A. With the PLS regression algorithm, which is the default in WarpPLS, this rarely happens.

 

Nevertheless, this problem can still happen with WarpPLS, when indicators that are inversely correlated with one another are assigned to an LV; as in the example below.

 

Ind1: I feel bored

Ind2: I feel excited

LV: Perceived excitement

 

Simply staying away from PLS Mode B can solve this problem. If that doesn’t work, you can try to reverse or remove some of the indicators – the ones that are associated with odd weights. In the example above, the indicator that needs to be reversed is “Ind1”.

 

Also, you may want to use the anchor variable procedure (article and link below) to identify LV-indicator candidates. It explicitly relies on correlations among indicators, which are provided by WarpPLS.

 

Kock, N., & Verville, J. (2012). Exploring free questionnaire data with anchor variables: An illustration based on a study of IT in healthcare. International Journal of Healthcare Information Systems and Informatics, 7(1), 46-63.

 

http://www.scriptwarp.com/warppls/pubs/Kock_Verville_2012_IJHISI_FreeQuest.pdf

 

Ned

[Terans Gunawardhana]

I think, you have misunderstood correlations and path coefficients. Please recheck your correlation matrix from bootstrapping in SmartPLS and SPSS and get the same values. Path coefficients are different; if you need to find similar coefficients run PLS in SPSS. Both software provide same results. Only thing you have to compare like to like in any output.

Good Luck

Terans