Standardized or Unstandardized LV scores?

[yokemay]

Dear all,

I intend to test a two-way moderation on SPSS and currently, I'm a user of SmartPLS 2.0. I understand from previous postings in this forum that we need to copy-paste the LV scores from the PLS Algorithm results in SmartPLS so as to run regression and other analyses in SPSS.

Do I copy the standardized or unstandardized LV scores from the PLS Algorithm results?

Thank you very much.

Best,
May
PhD candidate in Information Management
Yuan Ze University, Taiwan

[Hengkov]

Hi,

Why you don't run a two-way interaction with SmartPLS? It was very easy to do than using SPSS. If using SPSS there are some problems that can interfere with the results of your analysis.

Greetings,

[yokemay]

Dear Hengkov,

Thank you for replying. The reason for using SPSS is that I can't create an interaction effect without establishing a connection from the Moderator to the Dependent variable in SmartPLS 2.0. Our concern is that this connection will affect the results of the interaction effect.

What is your advice or suggestions? Looking forward to your soonest reply as I need to report this finding as soon as possible.

Terima kasih,
May

[jmbecker]

You must include the moderator in your model. If you do not include the moderator your interaction effect will be biased and not meaningfull. There is no way that a model without the moderator makes sense.
There is a bunch of literature on this topic.

[yokemay]

Dear all,

APT as a moderator in SmartPLS

moderator in SmartPLS.JPG (23.26 KiB) Viewed 7260 times

Attached is a screenshot of our model in SmartPLS. We would like to test for moderation effect without the Moderator (APT) pointing (highlighted in yellow) to the variable, ATN. This is because we are afraid that this connection (the highlighted line) might affect the results of study.

I would be grateful if you could confirm that this connection doesn't have an impact on the analysis? I need this confirmation so as to inform my research advisor. Thank you very much.

Best,
May

[jmbecker]

The connection will most likely have an impact, but that is not bad. Without the moderator the interaction effect would simply be wrong. So you should include the moderator in your analysis!

If you get a significant interaction without the moderator but not with the moderator, then the interaction is probably just not there.

[Hengkov]

Hi,

I just have a simple answer to this problem. If you do not enter the variable M to Y, then this is a partial effects or unconditional. Conversely, if you enter X*M with X and M to Y, then this is the conditional effects. The simple moderation models Allows X's effect on Y to be a linear function of M. So, If you choose to retain XM in the models, X and M must be included as well, even if b1 and b2 are not statistically significant. Excluding X or M will bias the estimate of the moderation of X's effect by M. There are almost no circumstances in the which you would want to estimate Including a model of XM as a predictor without Also Including X and M. So when XM is in the models, keep the X and M in the models as well, Regardless of the outcome of an inferential test of b1 and / or b2.

Note: Using SPSS also must include X and M + X*M to test the moderation effects.

Best regards,

[yokemay]

Dear both,

Thanks for your prompt reply and brilliant comments. BTW, should I choose "standardize indicator values before multiplication" when I create moderating effect?

Standardize indicator values?

Standardize values.JPG (48.39 KiB) Viewed 7249 times

thanks again.

Best,
May

[jmbecker]

Yes, that is the preffered option.