negative PLS indicators

This forum is closed, and read-only.
Locked
jfuchs
PLS Junior User
Posts: 9
Joined: Tue May 19, 2015 1:53 pm
Real name and title: José Fuchs - Engeneering

negative PLS indicators

Post by jfuchs »

I have two doubts

first - if I have a lot of indicators for a formative construct ( 52 ) and some of them are negative, It's better delete them before running the PLS algorithm in a model ?
User avatar
Hengkov
PLS Super-Expert
Posts: 1599
Joined: Sun Apr 24, 2011 10:13 am
Real name and title: Hengky Latan
Location: AMQ, Indonesia
Contact:

Re: negative PLS indicators

Post by Hengkov »

Hi,

Formative indicators are strongly advised not more than five to one construct. Because if more than that will cause a lot of problems that are difficult to overcome. You can not remove the formative indicator without a strong argument.

Regards,
jfuchs
PLS Junior User
Posts: 9
Joined: Tue May 19, 2015 1:53 pm
Real name and title: José Fuchs - Engeneering

Re: negative PLS indicators

Post by jfuchs »

I will explain better to you - In the beginning there were 4 latent constructs with about 6 to 9 indicators per construct, and they are related to another first order construct (they formed this construct), but when I checked the loadings the r² for the first construct was very low.
So I decided change the reflexive relationship in a formative one, because the my theory provides this support.
So now I have just one construct, a formative one, and the indicators of that 4 initial constructs changing to in order to belong to this new construct.
So now I have all of the indicators pointing to this formative construct, and some of them are negative.
Is it better to delete this negative indicators ?
I runned the PLS algorithm and there is no significant differences between the two results.
Best regards
User avatar
Hengkov
PLS Super-Expert
Posts: 1599
Joined: Sun Apr 24, 2011 10:13 am
Real name and title: Hengky Latan
Location: AMQ, Indonesia
Contact:

Re: negative PLS indicators

Post by Hengkov »

Hi,

Why do not you make a 2nd order constructs (1st order reflective, 2nd order formative)?
It will be better than by combining four reflective constructs into a formative construct. It's clear there will be a negative sign, because there are questions your indicators are not appropriate for formative.

Greetings,
jfuchs
PLS Junior User
Posts: 9
Joined: Tue May 19, 2015 1:53 pm
Real name and title: José Fuchs - Engeneering

Re: negative PLS indicators

Post by jfuchs »

OK Hengkov, but I think there is a problem when we have to run the PLS software, because it doesn't accept second order constructs - they stay with red color.
We have simply to link all the first order constructs indicators on the second order construct, becoming a big framework.
Am I sure ?
Greetings
User avatar
Hengkov
PLS Super-Expert
Posts: 1599
Joined: Sun Apr 24, 2011 10:13 am
Real name and title: Hengky Latan
Location: AMQ, Indonesia
Contact:

Re: negative PLS indicators

Post by Hengkov »

Yes, use repeated indicator approach combined with a two-stage approach for handling hierarchical component model.

Best regards,
jfuchs
PLS Junior User
Posts: 9
Joined: Tue May 19, 2015 1:53 pm
Real name and title: José Fuchs - Engeneering

Re: negative PLS indicators

Post by jfuchs »

More one doubt - if after run the PLS algorithm we obtain Cronbach Alpha 0,55 in a reflexive consctruct, but composite reliability 0,75 - this is an indicator that we have maybe some multidimensional effects in the measurement construct scale ?
The scale is Dinamism ( reflexive, with ten indicators in the beginning of the analisys, but with three indicators after delete all low loadings ), which measures customer behavior and climatic conditions in
wine producers.
User avatar
Hengkov
PLS Super-Expert
Posts: 1599
Joined: Sun Apr 24, 2011 10:13 am
Real name and title: Hengky Latan
Location: AMQ, Indonesia
Contact:

Re: negative PLS indicators

Post by Hengkov »

Cronbach Alpha is conservative compared with the Composite Reliability liberal so that its value will always be smaller.
If you combine multiple constructs into a single, maybe it just happened.
jfuchs
PLS Junior User
Posts: 9
Joined: Tue May 19, 2015 1:53 pm
Real name and title: José Fuchs - Engeneering

Re: negative PLS indicators

Post by jfuchs »

What's the meaning of negative signal on one of the paths in a PLS model after running the PLS Alghorithm ?
This means that the relationship is the oposite between the constructs ?
And in this case the VAF mediation test is made with this negative signal or not ? If use negative signal case, the mediation will not occurs, because the result will be more than 1.
User avatar
Hengkov
PLS Super-Expert
Posts: 1599
Joined: Sun Apr 24, 2011 10:13 am
Real name and title: Hengky Latan
Location: AMQ, Indonesia
Contact:

Re: negative PLS indicators

Post by Hengkov »

Hi,

Yes, if there is a negative beta (but desired positive), then the opposite effect. This may occur because the scale of measurement forget reversed. For the calculation of fixed VAF using negative sign (direct suppression effect occurs), so the value of VAF will be more than 100, but still full mediation concluded.

Best regards,
Locked