Dear all,
having estimated my PLS model, I find that for some items of my reflective constructs PLS returns negative item loadings.
For formative constructs theory suggests to eliminate items with signs non-conforming to the theory-derived causal directions.
Does the same hold for reflective constructs? Or is there a general problem with my model if negative loadings show for reflective items.
So far I considered eliminating the resepctive items also, for the same reason as proposed for formative constructs. How do you handle such situation?
Thank you very much for your answer in advance.
Best regards
Florian
How to interpret negative loadings of reflective items
- Diogenes
- PLS Super-Expert
- Posts: 899
- Joined: Sat Oct 15, 2005 5:13 pm
- Real name and title:
- Location: São Paulo - BRAZIL
- Contact:
Hi Florian,
Sometimes people use reversed items in their Likert Scale to reduce the halo effect, then there is an explanation for the negative loading of this item. In these cases, we recode the answers like this: 1 is changed to 7; 2 to 6; 3 to 5… (or new = 8 – old).
After the recoding process the sign of the outer loading will be positive and the reliability will be computed properly.
In your case, there are some suggestions:
- If the absolute value of this outer loading is low (< 0,6), and you have at least 3 other items, the elimination of this item is a good option;
- If the absolute value of this outer loading is > 0,7, this item is improving the validity and reliability. You could invert the scale to have positive outer loading and correct values for the reliability. And include a good explanation for the reason of this negative outer loading (interpretation.
Best regards,
Bido
Sometimes people use reversed items in their Likert Scale to reduce the halo effect, then there is an explanation for the negative loading of this item. In these cases, we recode the answers like this: 1 is changed to 7; 2 to 6; 3 to 5… (or new = 8 – old).
After the recoding process the sign of the outer loading will be positive and the reliability will be computed properly.
In your case, there are some suggestions:
- If the absolute value of this outer loading is low (< 0,6), and you have at least 3 other items, the elimination of this item is a good option;
- If the absolute value of this outer loading is > 0,7, this item is improving the validity and reliability. You could invert the scale to have positive outer loading and correct values for the reliability. And include a good explanation for the reason of this negative outer loading (interpretation.
Best regards,
Bido
-
- PLS Junior User
- Posts: 9
- Joined: Tue May 19, 2015 1:53 pm
- Real name and title: José Fuchs - Engeneering
Re: How to interpret negative loadings of reflective items
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 ?
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 ?
-
- PLS Junior User
- Posts: 9
- Joined: Tue May 19, 2015 1:53 pm
- Real name and title: José Fuchs - Engeneering
Re: How to interpret negative loadings of reflective items
the second - after running the PLS algorithm and looking for the report I note one of the latent constructs with a Cronbach Alfa around 0,55 but with composite reliability 0,75 and all of other indicators are good, including AVE and f square.
This is a very important construct for my model, so I would like to know if I can keep it in the model despite this Alfa not very good score.
The path final coeficient between the constructs after bootstrapping was 2,577 ( not so bad, I think... because it's a very exploratory model )
This is a very important construct for my model, so I would like to know if I can keep it in the model despite this Alfa not very good score.
The path final coeficient between the constructs after bootstrapping was 2,577 ( not so bad, I think... because it's a very exploratory model )
- 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: How to interpret negative loadings of reflective items
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.
Greetings,
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.
Greetings,
-
- PLS Junior User
- Posts: 9
- Joined: Tue May 19, 2015 1:53 pm
- Real name and title: José Fuchs - Engeneering
Re: How to interpret negative loadings of reflective items
I'm doing a mediation test according Hair et al (2014) and a problem emerges : the VAF test is made with the tail Betas.
I have 3 Betas between three constructs: Beta 12 = 0,561, beta 23 = 0,640 and Beta 13 = - 0,111.
When I calculate the VAF mediation test, this negative signal of Beta 13 implies in a result of no mediation, but the theory says that mediation occurs.
Can I have to run this VAF test without the negative signal, only with the absolute value of Beta 13 ?
In this case mediation is comproved.
Best Regards
josé Fuchs
I have 3 Betas between three constructs: Beta 12 = 0,561, beta 23 = 0,640 and Beta 13 = - 0,111.
When I calculate the VAF mediation test, this negative signal of Beta 13 implies in a result of no mediation, but the theory says that mediation occurs.
Can I have to run this VAF test without the negative signal, only with the absolute value of Beta 13 ?
In this case mediation is comproved.
Best Regards
josé Fuchs
- 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: How to interpret negative loadings of reflective items
Hi,
Negative beta (direct effect) indicates that the mediator variables have supression effect. In the calculation of value VAF will be greater than 100.
Regards,
Negative beta (direct effect) indicates that the mediator variables have supression effect. In the calculation of value VAF will be greater than 100.
Regards,