loadings value

Frequently Asked Questions about the SmartPLS software. Please refer to this section first, if you have problems with the software.
lopez
PLS Junior User
Posts: 1
Joined: Mon Oct 29, 2007 9:45 am
Real name and title:

loadings value

Post by lopez » Tue Oct 14, 2008 10:54 am

I would like to know which is the minimum acceptable value of the "loadings" (0.6 or 0.7).

Thank you

User avatar
Diogenes
PLS Super-Expert
Posts: 905
Joined: Sat Oct 15, 2005 5:13 pm
Real name and title:
Location: São Paulo - BRAZIL
Contact:

Post by Diogenes » Tue Oct 14, 2008 5:18 pm

Hi Miguel,
It is recommended 0,7 because 0,7^2 = 0,49
49% (almost 50%) of the variance of the indicator is explained by its LV.

Remenbering that AVE also should be > 50% (all loading = 0,7; the AVE will be 49%).

In practical cases we have some indicators lower than 0,7, but others bigger than 0,7, with AVE > 0,5;
in this case is common to keep the indicator even when its loading < 0,7, mainly in cases where we have just 3 or 4 indicators.

Best regards.
Bido

NitinJain
PLS Junior User
Posts: 5
Joined: Mon Jan 21, 2008 9:45 pm
Real name and title:

Can Outer Loadings be negative for a reflective indicator

Post by NitinJain » Fri Oct 24, 2008 7:21 am

Hi,

Can Outer Loadings be negative for a reflective indicator.
In my case 3 out 6 indicators for a reflective dimension have a negative outler loading.

Thanks,
Nitin

User avatar
Diogenes
PLS Super-Expert
Posts: 905
Joined: Sat Oct 15, 2005 5:13 pm
Real name and title:
Location: São Paulo - BRAZIL
Contact:

Post by Diogenes » Fri Oct 24, 2008 11:37 am

Hi Nitin,
It could be a problem because this has a influence in the signal of the structural paths (could change the signal).
I suggest that:
1) Compare these results with the simple correlations between these indicators (the signals should be the same)
2) You could invert the scale of some indicators to have all of them with positive correlations.
3) After that, the results in SmartPLS will be more interpretable.
Best regards.
Bido

naggi
PLS Junior User
Posts: 5
Joined: Thu Oct 23, 2008 10:39 pm
Real name and title:

Re: loadings value

Post by naggi » Wed Nov 12, 2008 9:07 pm

lopez wrote:I would like to know which is the minimum acceptable value of the "loadings" (0.6 or 0.7).

Thank you
And these results should be taken from the "Cross loading" under "Quality criteria" section in the the regular (not Bootstrap) results? Just to be sure!

Thank you,

Pankaj

User avatar
Diogenes
PLS Super-Expert
Posts: 905
Joined: Sat Oct 15, 2005 5:13 pm
Real name and title:
Location: São Paulo - BRAZIL
Contact:

Post by Diogenes » Wed Nov 12, 2008 10:19 pm

Hi Nitin,
the simple correlations between the indicators could be assessed in Excel or SPSS.
The "outer loadings" or "cross loadings" will show the same results that you could see in the model.
Best regards,
Bido

chinasun
PLS User
Posts: 14
Joined: Thu Mar 20, 2008 1:41 pm
Real name and title:

Post by chinasun » Fri Nov 14, 2008 3:14 am

Diogenes wrote:Hi Nitin,
It could be a problem because this has a influence in the signal of the structural paths (could change the signal).
I suggest that:
2) You could invert the scale of some indicators to have all of them with positive correlations.

Bido
professor
How to invert the scale of indicators that have negative sign in PLS?

Thanks in advance!

User avatar
Diogenes
PLS Super-Expert
Posts: 905
Joined: Sat Oct 15, 2005 5:13 pm
Real name and title:
Location: São Paulo - BRAZIL
Contact:

Post by Diogenes » Fri Nov 14, 2008 11:18 am

Hi Sun,

with 5 point Likert scale, to have 1 --> 5; 2 --> 4...
Just do "New value = 6 - Old value"

with 7 point Likert scale, to have 1 --> 7; 2 --> 6...
Just do "New value = 8 - Old value"

With numeric value (measured variable) wwe could multiply the values by (-1).

Best regards,
Bido

chinasun
PLS User
Posts: 14
Joined: Thu Mar 20, 2008 1:41 pm
Real name and title:

Post by chinasun » Sat Nov 15, 2008 3:14 pm

Thank you , professor
Could you tell me the principal of inverting the numerical values,
or provide some papers about it ?

I am looking forward. Thanks in advanced!

sterdelias
PLS Junior User
Posts: 1
Joined: Sun Apr 12, 2009 6:22 am
Real name and title:

loadings

Post by sterdelias » Mon Nov 30, 2009 8:20 am

is it possible for loadings to exceed 1 ?

User avatar
Diogenes
PLS Super-Expert
Posts: 905
Joined: Sat Oct 15, 2005 5:13 pm
Real name and title:
Location: São Paulo - BRAZIL
Contact:

Post by Diogenes » Thu Jan 21, 2010 4:01 pm

Hi,

1) Some references about reverse coded item:
Nunnally, J. C. & Bernstein I. H. (1994). Psychometric Theory. 3rd Ed. New York: McGraw-Hill, Inc. (see p.326).
Netemeyer, R. G., Bearden, W. O., & Sharma, S. (2003). Scaling Procedures: Issues and Applications. Thousand Oaks: Sage Publications. (see p.99).
DeVellis, R. F. (2003). Scale Development: theory and applications. 2nd Ed. Thousand Oaks: Sage Publications. (see pp.91-92).
Spector, P. E. (1992). Summated Rating Scale Construction: an introduction. Newbury Park: Sage Publications. (see p.22).

2) Outer loading usually is standardized (like correlations between LV and its reflective indicators), for this reason, it is not expected that its value be greater than 1.
If you are using formative indicators, the outer weights could have values greater than 1 because the multicollinearity between them.

Best regards,

Bido

User avatar
Diogenes
PLS Super-Expert
Posts: 905
Joined: Sat Oct 15, 2005 5:13 pm
Real name and title:
Location: São Paulo - BRAZIL
Contact:

Post by Diogenes » Sun Jan 31, 2010 10:41 pm

Hi,

A classical reference about this:

LIKERT, Rensis. 1932. A Technique for the Measurement of Attitudes. Archives of Psychology, n.140, pp.1-50.
That was republished in Spanish in 1976:
WAINERMAN. C. H. (compil.). 1976. Escalas de medición em ciências sociales. Buenos Aires: Ediciones Nueva Visión. pp.199-260.

He recommends the inversion of the scale when the item-total correlation is negative (p252).

Best regards,

Bido

Melvin
PLS Junior User
Posts: 3
Joined: Thu Jan 21, 2010 8:03 am
Real name and title:

Post by Melvin » Mon Mar 22, 2010 3:20 am

Diogenes wrote:Hi Miguel,
It is recommended 0,7 because 0,7^2 = 0,49
49% (almost 50%) of the variance of the indicator is explained by its LV.

Remenbering that AVE also should be > 50% (all loading = 0,7; the AVE will be 49%).

In practical cases we have some indicators lower than 0,7, but others bigger than 0,7, with AVE > 0,5;
in this case is common to keep the indicator even when its loading < 0,7, mainly in cases where we have just 3 or 4 indicators.

Best regards.
Bido
Hi, Professor Diogenes Bido,

Since, i'm not good at statistic. now i know the standard and i wanna know about others like AVE and T-value's Standard. And from where that i can get the reference about support such loading is acceptable. can professor give me some reference paper. thank you

Best regards
Melvin

iris_afandiphd
PLS Senior User
Posts: 28
Joined: Thu Jun 24, 2010 5:38 am
Real name and title:

Post by iris_afandiphd » Wed Jan 05, 2011 4:37 am

Hi Prof. Bido,

What happen if the outer loadings for INDICATORS are below 0.7?, what was the problem actually?, it is caused by most of responses across constructs are 1,2,3 , which group as Strongly Disagree etc..

Thanks

haslindar
PLS Junior User
Posts: 8
Joined: Sat Jan 30, 2010 3:40 am
Real name and title:

Post by haslindar » Thu Jan 06, 2011 2:58 am

Hi Prof Bido,

What happen if one of the variable's AVE is <0.5 while others are >0.5.

Thanks.

Post Reply