Is it simply taking average of the values of all its reflective indicators? I have read sth. like this in this forum, and would like to check its validity..
Thank you a lot for your time n kind help!
Regards,
Esther :)
how to calculate LV(with reflective indicators) score in PLS
-
estherxing
- PLS User
- Posts: 14
- Joined: Mon Nov 06, 2006 12:23 pm
- Real name and title:
-
Diogenes
- PLS Super-Expert
- Posts: 899
- Joined: Sat Oct 15, 2005 5:13 pm
- Real name and title:
- Location: São Paulo - BRAZIL
- Contact:
Hi Esther,
to compute LV scores it doesn´t matter if the indicators are reflectives or formatives !
STANDARDIZED
1) Standardize the MVs (z = (xi - mean)/ standard deviation)
2) Compute LV scores standardized using "Outer weights"
LV score = weight 1 * x1 standardized + weight 2 * x2 stand ....
UNSTANDARDIZED
1) It´s cool if you want LV scores in the same unit as yours MVs.
2) Now you have to use "Measurement model", observe that the sum of these weights is equal 1 (this is why some people say that LV score is a weighted sum of its indicators).
3) LV score = weight 1 * x1 raw + weight 2 * x2 ....
Best regards
Bido
to compute LV scores it doesn´t matter if the indicators are reflectives or formatives !
STANDARDIZED
1) Standardize the MVs (z = (xi - mean)/ standard deviation)
2) Compute LV scores standardized using "Outer weights"
LV score = weight 1 * x1 standardized + weight 2 * x2 stand ....
UNSTANDARDIZED
1) It´s cool if you want LV scores in the same unit as yours MVs.
2) Now you have to use "Measurement model", observe that the sum of these weights is equal 1 (this is why some people say that LV score is a weighted sum of its indicators).
3) LV score = weight 1 * x1 raw + weight 2 * x2 ....
Best regards
Bido
-
estherxing
- PLS User
- Posts: 14
- Joined: Mon Nov 06, 2006 12:23 pm
- Real name and title:
Thank you very much Prof. Bido! :)
But I realized measurement model specification is not available in M3, even if I added it in from 'preference' manual. May I know is it my wrong operation or the feature is not in place? However, there is a table in the report named 'Latent Variable Scores (unstandardised)', can I use those values as LV scores (if I want LV score is in same scale as MV) directly?
The purpose for me to get the LV score is to manually calcuate another variable score. In my framework, construct A, B and C each is measured by a set of reflective indicators, and construct D is theoretically calculated as weight1*score of LV A+weight2*score of LV B+weight3*score of LV C. Because this kind of calculation is not readily support in SmartPLS, thus I want to calculate it manually because I need to test the relationship of D with other LV in the model.
However, there is another problem relating to this. With different structural models, the scores of LV A, B and C (read from table 'Latent Variable Scores (unstandardised)') are different. So in this case, my way of calculating LV D seems not working. May I know do you have any suggestions regarding to my problem then, i.e. how to model LV D in the model?
Thank you a million!!!
But I realized measurement model specification is not available in M3, even if I added it in from 'preference' manual. May I know is it my wrong operation or the feature is not in place? However, there is a table in the report named 'Latent Variable Scores (unstandardised)', can I use those values as LV scores (if I want LV score is in same scale as MV) directly?
The purpose for me to get the LV score is to manually calcuate another variable score. In my framework, construct A, B and C each is measured by a set of reflective indicators, and construct D is theoretically calculated as weight1*score of LV A+weight2*score of LV B+weight3*score of LV C. Because this kind of calculation is not readily support in SmartPLS, thus I want to calculate it manually because I need to test the relationship of D with other LV in the model.
However, there is another problem relating to this. With different structural models, the scores of LV A, B and C (read from table 'Latent Variable Scores (unstandardised)') are different. So in this case, my way of calculating LV D seems not working. May I know do you have any suggestions regarding to my problem then, i.e. how to model LV D in the model?
Thank you a million!!!
-
viswadatta
- PLS Expert User
- Posts: 139
- Joined: Wed Jul 05, 2006 1:43 pm
- Real name and title:
- Location: Coimbatore, India
LV scores
LV scores need not be computed. They are available in the reports of the output itself
-
estherxing
- PLS User
- Posts: 14
- Joined: Mon Nov 06, 2006 12:23 pm
- Real name and title:
Hi Vivek,
Do you mean the Latent Variable Score (Unstandardized)? I have tested with different structural models, the scores differ in each model though the difference is not big. In this case, can i safely take scores measured in one of the models as its score to solve the problem i mentioned above?
Thank you for your time!
Do you mean the Latent Variable Score (Unstandardized)? I have tested with different structural models, the scores differ in each model though the difference is not big. In this case, can i safely take scores measured in one of the models as its score to solve the problem i mentioned above?
Thank you for your time!
-
jmbecker
- SmartPLS Developer
- Posts: 1284
- Joined: Tue Mar 28, 2006 11:09 am
- Real name and title: Dr. Jan-Michael Becker
Hi,
Of course you probably get different results for different structural models.
The scores of the LV’s are finally calculated as weighted mean of the indicators, the way Prof. Bido described.
However the calculation of the weights depends also on the structural model and therefore the final calculation of the LV scores. This is due to the iterative two-step estimation process of the scores (outer/inner approximation) till convergence of the weights. For further information on this topic the related literature (especially Tenenhaus et al. viewtopic.php?t=21) is hardly recommended.
Of course you probably get different results for different structural models.
The scores of the LV’s are finally calculated as weighted mean of the indicators, the way Prof. Bido described.
However the calculation of the weights depends also on the structural model and therefore the final calculation of the LV scores. This is due to the iterative two-step estimation process of the scores (outer/inner approximation) till convergence of the weights. For further information on this topic the related literature (especially Tenenhaus et al. viewtopic.php?t=21) is hardly recommended.
-
estherxing
- PLS User
- Posts: 14
- Joined: Mon Nov 06, 2006 12:23 pm
- Real name and title:
Dear Prof. Bido,
Thank you for your recommendation! But my problem is that construct D is not second order construct, but rather, it is at same 'level' as construct A, B and C, and calculated as D=weight1*A+weight2*B+weight3*C, where weight1, 2 and 3 are given by respondents in each case.
So what I have done is to model construct A, B and C as second order construct and calculate their scores using their respective second order model. Meaning, I construct a single model using A1-3 (sub-dimensions of A) as first layer and A as second layer, and re-assign indicators of A1-3 to A following Wold (1982). I record down the score value for A. And I do the same thing to B and C. After that, using Excel to manually calculate the score of D. Then use the calculated score of D to test its relationships with other constructs in my framework.
May I know is this a valid approach?
Thank you so much for your kind help!
Fruitful day,
Esther :)
Thank you for your recommendation! But my problem is that construct D is not second order construct, but rather, it is at same 'level' as construct A, B and C, and calculated as D=weight1*A+weight2*B+weight3*C, where weight1, 2 and 3 are given by respondents in each case.
So what I have done is to model construct A, B and C as second order construct and calculate their scores using their respective second order model. Meaning, I construct a single model using A1-3 (sub-dimensions of A) as first layer and A as second layer, and re-assign indicators of A1-3 to A following Wold (1982). I record down the score value for A. And I do the same thing to B and C. After that, using Excel to manually calculate the score of D. Then use the calculated score of D to test its relationships with other constructs in my framework.
May I know is this a valid approach?
Thank you so much for your kind help!
Fruitful day,
Esther :)