Calculate factor scores manually for additional time period

[Tina-19]

Hi All,

I am able to calculate factor scores manually for one time period, but
would like to calculate orthogonal factor scores for an additional time
period. Would that be possible by retaining the same grouping of variables
as in the original factor groupings?
Any information on this would be very helpful.

Thank you,

Moumita

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[William Dudley WNDUDLEY]

This is a follow up to a question about planned comparisons in and MANOVA.

I have a 2 (within) X 3 (between group) design which all factors are binary
We originally ran the analyses as two univariate ANOVA's but the reviewers wanted to see a MANOVA.
The two outcomes are different aspects of a larger instrument .

We are able in the univariate to demonstrate some weak but significant two way interactions (e.g. gender by school level (high school versus college).
We examined the nature of these interactions using  t-tests (thus comparing males and females college students on math and male females in high-school on math).

Although I have tried to figure out how to run these t-tests within the larger repeated measure ANOVA, I cannot figure out how to do the contrasts.
Can someone either show me how to get started on this (I am not suer if I use LMATRIX of MMATRIX) OR

Or perhpas I just stay with the t-tests (and if so if there a literature on this?)

thanks in advance

Here is the GLM for the analyses



GLM Math Verbal BY schoolgrp gender status
  /WSFACTOR=math 2 Polynomial
  /METHOD=SSTYPE(3)


William N. Dudley, PhD
Associate Dean for Research
The School of Health and Human Performance Office of Research
The University of North Carolina at Greensboro
126 HHP Building, PO Box 26170
Greensboro, NC 27402-6170
VOICE 336.2562475
FAX [phone]

[Art Kendall]

Assuming that you are creating scale scores, that the items were on the
same response scale, that you summed (averaged) only items loading over
something like .4 and not more than something like .25 on another
factor, simply use the same scoring key.
e.g.,
compute scale1 = mean(item1, item5, ...).
compute scale2 = mean(item11, item22 ...).

Art Kendall
Social Research Consultants

Tina wrote:

> Hi All,
>
> I am able to calculate factor scores manually for one time period, but
> would like to calculate orthogonal factor scores for an additional time
> period. Would that be possible by retaining the same grouping of variables
> as in the original factor groupings?
> Any information on this would be very helpful.
>
> Thank you,
>
> Moumita
>
> =====================
> To manage your subscription to SPSSX-L, send a message to
> [hidden email] (not to SPSSX-L), with no body text except the
> command. To leave the list, send the command
> SIGNOFF SPSSX-L
> For a list of commands to manage subscriptions, send the command
> INFO REFCARD
>
>
>

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[Ruben Geert van den Berg]

Dear Tina,
 
I'm not sure whether my response will be of any use whatsoever but a couple of months ago, I had the following situation: factor analysis was carried out on some 2500 respondents. Later, factor scores had to be computed on the spot for a couple of entirely new respondents without any modification of the original factor analysis results. I first believed that this was technically impossible but I later understood how SPSS (f### PASW!) does the trick:
 
Assuming Varimax rotation and no missing values (perhaps your could impute these if they are present), three matrices can be obtained:
 
1) The correlation matrix of the (observed) variables in the data which enter the analysis
2) The correlation matrix between these observed variables and the factors (this is simply the factor loading matrix)
3) The correlation matrix between the factors (in this case the factors are uncorrelated so this is an identity matrix)
 
By concatenating like this
 
1 2
2' 3

the correlation matrix among all observed variables AND factors can be constructed. Then regression analysis is carried out upon this correlation matrix and the predicted values are saved as factor scores (assuming method = regression, not Bartlett). When you run regression analysis after saving factor scores, you'll have a perfect Rsquared simply because the factor scores have been derived from this very regression analysis. The unstandardized betas that yield this perfect Rsquare enable you to manually calculate factor scores without any modification of the original PCA.
 
HTH,
 
Ruben v.d. Berg


 

> Date: Thu, 23 Apr 2009 10:47:11 -0400
> From: [hidden email]
> Subject: Calculate factor scores manually for additional time period
> To: [hidden email]
>
> Hi All,
>
> I am able to calculate factor scores manually for one time period, but
> would like to calculate orthogonal factor scores for an additional time
> period. Would that be possible by retaining the same grouping of variables
> as in the original factor groupings?
> Any information on this would be very helpful.
>
> Thank you,
>
> Moumita
>
> =====================
> To manage your subscription to SPSSX-L, send a message to
> [hidden email] (not to SPSSX-L), with no body text except the
> command. To leave the list, send the command
> SIGNOFF SPSSX-L
> For a list of commands to manage subscriptions, send the command
> INFO REFCARD

---

Express yourself instantly with MSN Messenger! MSN Messenger

[Fernando Cartwright]

Moumita,

The easiest method of doing this is to do the following (assuming you have saved the syntax from your previous exercise):
 
1. produce a variable, V, that indicates if the records are from the original time period or the new time period (use 1 for the old time period and 0 for the new time period)

2. If you are using a weight variable, calculate a new weight variable by multiplying the indicator variable, V, with your original weight variable. If you were not previously using a weight variable, then your new weight variable will be variable V. The syntax for this is: WEIGHT BY v.

3. rerun the same syntax that produced your original results.

Because the new cases have a weight of 0 for the calculation of the factor analysis coefficients, the coefficients used to produce the factror scores will be the same as before. However, since they are on the same data file, the estimated scores will be produced for them the same as for the other cases.

Regards,

Fernando Cartwright

Tina-19 wrote

Hi All,

I am able to calculate factor scores manually for one time period, but
would like to calculate orthogonal factor scores for an additional time
period. Would that be possible by retaining the same grouping of variables
as in the original factor groupings?
Any information on this would be very helpful.

Thank you,

Moumita