Different results between SPSS and SAS for principal component analysis

[UniformVariance]

I'm trying to use SAS to replicate a principal component analysis in SPSS, but I get different results.

The data come from Task 18.4 in Chapter 18 of this website:

http://milton-the-cat.rocks/home/dsus_alex.html

I have attached both the SPSS and the SAS data files for this analysis.

I used the following code in SAS:

proc factor
     data = z1
          n = 3;
     ods output FactorPattern = z2;
run;

Here is the Factor Pattern matrix that I get.  I sorted the results by Factor1 in descending order.

Label	Factor1	Factor2	Factor3
Dramatic	0.83152	-0.00401	0.20573
Manipulative	0.75924	0.15946	0.15446
Arrogant	0.66808	0.14868	-0.05109
Eccentric	0.47807	0.41197	0.12763
Mistrustful	0.13197	0.70229	0.29149
Passive-Agressive	-0.11345	0.52411	-0.35355
Volatile	-0.24119	0.57074	0.50051
Perfectionist	-0.31997	0.00441	-0.06086
Detached	-0.34426	0.59224	-0.47549
Dependent	-0.40245	-0.2419	0.6669
Cautious	-0.73911	0.27281	0.25414

Here is the rotated component matrix from SPSS.

As you can see, there is no exact match.  Furthermore, some of the values are entirely different.  Why is that?

[Ksharp]

That is factor analysis ,NOT principal component analysis, Try  PRINCOMP Procedure .

[UniformVariance]

1) Doesn't PROC FACTOR perform PCA?

https://support.sas.com/documentation/cdl/en/statug/63347/HTML/default/viewer.htm#statug_factor_sect...

2) I searched "PROC PRINCOMP vs PROC PCA" on Google, but I don't see any results.  How do they differ?

[Ksharp]

That is factor analysis ,NOT principal component analysis, Try PRINCOMP Procedure

[PaigeMiller]

I think PROC FACTOR defaults to principal components analysis when you use it that way.

You can't compare rotated components to un-rotated components.

[UniformVariance]

Ah - thanks, Paige.  How, then, do I get the rotated components?

[PaigeMiller · Accepted solution]

PROC FACTOR documentation:

https://documentation.sas.com/?cdcId=pgmmvacdc&cdcVersion=9.4&docsetId=statug&docsetTarget=statug_fa...

Use the ROTATE= option

Although at that point, I don't know if the rotated factors from SAS will match the rotated factors from SPSS.

[UniformVariance]

That worked!  Thanks, Paige!

ods exclude all;
proc factor
	data = z1
		n = 3
		rotate = varimax;
	ods output OrthRotFactPat = z2;
run;
ods exclude none;



proc sort
	data = z2
		out = z3 (drop = variable);
	by descending 
		factor1;
run;



proc print
	data = z3 noobs;
run;

Label	Factor1	Factor2	Factor3
Dramatic	0.83344	-0.01713	-0.19708
Manipulative	0.78543	0.08145	-0.04691
Arrogant	0.67687	-0.04548	0.10403
Eccentric	0.54909	0.2988	0.1542
Mistrustful	0.27123	0.68113	0.24104
Passive-Agressive	-0.05357	0.17009	0.61706
Volatile	-0.09984	0.79007	0.0159
Detached	-0.27916	0.18104	0.76462
Perfectionist	-0.3187	0.02075	0.06406
Dependent	-0.37995	0.33545	-0.63903
Cautious	-0.65933	0.49985	0.02697

[ballardw]

And depending on the particular procedure you can get slightly different results inside the same system by changing the order of variables on model statements.

When I worked in an SPSS shop we had on process that we worked with to identify key elements. After we got a candidate list the approach had us change the order of variables several times because several of the variables would usually change levels of "importance" depending on which position in the code they occupied

(not role but position:

model y = a b c d e ; would yield different results for the variables than model y= c d e a b;  to use SAS generic code)

[PaigeMiller]

---

@ballardw wrote:

And depending on the particular procedure you can get slightly different results inside the same system by changing the order of variables on model statements.

---

This is not true for Principal Components Analysis in SAS. Although I have never done PCA in SPSS, there is nothing the PCA algorithm that was would cause an order of variable dependency, so I can't imagine any correctly programmed PCA algorithm to produce order effects, other than perhaps round-off error in the last decimal places.

Example:

title "ORDER OF VARIABLES: Age Height Weight";
proc princomp data=sashelp.class;
	var age height weight;
run;

title "ORDER OF VARIABLES: Weight Height Age";
proc princomp data=sashelp.class;
	var weight height age;
run;