ssd sem results differ from raw data sem results stata

[Saul Gabriel]

Using two different datasets, I cannot get stata to give me comparable efa/sem results when using summary statistics vs. raw data. I am not sure what I am missing...
The raw data is from:
use http://www.stata-press.com/data/r15/sem_1fmm

The summary data and basic efa syntax are as follows:

clear
ssd init x1 x2 x3 x4
ssd set obs 500
ssd set correlations 1.0 \ .4537 1.0 \ .4491 .5091 1.0 \ .6525 .6801 .6975 1.0
ssd set means 99.518 99.954 99.052 94.474
ssd set sd 14.35402 14.1939 14.26395 70.11603

[OR]

clear
ssd init x1 x2 x3 x4
ssd set obs 500
ssd set means 99.518 99.954 99.052 94.474
ssd set sd 14.35402 14.1939 14.26395 70.11603
#delimit;
ssd set correlations
1.0 \
.4537 1.0 \
.4491 .5091 1.0 \
.6525 .6801 .6975 1.0 ;
#delimit cr

ssd status
ssd list

save "C:\Users\...\Desktop\_EFA.dta"

cd "C:\Users\...\Desktop"
use _EFA

factor x1-x4, ml factors(1)

*Question: what about using ssd or my code leads stata (v. 15.1) to produce not even remotely close results?

[Weiwen Ng]

You can certainly feed the factor command a correlation matrix. If you ran factor after inputting summary statistics data, though, I'm pretty sure it treats the data as if were real data. Here is a listing of the actual data after you input the SSD characteristics:

     +-----------------------------------------------------------+
     | _group   _type         x1        x2         x3         x4 |
     |-----------------------------------------------------------|
  1. |      1       1     99.518    99.954     99.052     94.474 |
  2. |      1       2   14.35402   14.1939   14.26395   70.11603 |
  3. |      1       3          1     .4537      .4491      .6525 |
  4. |      1       4      .4537         1      .5091      .6801 |
  5. |      1       5      .4491     .5091          1      .6975 |
     |-----------------------------------------------------------|
  6. |      1       6      .6525     .6801      .6975          1 |
     +-----------------------------------------------------------+

So, that's what the factor command is running an EFA on, and indeed, you see that you had 6 observations in your factor output. You want to input your correlations as a matrix, and use factormat instead.

[Saul Gabriel]

Thank you Weiwan. This indeed worked. I noticed that the sample size, using the previous matrix input method, was off...