|
|
 |
| How to NOT estimate intercepts of sca... |
 
|
| Message/Author |
|
|
| Anh Hua posted on Wednesday, September 16, 2020 - 2:18 pm
|
|
|
Dear all,
I just looked at the CFA example from mplus user guide and ran a simple one-factor CFA model, like this:
VARIABLE:
NAMES = PTRS_1 PTRS_2 PTRS_3 PTRS_4 PTRS_5 PTRS_6 PTRS_7 PTRS_8 PTRS_9 PTRS_10
PTRS_11 PTRS_12 PTRS_13 PTRS_14 PTRS_15 PTRS_16 PTRS_17 PTRS_18 PTRS_19 PTRS_20
PTRS_21 PTRS_22 PTRS_23 PTRS_24;
MISSING=.;
MODEL: one_factor BY PTRS_1-PTRS_24;
OUTPUT: standardized;
Mplus by default, seems to estimate the intercepts of the items whereas this doesn't happen in AMOS or lavaan. My question: how do I tell mplus NOT to estimate the intercepts of the items? I ask because I want the number of parameters to be consistent so I can compare the results across softwares.
Thank you so much for your help. |
|
|
|
|
In the Analysis command, you add
Model = Nomeanstructure; |
|
|
| Anh Hua posted on Thursday, September 17, 2020 - 2:40 am
|
|
|
Dear Dr. Mthen,
Thank you so much for your suggestion. Unfortunately, when I added Model=Nomeanstructure, I got a message saying:
*** WARNING in ANALYSIS command
MODEL=NOMEANSTRUCTURE is not allowed in conjunction with TYPE=MISSING.
Request for MODEL=NOMEANSTRUCTURE will be ignored.
I then looked further and saw that Dr. Linda Muthen, at one point on the forum, suggested this to someone with the same problem: "If you want to exclude the means, you cannot use TYPE=MISSING which is the default. Add LISTWISE=ON; to the DATA command." I tried it too just to see if it worked, but I got the following warning instead:
*** WARNING in ANALYSIS command
MODEL=NOMEANSTRUCTURE is not allowed in conjunction with INFORMATION=OBSERVED.
Request for MODEL=NOMEANSTRUCTURE will be ignored.
1 WARNING(S) FOUND IN THE INPUT INSTRUCTIONS
I preferred to use pairwise deletion as this is more optimal. How can I use all the available data in my analysis and be able to ask Mplus to turn off the default of estimating intercepts?
Thank you for any additional advice you can give. |
|
|
|
|
If you have missing data, you should analyze your means as well. That is the optimal approach using FIML, i.e., using all available information.
If you are concerned about different number of parameters, just subtract the number of means. |
|
|
| Anh Hua posted on Monday, September 21, 2020 - 8:57 am
|
|
|
Okay. Sounds like a good plan. Thank you, as always, for your help, Dr. Muthen!
Anh |
|
|
| Back to top |
|
|
