|
|
 |
| Cluster/REPSE=BOOTSTRAP (w 3-way inte... |
 
|
| Message/Author |
|
|
|
|
1) Why do I get: The BOOTSTRAP option with weights is only allowed with TYPE=COMPLEX when replicate weights are present or when REPSE=BOOTSTRAP is requested?
CLUSTER IS TeachID;
Analysis: TYPE = COMPLEX;
REPSE=BOOTSTRAP;
BOOTSTRAP=500;
Model:
DV1 ON CoV1 CoV2 IV1 IV2 IV1xIV2;
[DV2] (b0);
DV2 ON CoV1 CoV2
IV1 (b1)
IV2 (b2)
IV1xIV2 (b3);
Insert relevant WITH statements
MODEL CONSTRAINT:
NEW(SIMP_LO SIMP_MED SIMP_HI);
!Numbers represent IV2 at -1SD, Mean, +1SD
SIMP_LO = b1 + b3*12.3306;
SIMP_MED = b1 + b3*16.4112;
SIMP_HI = b1 + b3*20.4918;
!5 is the min possible value of IV1 and 25 the max
PLOT(LOMOD MEDMOD HIMOD);
LOOP(XVAL,5,25,0.1);
LOMOD = (b0 + b2*12.3306) + (b1 + b3*12.3306)*XVAL;
MEDMOD = (b0 + b2*16.4112) + (b1 + b3*16.4112)*XVAL;
HIMOD = (b0 + b2*20.4918) + (b1 + b3*20.4918)*XVAL;
PLOT:
TYPE = plot2;
2) Is the plot statement correct? I used the following as a resource to write my syntax: http://www.offbeat.group.shef.ac.uk/FIO/mplusmedmod.htm |
|
|
|
|
1. The BOOTSTRAP option is available with TYPE=COMPLEX only with replicate weights.
2. It depends on what you want. Run it and see if you get what you want. |
|
|
| namer posted on Tuesday, February 07, 2017 - 7:39 am
|
|
|
Hello,
I am running a ZIP regression and making use of cluster-robust standard errors, whilst including various control variables all on a small experimental sample. However, I've been asked to bootstrap these models because they are underpowered. However, the complex models produce the same results as simple bivariate analyses.
Is it correct that given the use of TYPE=COMPLEX I will have to use replicate weights, but I would also need to generate them first?
Second, is this possible the cluster command?
Third, is this a fruitful approach to addressing the low power in the complex additional models? Or would you advise something else.
Thanks kindly in advance! |
|
|
|
|
1. Yes. You need replicate weights with bootstrap and complex. You can get that from Mplus by tricking the program: just run a regular regression (the ZIP regression is not available with complex and bootstrap) using this command
ANALYSIS: TYPE=COMPLEX; REPSE=BOOTSTRAP; BOOTSTRAP=50;
savedata: file is 1.dat; save=repw;
Then you will need to run Mplus 50 times for each weight and combine the results manually following section 2.1
https://www.statmodel.com/download/Resampling_Methods5.pdf
2. Yes
3. This is a research question I won't be able to give you a definite answer. I doubt that the answer is yes. I found this discussion enlightening
http://stats.stackexchange.com/questions/112147/can-bootstrap-be-seen-as-a-cure-for-the-small-sample-size |
|
|
| Back to top |
|
|
