mvdcmp: using iweight

[Francisco Rowe]

Hi,
I am using mvdcmp to conduct a regression-based decomposition: mvdcmp is a command developed by Powers, Yoshioka and Yun (2011) for Multivariate Decomposition for Nonlinear Response Models. The command is described to support specification of weights. However, it does not seem to be applying my specified iweights. When I run a regression model using iweight, the number of observations reported by Stata is 2,556. But when I use mvdcmp, the number of observations reported is 2,514 and R component of the decomposition is 0.0108 -which should instead be 0.833.

Any ideas if the command is applying the specified weights? or why does the reported R differ from the actual value of 0.833?

Powers, D, Yoshioka, H and Yun, M 2011, 'mvdcmp: Multivariate decomposition for nonlinear response models', The Stata Journal, vol. 11, no. 4, pp. 577-88.

Thanks in advance,

Francisco.

[Steve Samuels]

I know little about mvdcmp.

The help for mvdcmp states that it allows pweights and fweights. When I tried a logit command with iweights, the program stopped with an error "iweight not allowed" but regress did not give an error. I'm curious about the nature of your weight variable and why you require iweights.

More knowledgeable responses will have to come from other list members. You are not likely to interest them if you don't show the exact command you use and all of the Stata results. This is a request in the FAQ Section 12 (http://www.statalist.org/forums/help.) For readability, It is essential that you format the commands and output between code delimiters; push the "A" button (for Advanced Editing) at the top right of the message-edit window and hit the link symbol. See http://www.statalist.org/forums/help...ntent_advanced.