Hello,
I am doing decomposition analysis with a survey data, and I am stuck.
I need to use svy to estimate numbers at the population level, so I use the below code.
wgt_all_byphase -> sampling weight variable. different for each year (2005 / 2014).
There are,
(2005) 32 strata & 200 PSUs
(2014) 25 strata & 384 PSUs
In my decomposition model, I am comparing a group in 2005 with a group in 2014.
I used Oaxaca command (there are more variables in the model, but I deleted them for simplicity).
wgt_dec -> outcome variable, dichotomous (1/0)
group_phase36 -> dichotomous (1/0), 1 if 2014, 0 if 2005.
domain_allage_f -> female=1
The code works. The question I have is for the decomposition results.
It seems like Stata considers both years in the same group and uses 57 strata and 584 PSUs, instead of 32/200 vs. 25/284, and decompose the difference though for logistic regression results, it separates two years and applies different sampling weights.
Is this a problem?
If yes, is there any way that I can avoid this issue?
And I wonder if it would be ok if I don't use svy option and use pweight & vce(cluster) instead to circumvent this issue.
I would appreciate any comments.
Thanks.
I am doing decomposition analysis with a survey data, and I am stuck.
I need to use svy to estimate numbers at the population level, so I use the below code.
Code:
use knhanes_0514, clear svyset psu [pweight=wgt_all_byphase], strata(kstrata)
There are,
(2005) 32 strata & 200 PSUs
(2014) 25 strata & 384 PSUs
In my decomposition model, I am comparing a group in 2005 with a group in 2014.
I used Oaxaca command (there are more variables in the model, but I deleted them for simplicity).
Code:
oaxaca wgt_dec bmi_over2, by(group_phase36) logit svy(,subpop(domain_allage_f)) noisily detail swap
group_phase36 -> dichotomous (1/0), 1 if 2014, 0 if 2005.
domain_allage_f -> female=1
The code works. The question I have is for the decomposition results.
It seems like Stata considers both years in the same group and uses 57 strata and 584 PSUs, instead of 32/200 vs. 25/284, and decompose the difference though for logistic regression results, it separates two years and applies different sampling weights.
Is this a problem?
If yes, is there any way that I can avoid this issue?
And I wonder if it would be ok if I don't use svy option and use pweight & vce(cluster) instead to circumvent this issue.
I would appreciate any comments.
Thanks.