| Re: Disability in HR with characteristics from IR and MR [message #26890 is a reply to message #26888] |
Wed, 17 May 2023 08:00  |
Bridgette-DHS
Messages: 3043
Registered: February 2013
|
Senior Member |
|
|
Following is a response from Senior DHS staff member, Tom Pullum:
Although matching may seem like a good strategy, I would not recommend it. You would be subsampling the respondents who do not have a disability. One issue then is that there would be multiple potential matches, and someone else who tried the same thing might match different respondents and get different results. That is, there is a problem of replicability. A second issue is that by keeping respondents with a disability and an equal number of respondents without a disability, you would be throwing out approximately 98% of the sample. You would lose a lot of statistical power, more than what you would have gained from the matched pairs strategy. I recommend just using logit regression, with utilization of services as a binary outcome variable, disability as the predictor of interest, and the variables you describe as control variables. District is nested in state, and there are a huge number of districts, so you may want to have a random effect for district.
|
|
|
|
The DHS Program User Forum
Members
Search
Help
Register
Login
Home
