Hello,
I have a complex survey design and need to use svyset to set specifications, but also need to account for correlated errors due to repeated measures in pairs of years in my dataset.
Context: The MEPS data collecting process uses panels of participants. Each panel spans a 2-year time period, so Panel 19 would cover the same group of individuals from 2014-2015.
Each panel consists of 5 rounds of data collection, so in Panel 19 Rounds 1, 2, and ~3 take place in 2014 and Round ~3, 4, and 5 take place in 2015. Hopefully this makes sense, see figure below.
I am using the Medical Expenditure Panel survey's (MEPS) full-year datasets for 2007-2015. The full-year datasets (like 2015 shown in gray above): these only focus on 1 year and incorporate individuals from two overlapping panels (Panels 19 and 20). Because of the overlapping panels, I have some of the same people from 2014's full-year dataset also showing up in 2015's full year dataset (this would be the case for all of my previous years as well, I'm just using 2014 and 2015 as an example).
This creates a correlated error issue. I would like to use cluster robust standard errors to account for this, but cannot figure out how to adjust my syntax to include this. Currently I'm using: svyset VARPSU [pweight=SAQWTF], strata(VARSTR) singleunit(centered) || YEAR
but do not think this is the right route. Either way, this is not working because I do not have a finite population correction (FPC) Stata is returning a note saying: Stage 1 is sampled with replacement; further stages will be ignored for variance estimation.
I appreciate any assistance or feedback that might help. Thanks!
I have a complex survey design and need to use svyset to set specifications, but also need to account for correlated errors due to repeated measures in pairs of years in my dataset.
Context: The MEPS data collecting process uses panels of participants. Each panel spans a 2-year time period, so Panel 19 would cover the same group of individuals from 2014-2015.
Each panel consists of 5 rounds of data collection, so in Panel 19 Rounds 1, 2, and ~3 take place in 2014 and Round ~3, 4, and 5 take place in 2015. Hopefully this makes sense, see figure below.
I am using the Medical Expenditure Panel survey's (MEPS) full-year datasets for 2007-2015. The full-year datasets (like 2015 shown in gray above): these only focus on 1 year and incorporate individuals from two overlapping panels (Panels 19 and 20). Because of the overlapping panels, I have some of the same people from 2014's full-year dataset also showing up in 2015's full year dataset (this would be the case for all of my previous years as well, I'm just using 2014 and 2015 as an example).
This creates a correlated error issue. I would like to use cluster robust standard errors to account for this, but cannot figure out how to adjust my syntax to include this. Currently I'm using: svyset VARPSU [pweight=SAQWTF], strata(VARSTR) singleunit(centered) || YEAR
but do not think this is the right route. Either way, this is not working because I do not have a finite population correction (FPC) Stata is returning a note saying: Stage 1 is sampled with replacement; further stages will be ignored for variance estimation.
I appreciate any assistance or feedback that might help. Thanks!
Comment