Saturday, December 16, 2017

Analyzing MEPS Survey Data in STATA

Analyzing MEPS Survey Data in STATA

Question:    One analyst examines MEPS data on total health expenditures by taking the mean of all observations without considering any differences across observations in the sample.   A second analyst uses a statistical procedure, which incorporates information on the MEPs sample.

The MEPS survey includes stratification, clustering, multiple stages of selection, disproportionate sampling, and reflects both adjustments for survey non-response and controls from the Current Population Survey.

Present the code the two analysts used to estimate means of key expenditure and key health care utilization variables using STATA.

How do point estimates of the mean expenditure and mean utilization variables differ for the two procedures?

How do standard error estimates differ for the two procedures?

Discuss implication on the choice of procedures.

Code for two procedures:

The STATA code used to estimate simple averages for the expenditure variables is presented below.

mean TOTEXP15 ERTEXP15 IPTEXP15 RXEXP15 DVTEXP15

The STATA code needed for the complex survey design involves a statement declaring the survey design and a statement invoking the procedure.

svyset varpsu [pweight=PERWT15F], strata(varstr) vce(linearized) singleunit(missing)

svy linearized : mean TOTEXP15 ERTEXP15 IPTEXP15 RXEXP15 DVTEXP15
Results for Expenditures and Utilization Variables:


Impact of MEPS Survey Design on Expenditure and Utilization Estimates
Expenditure Variables
Random Unweighted Sample
MEPS Survey Design
Mean
Std. Err.
Mean
Std. Err.
Total
4171.1
74.6
4977.6
134.4
ER
190.7
6.6
212.2
12.6
In Patient
1146.9
49.9
1326.0
79.1
RX
965.7
24.7
1160.5
48.4
Dentist
241.2
4.7
298.4
7.6
Utilization (Number of Visits)
Random Unweighted Sample
MEPS Survey Design
Mean
Std. Err.
Mean
Std. Err.
ER
0.202
0.0033
0.203
0.0055
In Patient
0.082
0.0021
0.091
0.0035
RX
9.328
0.1078
10.611
0.2112
Dentist
0.793
0.0080
0.952
0.0166


A second table comparing random unweighted means to survey means for all variables and random unweighted standard errors to survey standard errors for all expenditure and utilization variables was created to facilitate analysis.





Comparisons of Random Unweighted Statistics to
MEPS Survey Design Statistics



Random Mean to Survey Mean
Random Standard Error to Survey Standard Error
Total
1.19
1.80
ER
1.11
1.91
In Patient
1.16
1.58
RX
1.20
1.96
Dentist
1.24
1.63
Utilization Variables
Random Mean to Survey Mean
Random Standard Error to Survey Standard Error
ER
1.00
1.67
In Patient
1.10
1.69
RX
1.14
1.96
Dentist
1.20
2.07



Observations:

The expenditure and utilization means are larger when the survey design information is included.  The range of difference is 11 to 24 percent for expenditure variables and 0 to 20 percent for utilization variables.

The standard error of all estimates is much larger when the MEPS survey design procedure is used.  The differences range from 58 percent to 96 percent for expenditure variables and from 67 to 107 percent for utilization variables.

Concluding Thought: Analysts who use simple averages rather than averages that incorporate information in the MEPS survey design will understate health care expenditure and utilization.   More importantly, the simple averages understate the standard error or overstate the precision of their estimates.   This could lead analysts to incorrectly reject a null hypothesis.


Authors Note: I am using the MEPS databases to analyze health care trends and provide insight into different policy proposals.   The following post looks at data from both the 2005 and 2015 MEPS surveys.

The statistics show that growth in RX expenditures have outpaced growth in total health care expenditures.  Interestingly, there was little increase in the utilization of prescription drugs. 





This analysis leads me to the conclusion that it will be hard to control health care costs and fix insurance markets without restraining growth in drug prices.

No comments:

Post a Comment