|
Both MEPS and MCBS
use a multistage cluster probability sample as an efficient strategy
for data collection. Because it would be inordinately expensive
to travel the entire US to interview a random sample of individuals
or households across the United States, the multistage sampling
strategy begins by selecting representative geographic units for
sampling. These geographic primary sampling units are known as
PSUs. Strata within PSUs are then identified for random sampling,
allowing oversampling to assure adequate sample size of specific
populations of interest. The MCBS oversamples individuals under
age 65 (disabled) and age 85 or older (the oldest old).
This table shows the
distribution by age of individuals in the MCBS sample frame compared
to their distribution in the population.
Mid-1990's
Percent of Medicare Beneficiaries by Age
Group in the Entire Medicare Population and in MCBS
|
|---|
Age
Group
| Medicare
Enrollment
| MCBS
Sample Frame
|
|---|
0-44
| 4.5
| 7.8
|
45-64
| 12.3
| 8.9
|
65-69
| 23.9
| 19.1
|
70-74
| 21.9
| 15.4
|
75-79
| 16.5
| 16.6
|
80-84
| 11.2
| 17.3
|
85
and older
| 9.6
| 14.8
|
|
Compare the proportion
of individuals in the frame age 80-84 and 85+ to the proportion
in the population to see the result of oversampling. The NHIS,
from which the MEPS sample is selected, oversamples African Americans
and Hispanics. Beginning with the 1997 panel the MEPS oversamples
policy-relevant population subgroups. The subgroups initially
targeted for oversampling include adults with functional impairments,
children with functional limitations in their activities, individuals
aged 18-64 who are predicted to have high levels of medical expenditures,
and individuals with family income less than 200 percent of the
poverty level. These sampling strategies affect the variance of
estimates. Analyses need to take into account the potential design
effects of this sampling strategy. Click here
for more information on cluster and stratified samples.
Variables indicating
the PSU and the strata that a given respondent belonged to are
included in the dataset. When using the appropriate software to
analyze these survey data the researcher must specify both of
these variables (PSU and strata) in the programming statements.
|