Ron PrevostU.S. Census Bureau

NAWRS 46th Annual WorkshopAugust 22, 2006

The Medicaid Differential Project andPreliminary Results

2

Overview

• Phase I & II Study Results – Continuation of Dave Baugh’s Presentation

• Medicaid Study Elements - Phases III & IV

• The Census Bureau’s Administrative Records Infrastructure

• New and Emerging Applications of Administrative Records

3

Project Collaborators and Co-Authors

Collaborators:• US Census Bureau Collaborators:

• Sally Obenski• Ron Prevost• Dean Michael Resnick• Marc Roemer

• Office of the Assistant Secretary for Planning and Evaluation:• Rob Stewart• George Greenberg• Kate Bloniarz

Coauthors:• Centers for Medicare and Medicaid Services

• Dave Baugh• Gary Ciborowski

• State Health Access Data Assistance Center• Kathleen Thiede Call• Gestur Davidson• Lynn Blewett

• Srand Corporation• Jacob Klerman

4

Preliminary explanations we are reporting on today

Universe differences: • Enforce CPS group quarter definitions on MSIS

where we have administrative data address information

• Look for duplicate persons in different states or same state

Measurement error:• Link CPS respondents to MSIS data for CY 2000

to examine survey reports of enrollees• Understand the covariates of misreporting

5

Building a common universe

MSISFrame

CPS Sampling Frame

Group quarters, dead,

not a valid record, in two

states

Not a valid record

In CPS universe and in MSIS universe

6

Preparing MSIS data for comparison and linking to CPS

• Removed MSIS dual eligible cases defined as a “group quarter” by Census

• Ran the 2000 MSIS data through Census Bureau’s Person-ID validation system

• A record is “valid” if in the appropriate format and demographic data is consistent

• Removed duplicate valid records

• Removed those MSIS enrollees not enrolled in “full benefits”

7

Matching the CPS universe

Number of MSIS Medicaid Records in 2000:

44.3 M (total MSIS records)

- 1.5 M (records in more than one state or group quarter)

- 4.0 M (partial Medicaid benefits)

38.8 M (the target Medicaid total)

8

Sample loss in the 2000 MSIS and 2001 CPS linking

MSIS:9% of all MSIS records did not have a valid

record and were not eligible to be linked to the CPS

CPS: 6.1% (respondents’ records not validated)

+ 21.5% (respondents refused to have their ______ data linked)27.6% (total not eligible to be linked to MSIS)

9

The matched CPS-MSIS respondents with reported data only

12,341 CPS person records matched into the MSIS• 1,906 records had imputed or edited CPS data (15.5% of

total).

Focusing on only those with explicitly reported data:60% (responded they had Medicaid)

9% (responded some other type of public coverage but not Medicaid)

17% (responded some type of private coverage, but not Medicaid)

15% (responded they were uninsured)101% (over 100% due to rounding)

10

What factors are associated with measurement accuracy/error?

• Length of time enrolled in Medicaid

• Recency of enrollment in Medicaid

• Poverty status impacts Medicaid reporting but does not impact the percent reporting they are uninsured

• Adults 18-44 are less likely to report Medicaid enrollment

• Adults 18-44 more likely to report being uninsured

• Overall CPS rate of those with Medicaid reporting that they are uninsured is higher than other studies

• Overall CPS rate of those with Medicaid reporting Medicaid is lower than other studies

11

Explanations of the undercount revisited: work remaining to be done

Phase III: Measure Universe Differences:– Use 7 Medicaid state files with name and address

information to understand the impact of MSIS non-validation (one of the states is CA)

– Use enhanced MSIS data to further analyze the CPS sample frame coverage

Phase IV: Assess Measurement Error:– Compare measurement error in the CPS to the

National Health Interview Survey (NHIS) by linking the NHIS to the MSIS

– Compare measurement error in CPS to state survey experiments

12

Continued

Administrative and survey data processing, editing and imputation– Evaluate how well the CPS edits and imputations work at both

the micro level and the overall macro level

– Evaluate additional state-level Medicaid data

Survey sample coverage error and survey nonresponse bias– Link the address data from the 7 states to the Census Bureau’s

Master Address File to determine sample coverage problems

– Assess whether those addresses with a Medicaid enrollee are more likely to not participate in Census Bureau surveys

13

Study Conclusions

We have presented preliminary results that are subject to change after further investigation

At the moment we conclude that survey measurement error is playing the most significant role in producing the undercount– Some Medicaid enrollees answer that they have other

types of coverage and some answer that they are uninsured

The overall goal of the project is to improve the CPS for supporting health policy analysis– Especially refining estimates of the uninsured

14

The Census Bureau’s Mandate for Administrative Records Use

Title 13, Section 6: • Use administrative records information as

extensively as possible in lieu of conducting direct inquires  

Census Bureau Strategic Plan:• Reduce reporting burden and minimize cost to

taxpayer by acquiring and developing high-quality data from sources maintained by other government and commercial entities

15

Safeguarding Administrative Records at the Census Bureau

Consistent Application of Policies• To ensure that projects have the appropriate legal

authorization, comply with existing data agreements, and provide adequate controls to protect confidentiality and privacy

Administrative Controls• Numerous levels of approval • Need-to-know access• Removal of identifiable information• Administrative Records Tracking System• Security and confidentiality training

16

Census Bureau Programs that Use Administrative Records

Economic Directorate’s Business Register Intercensal Estimates Master Address File Small Area Income and Poverty Estimates Longitudinal Employer-Household Dynamics National Longitudinal Mortality Study Current and Periodic Demographic Surveys

17

The Census Bureau’s Administrative

Records Program Evolution

Today

Program begins

mid 1990s

July 1993 1996

AR Research staff created

Survey launched to gather info on potential AR files

Early 1990s

Statistical uses of AR

conference held

July 1993

1999/2000

Projects included AREX 2000 and the

1999 StARS prototype

1999

Centralized program emerges

Data Stewardship

program begins

2001

AR Test for 2000

Census

Race Model Addresses

Quality Concerns

PVS Increases Linking

Capacity

Infrastructure investments

allow new interagency

collaborations

2004

2002 2005

18

The Census Bureau’s Administrative Records Infrastructure

• STARS National Files including:- IRS personal tax returns, salary reports, and information

returns- Medicare & Medicaid- Indian Health Service- HUD files (TRACS, MTCS, PIH, and CHUMS)- Selective Service- Supplemental Security Income- Unemployment Insurance Wage Files (selected states)- MAF – National Change Of Address File

• Social Security Numident File • Creation of the Census Numident file• Look-up file that provides demographic data

19

Current and Emerging Census Bureau Demographic Applications (1)

Tested for assigning missing characteristics to census records

Used to assist coverage improvement operations target areas requiring follow-up

Enhanced the decennial Group Quarters frame with commercial and state address lists

Reducing ACS small area variance with model-assisted estimation based on AR

20

Current and Emerging Census Bureau Demographic Applications (2)

Using integrated data sets to better understand the differences between survey and administrative data Improve survey instrument design, editing, imputation, and

weighting

Provide statistics to improve an agency’s approach to measure program effectiveness

Provide measures to improve agency micro-simulation models

Identify areas for agency-targeted program outreach

21

Current and Emerging Census Bureau Demographic Applications (3)

Reacting to disaster and other near-real time requirements Katrina’s effect on the federal statistical system and our lack

of current response data highlighted need

Acquired the USPS National Change of Address File and FEMA’s emergency management and flood insurance files

Developing next generation StARS – near real-time measurements

22

Conclusions• Integrated data architectures are the future of

American statistics

• As the demand for data increases and budgets decrease data re-use many be the only cost-effective option

• Technical and policy related challenges must be addressed

• This approach will support evidence-based public policy research and decisions.

23

Contact Information

Ronald C. PrevostU.S. Census BureauWashington D.C. 20233-8100

Email: Ronald.C.Prevost@Census.GovPhone: (301) 763-2264Cell: (202) 641-2246