Call Girls Begur Just Call 👗 7737669865 👗 Top Class Call Girl Service Bangalore
Who gets it right
1. Who gets it right? Characteristics
associated with accurate reports of
health insurance coverage
Kathleen Call, State Health Access Data Assistance Center
Angela Fertig, University of Minnesota
Joanne Pascale, US Census Bureau
Don Oellerich, Office of the Assistant Secretary for Planning and Evaluation, HHS
AcademyHealth Research Meeting
Seattle, WA
June 25, 2018
2. Goals of this study
Describe correlates of accurate reports of insurance
coverage in two commonly used census surveys:
Current Population Survey ASEC (CPS)
American Community Survey (ACS)
Identify variation in correlates of accurate reporting of
coverage by
type of insurance (public or private) and
survey (ACS and CPS)
3. Why do correlates of accuracy matter?
Results can inform
Survey design
Editing or imputation routines
Adjustments to population estimates of coverage for policy
simulation and evaluation
4. Who gets it right?
What is known is limited to Medicaid reporting
Most accurate:
Adults reporting for children vs adults
Low income, unemployed, low education
Shared coverage
i.e., respondent shares same coverage as other HH members
Received medical care
Recency, intensity of coverage
Here we expand to private insurance
5. Data: CHIME validation study
Start with enrollment records from a private health plan
that offers multiple coverage types
Medica Health Plan (MHP) in Minnesota
Use records as sample and randomly assign to different
survey treatments
Current Population Survey ASEC (CPS)
American Community Survey (ACS)
Compare estimates/indicators of coverage type:
Survey estimates versus enrollment records
Difference in surveys and records across CPS and ACS
6. CHIME survey methods
15-minute phone survey conducted in Spring, 2015
Content: questions from both CPS and ACS:
Demographics
Labor force
Government program participation (food stamps, WIC, etc.)
Health insurance randomization
Stratified sample: oversampled public, undersampled ESI weight
data to Medica population totals
22% response rate (AAPOR RR4)
Data collected on all household members
Individuals in surveys matched to enrollment records: at least one
person matched in 87% of households
Final matched dataset: 3,823 people
ACSCPS
1,989 received CPS
1,834 received ACS
7. Potential correlates of accurate reporting
From CHIME survey:
Covered individual characteristics
Age, health status
Any services in past 6 months (claims data, public only)
Respondent characteristics
Gender, education and employment status, employer size,
family income, and
Policy holder status (claims data)
Family/HH characteristics
Income as % poverty
Any SSI/TANF or food stamp participation (Medicaid only)
8. Potential correlates continued
From administrative records:
Complexity of survey reporting task
Shared coverage
Proxy-report in multi-person HH w/ different coverage
Proxy-report in multi-person HH w/ same coverage
self-report in multi-person HH
Self-report in one-person HH
Recency of coverage
past 6 mos, 7-17 months, 18 months or more
Receipt of subsidy (marketplace only)
9. Reporting accuracy by insurance type
and survey treatment CPS
ACS
78.4%
85.6%*
62.9%
83.7%* 83.2% 83.2%
77.1%
69.5%*
0.0%
10.0%
20.0%
30.0%
40.0%
50.0%
60.0%
70.0%
80.0%
90.0%
CPS
n=629
ACS
n=527
CPS
n=178
ACS
n=153
CPS
n=496
ACS
n=533
CPS
n=331
ACS
n=327
Non-Group Marketplace Medicaid MinnesotaCare
* Indicates a significant difference between CPS and ACS p < .05 or better.
10. 0.0 1.0 2.0 3.0 4.0 5.0 6.0 7.0 8.0 9.0 10.0 11.0 12.0 13.0 14.0 15.0 16.0 17.0 18.0 19.0 20.0
Age less than 19 vs age 35+
Age 19-34 vs age 35+
Health status fair/poor vs excellent/v.good/good/D/R
^Had any health care claims in last 6 months
Male vs female
High school or less vs bachelor or higher
Some college, associate vs bachelor degree or higher
Less than full time, full year vs full time, full year in 2014
Not working/D/R vs full time, full year in 2014
Employer size >50 vs 50 or fewer
^Respondent is policyholder
Family income in 2014 <200% vs > 200% FPG/D/R
Household receives TANF or SSI
Household receives SNAP
^Self report in one person HH vs REF
^Self report in multi-person HH vs REF
^Proxy report in multi-person HH with same coverage vs REF
^Missing (respondent didn't match claims data)
^PIT coverage obtained in last 6 months vs 18+ months ago
^PIT coverage obtained 7-17 months ago vs 18+ months ago
^Received a premium subsidy
CPS
ACS
Odds of accurate Medicaid reporting
Personal-level characteristics
Respondent-level characteristics
Family-level characteristics
Task complexity
REF=Proxy report in multi-person HH with different coverage
^ Based on administrative records data; all other indicators are from survey data.
LOWER ODDS OF
ACCURATE REPORTING
HIGHER ODDS OF
ACCURATE REPORTING
11. 0.0 1.0 2.0 3.0 4.0 5.0 6.0 7.0 8.0 9.0 10.0 11.0 12.0 13.0 14.0 15.0 16.0 17.0 18.0 19.0 20.0
Age less than 19 vs age 35+
Age 19-34 vs age 35+
Health status fair/poor vs excellent/v.good/good/D/R
^Had any health care claims in last 6 months
Male vs female
High school or less vs bachelor or higher
Some college, associate vs bachelor degree or higher
Less than full time, full year vs full time, full year in 2014
Not working/D/R vs full time, full year in 2014
Employer size >50 vs 50 or fewer
^Respondent is policyholder
Family income in 2014 <200% vs > 200% FPG/D/R
Household receives TANF or SSI
Household receives SNAP
^Self report in one person HH vs REF
^Self report in multi-person HH vs REF
^Proxy report in multi-person HH with same coverage vs REF
^Missing (respondent didn't match claims data)
^PIT coverage obtained in last 6 months vs 18+ months ago
^PIT coverage obtained 7-17 months ago vs 18+ months ago
^Received a premium subsidy
CPS
ACS
Odds of accurate Medicaid reporting
REF=Proxy report in multi-person HH with different coverage
^ Based on administrative records data; all other indicators are from survey data.
Personal-level characteristics
Respondent-level characteristics
Family-level characteristics
Task complexity
LOWER ODDS OF
ACCURATE REPORTING
HIGHER ODDS OF
ACCURATE REPORTING
12. REF=Proxy report in multi-person HH with different coverage and missing (respondent didn’t match)
^ Based on administrative records data; all other indicators are from survey data.
0.0 1.0 2.0 3.0 4.0 5.0 6.0 7.0 8.0 9.0 10.0 11.0 12.0 13.0 14.0 15.0 16.0 17.0 18.0 19.0 20.0
Age less than 19 vs age 35+
Age 19-34 vs age 35+
Health status fair/poor vs excellent/v.good/good/D/R
^Had any health care claims in last 6 months
Male vs female
High school or less vs bachelor or higher
Some college, associate vs bachelor degree or higher
Less than full time, full year vs full time, full year in 2014
Not working/D/R vs full time, full year in 2014
Employer size >50 vs 50 or fewer
^Respondent is policyholder
Family income in 2014 <200% vs > 200% FPG/D/R
Household receives TANF or SSI
Household receives SNAP
^Self report in one person HH vs REF
^Self report in multi-person HH vs REF
^Proxy report in multi-person HH with same coverage vs REF
^Missing (respondent didn't match claims data)
^PIT coverage obtained in last 6 months vs 18+ months ago
^PIT coverage obtained 7-17 months ago vs 18+ months ago
^Received a premium subsidy
CPS
ACS
Odds of accurate MNcare reporting
Personal-level characteristics
Respondent-level characteristics
Family-level characteristics
Task complexity
LOWER ODDS OF
ACCURATE REPORTING
HIGHER ODDS OF
ACCURATE REPORTING
13. 0.0 1.0 2.0 3.0 4.0 5.0 6.0 7.0 8.0 9.0 10.0 11.0 12.0 13.0 14.0 15.0 16.0 17.0 18.0 19.0 20.0
Age less than 19 vs age 35+
Age 19-34 vs age 35+
Health status fair/poor vs excellent/v.good/good/D/R
^Had any health care claims in last 6 months
Male vs female
High school or less vs bachelor or higher
Some college, associate vs bachelor degree or higher
Less than full time, full year vs full time, full year in 2014
Not working/D/R vs full time, full year in 2014
Employer size >50 vs 50 or fewer
^Respondent is policyholder
Family income in 2014 <200% vs > 200% FPG/D/R
Household receives TANF or SSI
Household receives SNAP
^Self report in one person HH vs REF
^Self report in multi-person HH vs REF
^Proxy report in multi-person HH with same coverage vs REF
^Missing (respondent didn't match claims data)
^PIT coverage obtained in last 6 months vs 18+ months ago
^PIT coverage obtained 7-17 months ago vs 18+ months ago
^Received a premium subsidy
Odds of accurate Non-group reporting
CPS
ACS
REF=Proxy report in multi-person HH with different coverage and missing (respondent didn’t match)
^ Based on administrative records data; all other indicators are from survey data.
Personal-level characteristics
Respondent-level characteristics
Family-level characteristics
Task complexity
LOWER ODDS OF
ACCURATE REPORTING
HIGHER ODDS OF
ACCURATE REPORTING
14. 0.0 1.0 2.0 3.0 4.0 5.0 6.0 7.0 8.0 9.0 10.0 11.0 12.0 13.0 14.0 15.0 16.0 17.0 18.0 19.0 20.0
Age less than 19 vs age 35+
Age 19-34 vs age 35+
Health status fair/poor vs excellent/v.good/good/D/R
^Had any health care claims in last 6 months
Male vs female
High school or less vs bachelor or higher
Some college, associate vs bachelor degree or higher
Less than full time, full year vs full time, full year in 2014
Not working/D/R vs full time, full year in 2014
Employer size >50 vs 50 or fewer
^Respondent is policyholder
Family income in 2014 <200% vs > 200% FPG/D/R
Household receives TANF or SSI
Household receives SNAP
^Self report in one person HH vs REF
^Self report in multi-person HH vs REF
^Proxy report in multi-person HH with same coverage vs REF
^Missing (respondent didn't match claims data)
^PIT coverage obtained in last 6 months vs 18+ months ago
^PIT coverage obtained 7-17 months ago vs 18+ months ago
^Received a premium subsidy
Odds of accurate Marketplace reporting
CPS
ACS
REF=Proxy report in multi-person HH with different coverage and missing (respondent didn’t match)
^ Based on administrative records data; all other indicators are from survey data.
Personal-level characteristics
Respondent-level characteristics
Family-level characteristics
Task complexity
LOWER ODDS OF
ACCURATE REPORTING
HIGHER ODDS OF
ACCURATE REPORTING
15. Summary of results
Consistent with prior research public reporting is more
accurate among
less social/structurally advantaged:
Low income and education
those with experience with other social programs and who likely
need care
TANF/SSI and SNAP recipients
Fair/poor self-reported health
Variation across public programs
For Medicaid, family-level characteristics matter
For MNcare, respondent-level characteristics matter
16. Summary continued
For ACS, private reporting is more accurate among
more social/structurally advantaged
Males, higher income
those less likely to have ESI offer
Part-time/part-year, modest educational attainment
those with less task complexity
Living alone and reporting for self
those receiving a subsidy in Marketplace plan
For CPS, private reporting is more accurate among
those age 35 plus vs age 19-34
those with long duration of same coverage
17. Conclusions
CHIME is first look at correlates of accurate reporting for ACS,
CPS redesign, direct purchase and marketplace
Although significant correlates are sparse, there patterns:
CHIME results for public insurance are consistent with past research
For both public and private insurance:
characteristics of accurate reporting match likely enrollees
lends confidence in editing/imputation routines and use of survey data
for policy simulations
Correlates of private reporting accuracy vary by survey:
For ACS, respondent-level characteristics matter, more significant
correlates
For CPS, fewer significant correlates
Next steps: Refine regression models; look at other accuracy
metrics beyond undercount
Survey reports of source of health insurance have measurement error – this is well documented.
Goal of this study is to describe factors associated with accurate reports of coverage for two popular Census surveys: CPS and ACS
And see if there are differences in factors associated with accurate reporting based on the type of insurance or based on the survey
Ideally, measurement error is randomly distributed, not systematic
However, to the extent it is not random, knowing the factors associated with error can
Provide clues for redesign of surveys
can be accounted for in editing and imputation routines
Or can be accounted for policy simulations and evaluations.
For example
The CPS redesign was based on years of analysis on challenges with the way HI questions are asked
The TRIM models by UI attempt to correct for reporting error
Estimates of cost impact of expansion of Medicaid eligibility to adults w/o children will be flawed if adults w/o children are less accurate reporters of insurance status – but knowing this propensity for error can be accounted for in models
To date validation studies have focused on the Medicaid population – all the information we have about correlates of accurate reporting is limited to public coverage.
Research consistently shows most accurate reporting is among:
Adults reporting for children vs other adults
People that are low income, unemployed, and lower education more accurately report Medicaid
Some evidence that – respondents are more accurate reporters of Medicaid for other household members when the respondent shared that coverage with the other members of the household, but results are mixed.
Using claims data, we also know that reporting is more accurate among those
who received HC services
who had coverage close to the date of the survey
and who had that coverage for a longer period of time.
This study expands what is known to private insurance.
We look at characteristics of the covered individual, the respondent and family or HH, most of which are based on survey data.
For example, for the covered person we have age and HS from the survey
reports of any claims in the 6 months prior to the survey are from admin records –this measure not available for NonG, MKt -- only public programs
(ACG/RUB health status and utilization)
Characteristics of the respondent include…. All from the survey.
Dropped R/E, marital status– NS in bivariate tests.
We also know from the claims data, whether the R is the policy holder.
For the family/hh we have income as a % of poverty and
Reports of SSI/TANF or Food Stamp participation for any HH members (limited to Medicaid).
program participation was too sparse for other coverage types (8% or less for NonG, Mkt and Mncare compared to 16% and 34% or more for Medicaid).
This set of correlates are all based on claims data and can be thought of as factors that may increase the complexity of reporting coverage.
First is a measure of shared coverage that I introduced earlier.
This is coded as follows from the most to least complex: READ…
Next is recency: or the count of consecutive months with same coverage
coded as having same coverage from the month of the survey to the past 6 month, the past 7-17 months or the entire 18 month observation period or more.
Finally for marketplace plans, we have an indicator of whether the policyholder received a subsidy.
Before turning to the correlates of reporting accuracy I briefly present accuracy levels. This is the only 1 of the 3 metrics introduced by Angie –how many get it right and the extent to which coverage is under-reported.
Blue bars are CPS, red are ACS
On the far left is NonG, followed by….
For nonG and MKt, reporting accuracy is higher for ACS, the two surveys have the same level of accuracy for Medicaid and CPS reporting is more accurate for MNcare.
Logit regressions are set up as 1 for accurate reporting of coverage type, 0 for inaccurate (most report a different source of coverage, a small minority do not report any coverage)
So on to the correlates… Let me provide an orientation to this slide – and the next 4 slides that use the same structure
This is a plot of the OR and Cis for each correlate in the model examining correlates of accurate Medicaid reporting for the ACS and CPS separately.
Again Blue = CPS, Red = ACS
These variables are organized by…
Covered individual- , respondent- and family-level characteristics, and measures of task complexity – all of which are from admin data as represented by the hats^.
OR less than 1.0 indicate lower odds of accurate reporting of Medicaid
OR above 1.0 indicate higher odds of accurate reporting
Can see that a number of variables have wide confidence intervals and many overlap the 1.0 line indicating NS
Same chart again but this time I only present significant correlates of accurate Medicaid reporting. Lots of white space right.
Again blue represents ORs for CPS and Red represents ORs for ACS
50/50 split in the number of signif correlates for the ACS and CPS– with most falling under family-level characteristics.
ACS respondents with a HS degree or less are less accurate reporters of Medicaid.
By contrast those is low income HHs are more accurate for the ACS
Reporting is less accurate in CPS HHs where someone is reported to receive TANF and/or SSI but for ACS HHs reporting accuracy is higher if a member of the HH receives TANF and/or SSI
For the CPS, accuracy is higher when someone in the HH receives SNAP/foodstamps
Finally, looking at task complexity…
Expect that those reporting for self, and those proxy reporting for those with the same coverage will be most accurate.
This expectation is not met! Instead in CPS, R reporting for self in 1 person HH have lower odds of accurate reporting than those in more complex HHs
Here we look at correlates of accurate MNcare reporting.
MNcare is now MN’s BHP – it is a small public program compared to Medicaid but has existed in MN since 1990s. It is a public program but enrollees pay a sliding fee premium each month. Only adults are enrolled in this program.
50/50 split between ACS/CPS in signif correlates – most falling under respondent level characteristics
Find that
For the ACS, accurate reporting of Mncare is higher for persons with fair/poor health status – consistent with prior research on Medicaid where those who need or use HC are more accurate
males respondents are less accurate reporters of Mncare in the ACS treatment
ACS and CPS reporting of MNcare is more accurate for those with some college compared to college degree
And those with lower family income are more accurate for CPS
We included a measure of the respondent’s employer size, hypothesizing that those working for large employers would be less likely to qualify for and report public cov.
Odds of accurately reporting Mncare in CPS is higher for respondents working for larger employer; opposite is true in ACS.
The next two slides represent firsts -- exploration of factors associated with accurate reporting of private insurance.
For Non-group insurance we see very little overlap in correlates for ACS and CPS, with more signif correlates for ACS than CPS
For CPS there are 3 signif ORs
odds of accurate reporting is lower for young adults (19-34) than older adults (35+)
and looking down at the bottom, the odds of accurate reporting of NonG insurance is greater for those enrolled 18 months or more.
Only regression where recency/duration of coverage is significant
For ACS there are 5 signif ORs – clustered in the respondent and family domains
odds of accurate reporting of NonG is higher for males, and those living in higher income HHs – social advantage
Also higher for those working less than FTFY (compared to FTFY) with some college or associates degree (compared to bachelor college degree or more)
CHIME is only validation study to include Marketplace enrollees
Very small sample so fewer significant correlates
Again, correlates for ACS and CPS do not overlap.
For CPS there is 1 signif OR
Consistent with NonG – the odds of accurate reporting of MKT is lower for young adults (19-34) than older adults (35+)
For ACS there are 2 signif ORs
odds of accurate reporting of Mkt is lower for those working less than FTFY (compared to FTFY) (social advantage)
those receiving a subsidy in Mkt plan, according to admin records, have 5x greater odds of reporting accurately reporting Mkt
Prior research generally indicates that those with greater social disadvantage and participating in other benefit programs are better reporters of Medicaid.
Generally speaking, the CHIME results are consistent with past research
For Medicaid reporting, significant correlates of accurate reporting were family-level – income, program participation and the like. Makes sense given this is a program for low-income children and their families.
For MNcare, significant correlates of accurate reporting were respondent-level – gender, education, employer size… not sure what to make of this.
For the ACS report of private insurance is more accurate for
those more advantaged,
those less likely to have an offer of insurance from their employer
and those who only have to report for themselves
for the CPS private reporting is more accurate for older adults and those who have the same coverage for 18 months or more.
First look…
From a reporting/measurement error perspective, the goal is to have random vs systematic pattern of error. Significant correlates are sparse (Lots of white space in slides 11 -14) which is encouraging. However there is a pattern.
Good news is that our findings are consistent with past research for public programs.
And characteristics of accurate report match likely public and private enrollees which lends confidence in editing/imputation routines and use of survey data for policy simulations
Correlates vary by survey for private coverage
for the ACS, respondent level characteristics matter and there are more signif correlates whereas only a few ORs are significant for the CPS.
This may imply that the CPS redesign worked, making it is easier for people to complete regardless of their personal or family characteristics.
Next steps