The Errors in Reporting Medicare Coverage: A Comparison of Survey Data and Administrative Records

Abstract

Objectives: We examine survey reporting of Medicare coverage of the older population by evaluating discordance between survey responses and administrative records. Method: We link data from the 2014 Current Population Survey Annual Social and Economic Supplement (CPS ASEC) and 2014 Medicare Enrollment Database to evaluate the extent to which individuals misreport Medicare coverage in the CPS ASEC. Using regression analyses, we assess factors associated with misreporting. Results: We find the CPS ASEC undercounts the population aged 65 years and older with Medicare by 4.5%. Misreporting of Medicare coverage is associated with citizenship status, immigration year of entry, employment, coverage of other household members, and imputation of Medicare responses. Adjusting for misreporting, Medicare coverage among older individuals increases from 93.4% to 95.6%. Discussion: The CPS ASEC underestimates Medicare coverage for the older population. Administrative records may be useful to evaluate and improve survey imputation of Medicare coverage when missing.

Keywords

Medicare survey measurement error administrative records health insurance

Introduction

As the U.S. population ages, meeting the health care needs of the older population will become increasingly challenging (Grady, 2011). A critical aspect of meeting this challenge is having an accurate understanding of the health insurance coverage of older individuals, as health insurance coverage has been shown to be directly associated with access to care and better health outcomes (Bernstein, Chollet, & Peterson, 2010; Institute of Medicine, 2002; Wilper et al., 2009).

The Current Population Survey Annual Social and Economic Supplement (CPS ASEC) is a primary source of data on health insurance coverage. Recent CPS ASEC data indicate that only 1.6% of individuals aged 65 years and older are uninsured, and approximately 93% have Medicare (Smith & Medalia, 2014). If the estimate of Medicare coverage is biased, this may affect official estimates of uninsurance as well as research that uses CPS ASEC data to study insurance coverage among the older population. For example, Schoen, Solís-Román, Huober, and Kelchner (2016) use CPS ASEC data to examine out-of-pocket medical expenses per Medicare recipient. The findings and policy recommendations of this work are dependent on the accuracy of Medicare coverage estimates.

Previous research has found that survey data undercount enrollment figures for some types of insurance, specifically Medicaid (Call, Davidson, Davern, Blewett, & Nyman, 2008; Davern et al., 2008; Davern, Klerman, Baugh, Call, & Greenberg, 2009; Klerman, Davern, Call, Lynch, & Ringel, 2009; Klerman, Ringel, & Roth, 2005; Lewis, Elwood, & Czajka, 1998; Noon, Fernandez, & Porter, 2016). We expect reporting of Medicare coverage to be more accurate than reporting of Medicaid coverage. In most cases, when individuals sign up for Medicare, they are then covered for life, whereas people often cycle on and off Medicaid coverage. Still, some older individuals may be misreporting their Medicare coverage and any potential misreporting has been unexplored. We, therefore, do not know the extent to which survey data may undercount Medicare coverage or the extent to which this may affect estimates of the uninsured population.

Our research evaluates the accuracy of reports of Medicare coverage among older individuals (aged 65 years and above) in the CPS ASEC and focuses on three main research questions.

Research Question 1: Does the CPS ASEC undercount the Medicare-enrolled population?

Research Question 2: To what extent do individuals misreport their Medicare coverage in the CPS ASEC, and what characteristics are associated with misreporting?

Research Question 3: How does misreporting affect estimates of Medicare coverage and the uninsured population among persons aged 65 years and older?

Background

Medicare

Medicare is a public health insurance program that was implemented in 1966 in an effort to improve health insurance coverage of the older population (Klees, Wolfe, & Curtis, 2010; Moon, 1996). Eligibility for Medicare depends on several rules including factors related to age, residency, marital history, and employment history (Centers for Medicare & Medicaid Services [CMS], 2012). Although there are exceptions for certain health and disability-related conditions, most people first become eligible for Medicare when they turn 65. Some individuals are automatically enrolled in Medicare when they turn 65, whereas others need to sign up (CMS, 2012).

According to recent statistics from the CPS ASEC, about 93% of individuals aged 65 years and older had Medicare in 2013 (Smith & Medalia, 2014). Some individuals had other insurance coverage along with Medicare—49% of individuals reported being covered by Medicare and private insurance, whereas around 6% were covered by Medicare and Medicaid—whereas others have Medicare as their only source of insurance (Smith & Medalia, 2014).

Medicare is often thought of as providing universal insurance coverage for the older population, but it does not (Birnbaum & Patchias, 2010; Gray, Scheinmann, Rosenfeld, & Finkelstein, 2006). Individuals who do not have Medicare coverage may be ineligible for coverage if they do not meet residency requirements. Others may be ineligible for coverage due to their own or their spouse’s lack of a work history. Research evaluating Medicare Part A (hospital insurance) coverage found that lack of coverage is particularly high for older individuals in certain states, namely, California, New York, Texas, and Florida (Birnbaum & Patchias, 2010). Another study focusing on New York City found that 16% to 20% of older individuals lacked Medicare, much higher than the nationwide 7% reported by the 2013 CPS ASEC (Gray et al., 2006). This study found lack of Medicare coverage to be particularly pronounced for immigrants; coverage was also found to be associated with country of origin, years in the United States, age at arrival, and current age (Gray et al., 2006).

Survey Reporting of Health Insurance

Health insurance research often relies on survey data. Along with the CPS ASEC, the National Health Interview Survey, the American Community Survey, Medical Expenditure Panel Survey (Household Component), and the Survey of Income and Program Participation are important sources of information on health insurance coverage and the uninsured. Thus, an understanding of the accuracy and reliability of health insurance data in surveys is critical for researchers and policy makers. Previous research has examined response error of Medicaid reporting, with recent results finding that the CPS ASEC undercounts the Medicaid-enrolled population by between 35% and 39% (Noon et al., 2016). Researchers have evaluated two types of response errors—false negatives (persons who reported not having Medicaid coverage in the CPS ASEC but were present in enrollment data) and false positives (persons who reported having Medicaid coverage in the CPS ASEC but were not present in enrollment data) and found that both types of response errors are not randomly distributed across the population (Noon et al., 2016).

Although we expect reporting of Medicare coverage to be more accurate than Medicaid, there are several reasons why survey respondents may misreport their Medicare coverage. Individuals with multiple types of insurance coverage may not accurately report their coverage when responding to a survey (Noon et al., 2016). Accuracy in reporting may also differ based on which aspects of Medicare individuals are enrolled. For example, Medicare Advantage (Medicare Part C) plans are run by private companies and offer hospital and medical coverage (CMS, 2012). Some individuals with Medicare Advantage may, therefore, report this coverage as private insurance instead of reporting their coverage as Medicare.

Correct reporting of Medicare coverage may also be tied to usage of health services—those who seek more medical care may be more aware of their insurance coverage and more likely to report it accurately. Persons with disabilities, for example, may need more medical care and, thus, be more aware of their health insurance coverage. Some individuals with disabilities may also have been eligible for Medicare prior to turning 65, and having Medicare coverage for many years may result in more accurate reporting. Previous research has found racial disparities in access to care, leading groups with greater access to care to report their Medicare coverage more accurately (Agency for Health Care Research and Quality, 2012).

Accuracy in reporting has been shown to be affected by survey design features (Bowling, 2005). For example, the person responding to the question may not know the health insurance coverage of each member of the household. Reporting may also be influenced by the insurance coverage of other members of the household. If multiple people in the household have Medicare, it may be more likely that the survey respondent is more aware of the coverage and more likely to report it. However, if respondents mistakenly report having Medicare coverage when they do not, they may be more likely to report this for another individual in the household. Also, if one older person in the household is enrolled in Medicare, the survey respondent may assume that all older persons in the household are enrolled in Medicare.

Misreporting may also occur due to confusion regarding the question and insurance options. Previous research has found that confusion between Medicare and Medicaid may lead to misreporting of Medicaid coverage in surveys—similarities in the names of the programs may lead people to report coverage through one program when their coverage is really through the other (Pascale, 2004). For Medicare, respondents with private health insurance may misreport their Medicare enrollment status. Factors such as educational attainment, which are associated with general difficulties in responding to surveys (Holbrook, Cho, & Johnson, 2006), may also result in misreporting of Medicare enrollment.

Data and Method

We use data from the 2014 CPS ASEC and the 2014 Medicare Enrollment Database (MEDB). The 2014 CPS ASEC data were collected between February and April 2014 for a sample of approximately 98,000 households (U.S. Census Bureau, 2014). The universe for the CPS ASEC is the civilian noninstitutional population; if a person is temporarily absent from the household, for instance, as a nonpermanent patient in a hospital, they count as part of the civilian noninstitutional population. Individuals living in institutional group quarters (such as nursing homes) are excluded from our analysis. We use survey weights that account for factors such as coverage and nonresponse (U.S. Census Bureau, 2006). We restrict our analysis to those aged 65 years and older at the time of the survey, resulting in a sample of about 24,000 unweighted individuals or 44.4 million people (weighted to the population). The 2014 CPS ASEC asks respondents to indicate their current health insurance coverage as well as when that coverage began and which months they held that coverage (Medalia, O’Hara, Rodean, Steinweg, & Brault, 2014). For our analysis, we use information on current coverage (i.e., coverage at the time of the survey). For the majority of people, Medicare responses were reported by respondents, but some cases were imputed due to a missing or invalid response.¹

The 2014 MEDB file is the CMS database of Medicare beneficiary enrollment information. We make some adjustments to account for universe differences between the CPS ASEC and MEDB file. We exclude individuals who live in group quarters, as the CPS ASEC includes only the civilian noninstitutionalized population. We also restrict the MEDB file to persons aged 65 years and older and enrolled in Medicare as of April 2014, with addresses in the 50 states or District of Columbia. We remove duplicate records and records that have a listed date of death on the MEDB file.

To estimate the Medicare undercount, we compare estimates of the Medicare-covered population from the CPS ASEC with counts of the enrolled population in the MEDB. To evaluate misreporting in the CPS ASEC, we link the CPS ASEC and MEDB files using unique, protected identifiers that are assigned to each file by another area of the Census Bureau. These identifiers are assigned through probabilistic matching techniques, which use personally identifiable information on the files such as name, date of birth, address, gender, and, in the case of MEDB, Social Security Number (see Wagner & Layne, 2014, for more details). After the assignment of these unique identifiers and before we receive the file, all personal information is removed to preserve confidentiality.

For the 2014 CPS ASEC, protected unique identifiers were assigned based on name, date of birth, gender, and address; 90.1% of individuals aged 65 years and older were assigned a unique identifier. For the remaining cases, the personal information provided may have been incomplete or did not match reference files and, therefore, a unique identifier could not be assigned. The 2014 MEDB file included Social Security Numbers, and as a result, more than 99.9% of cases were assigned a unique identifier. There is possibility of error in the assignment of unique identifiers and, in some cases, the wrong identifier may be assigned to a person. Previous research has examined false match rates for a previous vintage of the MEDB file and found error rates to be quite low, ranging between 0.005% and 1.174% (Layne et al., 2014). Thus, although our analysis may include some false matches, we expect that it is a small percentage of our sample. An evaluation of the consistency of assigning unique identifiers across different groups has found that the assignment process has biases that result in lower rates of assignment for some groups such as immigrants, recent movers, and low-income individuals (Bond, Brown, Luque, & O’Hara, 2014). To account for observations that cannot be linked because they were not assigned a unique identifier, we reweight the CPS data using adjustment factors based on characteristics we find to be associated with the assignment of unique identifiers. These characteristics include race, Hispanic origin, age, imputation status of age, sex, nativity, marital status, recent migration, and household income. Because the assignment of unique identifiers is so high on the MEDB file across characteristics, we do not use adjustment factors for that file.

Using the unique identifiers, we link the CPS ASEC and MEDB files and measure two types of reporting errors: false negatives and false positives. False negatives are defined as individuals who do not report having Medicare in the CPS ASEC but are found in the MEDB file, indicating Medicare enrollment. The universe for measuring false negatives is all records in the CPS ASEC–MEDB linked data. We measure the false negative rate using as the denominator all linked CPS ASEC–MEDB linked records. The numerator is the number of people in the linked records who do not report having Medicare. False positives are defined as individuals who report having Medicare in the CPS ASEC despite not being present in the MEDB file. The universe for measuring false positives includes all linkable records (i.e., individuals with unique identifiers) in the CPS ASEC that do not match to the MEDB file. We measure the false positive rate using as the denominator all people who truly are not enrolled, which includes correct reports of not having Medicare (true negatives) and false positive reports of Medicare coverage. The numerator is the total number of false positives, those not enrolled in Medicare who report that they are. This measurement of the false positive rate is consistent with Klerman et al.’s (2005) research on Medicaid underreporting in the CPS ASEC. Using this method, we are able to measure the differences in the population of people without Medicare between those who incorrectly report enrollment (false positives) and persons who correctly do not report having Medicare.

We use descriptive analysis and two logistic regressions to evaluate characteristics associated with both types of misreporting. One model assesses characteristics associated with reporting a false negative response among individuals in the CPS ASEC–MEDB linked data set and the second model assesses characteristics associated with false positive reporting among linkable CPS ASEC records that are not in the MEDB. We include sex, race and Hispanic origin, educational attainment, citizenship and year of entry, marital status, income to poverty ratio for the health insurance unit,² labor force participation, disability status, Medicare coverage of other individuals in the household, coverage through private insurance, and CPS ASEC imputation status of Medicare coverage as independent variables in each model.

The final step of our analysis is to calculate adjusted Medicare coverage and uninsured rates for the population aged 65 years and older, taking into account false negative and false positive misreporting. To estimate the adjusted Medicare coverage rate, we start with the CPS ASEC estimate of the Medicare-covered population and add false negative reports and subtract false positive reports. To estimate the adjusted uninsured rate, we start with the CPS ASEC estimate of the uninsured older population and subtract false negatives that did not report any other health insurance coverage and add false positives that did not report any other health insurance coverage. We make the assumption that those who falsely report having Medicare coverage and report no other types of health insurance coverage are uninsured. But it is important to note that some of these individuals may in fact have coverage—for example, people with Medicaid coverage who were confused by the similar program names and incorrectly reported Medicare instead. As a result, the adjusted uninsured rate may be lower than our calculation. Persons who report false negatives or false positives and who have other insurance coverage do not affect estimates of the uninsured.

Moreover, our analysis is focused on evaluating CPS ASEC Medicare coverage data relative to the MEDB; we do not take into account any potential coverage or quality issues with the MEDB itself. Specifically, we assume the MEDB includes all individuals with Medicare coverage as of April 2014 and that age, date of enrollment, residency in the United States, and date of death information on the file (which are used to create the data set we use in our analysis) are accurate. This may not always be the case, for example, previous research has found issues with date of death information in the MEDB (West, Devine, DeSalvo, & Condon, 2010),³ and, therefore, these assumptions should be kept in mind when considering our results.

Results

We first compare the CPS ASEC estimate of those with Medicare coverage with the count of Medicare enrollees from the MEDB file to determine whether the CPS ASEC undercounts the Medicare population. Table 1 shows that, in the 2014 CPS ASEC, there were 41.4 million individuals aged 65 years and older reported as having Medicare coverage. Enrollment data, however, show 43.4 million individuals aged 65 years and older are covered by Medicare,⁴ indicating that the CPS ASEC undercounts the Medicare population by 4.5%.

Table 1.

Medicare Undercount.

Number who report Medicare coverage in 2014 CPS ASEC		2014 MEDB count	2014 CPS ASEC Medicare undercount
Unweighted	Weighted	2014 MEDB count	Estimate	95% CI
22,500	41,420,000	43,380,000	4.5	[4.51, 4.52]

Source. 2014 CPS ASEC and 2014 MEDB linked data.

Note. All counts are rounded to be in compliance with the Census Bureau’s disclosure-avoidance policies. The MEDB count includes persons living in the United States aged 65 years and older who were enrolled in Medicare as of April 2014, and is adjusted to exclude individuals living in group quarters and those individuals on the file who are deceased. CPS ASEC = Current Population Survey Annual Social and Economic Supplement; MEDB = Medicare Enrollment Database; CI = confidence interval.

Next, we use the linked CPS ASEC and MEDB data to estimate the extent of misreporting among individuals in the CPS ASEC, as shown in Table 2. Of the approximately 42.5 million weighted cases in the linked CPS ASEC–MEDB file, 1.8 million (4.2%) were false negatives as they did not report Medicare coverage in the CPS ASEC. The universe for false positives is much smaller, as most linkable records aged 65 years and older in the CPS ASEC are found in the MEDB file. Of the 1.6 million weighted linkable records not found in the MEDB file, about 817,000 (51.7%) reported having Medicare coverage in the CPS ASEC and are considered false positives. Among all reports of Medicare in the CPS ASEC, the 817,000 Medicare reports coded as false positives account for only about 2% of the share of total Medicare reports. Unweighted numbers are shown for reference and show a similar pattern.

Table 2.

False Negative and False Positive Misreporting of Medicare Coverage in the 2014 CPS ASEC—Universe: Ages 65 and Older.

Calculation of false negative and positive errors	Unweighted		Weighted
Calculation of false negative and positive errors	Number	Percent	Number	Percent [95% CI]
False negative rate
Total CPS ASEC–MEDB matched records	21,000	100.0	42,470,000	100.0
Matched records that correctly report Medicare coverage	20,500	97.6	40,680,000	95.8 [95.4, 96.2]
Report false negative	850	4.0	1,790,000	4.2 [3.85, 4.7]
False positive rate
Linkable records not in MEDB	600	100.0	1,580,000	100.0
Linkable records that correctly report no Medicare coverage	300	50.0	760,000	48.1 [41.7, 54.7]
Report false positive	300	50.0	817,000	51.7 [45.3, 58.3]
False positive share of all Medicare reports
Linkable records that report having Medicare coverage	20,500	100.0	41,500,000	100.0
Linkable records that report having Medicare coverage and match to MEDB	20,500	100.0	40,680,000	98.0 [97.6, 98.4]
Report false positive	300	1.5	817,000	2.0 [1.6, 2.3]

Source. 2014 CPS ASEC and 2014 MEDB linked data.

Note. All counts are rounded to be in compliance with the Census Bureau’s disclosure-avoidance policies. CPS ASEC = Current Population Survey Annual Social and Economic Supplement; MEDB = Medicare Enrollment Database; CI = confidence interval.

In Table 3, we examine reported health insurance among people who are in the MEDB file but do not report Medicare coverage (i.e., our false negatives). Of the 1.8 million cases who were in the linked CPS ASEC–MEDB file but did not report having Medicare in the CPS ASEC, the majority (77.7%) reported having only private insurance coverage and about 20.1% of individuals did not report having any health insurance coverage. The remaining individuals reported having Medicaid coverage (alone or in combination with another type of coverage) or some other form of public coverage.

Table 3.

Reported Health Insurance Coverage Among False Negative Misreports in the 2014 CPS ASEC—Universe: Ages 65 and Older.

Reported health insurance coverage	Number	Percent	95% confidence interval
All false negatives	1,790,000	100.0
Medicaid coverage only	20,500	1.1	[0.4, 1.8]
Medicaid coverage and other insurance	17,000	0.9	[0.3, 1.6]
Other public insurance	5,300	0.3	[−0.1, 0.7]
Private insurance only	1,390,000	77.7	[73.3, 81.9]
Uninsured	359,000	20.1	[15.8, 24.2]

Source. 2014 CPS ASEC and 2014 MEDB linked data.

Note. Numbers are weighted. All counts are rounded to be in compliance with the Census Bureau’s disclosure-avoidance policies. CPS ASEC = Current Population Survey Annual Social and Economic Supplement; MEDB = Medicare Enrollment Database.

Next, we take a closer look at the characteristics associated with misreporting using descriptive and logistic regression analyses. First, we show descriptive characteristics of false negatives and false positives in Tables 4 and 5. We find that false negative rates are highest for those aged 65 years (19.3%) and generally decrease as age increases. Among those aged 70 years and higher, the false negative rate is only 1.9%. Individuals aged 65 years in the CPS ASEC may not yet be enrolled or be aware they are enrolled in Medicare at the time of the survey and, thus, be more likely to report not having Medicare coverage.⁵ We also find higher rates of false negatives for those employed in the labor force (15.2%) compared with those not in the labor force (2.0%) and for recent noncitizen entrants (17.1%) compared with native U.S. citizens (3.9%). Persons with disabilities have lower rates of false negatives (1.7%) compared with those without a disability (5.3%); and, those living in households where other individuals were reported as having Medicare are less likely to report false negatives. When Medicare coverage was imputed rather than reported, individuals have higher rates of false negative reporting (6.2% compared with 3.7%).

Table 4.

False Negative Misreporting of Medicare Coverage in the 2014 CPS ASEC by Characteristics—Universe: Ages 65 and Older.

Characteristics in CPS ASEC	Total CPS ASEC–MEDB matched		Report false negative
Characteristics in CPS ASEC	Number	Number	Percent	95% confidence interval
Total, aged 65 years and older	42,470,000	1,790,000	4.2	[3.8, 4.6]
65	3,010,000	581,000	19.3	[16.8, 21.8]
66	3,280,000	268,000	8.2	[6.6, 9.8]
67	3,070,000	185,000	6.0	[4.4, 7.6]
68	2,480,000	122,000	4.9	[3.3, 6.5]
69	2,320,000	106,000	4.6	[3.1, 6.1]
70+	28,310,000	531,000	1.9	[1.5, 2.3]
Sex
Male	18,790,000	926,000	4.9	[4.3, 5.5]
Female	23,690,000	869,000	3.7	[3.2, 4.2]
Race and Hispanic origin
Non-Hispanic White alone	33,450,000	1,260,000	3.8	[3.4, 4.2]
Non-Hispanic Black alone	3,560,000	193,000	5.4	[3.6, 7.2]
Non-Hispanic Asian alone	1,620,000	128,000	7.9	[4.3, 11.5]
Non-Hispanic Other	558,000	38,000	6.8	[4.2, 9.6]
Hispanic	3,280,000	173,000	5.3	[4.1, 6.5]
Educational attainment
No high school	7,050,000	181,000	2.6	[1.9, 3.3]
High school degree	14,680,000	513,000	3.5	[2.8, 4.2]
Some college	9,720,000	394,000	4.1	[3.4, 4.8]
Bachelor’s degree	6,260,000	341,000	5.4	[4.5, 6.5]
Graduate or professional degree	4,770,000	364,000	7.6	[6.2, 9.0]
Citizenship status and year of entry
Native	37,330,000	1,460,000	3.9	[3.5, 4.3]
Naturalized entered 10+ years ago	3,990,000	225,000	5.6	[3.9, 7.3]
Naturalized entered less than 10 years ago	97,500	4,600	4.7	[−1.4, 10.8]
Noncitizen entered 10+ years ago	965,000	89,000	9.2	[6.0, 12.4]
Noncitizen entered less than 10 years ago	87,500	15,000	17.1	[4.8, 29.2]
Marital status
Now married	24,940,000	1,160,000	4.7	[4.2, 5.2]
Widowed, separated, or divorced	15,870,000	496,000	3.1	[2.4, 3.8]
Never married	1,670,000	138,000	8.3	[5.5, 11.1]
Income to poverty ratio of health insurance unit
Less than 100%	5,930,000	359,000	6.1	[4.4, 7.8]
100%-149%	5,870,000	102,000	1.7	[1.0, 2.4]
150%-199%	5,090,000	80,000	1.6	[1.1, 2.1]
200% or higher	25,590,000	1,250,000	4.9	[4.4, 5.4]
Labor force participation
Employed	7,020,000	1,070,000	15.2	[13.8, 16.8]
Unemployed	299,000	29,000	9.7	[4.5, 14.9]
Not in labor force	35,160,000	692,000	2.0	[1.6, 2.4]
Disability
Has a disability	12,920,000	215,000	1.7	[1.3, 2.1]
Does not have a disability	29,550,000	1,580,000	5.3	[4.7, 5.9]
SSI receipt
Yes	1,230,000	14,000	1.1	[0.4, 2.0]
No	41,240,000	1,780,000	4.3	[3.9, 4.7]
Other household members have Medicare
Yes	20,860,000	459,000	2.2	[1.9, 2.5]
No	21,610,000	1,340,000	6.2	[5.5, 6.9]
Imputation status of Medicare response
Explicitly reported	33,950,000	1,260,000	3.7	[3.3, 4.1]
Imputed	8,530,000	532,000	6.2	[5.4, 8.0]

Source. 2014 CPS ASEC and 2014 MEDB linked data.

Table 5.

False Positive Misreporting of Medicare Coverage in the 2014 CPS ASEC by Characteristics—Universe: Ages 65 and Older.

Characteristics in CPS ASEC	Linkable records not in MEDB		Report false positive
Characteristics in CPS ASEC	Number	Number	Percent	95% confidence interval
Total, aged 65 years and older	1,580,000	817,000	51.7	[45.3, 58.3]
65	330,000	117,000	35.5	[25.3, 45.9]
66	271,000	148,000	54.6	[37.6, 71.4]
67	107,000	39,500	36.9	[19.0, 55.4]
68	92,000	46,500	50.5	[25.7, 75.7]
69	101,000	33,500	33.2	[14.8, 51.4]
70+	676,000	432,000	63.9	[52.4, 75.4]
Sex
Male	777,000	358,000	46.1	[36.0, 56.2]
Female	801,000	459,000	57.3	[50.0, 64.6]
Race and Hispanic origin
Non-Hispanic White alone	869,000	474,000	54.5	[47.1, 62.1]
Non-Hispanic Black alone	178,000	99,500	55.9	[40.7, 71.3]
Non-Hispanic Asian alone	238,000	121,000	50.8	[29.9, 72.1]
Non-Hispanic Other	15,500	7,300	47.1	[13.2, 81.4]
Hispanic	278,000	115,000	41.4	[27.5, 55.1]
Educational attainment
No high school	334,000	237,000	71.0	[62.6, 79.6]
High school degree	423,000	224,000	53.0	[41.4, 64.4]
Some college	389,000	185,000	47.6	[31.0, 63.8]
Bachelor’s degree	274,000	114,000	41.6	[26.1, 57.1]
Graduate or professional degree	157,000	56,500	36.0	[18.1, 54.1]
Citizenship status and year of entry
Native	935,000	505,000	54.0	[46.8, 61.2]
Naturalized entered 10+ years ago	248,000	118,000	47.6	[23.8, 71.0]
Naturalized entered less than 10 years ago	27,500	16,500	60.0	[24.8, 94.4]
Noncitizen entered 10+ years ago	193,000	76,500	39.6	[25.3, 53.7]
Noncitizen entered less than 10 years ago	174,000	102,000	58.6	[47.4, 69.8]
Marital status
Now married	856,000	419,000	48.9	[41.4, 56.4]
Widowed, separated, or divorced	592,000	336,000	56.8	[45.2, 68.4]
Never married	129,000	62,000	48.1	[20.6, 75.0]
Income to poverty ratio of health insurance unit
Less than 100%	503,000	312,000	62.0	[51.8, 72.4]
100%-149%	112,000	84,500	75.4	[62.3, 89.3]
150%-199%	116,000	89,000	76.7	[60.2, 92.6]
200% or higher	847,000	331,000	39.1	[29.4, 48.8]
Labor force participation
Employed	595,000	185,000	31.1	[21.0, 41.2]
Unemployed	33,500	8,200	24.5	[0, 49.3]
Not in labor force	949,000	624,000	65.8	[56.6, 74.8]
Disability
Has a disability	346,000	271,000	78.3	[69.6, 87.0]
Does not have a disability	1,230,000	546,000	44.4	[36.6, 52.0]
SSI receipt
Yes	77,000	77,000	100.0	[100.0, 100.0]
No	1,500,000	740,000	49.3	[42.8, 55.8]
Other household members have Medicare
Yes	402,000	303,000	75.4	[66.5, 84.5]
No	1,180,000	513,000	43.5	[36.1, 51.3]
Imputation status of Medicare response
Explicitly reported	931,000	290,000	31.1	[25.0, 37.2]
Imputed	646,000	527,000	81.6	[75.3, 87.9]

Source. 2014 CPS ASEC and 2014 MEDB linked data.

Table 5 presents the unlinked records, which comprise our false positive universe and examines the characteristics of the population not enrolled in Medicare but incorrectly reporting Medicare enrollment. We find lower rates of false positive reporting among those with higher levels of education—for example, 36.0% of those with a graduate degree or higher report a false positive compared with 71.0% of those with no high school degree. We find higher false positive reporting among persons not in the labor force compared with those in the labor force, as well as for persons with disabilities and those living in households where another individual was reported as having Medicare coverage. As with false negatives, false positive reporting is higher in cases where the CPS ASEC report of Medicare coverage requires imputation relative to those whose response is as reported. In fact, of all the false positive responses, most (527,000 or 64.5%) were imputed values rather than reported responses.

Next, we turn to our two logistic regressions evaluating the characteristics associated with, first, false negative responses and, second, false positive responses. Odds ratio results for both models are shown in Table 6. We find that minorities are generally more likely to report false negatives than non-Hispanic Whites. Noncitizens are more likely to report false negatives compared with natives. This is particularly true for recent noncitizen entrants who are more than 5 times more likely to report false negatives compared with natives. This may be a result of confusion regarding the different types of insurance or a misunderstanding of the survey question. Recent immigrants may have more difficulty answering the question and may be less familiar with the various types of insurance, resulting in higher rates of false negative reporting errors. Noncitizen is not significant for false positive reporting.

Table 6.

Weighted Logistic Regression Results: False Negative and False Positive Misreporting of Medicare Coverage in the 2014 CPS ASEC—Universe: Ages 65 and Older.

Variables (reference category)	Model 1: False negative response	Model 2: False positive response
Variables (reference category)	Odds ratios	Odds ratios
Sex (vs. male)
Female	1.15	0.76
Race and Hispanic origin (vs. non-Hispanic White)
Non-Hispanic Black alone	1.69**	0.86
Non-Hispanic Asian alone	1.47	0.63
Non-Hispanic Other	2.22***	0.93
Hispanic	1.49*	0.3*
Educational attainment (vs. high school degree)
No high school	0.67*	3.43**
Some college	1.01	1.88
Bachelor’s degree	1.09	2.23
Graduate or professional degree	1.3	1.92
Citizenship/year of entry (vs. native)
Naturalized entered 10+ years ago	1.23	1.12
Naturalized entered less than 10 years ago	1.05	0.53
Noncitizen entered 10+ years ago	2.56***	0.56
Noncitizen entered less than 10 years ago	5.48**	0.45
Marital status (vs. now married)
Widowed, separated, or divorced	0.41***	1.7
Never married	0.94	1.15
Income to poverty ratio of the health insurance unit (vs. 200%+)
Less than 100%	2.97***	3.09*
100%-149%	0.82	4.34*
150%-199%	0.63*	6.97***
Labor force participation (vs. employed)
Unemployed	0.6	0.1
Not in labor force	0.13***	3.89**
Disability status (vs. no disability)
Has a disability	0.49***	4.16**
Medicare status of other household members (vs. no others with Medicare)
Others in household have Medicare	0.25***	6.29***
Private insurance (vs. no private insurance)
Has private insurance	1.96***	0.34**
Imputation of Medicare response (vs. not imputed)
Imputed	1.36*	25.96***
–2 log likelihood	14,870,000	2,180,000
N (weighted)	42,470,000	1,580,000

Source. 2014 CPS ASEC and 2014 MEDB linked data.

p < .05. **p < .01. ***p < .001.

Individuals who are not in the labor force, have a disability, or live in a household with others who have Medicare have lower odds of reporting false negatives (odds ratios ranging from 0.13 to 0.49) and higher odds of reporting false positives (odds ratios ranging from 3.89 to 6.29) relative to their counterparts. It is not surprising that we see lower odds of false negative reporting for these groups. Compared with those still in the labor force, retired individuals may be less likely to have access to employer-based coverage and more likely to depend on Medicare and, therefore, more likely to accurately report their Medicare coverage. Persons with disabilities may be more likely to seek medical care and, thus, more aware of their coverage. In addition, Medicare eligibility criteria allow some disabled individuals to be eligible for Medicare prior to turning 65 years, and longer term coverage may also be associated with more awareness and accuracy in reporting. Living in a household with others reported as having Medicare coverage also suggests awareness of the Medicare program and a higher likelihood of reporting Medicare coverage. It is interesting that we also see higher rates of false positives for these groups, indicating higher rates of reporting Medicare coverage even when not enrolled.

Individuals who are a part of a family⁶ with an income to poverty ratio of less than 100% have higher odds of false negative reporting compared with those with an income to poverty ratio of more than 200%. For those whose family’s income to poverty ratio is less than 200%, we find higher odds of reporting false positives relative to those with an income to poverty ratio of more than 200%. This is particularly true for those with an income to poverty ratio between 150% and 199%, who are close to 7 times more likely to report a false positive compared with the reference group of 200% or higher. One possible explanation for this result may be health insurance coverage for the different income to poverty groups. People in the 150% to 199% income to poverty group may be less likely than those in the lowest income group to have Medicaid coverage and less likely than those in the highest income group to have private insurance. Thus, they may be more likely to report having Medicare both when they have it (leading to lower odds of reporting false negatives) and when they do not (leading to higher odds of reporting false positives).

Having private insurance is associated with higher odds of reporting false negatives and lower odds of reporting false positives. As discussed earlier, people with dual coverage may not accurately report each insurance type. People who have both private insurance and Medicare may rely on their private insurance as their primary source of coverage, and, thus, may be choosing to select only this insurance in their CPS ASEC response. Meanwhile, those without private insurance may be more likely to depend on their Medicare coverage and, thus, report it.

Individuals whose Medicare coverage response was imputed in the CPS ASEC are more likely to have both types of reporting errors compared with those whose response was reported. This is particularly true for false positive reporting—imputed responses are 26 times more likely to be false positives compared with responses that are reported. Further analysis (not shown) indicates that across all characteristics evaluated, false positive rates are higher when Medicare coverage is imputed compared with reported responses. This finding suggests current imputation procedures are often assigning Medicare coverage to individuals in the CPS ASEC who do not have Medicare coverage according to MEDB enrollment records.

Finally, Table 7 shows the process of calculating adjusted rates using our measures of false negative and false positive errors. Taking into account these errors, the Medicare rate for the population aged 65 years and older increases from 93.4% to 95.6% once we take into account misreporting, an increase of 2.2 percentage points. Adjusting for misreporting of Medicare coverage does not have a significant impact on the estimate of the uninsured population aged 65 years and older.

Table 7.

Adjusted Insured, Medicare Coverage, Uninsured Rates—Universe: Ages 65 and Older.

	CPS ASEC population
	Number	Percent	95% Confidence interval
Total, aged 65 years or higher	44,360,000	100.0	—
Report insured in CPS ASEC	43,720,000	98.6	[98.4, 98.8]
Report Medicare in CPS ASEC	41,420,000	93.4	[93.0, 93.8]
Report uninsured in CPS ASEC	641,000	1.4	[1.2, 1.6]
Total, false negative with other coverage	1,435,000
Total, false negative without other coverage	359,000
Total, false positive with other coverage	510,000
Total, false positive without other coverage	307,000
Adjusted Medicare rate	42,400,000	95.6	[95.0, 96.2]
(Report Medicare + false negative with or without other coverage − false positive with or without other coverage)	42,400,000	95.6	[95.0, 96.2]
Adjusted uninsured rate	589,000	1.3	[0, 4.2]
(Report uninsured − false negative without other coverage + false positive without other coverage)	589,000	1.3	[0, 4.2]

Source. 2014 CPS ASEC and 2014 MEDB linked data.

Conclusion

We examined Medicare reporting of the older population in the CPS ASEC. By comparing CPS ASEC estimates of the Medicare-enrolled population to administrative enrollment records, we find that the CPS ASEC undercounts the Medicare-enrolled older population by about 4.5%. Although notable, this undercount is much smaller than previous research on the Medicaid undercount, suggesting that Medicare coverage is more accurately reported and covered in surveys relative to Medicaid. Using linked data, we find both false negative and false positive reporting of Medicare coverage in the CPS ASEC, and these errors are not randomly distributed. Similar to Medicaid reporting research, having multiple types of insurance is associated with misreporting Medicare coverage. Those with private insurance are less likely to report having Medicare, and we find the majority of those who reported false negatives were reported as having private health insurance coverage only. We also find that Medicare coverage of others in the household is associated with accuracy in reporting, which is also consistent with research on Medicaid reporting (Noon et al., 2016; Pascale, Roemer, & Resnick, 2009). Usage of services also seems to be a factor—individuals with disabilities, whom we expect seek more care, are also more likely to report false negatives. We also find that noncitizens and recent entrants, groups that are less likely to have Medicare coverage (Gray et al., 2006), are more likely to report false negatives. Finally, imputation status of the Medicare coverage survey response is strongly associated with measurement error. Taking into account these misreporting errors, we estimate the adjusted Medicare coverage rate for the population aged 65 years and older to be 95.6%, 2.2 percentage points higher than the CPS ASEC estimate of 93.4%, and we find a modest decline in the uninsured rate among the population aged 65 years and older.

As researchers and policy makers often rely on survey data to understand health insurance coverage, our findings provide important information on the accuracy of survey-based estimates of Medicare coverage. Overall, most people accurately report their Medicare coverage in the CPS ASEC, making it a useful resource for researchers studying Medicare coverage among older individuals. However, researchers should be cautious about using these data to study certain subpopulations for which false negative and positive reporting is notably higher—for example, recent immigrants, those who are currently employed, and individuals who have recently become eligible for Medicare (i.e., 65-year-olds). Future research should also disentangle the effect of enrollment in different Medicare programs for explaining survey misreporting on Medicare enrollment.

Our results also suggest further avenues of research for the U.S. Census Bureau. In particular, our finding that misreporting is significantly higher for those whose Medicare coverage was imputed in the CPS ASEC highlights the need for evaluation of current imputation procedures and the potential use of administrative records to determine Medicare coverage for those with a missing response. Information from the MEDB may be useful for testing possible improvements to the imputation procedures used to assign health insurance coverage in household surveys and should be further explored. Knowledge of the extent of the Medicare undercount and factors associated with misreporting Medicare enrollment status may, therefore, contribute toward improvements in the measurement of Medicare coverage in surveys such as the CPS ASEC.

These findings may also be of use to Medicare program administrators and efforts targeted at beneficiary education and outreach.⁷ Resources devoted toward the administration of Medicare, and outreach efforts to educate participants, both depend on the knowledge of who is and is not enrolled in the program. If there are segments of the population that have a lower awareness of their Medicare enrollment status and incorrectly report not being enrolled, or incorrectly report being enrolled, then this could lessen the effectiveness of outreach efforts.

Footnotes

Authors’ Note

This article is released to inform interested parties of research and to encourage discussion. The views expressed are those of the authors and not necessarily those of the U.S. Census Bureau. The information in this article has been approved for release, with Disclosure Review Board Number CBDRB-FY18-250.

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The authors received no financial support for the research, authorship, and/or publication of this article.

Notes

References

Agency for Health Care Research and Quality. (2012). National health care disparities report (Publication No. 12-0006). Retrieved from https://www.ahrq.gov/sites/default/files/wysiwyg/research/findings/nhqrdr/nhdr11/nhdr11.pdf

Bernstein

Chollet

Peterson

(2010). How does insurance coverage improve health outcomes (Mathematica Policy Research Inc. Issue Brief). Washington, DC: Mathematica Policy Research.

Birnbaum

Patchias

E. M.

(2010). Measuring coverage for seniors in Medicare Part A and estimating the cost of making it universal. Journal of Health Politics, Policy, and Law, 35, 49-62.

Bond

Brown

J. D.

Luque

O’Hara

(2014, April). The nature of bias when studying only linkable person records: Evidence from the American Community Survey (Center for Administrative Records Research and Applications Working Paper No. 2014-08). Washington, DC: U.S. Census Bureau.

Boudreaux

Turner

(2011). Modifications to the imputation routine for health insurance in the CPS ASEC: Description and evaluation (State Health Access Data Assistance Center Working Paper). Minneapolis: University of Minnesota. Retrieved from www.census.gov/library/working-papers/2011/demo/2011-Boudreaux-Turner-01.html

Bowling

(2005). Mode of questionnaire administration can have serious effects on data quality. Journal of Public Health, 27, 281-291.

Call

K. T.

Davidson

Davern

Blewett

L. A.

Nyman

(2008). The Medicaid undercount and bias to estimates of uninsurance: New estimates and existing evidence. Health Services Research, 43, 901-914.

Centers for Medicare & Medicaid Services. (2012). Enrolling in Medicare Part A and Part B (CMS Product No. 11036). Baltimore, MD. Retrieved from https://www.medicare.gov/Pubs/pdf/11036-Enrolling-Medicare-Part-A-Part-B.pdf

Davern

Call

K. T.

Ziegenfuss

Davidson

Beebe

T. J.

Blewett

L. A.

(2008). Validating Health Insurance Coverage Survey estimates: A comparison between self-reported coverage and administrative data records. Public Opinion Quarterly, 72, 241-259.

10.

Davern

Klerman

J. A.

Baugh

D. K.

Call

K. T.

Greenberg

G. D.

(2009). An examination of the Medicaid undercount in the Current Population Survey: Preliminary results from record linking. Health Services Research, 44, 965-987.

11.

Department of Health and Human Services. (2015). HHS FY2015 budget in brief. Retrieved from http://www.hhs.gov/about/budget/fy2015/budget-in-brief/cms/program-management/index.html

12.

Grady

P. A.

(2011). Advancing the health of our aging population: A lead role for nursing science. Nurse Outlook, 59, 207-209.

13.

Gray

B. H.

Scheinmann

Rosenfeld

Finkelstein

(2006). Aging without Medicare? Evidence from New York City. Inquiry, 43, 211-221.

14.

Holbrook

Cho

Y. I. K.

Johnson

(2006). The impact of question and respondent characteristics on comprehension and mapping difficulties. Public Opinion Quarterly, 70, 565-595.

15.

Institute of Medicine. (2002). Care without coverage: Too little too late. Washington, DC: National Academies Press.

16.

Klees

B. S.

Wolfe

C. J.

Curtis

C. A.

(2010). Brief summaries of Medicare and Medicaid title XVIII and title XIX of the Social Security Act. Baltimore, MD: Department of Health and Human Services Centers for Medicare & Medicaid Services.

17.

Klerman

J. A.

Davern

Call

K. T.

Lynch

Ringel

J. D.

(2009). Understanding the Current Population Survey’s insurance estimates and the Medicaid “undercount.” Health Affairs, 28, 991-1001.

18.

Klerman

J. A.

Ringel

J. S.

Roth

(2005). Under-reporting of Medicaid and welfare in the Current Population Survey. Santa Monica, CA: RAND.

19.

Layne

Wagner

Rothhaas

(2014). Estimating Record Linkage False Match Rage for the Person Validation System (Center for Administrative Records Research and Applications Working Paper No. 2014-02). Washington, DC: US Census Bureau. Retrieved from https://www.census.gov/library/working-papers/2014/adrm/carra-wp-2014-02.html

20.

Lewis

C. H.

Elwood

Czajka

(1998). Counting the uninsured: A review of the literature. Washington, DC: The Urban Institute.

21.

Medalia

O’Hara

Rodean

Steinweg

Brault

(2014). Changing the CPS health insurance questions and the implications on the uninsured rate: Redesign and production estimates (Social Economic and Housing Statistics Division Working Paper 2014-16). Washington, DC: U.S. Census Bureau.

22.

Moon

(1996). What Medicare has meant to older Americans. Health Care Financing Review, 18(2), 49-59.

23.

Noon

J. M.

Fernandez

L. E.

Porter

S. R.

(2016, December). Response error and the Medicaid undercount in the Current Population Survey (Center for Administrative Records Research and Applications Working Paper No. 2016-11). Washington, DC: U.S. Census Bureau. Retrieved from https://www.census.gov/content/dam/Census/library/working-papers/2016/adrm/carra-wp-2016-11.pdf

24.

Pascale

(2004, May). Medicaid and Medicare reporting in surveys: An experiment on order effects and program definitions. Proceedings of the American Association for Public Opinion Research (AAPOR), American Statistical Association. Retrieved from http://www.amstat.org/sections/srms/Proceedings/y2004/files/Jsm2004-000594.pdf

25.

Pascale

Roemer

M. I.

Resnick

D. M.

(2009). Medicaid underreporting in the CPS: Results from a Record Check Study. Public Opinion Quarterly, 73, 497-520.

26.

Schoen

Solís-Román

Huober

Kelchner

(2016). On Medicare but at risk: A state-level analysis of beneficiaries who are underinsured or facing high total cost burdens (The Commonwealth Fund issue brief). Retrieved from http://www.commonwealthfund.org/publications/issue-briefs/2016/may/on-medicare-but-at-risk

27.

State Health Access Data Assistance Center. (2012, March). Defining “family” for studies of health insurance coverage (Issue Brief No. 27). Minneapolis: University of Minnesota

28.

Smith

J. C.

Medalia

(2014). Health insurance coverage in the United States: 2013 (Current Population reports). Washington, DC: U.S. Census Bureau.

29.

U.S. Census Bureau. (2006). Design and methodology (Current Population Survey Technical Paper 66). Washington, DC: Author. Retrieved from http://www.census.gov/prod/2006pubs/tp-66.pdf

30.

U.S. Census Bureau. (2014). Current Population Survey, 2014 ASEC technical documentation. Washington, DC: Author. Retrieved from http://www2.census.gov/programs-surveys/cps/techdocs/cpsmar14.pdf

31.

Wagner

Layne

(2014, July). The Person Identification Validation System (PVS): Applying the Center for Administrative Records Research and Applications’ (CARRA) record linkage software (Center for Administrative Records Research and Applications Working Paper No. 2014-01). Washington, DC: U.S. Census Bureau. Retrieved from https://www.census.gov/library/working-papers/2014/adrm/carra-wp-2014-01.html

32.

West

Devine

DeSalvo

Condon

(2010, January). The use of Medicare enrollment data in the 2010 demographic analysis estimates (Population Division Working Paper No. 89). Washington, DC: U.S. Census Bureau. Retrieved from https://www.census.gov/population/www/documentation/twps0089/twps0089.pdf

33.

Wilper

A. P.

Woolhandler

Lasser

K. E.

McCormick

Bor

D. H.

Himmelstein

D. U.

(2009). Health insurance and mortality in US adults. American Journal of Public Health, 99, 2289-2295.