Reporting accuracy of Social Security benefits and its implications in the Health and Retirement Study

Abstract

This paper examines whether and to what extent the amount of Social Security benefits of older survey respondents in the Health and Retirement Study (HRS) are reported accurately. Inaccurate reporting leads to biased estimates of gross Social Security benefits, affecting estimates of elderly well-being, including the proportion of beneficiaries classified as poor or near poor. Our findings indicate that 73% of HRS respondents report only the net amount of Social Security benefits they receive, excluding Medicare premiums. The implication is that Social Security benefits in the HRS are underestimates of the true gross benefits. Therefore, the HRS data overestimate the proportion of the elderly respondents who are poor or nearly poor. Finally, even after correcting for gross benefits, Social Security income comprises at least 50% of the total family income for about half of elderly respondents.

Keywords

Reporting accuracy Social Security benefits elderly household income poverty Medicare premiumsJEL codes: D31 H31 H55 I32 J14

1. Introduction

Survey respondents are commonly asked about the income they receive in a given year from different sources (such as earnings, Social Security benefits, pensions, assets, etc.). However, the accuracy of these reports depends upon both the design of the survey questions and the respondents’ reporting error. Therefore, it is of great interest to assess the extent and the magnitude of respondents’ reporting accuracy of their Social Security benefits. This is an important issue for researchers because estimates regarding Social Security incomes in retirement are commonly based on respondents’ self-reported information, and particularly since for about half of elderly beneficiaries, Social Security income comprises at least 50% of their family income [1]. Hence, if benefits are misreported, researchers interested in retirement income security and elderly wellbeing are likely to produce inaccurate estimates. It is also important for policy makers because underreporting of Social Security benefits will affect both the proportion of elderly beneficiaries classified as poor or nearly poor, and the estimates of the share of income received from Social Security.

The total amount of Social Security benefits includes the amount paid directly to the beneficiary and the amount deducted from gross benefits and paid for Medicare premiums. Thus, when asked, survey respondents can report either the total amount of Social Security benefits or the net amount excluding Medicare premiums. Previous research [2] has shown that respondents in the Survey of Income and Program Participation (SIPP) in 2009 underreported their Social Security benefits on average by an annual amount of $1,000. In contrast, Social Security income reported in the 2010 Current Population Survey (CPS) closely corresponded to the amount in the administrative records mainly because the CPS corrected the respondents’ reported Social Security income by adding the amount of Medicare premiums.1 2

This paper assesses the accuracy of reported Social Security benefits among elderly respondents in the Health and Retirement Study (HRS), a longitudinal study with comprehensive interviews of respondents aged 51 and older and their spouses of any ages. Because of the survey design and data processing, the HRS data omit a portion of the Social Security income received by respondents 65 and over, since respondents are asked to report the amount paid to them after any deductions.3 Consequently, it is plausible that respondents report the net amount of benefits received and ignore the amount deducted for Medicare premiums. Therefore, Social Security benefits in the HRS would underestimate the total (gross) benefits. This is concerning particularly since gross Social Security benefits are included in the estimation of official poverty rates. Hence, we assess whether and to what extent elderly respondents report their gross or net benefits and compare reported amounts to those in their administrative records, and then document the impact on estimates of elderly poverty rates and of the relative importance of Social Security income.

We find that, consistent with the question asked, about 73% of elderly HRS respondents with matched administrative records reported the amount of net benefits received (excluding Medicare premiums). Of the remaining respondents, about 16% and 11%, respectively, under-reported and over-reported their net benefits. Furthermore, the mean difference between self-reported and net benefits in the payment records is not statistically significant. As expected, given the HRS question and the respondents’ reporting of their net benefits, Social Security benefits in the HRS underestimate the true gross benefits, which lead to upward biased estimates of the proportion of elderly respondents that are poor or near poor, and also of the elderly’s reliance on Social Security income.

In the next section, we describe the data and our analysis approach, followed by a discussion of the findings in the result section. Finally, we conclude by offering suggestions for improvement in the HRS data.

2. Data and methodology

The data for this study come from the Health and Retirement Study (HRS), a longitudinal, nationally representative sample of Americans over the age of 50 and their spouses, first interviewed in 1992 with follow-up interviews conducted every other year. The HRS is the major data source for research of aging and retirement issues in America [11]. The depth and breadth of the HRS is unparalleled by other studies. It collects comprehensive information about respondents’ socio-economic characteristics, employment, health, disability, pensions, income and wealth. Moreover, it provides detailed information on income from different sources, separately for the respondent and the spouse (if coupled), such as from Social Security, earnings, private pensions, welfare programs, capital income, and other sources. Based on survey reports, RAND creates an annual measure for each income component, and a measure of the total annual household income. The latter variable is then used to generate a measure of the poverty status for each respondent according to the Census definitions of poverty threshold by family composition.

The main variable of interest in this analysis is Social Security income as reported by respondents in the 2012 interview. Specifically, HRS respondents are asked the following question (NQ085): “About the Social Security income that you (yourself) receive, how much was the Social Security check or the amount deposited directly into your account last month? We want the amount after any deductions”. Then, married respondents are asked the same question about their spouse’s Social Security benefits. Given the wording of this question, we hypothesize that most respondents report the net benefit amount received and ignore Medicare premiums, leading to an underestimate of respondents’ gross Social Security benefits. It is also plausible that some respondents may over- or under-report their net benefit amount or may even report the gross benefit amount.

To assess the true (whether net or gross) amount of Social Security benefits that survey respondents received, we use information from the Social Security administrative records. One of the advantages of the HRS is that respondents are asked to provide consent to merge their earnings and benefits records from Social Security Administration with their survey information for research purposes. For respondents who have claimed Social Security benefits, these restricted records contain detailed information about the type of benefits they receive, when they claimed benefits, the monthly amount of benefits they actually received from Social Security, and the amount of Medicare premiums paid to the Center for Medicare and Medicaid Services (CMS). We restrict our sample to respondents aged at least 65 and one month old at the beginning of interview in 2012 consisting of 10,713 observations.4 Of this total sample of elderly, about 56% (or 6,059 observations) have a match to their Social Security benefits record in year 2012. We use information for year 2012 because as of this writing, it is the latest year with matched Social Security administrative data available for the HRS sample.

For respondents with matched Social Security benefits records, we examine their reporting accuracy regarding the receipt and the amount of Social Security benefits by comparing their self-reported benefits with information available in their own administrative records. As noted above, the total Social Security benefit is comprised of the amount that is paid directly to the beneficiary and the amount that is paid for Medicare premiums (mainly Part B). For the majority of elderly beneficiaries, the amount of Social Security benefits received excludes the amount of Medicare Part B premiums withheld.5 For the sample of respondents with a match, the true amount of their Social Security benefits is available in the Social Security records from the Payment History Update System (PHUS) file.6 The unique characteristic of the PHUS is that it contains for each month both the amount of benefits paid to the beneficiary (either by check or as direct deposit to a bank account) and the amount of Medicare premiums (deducted from the beneficiary’s total Social Security benefits) paid by the Department of Treasury to the CMS. From these records, for each respondent we derive two annual measures for year 2012: the net amount and the gross amount of Social Security benefits. The annual net benefit amount is the sum of direct payments made to the beneficiary in each of the twelve months during the calendar year 2012, whereas the annual gross benefit amount is the sum of the direct payments and the Medicare premium payments made in each of the twelve months in 2012. We compare these measures to the self-reported annual Social Security benefit amounts in the RAND-HRS public file.7 For respondents in the 2012 wave who were interviewed during year 2012 we use their annual Social Security benefits amount.8 For those interviewed during February through May of year 2013, since they report monthly benefits received in 2013, we adjust their self-reported annual benefits to 2012 dollars using the 1.7% cost of living adjustment for Social Security benefits that took effect in January 2013 (i.e., we multiply by a factor of 0.983).

Table 1
Selected characteristics of respondents aged 65 and older in 2012, by status of match to their Social Security benefit records

Selected characteristics	With match ${}^{1}$	Without match ${}^{2}$
	(1)	(2)
Female (%)	55.6	56.8
Mean age (in years)	74.2	75.2 ${}^{\rm a}$
Age categories (%)
65–69	34.5	28.8 ${}^{\rm a}$
70–74	23.1	24.8 ${}^{\rm c}$
75–79	18.2	18.5
80–84	13.1	13.5
85+	11.1	14.5 ${}^{\rm a}$
Married (%)	59.4	58.4
Race (%)
Non-Hispanic white	84.1	78.3 ${}^{\rm a}$
Non-Hispanic black	7.4	10.1 ${}^{\rm a}$
Non-Hispanic other	2.3	2.5
Hispanic	6.2	9.1 ${}^{\rm a}$
Education (%)
Less than high school	16.2	21.7 ${}^{\rm a}$
High school graduate	36.2	35.7
Some college	22.6	21.8
College graduate	25.0	20.8 ${}^{\rm a}$
Health status (%)
Very good/excellent	39.8	38.3
Good	33.5	31.6 ${}^{\rm b}$
Fair/poor	26.6	29.9 ${}^{\rm a}$
Poverty status ${}^{3}$ (%)
Less than 1.0	9.1	10.2 ${}^{\rm c}$
1.0–1.99	20.4	22.1 ${}^{\rm b}$
2.0–3.99	32.1	33.3
4.0+	38.4	34.4 ${}^{\rm a}$
Total household income ($)
Mean	59,691	60,277
Median	38,612	35,476
Total household wealth ($)
Mean	508,357	516,331
Median	211,500	206,000
Self-reported medicare/medicaid coverage (%)
Covered by medicare	95.6	94.0 ${}^{\rm a}$
Covered by medicare part B	87.6	85.8 ${}^{\rm a}$
Covered by medicaid	9.2	9.2
N of Obs.	6,059	4,654

Then, we examine to what extent respondents’ report of net versus gross benefits affects the elderly poverty rate and their reliance on Social Security benefits.9 For this purpose, for respondents with a match, we replace the self-reported net Social Security benefits with the amount of gross benefits in their payment records. For respondents without a match, we add to their self-reported benefits an amount of $1,200 representing the typical Medicare premium amount paid in year 2012. Similarly, we adjust spouse’s Social Security benefits (if married and spouse is older than 65). Then, we adjust their total household income accordingly. Thus, for example, for couples without a match where both spouses are aged 65 or older and report receiving Social Security benefits, assuming that each spouse pays the standard amount of Medicare premiums, their total annual household income in 2012 would increase by $2,400 because of this adjustment. In addition, to get a more accurate estimate of poverty rate and reliance on Social Security benefits, we correct the amount of total household income to include the amount of potential withdrawals or distributions from IRAs or 401(k) type accounts.10 Dushi et al. [1] show that not accounting for such distributions will result in biased estimates of reliance on Social Security income. We follow a similar approach here and, based on respondents’ age and self-reported tax advantaged retirement account balances, calculate the stream of annual income that can be withdrawn from them. Since people at and over age 70½ are required by law to draw down their retirement account balances, we assume that they potentially have received such distributions even if they are not reported as income in the survey. To estimate the required distributions for each respondent aged 70 and over, we use the IRS Required Minimum Distribution (RMD) factor based on each respondent’s age and life expectancy. Previous research has shown that few people withdraw money from IRAs or 401(k) type accounts before the age of RMD requirement [14, 15, 16]. However, in order to maintain consistency in measuring household income from retirement accounts among older households, we derive the distribution factors for people aged 65–69 by extrapolating from the IRS RMD factor of those aged 70 and over, and then estimate their retirement accounts withdrawals.11

3. Results

In this section, we discuss our sample characteristics, followed by findings with respect to reporting accuracy of Social Security benefits and its implication for estimates of elderly poverty rate and elderly reliance on Social Security income. Section 4 concludes with some recommendations.

Table 2
Distribution of self-reported Social Security benefits among respondents aged 65 and older in 2012, by status of match to their benefits records

	Respondents self-reported Social Security benefits
	With match ${}^{1}$	Without match ${}^{2}$
	(1)	(2)
Percentiles
10th	1,112	1,116
25th	7,680	7,200
50th	12,192	12,000
75th	16,800	16,800
90th	21,024	21,600
Mean values	12,171	12,276
St. Dev.	7,123	7,739
N of Obs.	6,059	4,654

3.1 Sample characteristics and receipt of Social Security benefits

Table 1 provides demographic characteristics of the samples with and without matched records. The two samples are similar regarding gender, marital status, household income and wealth, and the proportion of respondents that report Medicaid coverage. However, the two samples differ regarding several dimensions. Compared to respondents without a match, those with a match are significantly more likely to be younger (ages 65–69), non-Hispanic whites, college graduates, and in good health; to be covered by Medicare as well as Medicare Part B; and less likely to be in poverty or near poverty.

When comparing the distribution of the survey-reported Social Security benefits separately for respondents with and without matched records we observe some small differences (Table 2). At the 25 ${}^{\rm th}$ percentile, those with a match report slightly higher benefits than those without a match (by about $500 or 6.3%) whereas at the median (50 ${}^{\rm th}$ percentile) their benefit is higher by only about $200 or 1.6%. In contrast, at the 90 ${}^{\rm th}$ percentile, respondents with a match report lower benefits than those without a match (about $600 or 2.7%). However, the mean Social Security benefits are not significantly different between the two groups. Given that respondents with a match differ from those without a match in many observable characteristics, the overall findings regarding reporting accuracy among matched respondents may not be generalizable for the non-matched respondents.

We now focus on the sample with matched records and compare the amounts of Social Security benefits as reported by respondents and as shown in the payment (PHUS) records. Table 3a shows that only 7.9% of respondents reported benefits that are inconsistent with administrative records. Out of those, 3.7% reported receiving a positive amount of Social Security benefits but the PHUS records indicate no payment was issued by the Department of the Treasury (a Type I error; all but one of these respondents have both Medicare payment and direct payment in the PHUS record equal to zero). The remaining 4.2% of respondents reported receiving no benefits whereas the PHUS records indicate that a payment was made (a Type II error).

Table 3a
Receipt of Social Security benefits as reported by respondents at the 2012 interview and as reported in the PHUS records, among respondents with a match aged 65 and over

	Zero	Positive	Total %	Total N
	PHUS net benefit amount ${}^{1}$
Social Security benefits	(1)	(2)	(3)	(4)
Self-reported benefits (cell %)
Zero	5.0	4.2	9.2	403
Positive	3.7	87.1	90.8	5,656
Total %	8.7	91.3	100
Total N	467	5,592		6,059

Table 3b

Distribution of Social Security benefit amounts in 2012 as reported by respondents and as reported in the PHUS records, among respondents with a match aged 65 and over

	Panel A. PHUS net annual benefit amount ${}^{1}$
	Zero	1–6000	6001–	12001–	18001–	24001+	Total %	Total N
			12000	18000	24000
Social Security benefits	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
Self-reported annual benefit amounts (cell %)
Zero	5.0	0.7	1.4	1.2	0.7	0.2	9.2	403
1–6000	1.0	5.6	2.3	0.8	0.3	0.0	9.9	663
6001–12000	1.9	1.0	22.7	3.6	0.7	0.2	30.0	1,917
12001–18000	0.7	0.1	1.5	27.3	3.4	0.2	33.2	2,059
18001–24000	0.1	0.1	0.4	2.0	10.4	0.8	13.7	778
24001+	0.1	0.1	0.3	1.0	0.6	2.0	4.1	239
Total %	8.7	7.6	28.5	35.8	16.0	3.4	100
Total N	467	493	1804	2,243	881	171		6,059
	Panel B. PHUS gross annual benefit amount ${}^{2}$
Self-reported annual benefit amounts (cell %)
Zero	4.9	0.8	1.3	1.2	0.8	0.3	9.2	403
1–6000	1.0	4.2	3.7	0.8	0.3	0.0	9.9	663
6001–12000	1.9	0.5	18.9	7.7	0.9	0.2	30.0	1,917
12001–18000	0.7	0.1	1.0	23.0	8.0	0.4	33.2	2,059
18001–24000	0.1	0.0	0.4	1.5	9.8	1.9	13.7	778
24001+	0.1	0.1	0.3	0.7	0.8	2.2	4.1	239
Total %	8.7	5.6	25.4	34.8	20.6	5.0	100
Total N	462	353	1633	2,179	1181	251		6,059

Table 3b provides the cross-tabulation of the distribution of the self-reported amount of benefits compared separately to the net and the gross amount of benefits in the PHUS. In each panel, the row and column benefit categories are created using the annual amounts of benefits. To put into perspective, within each broad category the upper and lower limits differ by $6,000 annually, or by $500 per month.

Panel A shows that, out of the total sample, 73% of respondents (the sum of the boldfaced percentages in the diagonal) reported benefits falling within the same broad categories as the net benefits in the payments records, suggesting that their self-reported benefits are close to the true net amount of Social Security payments. Of the remaining sample, 16.3% of them (the sum of all cells above the diagonal) under-reported their benefits. In other words, administrative records indicate that actual net benefits paid are higher than self-reported benefits, suggesting that these respondents clearly misreported their benefits. The other 10.7% of respondents (the sum of all cells below the diagonal) over-reported their benefits. This partly reflects the Type I error of respondents reporting non-existent Social Security benefits or reporting Supplemental Security Income instead (3.7%), whereas others may have over-reported their benefits, perhaps by including the amount of Medicare premiums.

Table 4

Distributions of Social Security benefits in 2012 as reported by respondents at the interview, as reported in the PHUS records, and of their differences, among respondents with a match aged 65 and over

Percentiles	Distribution of respondents’ survey-reported Social Security benefits	Distribution of PHUS net Social Security benefits ${}^{1}$	Distribution of PHUS gross Social Security benefits ${}^{2}$	Distribution of the net difference in Social Security benefits (1)–(2) ${}^{3}$	Distribution of the gross difference in Social Security benefits (1)–(3) ${}^{4}$	Distribution of the net ratio [(1)–(2)]/(2) ${}^{5}$	Distribution of the gross ratio [(1)–(3)]/(3) ${}^{6}$	Distribution of PHUS Medicare premiums ${}^{7}$
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
10th	1,112	1,990	2,780	$-$ 3,600	$-$ 4,726	0.00	0.01	0
25th	7,680	8,040	9,000	$-$ 840	$-$ 2,147	0.00	0.06	900
50th	12,192	12,960	14,040	$-$ 36	$-$ 1,224	0.04	0.11	1,200
75th	16,800	16,920	18,120	168	$-$ 60	0.20	0.25	1,200
90th	21,024	20,520	21,840	3,075	2,040	0.94	0.90	1,680
Mean	12,171	12,424	13,439 ${}^{\rm a}$	$-$ 253	$-$ 1,268	1.46	0.77	1,015
values
St. Dev.	7,123	6,657	6,920	5,162	5,253	6.45	2.82	622
N of Obs.	6,059

Notes: Authors’ calculations using data from the 2012 wave of the Health and Retirement Study (RAND-HRS user-friendly data file version P) matched with administrative data from the Social Security benefit records – the Payment History Update System (PHUS). The amount of self-reported Social Security benefits in column (1) is from the RAND data file, whereas benefits in columns (2) and (3) are from the PHUS administrative records. Sample consists of respondents aged at least 65 and one month old at the beginning of the interview and with a match to their PHUS payment records in 2012. Estimates are weighted using respondents weight from the 2012 interview. Number of observations is unweighted. The percentile categories are specific to each column. ${}^{1}$ The net annual benefit amount is the sum of monthly benefits paid to the beneficiary by check or as direct deposit to their bank account during year 2012. In other words, it excludes the amount of Medicare premiums paid on their behalf to the CMS during year 2012. ${}^{2}$ The gross annual benefit amount is the sum of monthly benefits paid to the beneficiary (by check or as direct deposit to their bank account) and the monthly amount of Medicare premiums paid on their behalf to the CMS during year 2012. ${}^{3}$ The net difference in Social Security benefits is calculated as the respondent’s self-reported annual benefit amount minus annual net benefit amount in the PHUS records. ${}^{4}$ The gross difference in Social Security benefits is calculated as the respondent’s self-reported annual benefit amount minus annual gross benefit amount in the PHUS records. ${}^{5}$ The net ratio is calculated as the absolute difference in Social Security benefits (self-reported minus net PHUS amount) relative to the net PHUS benefit amount. ${}^{6}$ The gross ratio is calculated as the absolute difference in Social Security benefits (self-reported minus gross PHUS amount) relative to the gross PHUS benefit amount. ${}^{7}$ The annual amount of Medicare premium is the sum of monthly premiums paid to the CMS on behalf of the beneficiaries during year 2012. ${}^{\rm a}$ denotes that the differences at the mean between column (1) vs. (3) and between (2) vs. (3) are statistically significance at the 1 percent level.

Compared to Panel A, the cross-tabulation in Panel B of self-reported benefits and gross benefits (including Medicare premium) in the payment records reveals a shift in higher cell-proportions toward the upper right corner. Thus, a lower proportion of HRS respondents have self-reported benefits in the same broad categories as those from the PHUS records (63%, the sum of all cells in the diagonal, versus 73% in Panel A). Furthermore, the proportion of respondents that seem to have under-reported gross benefits (cells above the diagonal) substantially increased to 28.3% (versus 16.3% in Panel A). Finally, the proportion of respondents that over-reported benefits (cells below the diagonal, Panel B) decreased to 9.0% (versus 10.7% in Panel A); i.e., they reported benefits that are much greater than the gross benefits in the payment records. In sum, these findings suggest again that majority of respondents reported net benefit, consistent with the question asked in the HRS, and not gross benefits.

3.2 Benefit amounts: Self-reported versus administrative record information

In Table 4 we examine the univariate distribution of the HRS reported benefits (col. 1) and of the net and gross benefits (cols 2 and 3, respectively) from the administrative payment records. Columns 4 and 5 report distributions of the difference between the survey- and PHUS-reported benefits measured at the individual level, whereas columns 6 and 7 show the distribution of the absolute difference relative to the amount of benefits in the PHUS record. Column 8 shows the distribution of Medicare premium as reported in the payment records.

Table 4 indicates that, throughout the distribution except in the 90 ${}^{\rm th}$ percentile, the survey-reported benefits (col. 1) are lower than the net benefits (col. 2) paid to the respondents after Medicare premium deductions. The mean value of survey-reported benefits is only $-$ $253 lower than the mean value of net benefits of $12,424 (a difference that is not statistically significant). Column 4 indicates that, while the difference at the median is only $-$ $36, there are substantial differences ranging from an underestimate of annual benefits of $-$ $3,600 at the 10 ${}^{\rm th}$ percentile to an overestimate of benefits of $3,075 at the 90 ${}^{\rm th}$ percentile. In addition, throughout the distribution, the survey-reported benefits are also lower than the actual gross benefits (col. 1 vs. col. 3). The mean of survey-reported annual Social Security benefits is $12,171 compared to the gross benefits of $13,439, and the mean difference of $-$ $1,268 (col. 5), or 9.4% relative to the mean of gross benefits, is statistically significant at the 1 percent level; the median difference is $-$ $1,224. These differences (both the mean and the median) are about equal to the median value of the Medicare premium paid on behalf of the beneficiaries in 2012 (col. 8); the standard annual Medicare Part B premium was about $1,200 in 2012. Columns 6 and 7 indicate that the absolute differences as proportions of the net and the gross benefits increase from 4% and 11% at the median, to 20% and 25% at the 75 ${}^{\rm th}$ percentile, and to 90% and 94% at the 90 ${}^{\rm th}$ percentile, suggesting that at least a quarter of respondents have misreported benefits by 20% or more.

Table 5a
Distribution of the absolute and relative difference between self-reported and PHUS annual Social Security benefits, among respondents with a match aged 65 and over

Absolute difference categories	Panel A. absolute difference
Absolute difference categories	Net difference ${}^{1}$		Gross difference ${}^{2}$
	Total %	Total N	Total %	Total N
	(1)	(2)	(3)	(4)
Both zero	5.0	228	5.0	224
Both positive and equal	5.4	326	1.4	80
Absolute difference between
1–1100	53.3	3,265	19.1	1,180
1101–1300	2.9	179	19.5	1,153
1301–1800	5.4	334	14.9	932
1801–2400	3.8	234	10.1	633
2401–3600	5.3	345	9.0	568
3601–4800	3.6	237	4.5	301
4801–9600	8.0	485	9.0	552
9601+	7.4	426	7.6	436
Total N	100	6,059	100	6,059
	Panel B. relative difference (ratio)
Relative difference categories	Net ratio ${}^{3}$		Gross ratio ${}^{4}$
Zero	10.4	554	6.4	304
Plus/minus 1 to 5%	42.2	2,555	13.6	797
Plus/minus 6 to 10%	13.0	798	24.0	1,452
Plus/minus 11 to 25%	12.5	809	31.5	1,994
Plus/minus more than 25%	21.9	1,343	24.5	1,512
Total	100	6,059	100	6,059

Table 5b

Sample distribution of the annual amount of Medicare premium in 2012, among respondents with a match aged 65 and over, by type of health insurance coverage

	Panel A: Distribution by Medicare premiums paid by type of health
	insurance coverage
			Self-reported coverage
PHUS Medicare part B annual premium paid	All		Medicare ${}^{1}$		Medicare part B (%) ${}^{2}$		Medicaid ${}^{3}$
PHUS Medicare part B annual premium paid	%	N	%	N	Yes	No	%	N
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
Zero	20.3	1,217	18.00	1,008	14.5	60.9	69.9	488
1–1100	6.7	368	6.4	349	6.3	9.4	7.0	57
1101–1200	58.1	3,557	60.2	3,515	62.8	25.6	18.8	128
1201–1680	10.0	617	10.4	613	11.0	2.8	3.4	22
1681–2400	3.7	230	3.8	228	4.1	0.9	0.9	4
2401–3120	0.9	49	0.9	49	1.0	0.2	0.2	1
3121+	0.4	21	0.4	20	0.4	0.3	0.0	0
Total %	100		100		100	100	100
Total N		6,059		5,862	5,410	649		700

	Panel B: Medicare part B premium and income thresholds ${}^{4}$
	2012 individual monthly premium	2012 income thresholds by premium level
		Individual	Couples
Standard premium	$99.90	< $85,000	< $170,000
Level 1	$139.90	$85,001–107,000	$170,001–$214,000
Level 2	$199.80	$107,001–160,000	$214,001–$320,000
Level 3	$259.70	$160,001–214,000	$320,001–$428,000
Level 4	$319.70	$214,000+	$428,000+

3.3 The magnitude of the difference between self-reported and payment benefit records

How different are the survey-reported benefits from the benefits actually paid? For each respondent, we calculate the absolute difference between the annual self-reported benefits and the annual net and gross benefits in the payment records. Table 5a, Panel A, reveals that the distribution of the differences between the self-reported and net benefits is less dispersed (or smaller differences) than when it is compared to the gross benefit amounts (direct payment plus Medicare premium). Thus, for about 63% of respondents (col. 1), the absolute net difference of annual benefits is less than $1,101, or a difference of less than 100 dollars per month. In contrast, for only about 25% of respondents, the absolute gross difference is less than $1,101, whereas for the remaining (about 75%), the difference is greater than that (col. 3).

Another way to gauge the difference between self-reported and true Social Security benefits is to look at the distribution of the relative difference – the ratio of the absolute difference to the benefit amounts in the payment records – separately for net and gross benefits. The relative difference is within 5% of the true net benefits for about 53% of respondents (Panel B, col. 1), compared to only 20% of respondents when using gross benefits (Panel B, col. 3).12 Nevertheless, for a substantial proportion of respondents (22% and 25%, respectively in cols 1 and 3), reported benefits differ from the true net or gross benefits by more than 25%. To summarize, these findings show that the majority of respondents report quite accurately the net amount of Social Security benefits received.

Table 5b, Panel A, shows the distribution of the annual amount of Medicare premiums deducted from the benefits in 2012 according to the payment records for the overall sample and separate subgroups, by premium level categories. In 2012, the standard Part B premium for most beneficiaries was $99.90 (Panel B) totaling to an annual amount of about $1,200 [18]. Depending on the beneficiary’s income level and filing status, the monthly premium increases to $139.9 (Level 1), $199.90 (Level 2), $259.70 (Level 3) and to $319.70 (Level 4). We use these monthly premium levels to create the annual categories in Panel A. About 20% of the overall sample paid zero Medicare premium in 2012 according to the payment records, whereas about 7% paid an annual amount of less than $1,100 (Panel A, col. 1). Note that annual premiums of less than $1,100 may be due to partial year of receipt of Social Security benefits and of Medicare Part B coverage. The majority of the sample (about 58%) had annual payments between $1,101 and $1,200, i.e., they paid the standard monthly premium in 2012. Of the remaining sample, 10% paid between $1,201 and $1,680 annually in Medicare premiums or around $139.9 per month (Level 1), whereas only 5% of respondents paid annual Medicare premiums of above $1,681 or more than $140 per month. A similar pattern occurs among respondents who self-reported Medicare coverage (col. 3).

Table 6
Mean and median values of the net and the gross difference between respondents- and PHUS-reported Social Security benefits, of the Medicare premiums, and the sample distribution by Medicare premium paid in 2012, among respondents with a match aged 65 and over, by selected characteristics

	Net difference		Gross difference		PHUS Medicare		% of sample paying
	($) ${}^{1}$		($) ${}^{2}$		premiums ($) ${}^{3}$		Medicare premium ${}^{4}$
	Mean	Median	Mean	Median	Mean	Median	Zero	Standard	Levels 1–4
Selected characteristics	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
All	$-$ 253	$-$ 36	$-$ 1,268	$-$ 1,224	1,015	1,200	20	65	15
Gender
Female	245	$-$ 12	$-$ 728	$-$ 1,200	973	1,200	24	62	14
Male	$-$ 878	$-$ 60	$-$ 1,946	$-$ 1,260	1,068	1,200	16	69	15
Age categories
65–69	$-$ 1,291	$-$ 36	$-$ 2,159	$-$ 1,212	868	1,200	28	60	12
70–74	$-$ 14	$-$ 48	$-$ 1,086	$-$ 1,248	1,072	1,200	18	65	17
75–79	427	$-$ 12	$-$ 671	$-$ 1,212	1,098	1,200	17	66	17
80–84	565	$-$ 24	$-$ 529	$-$ 1,224	1,094	1,200	16	69	15
85+	393	$-$ 36	$-$ 732	$-$ 1,236	1,125	1,200	14	71	16
Marital Status
Married	$-$ 244	$-$ 24	$-$ 1,304	$-$ 1,224	1,060	1,200	17	68	15
Not-married	$-$ 268	$-$ 36	$-$ 1,217	$-$ 1,222	949	1,200	26	60	14
Race
Non-Hispanic white	$-$ 209	$-$ 24	$-$ 1,275	$-$ 1,232	1,067	1,200	17	67	16
Non-Hispanic black	$-$ 749	$-$ 60	$-$ 1,550	$-$ 1,200	801	1,200	34	57	9
Non-Hispanic other	$-$ 844	$-$ 60	$-$ 1,684	$-$ 1,210	841	1,200	29	58	13
Hispanic	$-$ 6	$-$ 60	$-$ 646	$-$ 960	640	800	44	51	6
Education
Less than high school	20	$-$ 36	$-$ 760	$-$ 1,176	780	1,200	35	56	10
High school graduate	$-$ 199	$-$ 36	$-$ 1,231	$-$ 1,236	1,032	1,200	18	70	13
Some college	$-$ 368	$-$ 48	$-$ 1,451	$-$ 1,236	1,083	1,200	14	70	16
College graduate	$-$ 397	$-$ 12	$-$ 1,478	$-$ 1,214	1,082	1,200	20	60	20
Health status
Very good/excellent	$-$ 10	$-$ 24	$-$ 1,070	$-$ 1,224	1,060	1,200	18	65	17
Good	$-$ 307	$-$ 26	$-$ 1,347	$-$ 1,224	1,040	1,200	18	68	14
Fair/poor	$-$ 543	$-$ 48	$-$ 1,460	$-$ 1,223	917	1,200	27	60	13
Poverty status
Less than 1.0	$-$ 1,351	$-$ 100	$-$ 1,970	$-$ 1,188	619	100	49	41	10
1.0–1.99	$-$ 224	$-$ 36	$-$ 1,151	$-$ 1,200	927	1,200	25	63	12
2.0–3.99	$-$ 183	$-$ 36	$-$ 1,315	$-$ 1,236	1,131	1,200	10	75	15
4.0+	$-$ 68	$-$ 20	$-$ 1,127	$-$ 1,224	1,058	1,200	20	63	17
Total household income
quintiles
First (lowest)	$-$ 966	$-$ 60	$-$ 1,640	$-$ 1,200	674	900	44	47	9
Second	$-$ 149	$-$ 24	$-$ 1,173	$-$ 1,212	1,024	1,200	18	67	14
Third	$-$ 173	$-$ 36	$-$ 1,321	$-$ 1,248	1,148	1,200	8	77	15
Fourth	$-$ 36	$-$ 36	$-$ 1,152	$-$ 1,236	1,116	1,200	12	74	14
Fifth (highest)	$-$ 112	$-$ 12	$-$ 1,154	$-$ 1,224	1,041	1,200	23	58	19
Total household wealth
quintiles
First (lowest)	$-$ 329	$-$ 24	$-$ 1,045	$-$ 1,164	716	1,200	40	50	10
Second	$-$ 680	$-$ 60	$-$ 1,612	$-$ 1,236	932	1,200	23	65	12
Third	$-$ 342	$-$ 48	$-$ 1,397	$-$ 1,236	1,056	1,200	14	75	11
Fourth	75	$-$ 24	$-$ 1,022	$-$ 1,224	1,097	1,200	14	71	15
Fifth (highest)	$-$ 89	$-$ 24	$-$ 1,285	$-$ 1,248	1,196	1,200	14	62	23

Table 6, continued
	Net difference		Gross difference		PHUS Medicare		% of sample paying
	($) ${}^{1}$		($) ${}^{2}$		premiums ($) ${}^{3}$		Medicare premium ${}^{4}$
	Mean	Median	Mean	Median	Mean	Median	Zero	Standard	Levels 1–4
Selected characteristics	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
Has Medicare coverage
Yes	$-$ 194	$-$ 36	$-$ 1,242	$-$ 1,224	1,048	1,200	18	67	15
No	$-$ 1,545	$-$ 24	$-$ 1,843	$-$ 460	298	0	70	27	3
Has Medicaid coverage
Yes	219	0	$-$ 121	$-$ 36	340	0	70	26	4
No	$-$ 301	$-$ 36	$-$ 1,384	$-$ 1,236	1,083	1,200	15	69	16
Has Medicare part B
Yes	$-$ 133	$-$ 36	$-$ 1,231	$-$ 1,236	1,098	1,200	15	69	16
No	$-$ 1,099	$-$ 32	$-$ 1,534	$-$ 472	435	0	61	35	4
N of Obs.	6,059

Among those who report coverage by Medicare Part B (col. 5), a majority (69%; 6.3% $+$ 62.8%) paid the standard premium amount and only about 15% paid zero in Part B premiums. In contrast, among those reporting no Medicare Part B coverage (col. 6), 61% paid no Part B premium in 2012 and 35% actually paid the standard premium. Note that the distribution for this latter group is closer to that of respondents covered by Medicaid (col. 7). Thus, about 70% of those reporting Medicaid coverage had no Part B Medicare premium deducted from their Social Security benefits, probably because the premiums are paid by the Medicaid program. Of the remaining sample, around 26% paid the standard premium amount of up to $1,200 annually, and a trivial proportion (4.5%) paid more than the standard amount.

Table 7

Distribution of poverty status in 2012 according to the RAND-HRS poverty measure and according to our adjusted measure of poverty, among respondents aged 65 and over, by match status

Poverty status using RAND data ${}^{1}$ (cell %)	Poverty status using RAND adjusted data ${}^{2}$
Poverty status using RAND data ${}^{1}$ (cell %)	< 1.0	1.0–1.99	2.0–3.99	4.0+	Total %	Total N
	(1)	(2)	(3)	(4)	(5)	(6)
	All respondents
< 1.0	7.9	1.7	0.0	0.0	9.6	1,077
1.0–1.99	0.0	16.9	4.0	0.1	21.1	2,331
2.0–3.99	0.0	0.0	28.0	4.7	32.6	3,477
4.0+	0.0	0.0	0.0	36.7	36.7	3,374
Total %	7.9	18.6	32.0	41.5	100.0
Total N	884	2,071	3,459	3,845		10,259
	Respondents without a matched benefits record
< 1.0	8.3	1.9	0.0	0.0	10.2	485
1.0–1.99	0.0	17.6	4.4	0.1	22.1	1,053
2.0–3.99	0.0	0.0	28.0	5.3	33.3	1,491
4.0+	0.0	0.0	0.0	34.4	34.4	1,358
Total %	8.3	19.5	32.4	39.8	100.0
Total N	380	958	1,478	1,571		4,387
	Respondents with matched benefit record
< 1.0	7.6	1.5	0.0	0.0	9.1	592
1.0–1.99	0.0	16.4	3.8	0.2	20.4	1,278
2.0–3.99	0.0	0.0	27.9	4.2	32.1	1,986
4.0+	0.0	0.0	0.0	38.4	38.4	2,016
Total %	7.6	17.9	31.7	42.8	100.0
Total N	504	1,113	1,981	2,274		5,872

Notes: Authors’ calculations using data from the 2012 wave of the Health and Retirement Study (RAND-HRS user-friendly data file version P) matched with administrative data from the Social Security benefit records – the Payment History Update System (PHUS). Sample consists of respondents aged at least 65 and one month old at the beginning of the interview. Estimates are weighted using respondents weight from the 2012 interview. Cell proportions may not add to 100% due to rounding. Reported numbers of observations are unweighted. ${}^{1}$ The poverty status categories are derived from the RAND data file. More specifically they are derived from the RAND created variable called poverty ratio (H11INPOVR) calculated as the ratio of the total household income (H11ITOT) to the Census-given poverty threshold (H11POVTHR) according to each respondent’s family composition in 2012; it excludes members of the household that are in institution. ${}^{2}$ These poverty status categories are derived after we adjust total household income variable in the RAND data file. For majority of the sample the variable total household income (H11ITOT) is the same as variable H1POVHHI (which provides the household income to be compared to the poverty threshold). The two variables differ only for those households where at least a member is in an institution. Therefore, for the purpose of estimating the poverty, we adjust the latter variable (H11POVHHI), which excludes the income of the institutionalized household member(s), by adding the annual amount of Medicare premium only for the non-institutionalized household members that are aged at least 65 and one month old, and the amount of potential withdrawals from retirement accounts. Then we recalculate the poverty ratio using the same threshold variable (H11POVTHR) as provided in the RAND data. To be more specific about our adjustment, for respondents with a matched record we replace the part of self-reported Social Security income with the gross Social Security benefits that is available in the PHUS records (which includes the amount of Medicare premium). For respondents without a matched record we add $1,200, which is the typical amount of Medicare premium for the majority of older people in year 2012. Similarly, for each spouse in a coupled household for whom we do not have a matched record we add $1,200 to their household income. For couples where both spouses have a matched record we add the actual Medicare premium from their respective records. If one of the spouses has no matched record we add $1,200 for that spouse. For those coupled unit where the spouse is under 65 years old, we do not add the Medicare premium unless the spouse has a matched record and the Medicare premium is positive, which most likely is due to the spouse being a disability beneficiary. Finally, we also add to the adjusted total household income the amount of potential withdrawals from retirement accounts calcualted using IRS RMD guidelines (see discussed in the data section).

Both the gross and the net differences between the survey-reported and the true benefits in payment records vary by socio-economic and demographic characteristics (Table 6). At the median, the underreporting of the net Social Security benefits varies little by demographic characteristics (col. 2). In contrast, at the mean (col. 1), the magnitude of underreporting of net benefits is greater than the overall sample mean (of $-$ $253) among several groups. These include men, beneficiaries aged 65–69, Non-Hispanic blacks and other races (mainly Asian), those with some college or a college degree, those in good or poor/fair health, those with income less than the poverty threshold, those in the bottom of household income and wealth, and those reporting no Medicare coverage. Surprisingly, beneficiaries aged 75 and older, women, and those reporting Medicaid coverage over-reported their benefits; this finding may reflect outliers since, at the median, they also underreported their benefits. While the average amount of underreporting of gross benefits (col. 3) is larger than that for net benefits (col. 1), similar patterns of underreporting are prevalent across demographic categories. The gross differences are almost proportional to the Medicare premiums for each category (cols 5 and 6).

In Table 6, columns 7 to 9 provide the distribution of the sample, and of each subgroup, by levels of Medicare premiums paid in 2012. Overall, 20% of the sample paid no Medicare premiums in 2012, whereas 65% paid the standard monthly amount of about $99.90, and only 15% paid more than that. There are substantial differences by demographic characteristics in terms of the proportion that pays nothing in Medicare premiums and the portion that pays more than the standard amount. As expected, the demographic subgroups that were more likely to have paid Medicare premiums above the standard amount are college graduates, those with income four times above the poverty threshold, and those with household income and wealth in the highest quintiles.

3.4 Implications of measurement bias of Social Security benefits in HRS

As the findings above show, a majority of the elderly accurately reported their net Social Security benefits. Hence, the HRS does a good job in measuring the net Social Security benefits, while at the same time it does not measure the gross amount of benefits. Consequently, two measures – the Social Security (OASI) benefit and the total household income – are underestimated in the HRS, suggesting that estimates that utilize these two variables, such as poverty rate and reliance on Social Security benefits, are likely inaccurate.

First, we assess the impact of underreporting of gross Social Security benefits on the elderly poverty rate. Table 7 presents the cross-distribution of poverty under two different poverty classifications, for the whole sample of elderly respondents and separately for those with and without a match to the administrative payment records. The rows of the table classify respondents by the poverty status measure available in the RAND-HRS data file (which uses self-reported net Social Security benefits). The columns classify respondents by a poverty measure that we calculated after adjusting the total household income to reflect gross Social Security benefits and potential withdrawals from the IRA accounts, while using the same poverty threshold for each family composition as the one in the RAND-HRS file. The key poverty categories – less than 1.0 and 1.0–1.99 – identify those in poverty and those near poverty, respectively. About 9.6% of elderly respondents are classified as being in poverty according to the RAND variable (row 1, col. 5). However, when using the adjusted income measure, the proportion of elderly in poverty decreases to 7.9% (row 5, col. 1), and the proportion of those near poverty decreases from 21.1% (row 2, col. 5) to 18.6% (row 5, col. 2). Thus, overall, proportion of those in poverty or near poverty decreases by about 4 percentage points. Similar patterns are evident when looking at the sample of respondents with and without a match (similar tabulations by marital and Medicaid coverage status are available upon request from the authors).

Table 8
Proportion of respondents residing in families with at least 50% and at least 90% of family income from Social Security benefits, in 2012

	50% or more	90% or more	Total N
Measures used	(1)	(2)	(3)
Survey-reported meaures ${}^{1}$	50.8	21.1	10,713
Survey-reported Social Security income	47.4	17.6	10,713
+ IRA withdrawals ${}^{2}$
Adjusted measures ${}^{3}$	47.3	17.1	10,713
Adjusted measures for matched sample ${}^{4}$	43.3	14.0	6,059

Notes: Authors’ calculations using data from the 2012 wave of the Health and Retirement Study (RAND-HRS user-friendly data file version P) matched with administrative data from the Social Security benefit records – the Payment History Update System (PHUS). Sample consists of respondents aged at least 65 and one month old at the beginning of the interview. Estimates are weighted using respondents weight from the 2012 interview. Reported numbers of observations are unweighted. ${}^{1}$ Estimates are derived as the ratio of self-reported household’s total Social Security benefits and total household income (H11ITOT) available in the RAND data file. The total household income includes earnings, pension and annuities, social security benefits, income from welfare programs, and other income for respondents and their spouses. ${}^{2}$ Estimates are derived as the ratio of self-reported amount of household’s total Social Security benefits (in the RAND data file, same as in row 1) and the adjusted denominator that equals the total household income (H11ITOT) available in the RAND file plus the amount of annual potential withdrawals from IRA accounts calculated using IRS RMD guidlines (see discussion in the data section). ${}^{3}$ Estimates are derived by adjusting both the numerator and the denominator variables in the RAND data file. First, the numerator – the amount of household income from Social Security is adjusted to reflect the gross benefits. More specifically, for people with a match we replace their self-reported Social Security benefits with gross benefits (annual direct deposit + Medicare premiums) in the PHUS payment records. For respondents without a match we add to their self-reported benefits an amount of $1200 that is the average Medicare premium observed in the PHUS records in 2012. For married couples we adjust the Social Security benefit amount of their spouse in the same way. Thus, for couples the amount of Social Security benefits is the sum of the corrected benefits received from both spouses. Second, we adjust the denominator – the total household income – to reflect the corrected amount of Social Security benefits as discussed above. Then, to this corrected measure of total household income we add the annual amounts of potential withdrawals from IRA accounts (same as in row 2). ${}^{4}$ These estimates use the same adjustments as those in row 3. The only difference is that here we restrict the sample to those with a match to their PHUS records in 2012.

Next, in Table 8 we provide estimates of the proportion of elderly respondents with at least 50% and at least 90% of their family income coming from Social Security (similar estimates by selected socio-economic characteristics are available upon request by the authors). To assess the impact of underreporting of gross Social Security benefits, we contrast the estimates using survey-reported measures (based on net Social Security benefits and total household income) with estimates based on our adjusted measure of household income (which includes gross amount of Social Security benefits and withdrawals from retirement accounts). Row 1 shows that, when using survey-reported income measures, 50.8% of elderly respondents receive at least 50% of their family income from Social Security, whereas 21.1% receive at least 90% of their income from Social Security. When we adjust household income to include potential withdrawals from retirement accounts, the respective proportions decrease to 47.4% and 17.6% (row 2).

In row 3, we report estimates where, in addition to the adjustment for withdrawals from retirement accounts (as in row 2), we also correct both the numerator (household Social Security benefits) and the denominator (total household income) to reflect the gross amount of Social Security benefits received by respondents and their spouses if married. Because this correction is similar in the numerator and denominator, we do not expect a substantial change in the proportion of respondents with at least 50% and 90% of their income from Social Security (rows 3 vs. 2). However, the estimates in row 3 are the most accurate estimates of the reliance on Social Security benefits among people aged 65 and over.

Compared to survey reported measures (row 1), after adjusting for the gross amount of Social Security benefits and income from retirement accounts, the overall proportion of elderly relying on Social Security income for 50% or more and 90% or more of their household income decreases by about 3–4 percentage points to 47.3% and 17.1% (row 3), respectively.13 Finally, for people with matched records reliance on Social Security is even lower, by 3–4 percentage points (row 4 vs. row 3). Note that the observed differences in estimates, between the two subgroups in Table 7 and the last two rows in Table 8, reflect the differences in observable characteristics between respondents with and without a match to payment records.

In sum, these findings confirm that measurement issues regarding Social Security income and household income in the HRS produce upwardly biased estimates of the elderly poverty rate and of their reliance on Social Security. It is worth noting, however, that even after our adjustments, about half of elderly respondents rely on Social Security benefits for at least 50% of their family income.

4. Conclusions and recommendations

Social Security benefits are the most important source of income for a considerable proportion of the elderly population. Thus, it is important for policy makers to have accurate estimates of income security and wellbeing for elderly beneficiaries. Commonly, such estimates are based on survey reports, which are subject to measurement and/or reporting errors. In this paper, we examine the extent and the magnitude of reporting accuracy of Social Security benefits in the HRS and the implications of errors on estimates of the poverty rate among the elderly and their reliance on Social Security income.

Our findings indicate that, consistent with the question asked, 73% of HRS respondents reported the net amount of Social Security benefits they received (excluding Medicare premium). At the mean, the self-reported benefits and the net benefits in the administrative payment records are not statistically significantly different, indicating that the majority of respondents are accurate about their net benefits. In contrast, the mean difference between the self-reported and the gross benefits in the payment records is statistically significant and almost equal to the mean Medicare premium, another indication that people report net benefits and not their gross benefits. Still, about 22% of respondents misreported their annual Social Security benefits by more than 25% (or by more than $2,400) as compared to their net benefits in the payment records. Findings also reveal that, due to the downward bias of the gross Social Security benefits and exclusion of IRA distributions, proportions of elderly respondents that are poor or nearly poor and estimates of their reliance on Social Security benefits, are biased upward by about 2 to 4 percentage points. Importantly, even after our correction of Social Security income that includes Medicare premiums and withdrawals from retirement accounts, for about half of elderly respondents in our sample, Social Security benefits remain a significant income source, comprising 50% or more of their family income.

Given the differences in observable characteristics between respondents with and without a match to the payment records, it is worth noting that the magnitude of reporting accuracy of Social Security benefits among those with a match may not necessarily apply to those without a match. Since respondents in the latter group are more likely to be poor or near poor, and have lower income, they would be more likely to have their Medicare premium paid by Medicaid. Therefore, their gross benefit amount may not differ from their net benefits, so their reliance on Social Security and their poverty rate are less likely to be affected by the omission of Medicare premium.

There are alternative ways to adjust survey-reported net Social Security benefits in the HRS so that a variable of the gross amount of Social Security benefits is also available in the public file for data users. However, the simplest way that we would recommend is to correct the survey-reported net Social Security benefits for respondents (and their spouse if married) who are aged 65 and over and who do not report Medicaid coverage, by just adding the standard (or when available the self-reported) amount of Medicare premiums to the self-reported benefit amount for the respective year. Once the Social Security income variable for both the respondent and the spouse is adjusted, then the total household income and poverty measures available in the RAND-HRS public file also need to be adjusted and made available in the public file.

Footnotes

Due to lack of publicly available documentation, we have no way of knowing about the method the Census Bureau used to impute the Medicare premiums in the CPS public data.

Evidence of respondent misreports are also found regarding government transfer programs [3], Social Security wealth, pension plans type and characteristics [4, 5, 6, 7], contributions to defined contribution plans [8, 9], and retirement income [].

The RAND HRS user-friendly public data file (version P used here, available at http://hrsonline.isr.umich.edu/) – the most commonly used source of data – contains consistently created variables across all waves/interviews. However, data users may miss the fact that reported Social Security benefits are net of any deductions.

Because respondents are asked about Social Security benefits received last month, we condition the sample to those aged at least 65 and one month old at the beginning of the interview, so that the benefit amount reported corresponds to the month after respondent reached age 65. In the remainder of the paper, we interchangeably use “65 and over” to refer to the sample aged at least 65 and one month old.

Part C and Part D premiums may also be deducted from the Social Security payment, but beneficiaries have to notify their insurance and/or prescription drug plan if they want the premiums to be deducted from their monthly Social Security payments. However, in the administrative records the field that contains the amount of Medicare premiums does not indicates whether Parts C and D premiums are also deducted from gross benefits.

For the remainder of this paper we use the terms PHUS records and payment records interchangeably.

For each respondent (and spouse, if married), the monthly Social Security benefits are annualized and made available in the RAND-HRS file. Since the annual amount of benefit is derived from the self-reported monthly benefits and the number of months the respondent report receiving such benefits in the given year, we refer to this annual amount of benefits as self-reported amount.

This sample also includes those respondents interviewed in January of 2013 since they reported the amount of benefit received in the last month (i.e., in December 2012).

Poverty rate is the ratio of total family income to poverty threshold for each Census defined family composition. Reliance on Social Security benefits is the ratio of household income from Social Security to total household income.

Beginning with the 2015 Current Population Survey, the U.S. Census Bureau measures these distributions as pension income and includes them in the official estimates of poverty and income [12, ].

All data analyses in this study were implemented using Stata/MP 14 for Windows (64 bit) on a HP Probook Computer [].

To put it into context, for a person receiving, for example, a monthly benefit of $1,000 (or 12,000 annually), the 5% relative difference would mean an error of $50 per month (or $600 annually).

To test the sensitivity of these results, we applied an alternative adjustment to gross Social Security benefits to the one in row 3, where for the subset of respondents without a match to PHUS records and who reported coverage under Medicaid we assigned a zero Part B Medicare premium. Our estimates show that, at the mean, results were similar to those in row 3. Alternative estimations based only on adjustments of Social Security benefits and excluding the IRA withdrawals, revealed similar results to those in row 1. These estimates are available upon request from the authors.

Acknowledgments

The opinions and conclusions expressed herein are solely those of the authors and do not necessarily representing the opinions or policy of the Social Security Administration or any agency of the Federal Government. We thank David Weir and Susann Rohwedder for helpful comments. All errors are our own.

References

Dushi

Iams

Trenkamp

. A note on the importance of Social Security benefits to the income of the aged population. Social Security Bulletin 2017; 77(2): 1-12.

Iams

Purcell

. Social Security income measurement in two surveys. Social Security Bulletin 2013; 73(3): 1-10.

Meyer

Mittag

. Using linked survey and administrative data to better measure income: Implications for poverty, program effectiveness and holes in the safety net. NBER Working Paper 2015; 21676.

Gustman

Steinmeier

. What people don’t know about their pensions and Social Security. William GG, Shoven JB, Warshawsky MJ, editors, Private Pensions and Public Policies. Washington, DC: Brooking Institution 2004; 57-119.

Gustman

Steinmeier

. Imperfect knowledge of social security and pensions. Industrial Relations 2005; 44(2): 373-397.

Mitchell

. Worker knowledge of pension provisions. Journal of Labor Economics 1988; 6(1): 28-39.

Chan

Stevens

. What you don’t know can’t help you: Knowledge and retirement decision making. NBER Working Paper 2003; 10185.

Dushi

Honig

. How much do respondents in the Health and Retirement Study know about their contributions to tax deferred contribution plans? A cross-cohort comparison. Journal of Pension Economics and Finance 2015; 14(3): 203-239.

Dushi

Iams

. The impact of response error on participation rates and contributions to defined contribution pension plans. Social Security Bulletin 2011; 70(1): 45-60.

10.

Bee

Mitchell

. Do older Americans have more income than we think? SESHD Working Paper 2017; 39. Available from: https://www.census.gov/.

11.

University of Michigan, Institute for Social Research, Health and Retirement Study. The Health and Retirement Study: Aging in the 21st Century. Challenges and Opportunities for Americans 2017. Available from: hrsonline.isr.umich.edu/sitedocs/databook/inc/pdf/HRS-Aging-in-the-21St-Century.pdf.

12.

Semega

, The effects of changes to the Current Population Survey Annual Social and Economic Supplement on Estimates of Income 2015. Available from: https://www.census.gov/.

13.

Mitchell

Renwick

. A comparison of official poverty estimates in the redesigned Current Population Survey Annual Social and Economic Supplement. SEHSD Working Paper 2014-35 2015. Available from: https://www.census.gov/.

14.

Holden

Schrass

. The role of IRAs in U.S. households’ saving for retirement, 2013. ICI Research Perspective 2016; 22(1). Available from: www.ici.org.

15.

Poterba

Venti

Wise

. The composition and drawdown of wealth in retirement. Journal of Economic Perspectives 2011; 25(4): 95-118.

16.

Poterba

Venti

Wise

. The drawdown of personal retirement assets. NBER Working Paper 2011; 16675.

17.

StataCorp, Stata Statistical Software: Release 14 (College Station, TX: StataCorp LP) 2014.

18.

Center for Medicare and Medicaid Services. Medicare Premiums and Deductibles for 2012 2011. Available from: https://www.cms.gov/.

Reporting accuracy of Social Security benefits and its implications in the Health and Retirement Study

Abstract

Keywords

1. Introduction

2. Data and methodology

Table 1 Selected characteristics of respondents aged 65 and older in 2012, by status of match to their Social Security benefit records

Table 2 Distribution of self-reported Social Security benefits among respondents aged 65 and older in 2012, by status of match to their benefits records

Table 3a Receipt of Social Security benefits as reported by respondents at the 2012 interview and as reported in the PHUS records, among respondents with a match aged 65 and over

Table 5a Distribution of the absolute and relative difference between self-reported and PHUS annual Social Security benefits, among respondents with a match aged 65 and over

Table 6 Mean and median values of the net and the gross difference between respondents- and PHUS-reported Social Security benefits, of the Medicare premiums, and the sample distribution by Medicare premium paid in 2012, among respondents with a match aged 65 and over, by selected characteristics

Table 8 Proportion of respondents residing in families with at least 50% and at least 90% of family income from Social Security benefits, in 2012

Footnotes

Acknowledgments

References

Table 1
Selected characteristics of respondents aged 65 and older in 2012, by status of match to their Social Security benefit records

Table 2
Distribution of self-reported Social Security benefits among respondents aged 65 and older in 2012, by status of match to their benefits records

Table 3a
Receipt of Social Security benefits as reported by respondents at the 2012 interview and as reported in the PHUS records, among respondents with a match aged 65 and over

Table 5a
Distribution of the absolute and relative difference between self-reported and PHUS annual Social Security benefits, among respondents with a match aged 65 and over

Table 6
Mean and median values of the net and the gross difference between respondents- and PHUS-reported Social Security benefits, of the Medicare premiums, and the sample distribution by Medicare premium paid in 2012, among respondents with a match aged 65 and over, by selected characteristics

Table 8
Proportion of respondents residing in families with at least 50% and at least 90% of family income from Social Security benefits, in 2012