Correcting for Underreporting of Government Transfers: A Regression-Based Approach With Preliminary Results
1 like3,893 views
The Congressional Budget Office (CBO) presented a workshop on correcting underreporting of government transfers, focusing on a regression-based approach with initial results. The CBO aims to provide a more accurate accounting of transfer income to reflect its role in reducing income inequality, particularly among the lower income distribution segments. Their methodology involves using reported data to estimate transfer receipt probabilities and adjust for underreporting in the Current Population Survey.
Correcting for Underreporting of Government Transfers: A Regression-Based Approach With Preliminary Results
- 1. Congressional Budget Office Presentation at a Workshop Organized by the Washington Center for Equitable Growth July 21, 2017 Bilal Habib Tax Analysis Division Correcting for Underreporting of Government Transfers: A Regression-Based Approach With Preliminary Results As developmental work for analysis for the Congress, the information in this presentation is preliminary and is being circulated to stimulate discussion and critical comment.
- 3. CONGRESSIONAL BUDGET OFFICE CBO regularly produces a report on the distribution of household income and federal taxes. This year, the agency is working to change its analytical framework to treat means- tested transfers as equivalent to taxes.
- 4. CONGRESSIONAL BUDGET OFFICE Old Analytical Framework Market Income Before-Tax Income After-Tax Income Government Transfers (Includes social insurance benefits and means-tested transfers) Federal Taxes
- 5. CONGRESSIONAL BUDGET OFFICE New Analytical Framework Market Income Market Income Plus Social Insurance Benefits Income After Taxes and Transfers (Equal to After-Tax Income) Social Insurance Benefits Federal Taxes Means-Tested Transfers
- 6. CONGRESSIONAL BUDGET OFFICE Explicit analysis of government transfers requires a complete accounting of transfer income. CBO’s tax model uses the Annual Social and Economic Supplement of the Current Population Survey (CPS) for transfer income data.
- 7. CONGRESSIONAL BUDGET OFFICE Underreporting of transfer income in the CPS has increased over time, as is well documented in Wheaton (2008), Meyer, Mok, & Sullivan (2009), and Moffitt & Scholz (2009). As a result, CPS-based analyses are likely to understate income growth at the bottom of the distribution and the role of transfers in reducing income inequality.
- 8. CONGRESSIONAL BUDGET OFFICE Reporting Rates in the CPS: Means-Tested Transfers, 1979–2013 CPS Recipients as a Percentage of Administrative Recipients 0 20 40 60 80 100 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 2011 2013 73 66 53 Medicaid SSI SNAP
- 9. CONGRESSIONAL BUDGET OFFICE CBO’s goal is to obtain a more complete (although partially imputed) accounting of income from government transfers in the CPS with enough precision for quintile- level distributional analysis.
- 10. CONGRESSIONAL BUDGET OFFICE The analysis is focused on three of the largest means-tested transfers—Medicaid, the Supplemental Nutrition Assistance Program (SNAP), and Supplemental Security Income (SSI)—and the two largest social insurance benefits—Social Security and Medicare.
- 11. CONGRESSIONAL BUDGET OFFICE Means-Tested Transfers
- 12. CONGRESSIONAL BUDGET OFFICE Researchers typically use three ways to correct for underreporting: • Administrative matching, • Rules-based simulation, and • Regression-based estimation.
- 13. CONGRESSIONAL BUDGET OFFICE Administrative matching offers near-perfect accounting, but administrative microdata are not widely available. Examples: Davern et al. (2009); Meyer and Sullivan (2008). Rules-based simulation offers precise estimates at the micro level, but requires a significant research investment. Example: Zedlewski and Giannarelli (2015). Regression-based estimation is tractable for multiple programs/years but is less precise at the micro level. Example: Moffitt and Scholz (2009).
- 14. CONGRESSIONAL BUDGET OFFICE CBO’s preliminary regression-based estimation has three steps: 1. Use reported data to estimate the probability of receipt for all units. 2. Impute transfer receipt based on estimated probabilities. 3. Assign transfer income to recipients.
- 15. CONGRESSIONAL BUDGET OFFICE Step 1. Predicted probabilities are estimated using a probit model with CPS- reported receipt as the dependent variable. Independent variables are based on program rules and other characteristics associated with program participation.
- 16. CONGRESSIONAL BUDGET OFFICE Individual characteristics include age, race, education, labor force status, disability, marital status, and receipt of other means-tested transfers. Household/family characteristics include income (as a percentage of the federal poverty level), income composition, household size and structure, and geography.
- 17. CONGRESSIONAL BUDGET OFFICE Step 2. Transfer receipt is imputed to nonreporters with the highest probability of receipt until the administrative total is reached. This process is repeated to match the targets for each category (e.g., children, elderly).
- 18. CONGRESSIONAL BUDGET OFFICE Percentage of Adults Receiving Benefits 0 10 20 30 40 50 0 25 50 75 100 125 150 175 200 225 250 275 300 Annual Family Income as a Percentage of the Federal Poverty Level Medicaid Recipiency Rates, by Income, 2010 CPS (Reported Only) CBO (Imputed Plus Reported) The Urban Institute’s Transfer Income Model
- 19. CONGRESSIONAL BUDGET OFFICE Percentage of Individuals Receiving Benefits SSI Recipiency Rates, by Income, 2010 0 5 10 15 20 25 0 25 50 75 100 125 150 175 200 225 250 275 300 Annual Family Income as a Percentage of the Federal Poverty Level CPS (Reported Only) CBO (Imputed Plus Reported) The Urban Institute’s Transfer Income Model
- 20. CONGRESSIONAL BUDGET OFFICE Distribution of SNAP Recipients, by Annual Household Income as a Percentage of the Federal Poverty Level, 1979–2013 Percentage of Recipients 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% 1979 1984 1989 1994 1999 2004 2009 CPS (Reported Only) 1979 1984 1989 1994 1999 2004 2009 CBO (Imputed Plus Reported) 100 90 80 70 60 50 40 30 20 10 0 49% or Less 50%–99% 150%–199% 200% or More 100%–149%
- 21. CONGRESSIONAL BUDGET OFFICE Distribution of SNAP Recipients, by Annual Household Income as a Percentage of the Federal Poverty Level, 2005–2013 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% 2005 2008 2011 The Urban Institute’s Transfer Income Model Percentage of Recipients 100 90 80 70 60 50 40 30 20 10 0 49% or Less 50%–99% 150%–199% 200% or More 100%–149% 2005 2008 2011 CBO (Imputed Plus Reported)
- 22. CONGRESSIONAL BUDGET OFFICE Step 3. Transfer income is assigned to recipients. The assignment methodology varies by program.
- 23. CONGRESSIONAL BUDGET OFFICE For SNAP and SSI, CBO derives the average benefit per household from reported values (by household size and income-to- poverty ratio). Those averages are then assigned to newly imputed recipients and are adjusted as needed to match administrative totals.
- 24. CONGRESSIONAL BUDGET OFFICE 0 500 1,000 1,500 2,000 2,500 3,000 3,500 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 2011 2013 Average Annual SNAP Benefits per Household, 1979–2013 Constant 2013 Dollars Administrative Data Reporting Households Newly Imputed Households
- 25. CONGRESSIONAL BUDGET OFFICE For Medicaid, CBO derives the average cost to the government per participant from administrative data (by eligibility category). Those averages are then assigned to all recipients (CPS “reported” values are overwritten).
- 26. CONGRESSIONAL BUDGET OFFICE Average Annual Cost per Medicaid Recipient, 1979–2013 0 1,000 2,000 3,000 4,000 5,000 6,000 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 2011 2013 Constant 2013 Dollars CPS Averages CBO Averages (From Administrative Data)
- 27. CONGRESSIONAL BUDGET OFFICE CBO’s regression-based approach has both strengths and limitations.
- 28. CONGRESSIONAL BUDGET OFFICE The approach is straightforward to implement and easily scalable across multiple programs. Distributional results are similar to rules-based methods.
- 29. CONGRESSIONAL BUDGET OFFICE It does not, however, account for false positives in the CPS, and assumes that nonreporters have the same characteristics as reporters. It has a limited ability to simulate different policy scenarios.
- 30. CONGRESSIONAL BUDGET OFFICE Social Insurance Benefits
- 31. CONGRESSIONAL BUDGET OFFICE CBO uses a different approach for imputing social insurance benefits. CBO does not perform any explicit distributional analysis of social insurance benefits, since they are included in the base income measure.
- 32. CONGRESSIONAL BUDGET OFFICE Receipt of social insurance benefits is difficult to model with a regression. It is dependent on life-cycle income/labor force participation, it is not means tested, and there are no income data for children in the CPS (which is important for imputing Social Security survivors’ benefits).
- 33. CONGRESSIONAL BUDGET OFFICE Social Security Reporting Rates in the CPS, 2001–2013 0 20 40 60 80 100 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 CPS Recipients as a Percentage of Administrative Recipients Disability Survivors Old Age 85 97 47
- 34. CONGRESSIONAL BUDGET OFFICE To impute Social Security benefits, CBO creates a pool of eligible recipients for each type of benefit and randomly assigns receipt until the administrative counts are matched. The average benefit for each benefit type is then assigned to new recipients and aligned to administrative totals as needed.
- 35. CONGRESSIONAL BUDGET OFFICE 0 20 40 60 80 100 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 2011 2013 Medicare Reporting Rates in the CPS, 1979–2013 CPS Recipients as a Percentage of Administrative Recipients 93
- 36. CONGRESSIONAL BUDGET OFFICE To impute Medicare benefits, CBO makes no change to reported recipients. CBO assigns the average cost to the government per participant to all recipients. Benefits from the Low Income Subsidy for Prescription Drug Coverage are allocated separately.
- 37. CONGRESSIONAL BUDGET OFFICE Preliminary Conclusions
- 38. CONGRESSIONAL BUDGET OFFICE Income Inequality, 1979–2013 Gini Index 0.35 0.40 0.45 0.50 Market Income Plus Reported Social Insurance Benefits Market Income Plus Imputed Social Insurance Benefits Plus Reported Means- Tested Transfers Market Income Plus Imputed Social Insurance Benefits Plus Imputed Means-Tested Transfers Market Income Plus Imputed Social Insurance Benefits 0.00 0.35 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 2011 2013 0.35 0
- 39. CONGRESSIONAL BUDGET OFFICE -8% -7% -6% -5% -4% -3% -2% -1% 0% 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 2011 2013 0 -1 -2 -3 -4 -5 -6 -7 -8 Change in Gini Index, 1979–2013 Percentage Change Relative to Market Income Plus Imputed Social Insurance Benefits Reported Means-Tested Transfers Imputed Means-Tested Transfers
- 40. CONGRESSIONAL BUDGET OFFICE Ratio of Means-Tested Transfers to Base Income Measure Means-Tested Transfer Rates, by Quintile, 1979–2013 0 0.1 0.2 0.3 0.4 0.5 0.6 Quintile 1 Quintile 2 Quintile 3 0 0.1 0.2 0.3 0.4 0.5 0.6 1979 1984 1989 1994 1999 2004 2009 Quintile 4 1979 1984 1989 1994 1999 2004 2009 Quintile 5 1979 1984 1989 1994 1999 2004 2009 All Quintiles CPS (Reported Only) CBO (Imputed Plus Reported)
- 41. CONGRESSIONAL BUDGET OFFICE 0 0.1 0.2 0.3 0.4 0.5 0.6 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 2011 2013 Means-Tested Transfer Rates, Bottom Quintile, 1979–2013 Ratio of Means-Tested Transfers to Base Income Measure CPS (Reported Only) CBO (Imputed Plus Reported)
- 42. CONGRESSIONAL BUDGET OFFICE Notes to Figures ■ Slides 7 and 32: The reporting rate equals the weighted sum of recipients in the CPS (including CPS imputations) divided by the number of recipients in the administrative data, adjusted for recipients outside the CPS sampling frame. Where administrative totals are available on a monthly basis, they have been converted to reflect the total number of program participants across the calendar year. ■ Slide 17: Adults are defined as individuals aged 18–64 who are not disabled. ■ Slides 17, 18, 19, and 20: Individuals or households are considered recipients if they participate in the program at any point during the calendar year.
- 43. CONGRESSIONAL BUDGET OFFICE Notes to Figures (Continued) ■ Slides 17 and 18: The Urban Institute’s Transfer Income Model (TRIM) is a microsimulation model that uses the CPS as a basis to simulate program rules for various transfer programs. It uses those rules to determine program eligibility, participation, and benefits. The current version, TRIM3, has publicly available imputations for most major welfare programs going back to 1993. For more details, see Zedlewski and Gianarelli (2015). ■ Slide 33: The reporting rate equals the weighted sum of recipients in the CPS (including CPS imputations) divided by the number of recipients in the administrative data. ■ Slide 35: Reported and imputed social insurance benefits include Social Security and Medicare. Reported and imputed means-tested transfers include Medicaid, SNAP, and SSI.
- 44. CONGRESSIONAL BUDGET OFFICE Notes to Figures (Continued) ■ Slide 38: Social insurance benefits include Social Security and Medicare. Means-tested transfers include Medicaid, SNAP, and SSI. ■ Slides 39 and 40: The base income for the CBO (Imputed Plus Reported) quintiles and means-tested transfer rates is market income plus imputed social insurance benefits. The base income for the CPS (Reported Only) quintiles and means-tested transfer rates is market income plus reported social insurance benefits. Means- tested transfers include Medicaid, SNAP, and SSI.
- 45. CONGRESSIONAL BUDGET OFFICE References ■ Bruce D. Meyer, Wallace K. C. Mok, and James X. Sullivan, The Under-Reporting of Transfers in Household Surveys: Its Nature and Consequences, Working Paper 15181 (National Bureau of Economic Research, July 2009), www.nber.org/papers/w15181. ■ Bruce D. Meyer and James X. Sullivan, Using Two-Sample Methods to Correct for Reporting Bias in Surveys, Working Paper 0902 (University of Chicago, December 2008), https://tinyurl.com/y8jjnqma (PDF; 176 KB). ■ Robert A. Moffitt and John Karl Scholz, Trends in the Level and Distribution of Income Support, Working Paper 15488 (National Bureau of Economic Research, November 2009), www.nber.org/papers/w15488.
- 46. CONGRESSIONAL BUDGET OFFICE References (Continued) ■ Laura Wheaton, Underreporting of Means-Tested Transfer Programs in the CPS and SIPP, (Urban Institute, February 6, 2008), https://tinyurl.com/yd4caq7n. ■ Sheila Zedlewski and Linda Gianarelli, TRIM: A Tool for Social Policy Analysis, (Urban Institute, May 2015), https://tinyurl.com/y7sbos8l.