effects of income imputation on traditional poverty estimates 1987-2007

33
Effects of Income Imputation on Traditional Poverty Estimates 1987-2007 The views expressed here are the authors and do not represent the official positions of their organizations. 1

Upload: zada

Post on 22-Jan-2016

23 views

Category:

Documents


0 download

DESCRIPTION

Effects of Income Imputation on Traditional Poverty Estimates 1987-2007. The views expressed here are the authors and do not represent the official positions of their organizations. Authors. Joan Turek, Brian Sinclair James and Bula Ghose, Department of Health and Human Services - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: Effects of Income Imputation on Traditional Poverty Estimates 1987-2007

Effects of Income Imputation on Traditional

Poverty Estimates1987-2007

The views expressed here are the authors and do not represent the official positions of their organizations.

1

Page 2: Effects of Income Imputation on Traditional Poverty Estimates 1987-2007

Authors

• Joan Turek, Brian Sinclair James and Bula Ghose, Department of Health and Human Services

• Charles Nelson and Edward Welniak, Bureau of the Census

• Fritz Scheuren, NORC

2

Page 3: Effects of Income Imputation on Traditional Poverty Estimates 1987-2007

Outline of Talk

• Handling nonresponse on the CPS• Effects of imputation on income

and poverty estimates• Official poverty vs. first quintile

measure -- demographic characteristics

• Summary of findings • Implications for new measure?

3

Page 4: Effects of Income Imputation on Traditional Poverty Estimates 1987-2007

Upward Trend in Nonresponse

0.00%5.00%

10.00%15.00%20.00%25.00%30.00%35.00%40.00%

Trend in the Percent of Total Dollar Income Imputed, All People With Positive Income

20.5%

32.3%

4

Page 5: Effects of Income Imputation on Traditional Poverty Estimates 1987-2007

5

Handling CPS Nonresponse

• Uses “Hot Deck” procedures• Imputation occurs at the person

level by income source• Assigns amounts from reporters

with similar characteristics• Imputation method consistent

over time

5

Page 6: Effects of Income Imputation on Traditional Poverty Estimates 1987-2007

Types of Imputation

Two types of non-response in ASEC: item and whole imputes

• Item impute: sample person or other household member fails to respond to a specific question

• Item imputes are based on responses to both the basic monthly survey and on the ASEC supplement 6

Page 7: Effects of Income Imputation on Traditional Poverty Estimates 1987-2007

Types of Imputes (Con.)

• Whole impute: Sample persons only responded to the basic labor force questions in the monthly survey -- entire supplement is imputed using the monthly survey

• More limited data on monthly survey—have labor force experience for last month and not last year 7

Page 8: Effects of Income Imputation on Traditional Poverty Estimates 1987-2007

Income Per PersonComparisons

8

Page 9: Effects of Income Imputation on Traditional Poverty Estimates 1987-2007

What Comparisons Tell Us

• Imputation has greatest effect at lower per person income levels

• Predictable consequences for poverty rates

• Shown for 2007, but generally true over time

9

Page 10: Effects of Income Imputation on Traditional Poverty Estimates 1987-2007

Poverty Trends: 1987-2007

Trends in Poverty Rates for People with Positive Income -- Item, Whole or No Imputes

0.00%

2.00%

4.00%

6.00%

8.00%

10.00%

12.00%

1987

1989

1991

1993

1995

1997

1999

2001

2003

2005

2007

Item

Whole

No Imputes

10

Page 11: Effects of Income Imputation on Traditional Poverty Estimates 1987-2007

Poverty Trends Summary

• No imputes -- highest poverty rates• Item imputes -- lowest poverty

rates• Growth in imputation rates has not

really changed the poverty distribution

11

Page 12: Effects of Income Imputation on Traditional Poverty Estimates 1987-2007

Income Type and Poverty Status

Next look at the percent of the total population with positive income:

• below the official poverty line by type of imputation at five year intervals and for 5-year average

• Compares this 5 year average to those not in poverty and to all persons with positive income

12

Page 13: Effects of Income Imputation on Traditional Poverty Estimates 1987-2007

Imputation Type and Poverty Status

Year None Item Whole Total Imputes

Impute types sum to 100%

100% Poverty

2007 63.0% 28.3% 9.7% 38.0%

2002 56.5% 31.9% 11.6% 43.5%

1997 66.3% 24.4% 9.3% 33.7%

1992 73.7% 14.4% 11.9% 26.3%

1987 77.4% 13.4% 9.2% 22.6%

5 year average 67.3% 22.3% 10.4% 32.7%

Not in Poverty

5 year average 58.2% 31.8% 10.0% 41.8%

All Persons

5 year average 59.1% 30.9% 10.0 40.9%

Percents are for number of persons with positive income

13

Page 14: Effects of Income Imputation on Traditional Poverty Estimates 1987-2007

Income Type and Poverty Status (Con.)

Percentage of all persons with positive income who are item imputed falling below the poverty line grew, but trend seemed to reverse in recent years

• Whole imputes are relatively stable -- ranging between 9 and 12 percent.

14

Page 15: Effects of Income Imputation on Traditional Poverty Estimates 1987-2007

• Overall trend has been toward more imputing –

• On average, less imputation for poverty population

• Not sure what recent reversal between 2002 and 2007 means for the long term.

15

Income Type and Poverty Status (Con.)

Page 16: Effects of Income Imputation on Traditional Poverty Estimates 1987-2007

Role of Imputes on Poverty Rates in 2007

• No imputes only 9.8%• Item Imputes only 6.1%• Whole imputes only 8.4%• All of Above 8.3%• Whole plus no imputes 9.6%• Item plus no imputes 8.3%• All, item imputes set to 0 35.1%• Item imputes only set to 0 51.7%

16

Page 17: Effects of Income Imputation on Traditional Poverty Estimates 1987-2007

What Imputation Does?

• In 2007, the poverty rate including no, item and whole imputes is 8.26%

• Without whole imputes the poverty rate is 8.25%

• When all item imputed amounts are set to zero, the poverty rate increases from 8.3% to 35.1%%

17

Page 18: Effects of Income Imputation on Traditional Poverty Estimates 1987-2007

What Imputation Does?

• When looking only at persons with item imputes and setting these imputes equal to zero, the poverty rate increases from 6.1% to 51.7%

• Most persons with item imputes are workers

• O’Hara finds more persons have item imputed rather than reported or whole imputed earnings up to approx. $30,000

18

Page 19: Effects of Income Imputation on Traditional Poverty Estimates 1987-2007

Official Poverty vs. Lowest Quintile

• Approximately 50% of the worst off 20% of the population are in official poverty estimate

• Only 2% of the population in the next quintile are in the official poverty estimate

• Family income is used in putting persons into a quintile – but counts are number of persons

19

Page 20: Effects of Income Imputation on Traditional Poverty Estimates 1987-2007

Official Poverty vs Lowest Quintile (Con.)

• How are the demographics of the poor affected by the poverty measure selected

• Averages were constructed for selected demographic characteristics from annual estimates at five year intervals for: 2007, 2002, 1997, 1992 and 1987

20

Page 21: Effects of Income Imputation on Traditional Poverty Estimates 1987-2007

Official Poverty vs. Lowest Quintile (cont.)

• Comparisons made by gender, race, family type, age, and education

• First chart shows poverty rates for males and females separately by official poverty measure and by lowest quintile

21

Page 22: Effects of Income Imputation on Traditional Poverty Estimates 1987-2007

Gender

22

Demographic category

Official Poverty

Lowest Quintile Standardized Official Poverty

Standardized Lowest Quintile

Not Imputed

Male 8.2% 15.6% 64.3% 68.0%

Female 12.7% 22.9% 100.0% 100.0%

Item Imputed

Male 5.3% 11.4% 66.5% 64.2%

Female 8.0% 17.7% 100.0% 100.0%

Whole Imputed

Male 7.6% 14.9% 66.8% 67.6%

Female 11.4% 22.1% 100.0% 100.0%

All

Males 7.2% 14.2% 64.9% 67.0%

Females 11.1% 21.2% 100.0% 100.0%

Page 23: Effects of Income Imputation on Traditional Poverty Estimates 1987-2007

Race

23

Demographic category

Official Poverty

Lowest Quintile Standardized Official Poverty

Standardized Lowest Quintile

Not Imputed White 8.6% 17.2% 36.4% 49.5% Black 23.7% 34.7% 100.0% 100.0%

Item Imputed White 5.8% 13.7% 44.2% 60.9% Black 13.1% 22.5% 100.0% 100.0%

Whole Imputed White 8.3% 17.2% 50.1% 65.6% Black 16.6% 26.2% 100.0% 100.0%

All White 7.7% 16.1% 38.9% 53.2% Black 19.8% 30.2% 100.0% 100.0%

Page 24: Effects of Income Imputation on Traditional Poverty Estimates 1987-2007

AgeDemographic

category Official Poverty

Lowest Quintile Standardized Official Poverty Ratio

Standardized Lowest Quintile

Ratio Not Imputed

18-44 11.2% 17.5% 137.9% 119.9% 45-64 8.1% 14.6% 100.0% 100.0%

65 and over 12.3% 35.8% 152.0% 245.6% Item Imputed

18-44 7.3% 12.6% 149.0% 135.5% 45-64 4.9% 9.3% 100.0% 100.0%

65 and over 7.8% 28.5% 157.5% 305.6% Whole Imputed

18-44 10.5% 17.5% 147.0% 134.6% 45-64 7.2% 13.0% 100.0% 100.0%

65 and over 10.8% 32.2% 151.5% 247.9% All

18-44 10.1% 16.2% 146.4% 128.6% 45-64 6.9% 12.6% 100.0% 100.0%

65 and over 10.3% 32.8% 150.1% 260.8%

24

Page 25: Effects of Income Imputation on Traditional Poverty Estimates 1987-2007

Single and Two Parent

Demographic category

Official Poverty

Lowest Quintile Standardized Official Poverty

Standardized Lowest Quintile

Not Imputed

Two Parent 7.3% 7.0% 19.7% 13.8%

Single Parent 36.9% 50.7% 100.0% 100.0%

Item Imputed

Two Parent 3.8% 3.6% 16.4% 9.7%

Single Parent 23.4% 37.4% 100.0% 100.0%

Whole Imputed

Two Parent 5.9% 6.0% 18.5% 13.1%

Single Parent 32.1% 46.0% 100.0% 100.0%

All

Two Parent 6.2% 5.9% 18.4% 12.5%

Single Parent 33.4% 47.2% 100.0% 100.0%

25

Page 26: Effects of Income Imputation on Traditional Poverty Estimates 1987-2007

EducationDemographic

category Official Poverty

Lowest Quintile Standardized Official Poverty

Standardized Lowest Quintile

Not Imputed Less HS 27.1% 46.5% 269.5% 235.1% Some HS 20.2% 31.3% 200.1% 158.4%

HS Graduate 10.1% 19.8% 100.0% 100.0% Some College 7.0% 14.5% 69.4% 73.5% College Grad 2.8% 6.5% 28.0% 32.9

Item Imputed

Less HS 17.2% 36.7% 238.5% 215.3% Some HS 14.0% 26.1% 193.4% 153.4%

HS Graduate 7.2% 17.0% 100.0% 100.0% Some College 5.4% 12.4% 75.1% 72.8% College Grad 2.6% 5.9% 36.3% 34.7%

26

Page 27: Effects of Income Imputation on Traditional Poverty Estimates 1987-2007

Education (cont.)

Demographic category

Official Poverty

Lowest Quintile Standardized Official Poverty

Standardized Lowest Quintile

Whole Imputed Less HS 19.8% 37.8% 206.8% 196.0% Some HS 16.6% 27.9% 173.6% 144.4%

HS Graduate 9.6% 19.3% 100.0% 100.0% Some College 7.8% 15.5% 81.4% 80.4% College Grad 3.9% 7.8% 40.6% 40.6%

All Less HS 24.2% 43.4% 263.7% 229.6% Some HS 18.2% 29.7% 199.3% 157.2%

HS Graduate 9.2% 18.9% 100.0% 100.0% Some College 6.6% 13.9% 71.6% 73.8% College Grad 2.8% 6.4% 31.0% 33.9%

27

Page 28: Effects of Income Imputation on Traditional Poverty Estimates 1987-2007

Summary of Findings

Who are viewed as poor, is influenced by measure used:

• In one instance (Gender) no difference in impact of non-response

• In another instance (Race) large differences are found

28

Page 29: Effects of Income Imputation on Traditional Poverty Estimates 1987-2007

Summary of Findings

• In still other cases, (age, family type, education) results are mixed:– More poor elderly in poverty when use

lowest quintile– fewer two parent families in poverty using

lowest quintile– more persons with education below high

school graduate in poverty using official poverty

29

Page 30: Effects of Income Imputation on Traditional Poverty Estimates 1987-2007

The Supplemental Measures??

• Money income, with many additional adjustments, will also be used to construct supplemental poverty measures

• How does the addition of these new elements, such as near money income, expenditure and tax estimates affect the overall pattern of nonresponse

• Will poverty trends remain stable over a long period of time? 30

Page 31: Effects of Income Imputation on Traditional Poverty Estimates 1987-2007

Next Steps• ASPE and Census are jointly

sponsoring a project that will match SSA, TANF and SSI records to the 2008 ASEC

• We will compare the incomes reported on these files to those on the ASEC by imputation type and other characteristics

• This will add an additional dimension to the retooling of CPS Poverty measures 31

Page 32: Effects of Income Imputation on Traditional Poverty Estimates 1987-2007

Sources

• Amy O’Hara, Allocated Values in Linked Files, Housing and Household Economic Statistics Division, U.S. Census Bureau. [email protected]

• Joan Turek, Fritz Scheuren, Charles Nelson, Edward Welniak Jr., Brian Sinclair-James, and Bula Ghose,

-  Effects of Imputation on CPS Income and Poverty Series: 1981-2007,  Papers and Proceedings of the American Statistical Association, August 2009

- Effects of Imputation on CPS Poverty Series: 1987 – 2007, Papers and Proceedings of the Federal Committee on Statistical Methodology, November 2009.

32

Page 33: Effects of Income Imputation on Traditional Poverty Estimates 1987-2007

Thank You!!

Contact

Dr. Joan Turek

[email protected]

33