33
A Comparison of Survey Reports Obtained via Standard Questionnaire and Event History Calendar Jeff Moore, Jason Fields, Joanne Pascale, Gary Benedetto, Martha Stinson, and Anna Chan U.S. Census Bureau American Association for Public Opinion Research May 14-17, 2009

A Comparison of Survey Reports Obtained via Standard Questionnaire and Event History Calendar Jeff Moore, Jason Fields, Joanne Pascale, Gary Benedetto,

Embed Size (px)

Citation preview

A Comparison of Survey Reports Obtained via Standard

Questionnaire and Event History Calendar

Jeff Moore, Jason Fields, Joanne Pascale,

Gary Benedetto, Martha Stinson, and Anna Chan

U.S. Census Bureau

American Association for Public Opinion Research

May 14-17, 2009

Overview

Background:- SIPP; SIPP “re-engineering”- event history calendar (EHC) methods

Goals and Design of the SIPP-EHC Field Test

Preliminary Results

Summary / Conclusions / Next Steps

SIPPSurvey of Income and Program Participation

- income/wealth/poverty in the U.S.; program participation dynamics/effects

- interviewer-administered; longitudinal

- panel length = 3-4 years

Key Design Feature:

- 3 interviews/year, 4-month reference pd.

SIPP Re-Engineering

Implement Improvements to SIPP- reduce costs- reduce R burden- improve processing system- modernize instrument- expand/enhance use of admin records

Key Design Change:- annual interview, 12-month reference

pd., event history calendar methods

EHC Interviewing

Human Memory- structured/organized- links and associations

EHC Exploits Memory Structure- links between to-be-recalled events

EHC Encourages Active Assistance to Rs- flexible approach to help elicit an

autobiographical “story”

Evaluations of EHC Methods

Many EHC vs. “Q-List” Comparisons- various methods- in general: positive data quality results

BUT, Important Research Gaps- data quality for need-based programs?- extended reference period?

Field Test Goals & Design

Basic Goal:Can an annual EHC interview collect data of comparable quality to standard SIPP?

Basic Design:EHC re-interview of SIPP sample households

Design Details (1)

Sample:SIPP 2004 panel interview cases

- reported on CY-2007 in waves 10-12

EHC re-interview in 2008, about CY-2007

Design Details (2)

SIPP Sample Cases in Two Sites- Illinois (all)- Texas (4 metro areas)

N = 1,096 Wave 10-11-12 Addresses(cooperating wave 11 households)

IL: 487TX: 609

Design Details (3)

EHC Questionnaire- paper-and-pencil- 12-month, CY-2007 reference period- subset of SIPP topics (“domains”)- month-level detail

Sample of Addresses, Not People- post-interview clerical match to SIPP

Design Details (4)

1096 initial sample addresses

Outcomes:

- 935 household interviews (91%)

- 1,922 individual EHC interviews (99%)

- 1,658 EHC Rs matched to SIPP (86%)

FINAL ANALYSIS SAMPLE: 1,620

Primary Evaluation

Compare SIPP and EHC Survey Reports - same people- same time period- same characteristics

Differences Suggest Data Quality Effects

(later: use administrative records for a moredefinitive data quality assessment)

Main Research Questions

1. Are responses to Qs about government programs and other characteristics affected by interview method (SIPP vs. EHC)?

2. Does the effect of interview method vary across calendar months (especially early in the year vs. late in the year)?

Preliminary Results

3 Government “Welfare” Programs: Food Stamps

Supplemental Security Income (SSI)Women Infants & Children (WIC)

4 Other Characteristics:MedicareSocial Securityemploymentschool enrollment

Results in ContextAlmost All SIPP and EHC Reports Agree

- all characteristics, all months

- in general: 97-98% likelihood that a respondent’s SIPP and EHC reports will agree

- worst case (employment): 92-94%

Disagreements are RARE EVENTS

Results Summary3 Patterns:

1. EHC = SIPP All Year equivalent data quality

0.0%

2.5%

5.0%

7.5%

10.0%

JAN FEB MAR APR MAY JUN JUL AUG SEP OCT NOV DEC

% Y

es

SIPP % yes

EHC % yes

SSI -- % Participation for Each Month of CY2007 According to the SIPP and EHC Reports

Analysis Summary

- no “main effect” for method

- no significant method difference in any month

WIC (Illinois Only) -- % Participation for Each Month of CY2007 According to the SIPP and EHC Reports

0.0%

2.5%

5.0%

7.5%

10.0%

JAN FEB MAR APR MAY JUN JUL AUG SEP OCT NOV DEC

% Y

es

SIPP % yes

EHC % yes

Analysis Summary

- no “main effect” for method

- no significant method difference in any month

Results Summary3 Patterns:

1. EHC = SIPP All Year SSI; WIC (IL)

2. EHC < SIPP All Yearreduced EHC data quality, but not due to longer recall period

15.0%

17.5%

20.0%

22.5%

25.0%

JAN FEB MAR APR MAY JUN JUL AUG SEP OCT NOV DEC

% Y

es

SIPP % yes

EHC % yes

MEDICARE -- % Covered in Each Month of CY2007 According to the SIPP and EHC Reports

Analysis Summary

- significant “main effect” for method

- method difference (SIPP > EHC) is constant across months

SOCIAL SECURITY -- % Covered in Each Month of CY2007 According to the SIPP and EHC Reports

15.0%

17.5%

20.0%

22.5%

25.0%

JAN FEB MAR APR MAY JUN JUL AUG SEP OCT NOV DEC

% Y

es

SIPP % yes

EHC % yes

Analysis Summary

- significant “main effect” for method

- method difference (SIPP > EHC) is constant across months

WIC (Texas Only) -- % Participation for Each Month of CY2007 According to the SIPP and EHC Reports

0.0%

2.5%

5.0%

7.5%

10.0%

JAN FEB MAR APR MAY JUN JUL AUG SEP OCT NOV DEC

% Y

es

SIPP % yes

EHC % yes

Analysis Summary

- significant “main effect” for method

- method difference (SIPP > EHC) is constant across months

FOOD STAMPS (Illinois Only) -- % Participation for Each Month of CY2007 According to the SIPP and EHC Reports

0.0%

2.5%

5.0%

7.5%

10.0%

JAN FEB MAR APR MAY JUN JUL AUG SEP OCT NOV DEC

% Y

es

SIPP % yes

EHC % yes

Analysis Summary

- significant “main effect” for method

- method difference (SIPP > EHC) is essentially constant across months

Results Summary3 Patterns:

1. EHC = SIPP All Year SSI; WIC (IL)

2. EHC < SIPP All Year Medicare; Social Security; WIC (TX); Food Stamps (IL)

3. EHC < SIPP, Early in the Year OnlyEHC data quality may suffer due

– to longer recall period

FOOD STAMPS (Texas Only) -- % Participation for Each Month of CY2007 According to the SIPP and EHC Reports

0.0%

2.5%

5.0%

7.5%

10.0%

JAN FEB MAR APR MAY JUN JUL AUG SEP OCT NOV DEC

% Y

es

SIPP % yes

EHC % yes

Analysis Summary

- no significant “main effect” for method

- BUT significant variation by month --

JAN-MAY: SIPP > EHC

later months: no difference (reversal?)

EMPLOYMENT -- % Working for Each Month of CY2007 According to the SIPP and EHC Reports

57.5%

60.0%

62.5%

65.0%

67.5%

JAN FEB MAR APR MAY JUN JUL AUG SEP OCT NOV DEC

% W

ork

ing

SIPP % w orking

EHC % w orking

Analysis Summary

- significant “main effect” for method (SIPP > EHC)

- BUT significant variation by month --

JAN-AUG (SEP): SIPP > EHC

later months: no difference

5.0%

7.5%

10.0%

12.5%

15.0%

JAN FEB MAR APR MAY JUN JUL AUG SEP OCT NOV DEC

% E

nrol

led

SIPP % enrolled

EHC % enrolled

SCHOOL ENROLLMENT -- % Enrolled in Each Month of CY2007 According to the SIPP and EHC Reports

Analysis Summary

- no significant “main effect” for method

- BUT significant variation by month

JAN-APR: SIPP > EHC

JUN-JUL: SIPP < EHC

AUG-DEC: no difference

Field Test Overall SummarySuccessful “Proof of Concept”

Overwhelming Finding: SIPP-EHC Agreement

Valuable Lessons to Inform Next Test- larger, broader sample- “correct” timing of field period- automated questionnaire

Specific Data Comparisons are Instructive

Results ImplicationsPattern 1. EHC = SIPP All Year

SSI; WIC (IL)

No evident problems; no reason for concern about data quality in a 12-month EHC interview

Results ImplicationsPattern 2. EHC < SIPP All Year

Medicare; Social Security; WIC (TX); Food Stamps (IL)

Problems with data quality in the EHC treatment, but probably not due to recall length

- less effective screening questions (no D.I.; fewer probes; no local labels)- different definitions

Likely fixes in CAPI

Results ImplicationsPattern 3. EHC < SIPP, Early in the Year Only

Food Stamps (TX); employment;school enrollment

Most cause for concern; longer recall period may cause reduced data quality in the earlier months of the year

Additional research:- why these characteristics?- understand Field Test time lag effects

.