Upload
others
View
0
Download
0
Embed Size (px)
Citation preview
Guterbock, et al: Dual-frame to Triple Frame AAPOR: May 15, 2011
UVa Center for Survey Research 1
Center for Survey Research
University of Virginia
From Dual-Frame
to Triple Frame:
An Assessment of Coverage Bias in a Telephone Survey Design
Combining RDD, Directory-Listed And Cell Phone Samples
Presented at AAPOR 2011 Phoenix, AZ May 15, 2011
Center for Survey Research
University of Virginia 2
Authors
Thomas M.Guterbock
University of Virginia
Abdoulaye Diop
Qatar University [email protected]
James M. Ellis
University of Virginia [email protected]
John Lee Holmes
University of Virginia [email protected]
Trung Kien Le
Qatar University [email protected]
Center for Survey Research
University of Virginia 3
Overview
• Why triple-frame?
• The 2010 Behavioral Study of the NCR
• Coverage and distribution of phone service (5 segments)
• Contrast of RDD, EWP and Cell Phone frames
• Calling efficiencies
• Comparing substantive results:
– triple frame vs. RDD+Cell
• Cost comparisons
• Conclusion
Guterbock, et al: Dual-frame to Triple Frame AAPOR: May 15, 2011
UVa Center for Survey Research 2
Center for Survey Research
University of Virginia
Why triple-frame?
From one frame, to two, to three
Center for Survey Research
University of Virginia 5
From RDD to Dual-frame • ―Traditional‖ list-assisted landline Random Digit Dialing
is beset with problems – Increasing under-coverage due to Cell Phone Only (CPO)
households
– Lower working-number rates
– Declining response rates
• Dual-frame telephone surveys are now standard for many survey organizations – Combining landline RDD and cell phone RDD frames
– Most often using an ‗overlap‘ or ‗all cell‘ design
– See AAPOR Task Force 2010 for full discussion
• Adding cell phones covers the CPO‘s and favorably alters reachability of dual-phone households
Center for Survey Research
University of Virginia 6
Consider: Electronic White Pages
(directory listed) sample
• Previous studies have compared EWP to landline
RDD sample in statewide, regional, local studies.
– Guterbock et al. 2003, Oldendick et al. 2004.
• Guterbock, Diop & Holian 2007 explored race and
other predictors of listedness in a survey of the
National Capital Region
• Caution: These studies pre-date the surge in CPO
households
Guterbock, et al: Dual-frame to Triple Frame AAPOR: May 15, 2011
UVa Center for Survey Research 3
Center for Survey Research
University of Virginia 7
EWP vs. RDD
• Substantive results are similar, in general
• Advantages of EWP:
– More efficient
– Lower cost
– Greater geographic specificity
• Disadvantages:
– Undercoverage of minorities, lower income, renters
– African Americans less likely to have listed numbers
– Blacks therefore under-represented in EWP samples
Center for Survey Research
University of Virginia 8
Who needs RDD?
• There are notable similarities between characteristics of
unlisted landline households and CPO households
– Minorities, the young, renters, lower incomes
• Could the gaps in the EWP frame (undercoverage) be filled
in by inclusion of the cell phone frame?
• We proposed in 2008: EWP+Cell as a dual-frame design
alternative to RDD+Cell
– Based on analysis of NHIS data through 2006
– Just published in Social Science Research
• We presented a favorable comparison of the two designs in
three county-based surveys in Virginia (2009)
Center for Survey Research
University of Virginia 9
From Dual Frame to Triple Frame
• CSR has completed ten telephone surveys that use a triple-frame design – 4 metro-area based; 6 county-based
– All cell phones included—no screening for CPO
– All in Virginia or DC metro area
• Two reasons for triple-frame design: – To allow comparison of RDD+Cell vs. EWP vs. Cell
– Most of these studies needed comparison to earlier years that used landline RDD exclusively
• We are at a transitional stage in telephone sampling – Triple-frame designs are a compromise between ‗standard‘ dual-
frame design and our proposed alternative of EWP+Cell
Guterbock, et al: Dual-frame to Triple Frame AAPOR: May 15, 2011
UVa Center for Survey Research 4
Center for Survey Research
University of Virginia 10
error cost
It‘s all about. . .
Center for Survey Research
University of Virginia
The 2009
Behavioral Survey of the
National Capital Region
12 12 12
2009 Survey of
Behavioral Aspects of
Sheltering and Evacuation
in the National Capital
Region
Sponsor: VDEM Funding: U.S. DHS
Guterbock, et al: Dual-frame to Triple Frame AAPOR: May 15, 2011
UVa Center for Survey Research 5
Center for Survey Research
University of Virginia 13
13 13 13
Survey Goal
• Collect information from residents of the National Capital Region that would predict behavior in the area in the event of an emergency.
• Included factorial experiment that varied features of a ―dirty bomb‖ attack by terrorists.
– Ask how residents would respond to specific ―shelter-in-place‖ scenarios
– What variables have the most effect on behavior?
– What patterns of evacuation and shadow evacuation should be expected?
– Where would the evacuees try to go?
• The resulting data are being used to inform the decisions made by administrators in the region and beyond.
– Details in CSR‘s report, 2010.
Center for Survey Research
University of Virginia 14
14 14 14
Features of the Survey
• In-depth survey: average interview length 28 minutes – Fully supported Spanish language interviews as needed
• 2,609 interviews conducted by CSR, Sept-Dec 2009.
• Triple-frame sample design: – 1269 Landline RDD completes – 898 EWP (directory listed) completes – 442 cell phone completes (no screening for CPO‘s)
• RDD sample was backmatched to addresses – Advance postcard sent to EWP and backmatched RDD cases
• Weighting by ownership, race, gender, geography, and type of telephone service
• Margin of error: +/- 2.3 percentage points – After weighting
Center for Survey Research
University of Virginia
Coverage and
distribution of phone service
Guterbock, et al: Dual-frame to Triple Frame AAPOR: May 15, 2011
UVa Center for Survey Research 6
Center for Survey Research
University of Virginia 16
What percentage of
landline phones are unlisted?
• Each respondent was asked whether their landline
is listed in the directory.
– Dual users reached by cell phone were asked to report
on whether their landline is listed
• Percent unlisted can be taken directly from the
RDD frame.
• 19.9% of landline RDD completes are unlisted.
– As in our other studies, a small portion of those in the
EWP frame report their phones to be unlisted.
Center for Survey Research
University of Virginia 17
What is the CPO percentage?
• NHIS has been used as the ‗gold standard‘ for
weighting by phone service.
• NHIS did not provide estimates for this
geography.
• We used the locally based method of estimation
described by Guterbock 2009.
• 31.4% of cell phone respondents were CPO‘s.
• Final estimate: 15.4% of telephone HH are CPO.
– After excluding landlines with unknown listed status
2009 NCR Telephone Universe
2 CELL + ULL
18.1%
4 CELL + LLL
61.0%
5 LLL ONLY
4.2%
3 ULL ONLY
1.3% 1 CELL ONLY
15.4%
Guterbock, et al: Dual-frame to Triple Frame AAPOR: May 15, 2011
UVa Center for Survey Research 7
What EWP+Cell would cover
2 CELL + ULL
18.1%
4 CELL + LLL
61.0%
5 LLL ONLY
4.2%
3 ULL ONLY
1.3% 1 CELL ONLY
15.4%
Center for Survey Research
University of Virginia 20
Little undercoverage
• Households with no cell phone and an unlisted
landline phone (ULL only), are but 1.3% of the
region‘s telephone households.
• These are covered by the landline RDD frame, but
not covered by the EWP frame.
• EWP frame underestimates unlisted percentage.
• We weight the unlisted percentage among all
landlines to 19.9% (the unlisted percent in RDD).
Center for Survey Research
University of Virginia
Contrasting the three frames
And the matched vs. unmatched
portions of landline RDD frame
Guterbock, et al: Dual-frame to Triple Frame AAPOR: May 15, 2011
UVa Center for Survey Research 8
Comparing calling efficiency
Landline RDD
EWP
Cell
RDD Matched Un-
matched
combined
Completes per
hour (CPH) .69 .39 .58 .84 .54
Hours per
complete (HPC) 1.45 2.56 1.71 1.19 1.85
n of completes 968 301 1269 898 442
Calling hours 1403 772 2175 1069 819
Comparing key demographics
Landline RDD
EWP
Cell
RDD
ACS
2008 Matched Un-
matched
combined
% renter 18.7 25.3 20.3 14.9 38.4 34.3
% African-
American 16.5 27.4 19.1 13.8 23.9 27.5
% never married 19.3 22.7 20.1 16.5 38.9 --
% age 18-25 3.3 5.4 3.8 1.6 22.0 11.0
n of cases 968 301 1269 898 442
Center for Survey Research
University of Virginia 24
Comparing substantive results
• How are survey results affected when a triple-
frame sample is used?
• We compare our triple-frame result with the
results we would have obtained with an
RDD+Cell design.
• Both designs are post-weighted to the same
control weights: – Ownership and race (joint distribution), gender, 8 counties, type of
telephone service (CPO, LLO, dual user reached by landline, dual
user reached by cell phone) and listed status
Guterbock, et al: Dual-frame to Triple Frame AAPOR: May 15, 2011
UVa Center for Survey Research 9
Center for Survey Research
University of Virginia 25
Comparing substantive results
Triple
frame
design
RDD+Cell Difference
Would leave
immediately in
maximum hazard
19.4% 19.8% -0.4%
Has an emergency kit
prepared 32.6% 32.6% 0.0%
Strongly agree it is
very important to live
in this particular area
32.5% 31.8% 0.7%
Center for Survey Research
University of Virginia
Let‘s compare 20 variables
• Would leave scene in a
dirty bomb attack
– Minimum hazard level
– Moderate
– Maximum
• Can trust most people
• Trust local government
• Has an emergency plan
• Has an emergency kit
• Has a meeting place
• Worry about attack
– 3 levels of hazard
• High perceived risk
– Property damage; injury
– 3 levels of hazard
• Agree/Strongly agree:
– Feel at home where I live
– I have a lot in common with
neighbors
– Important for me to live in
this area
Center for Survey Research
University of Virginia 27
Differences for 20 variables
0 0 0 0
1
14
5
0 0 0 0 0
2
4
6
8
10
12
14
16
-5 -4 -3 -2 -1 0 1 2 3 4 5
N
u
m
b
e
r
o
f
I
t
e
m
s
Raw Percentage Point Difference: Triple-frame - RDD+Cell
Margin of error +/- 2.3%
Guterbock, et al: Dual-frame to Triple Frame AAPOR: May 15, 2011
UVa Center for Survey Research 10
Center for Survey Research
University of Virginia
Cost comparison
Center for Survey Research
University of Virginia 29
Cost factors and assumptions
• Assume that a survey of like size had been carried out with
a ‗traditional‘ dual frame design
– Same number of cell phone completions as in our triple-frame
design
– 442 cell phone completes
– 2167 landline RDD completes
• RDD uses more sampled numbers
• RDD cost per sampled number is higher
– Due to extra charge for backmatching
• Assume postcard sent to backmatched RDD cases
• Cell phone completes get $10 incentive
Triple frame costs Land-line
RDD
EWP Cell
phone
RDD
Total
N of completes 1,269 898 442 2,609
CPH .58 .84 .54 .62
Calling hours 2,175 1,069 819 4,062
Cost of calling hours $69,590 $34,210 $26,193 $129,992
Sampled numbers 14,083 6,127 9,585 29,795
Cost of sampled numbers $2,110 $674 $1,150 $3,934
Postcards 7,631 6,127 -- 13,758
Cost of postcards $3,052 $2,451 -- $5,503
Incentive Cost -- -- $4,420 $4,420
Guterbock, et al: Dual-frame to Triple Frame AAPOR: May 15, 2011
UVa Center for Survey Research 11
RDD+Cell costs (projected) (Note: Assumes 54% Landline RDD
back-matched to directory listing and
mailed a postcard)
Land-line
RDD
Cell
phone
RDD
Total
N of completes 2,167 442 2,609
CPH .58 .54 .58
Calling hours 3,714 819 4,532
Cost of calling hours $118,835 $26,193 $145,028
Sampled numbers 24,049 9,585 33,634
Cost of sampled numbers $3,603 $1,150 $4,753
Postcards 13,031 -- 13,031
Cost of postcards $5,212 -- $5,212
Incentive Cost -- $4,420 $4,420
Triple frame vs.RDD+Cell costs Each design:
2,609 completes
Triple
Frame
RDD+
Cell
Diff
CPH .62 .58
Calling hours 4,062 4,532
Cost of calling hours $129,992 $145,028 $15,036
Sampled numbers 29,795 33,634
Cost of sampled numbers $3,934 $4,753 $819
Postcards 13,758 13,031
Cost of postcards $5,503 $5,212 - $291
Incentive Cost $4,420 $4,420 0
Total cost $143,849 $159,413 $15,564
Center for Survey Research
University of Virginia 33
Cost summary
• RDD+Cell would have cost: $159,413 ($61.10 per
completion)
• Triple frame design cost: $143,849 ($55.14 per
completion)
• Substitution of EWP for some of the landline
RDD frame saved $15,564 ($5.97 per completion)
or 10.8% of the Triple Frame total.
– Greater savings could have been realized if EWP
percentage were larger relative to landline RDD
Guterbock, et al: Dual-frame to Triple Frame AAPOR: May 15, 2011
UVa Center for Survey Research 12
Center for Survey Research
University of Virginia
Conclusion
Center for Survey Research
University of Virginia 35
error cost
It‘s all about. . .
Center for Survey Research
University of Virginia 36
Conclusion
• For representative general population results, we need to include cell phones in our telephone surveys
• When combined with the cell phone sample frame, EWP sample frames offer – greater efficiency than landline RDD
– lower cost
• There are good reasons to retain some landline RDD sample in the mix at this transitional stage – Allows direct measurement of unlisted percentage
• There is little or no loss of accuracy when EWP is substituted for some of the landline RDD frame – and cell phones are included in the design
• Cost savings are considerable (over 10% in this study)
Guterbock, et al: Dual-frame to Triple Frame AAPOR: May 15, 2011
UVa Center for Survey Research 13
Center for Survey Research
University of Virginia 37
Conclusion
Three frames are better than two!
RDD EWP Cell
Center for Survey Research
University of Virginia 38
References AAPOR Cell Phone Task Force. 2010. New Considerations for Survey Researchers When Planning and
Conducting RDD Telephone Surveys in the U.S. With Respondents Reached via Cell Phone Numbers. Available online at www.aapor.org.
Guterbock TM, 2009. ―Estimating Local Phone Service and Usage Percentages.‖: How to Weight the Data from a Local, Dual Frame Sample Survey Of Cell Phone and Landline Telephone Users in the United States.‖ AAPOR paper.
Guterbock TM, Diop A, Ellis JM, Le TK, & Holmes JLP, 2009. ―Who Needs RDD–Part II: An Assessment of Coverage Bias in Dual-Frame Designs That Combine Directory-Listed And Cell Phone Samples.‖ AAPOR poster, Hollywood FL.
Guterbock TM, Diop A, Ellis JM, Le TK, & Holmes JLP, 2011. "Who Needs RDD? Combining Directory Listings with Cell Phone Exchanges for an Alternative Telephone Sampling Frame". Social Science Research 40:3 (May): 860-872. Also presented as 2008 AAPOR paper, New Orleans.
Guterbock TM, Diop A, & Holian L, 2007. ―White pages, white people: Reasons for the low listed-phone rates of African-Americans.‖ AAPOR paper, Anaheim.
Guterbock TM, Hartman DE & Hubbard RA, 2003. ―RDD vs listed: An experimental study of coverage error, costs and non-response in a statewide telephone survey. AAPOR paper, Nashville.
Guterbock TM, Lambert JH, Bebel RA, Ellis JM, & Kermer DA, 2010. Population Behaviors in Dirty Bomb Attack Scenarios: A Survey of the National Capital Region. Prepared for the Virginia Department of Emergency Management. University of Virginia Center for Survey Research, April.
Oldendick, Robert W., et al. 2004. ―Differences in an RDD and list sample: An experimental comparison.‖ AAPOR paper, Phoenix.
Center for Survey Research
University of Virginia
From Dual-Frame
to Triple Frame:
An Assessment of Coverage Bias in a Telephone Survey Design
Combining RDD, Directory-Listed And Cell Phone Samples
Presented at AAPOR 2011 Phoenix, AZ May 15, 2011
Contact: [email protected]