Examination of Ipek Bilgen, Cheryl J. Recruitment Mailing ... 2.2 - Ipek Bilgen.pdf · HS grad or...
Transcript of Examination of Ipek Bilgen, Cheryl J. Recruitment Mailing ... 2.2 - Ipek Bilgen.pdf · HS grad or...
Examination of Recruitment Mailing Design Decisions for a Probability-based Research Panel
Ipek Bilgen, Cheryl J. Wiese, Lindsay Liebert, Erlina Hendarwan, Roopam Singh, Kanru Xia, Vicki J. Pineau, and J. Michael DennisCIPHER 2020, Washington DC,
February 26-27, 2020
2
Nonresponse during the recruitment stage is a source of risk to accuracy in estimates and panel representation in probability-based web/web-mostly panels
Mail recruitment is a large component of ABS-based web/web-mostly panels – Following visual design guidelines during mailing design is crucial – But… One size does not fit all (Dillman et al., 2014)
In this study, we examine the impact of key visual design decisions in recruitment mail materials and assess which design(s) work better for which part of the population
Introduction
3
AmeriSpeak Recruitment Methodology
Initial Recruitment Pre-notification postcard +5 days, 9 x 12 recruitment packet
w/ pre-incentive, letter, study brochure w/ privacy policy
+11 days, reminder postcard +18 days, reminder postcard In-bound CATI +25 days, Out-bound CATI
(call-outs to matched telephones)
NRFU Recruitment Federal Express study brochure,
letter, and enhanced pre-incentive In-person recruitment
4
Participants did not see uniqueness of AmeriSpeak in the materials. They saw it as another survey platform
The photos used to illustrate the points did not reinforce the instructions and may therefore distract from the instructional text
There was significant “visual noise” that may take away from the key information
Accordingly, the 2019 experiment designed to test how to effectively reduce the visual noise and simplify the content of the recruitment materials
Initial Qualitative Research Takeaways
5
Research Questions
What is the impact of each mailing design on the panel recruitment yield rate?
Do different mailing designs improve demographic representation of the panel sample?
How do different mailing designs impact the panel sample composition? One size fits all vs. tailor materials for households w/ different demographics.
What is the impact of recruitment mailing material design on AmeriSpeak Recruitment Rates and Panel Composition?
6
Overarching Hypotheses
Elimination of photos on mail recruitment materials will improve initial yield rates
Simplifying the mail recruitment material text will improve initial yield rates 8th grade reading level
Using formatting and layout to improve readability
Clear action items
The addition of a magnet as a non-monetary incentive will improve recruitment rates across all demographic groups
7
Pre-notification PostcardControl
Control Text
Treatment
8
What is AmeriSpeak?
Control: Photo + Full Text Envelope & Brochure Cover No Photo + Full Text Envelope & Brochure Cover
Photo + Simplified Text Envelope & Brochure Cover No Photo + Simplified Text Envelope & Brochure Cover
9
Initial Packet Cover Letter
Control Text
Magnet
Control Treatment
10
Reminder PostcardControl Treatment
11
0.37%(537)
2.53%(3,633)
0.65%(910)
2.99%(4,158)
6.41%Total Yield
Total Yielded9,238
(started survey)
Total Sampled144,198
Overall Recruitment Yield Rates – at a Glance
Reminder Postcard
Initial Packet Phone Follow-upPre-notification Postcard
12
Overall = 6.4%
Magnet = 5.6%No Graphics = 6.2%No Graphics + Magnet = 6.3%Simple Text + Magnet = 6.5%Simple Text = 6.8%Simple Text + No Graphics = 7.0%Simple Text + No Graphics + Magnet = 7.5%
Control = 5.8%
0% 1% 2% 3% 4% 5% 6% 7% 8% 9% 10%
Recruitment Yield Rate
Rec
ruitm
ent M
ailin
g D
esig
n
1.7 pp increase = 29% more than the control grp yield rate(p<.0001)
Recruitment Yield Rates by Treatment Group
NOTE: CATI outbound phone yield rates were very similar among the treatment groups (ranged between 2.8 to 3.1%)
2019 Panel Recruitment: Composition Analysis
14
2019 AmeriSpeak Panel Recruits
Unweighted Panel Distribution by Mailing Design and Gender
Control versus ST + NG: ** p=0.0044Control versus ST+NG+M: * p=0.0333
0%
20%
40%
60%
80%
100%
Male Female
Rec
ruitm
ent Y
ield
Rat
e Benchmark(CPS 2019)
Control
Simple Text +No Graphics
Simple Text +No Graphics +Magnet
New designs bring in more male
panelists
15
2019 AmeriSpeak Panel Recruits
Unweighted Panel Distribution by Mailing Design and Age
0%
10%
20%
30%
40%
50%
18-24 25-34 35-44 45-54 55-64 65-74 75+
Rec
ruitm
ent Y
ield
Rat
e Benchmark(CPS 2019)
Control
No Graphics +Magnet
Simple Text +No Graphics +Magnet
Control versus NG+M: + p=0.0946Control versus ST+NG+M: * p=0.0179
Design w/ no graphics and
magnet brings in younger panelists (18-24)
Design w/ simple text, no graphics, and
magnet brings in older panelists
(75+)
Overall, different designs appeal to
different age groups
(Tailored design)
16
2019 AmeriSpeak Panel Recruits
Unweighted Panel Distribution by Mailing Design and Race
Control versus NG: p=0.1059Control versus ST+NG+M: p=0.1444
0%
10%
20%
30%
40%
50%
NH Black NH Asian NH Other Hispanic NH, 2+
Rec
ruitm
ent Y
ield
Rat
e Benchmark(CPS 2019)
Control
No Graphics
Simple Text +No Graphics +Magnet
We still need NRFU to bring in
moreHispanic
Minorities
New designs bring in slightly
but not significantly more Asians
17
2019 AmeriSpeak Panel Recruits
Unweighted Panel Distribution by Mailing Design and Education
Control versus M: p=0.2152Control versus ST+NG+M: p=0.1353
0%
10%
20%
30%
40%
50%
No HS Diploma HS grad orequivalent
Some College BA or Above
Rec
ruitm
ent Y
ield
Rat
e Benchmark(CPS 2019)
Control
Magnet
Simple Text +No Graphics +Magnet
No significant impact of new design options on panelists w/ different education levels
18
2019 AmeriSpeak Panel Recruits
Unweighted Panel Distribution by Mailing Design and Income
Control versus NG: p=0.5319Control versus ST+NG+M: p=0.7862
0%
10%
20%
30%
40%
50%
Less than$25,000
$25,000 to$49,999
$50,000 to$74,999
$75,000 to$99,999
$100,000 ormore
Rec
ruitm
ent Y
ield
Rat
e Benchmark(CPS 2019)
Control
No Graphics
Simple Text +No Graphics +Magnet
No significant impact of new design options on panelists w/
different income levels
19
• Overall, the mailing design interventions significantly increase the yield rates• Different mailing design options appeal to different age groups
• Consideration of tailored design model during future recruitment
• All interventions resulted in a significantly larger proportion of male respondents• Removing graphics and using magnet resulted in a larger proportion of younger
respondents (ages 18-24)• We still need NRFU to represent some H2R groups (e.g. Hispanic minorities,
low educated, and very high income level groups) • Future direction – Investigating the use of auxiliary data during recruitment to
predict HH demographics to determine how we can best tailor mailing materials based on the results of these experiments
Summary and Discussion
Thank You!
21
AAPOR RR3 (2014-2018 AmeriSpeak Panel Recruits)
Description Response Rate (AAPOR RR3, Weighted)
Household ResponseRate due to Initial Recruitment 5.6%
Household Response Rate due to NRFU 28.6%
Household Response Rate 34.2%
2019 Recruitment RR (Initial Recruitment Rate since we did not have NRFU last year) = 5.7%
2014 – 2019 Overall Household Response Rate = 24.1%