the use of random digit dialing in household surveys: challenges and changes chris chapman 2008 ies...
TRANSCRIPT
The Use of Random Digit Dialing in Household Surveys: Challenges and
Changes
Chris Chapman2008 IES Research Conference
Washington, DCJune 11, 2008
Overview Two primary challenges in the use of traditional RDD collections for household surveys
• Declining response rates in traditional Random Digit Dialing (RDD) sample methodology
• Declining coverage rates in traditional RDD sample design studies
Analyses of alternative approaches for government RDD surveys
• Use of pre-phone contact mailed information materials
• Use of prepaid monetary incentives
• Study of staged, multimode collection methods
• Expanding coverage
National Household Education Surveys (NHES)
The analyses discussed today will be in context of work primarily done with NHES
NHES is a repeating household survey conducted by NCES
NHES has relied on RDD sampling approaches since its inception in 1991 through the latest collection in 2007
Sample is drawn from existing 100 banks starting with 130,000 to 475,000 numbers
Sample is restricted to non-business numbers and excludes cells phones
Topics AddressedEarly childhood education and nonparental care for infants, toddlers, preschoolers and students in early elementary grades
Parent and family involvement in the education of K-12 students
After school programs and activities for elementary and middle school children
Adult education for all civilian, non-institutionalized adults
Civic education and participation for K-12 students, parents and adults
School safety and discipline for K-12 students
National Household Education Surveys (NHES)
Typically relies on two stages of interviews
1. Screener interview to determine household and withinhousehold eligibility – generally need 45,000- 60,000 completed screener interviews per collection
2. Extended interviews collects detailed information about person(s) sampled within the household
Significant nonresponse problem in the Screener
1. Approximately 80% in the early 1990s
2. Approximately 50% now
Increasing Response Rates with RDD
Use of premailingConducted experiments showing that sending letters to households before calling helped improve first stage response rates
Use of monetary incentivesUse of small monetary incentives mailed to households before calling helped improve first stage response rates
In-person follow-upDuring recent bias study, found that in-person contact with phone interview refusal cases significantly boosted response rates
Premailing with RDDLetters about the purpose of the study and where to get information about it help boost first stage response rates
In the 2001 NHES, theresponse rate for matched households sent a letter was 75%response rate for matched households not sent a letter was 70%response rate for households that could not be matched was 55%
Challenge is matching randomly generated phone sample to mailing addresses – approximately 50 percent
See http://nces.ed.gov/pubs2005/2005071_2.pdf
Prepaid Incentives with RDD
In 2003, we tested monetary incentive options
Required that we had a mailing address – again had a 50% match rate
Tested no incentives against combinations of $2 and $5 incentives sent before the interview and as a refusal conversion technique
http://nces.ed.gov/pubs2006/2006066.pdf
Prepaid Incentives with RDD
42.1
38.9
42.4
42.1
40.9
45.2
46.6
45.4
47.1
69.9
69.7
69.1
68.9
67.9
66.7
63.7
69.5
67.3
64.148.8
30 35 40 45 50 55 60 65 70 75
10 - ($2/1st $2)
9 - ($5/1st $0)
8 - ($2/Priority $2)
7 - ($2/Priority $0)
6 - ($2/1st $0)
5 - ($0/Priority $2)
4 - ($0/Priority $0)
3 - ($0/1st $5)
2 - ($0/1st $2)
1 - ($0 brochure/1st $0)
Incentive group
Rate (%)
Unit responserate (%)
Ever refused rate (%)
adfafasfas adfafasfas
Prepaid Incentives with RDDOptimal approach for cost v. response rate effect was no initial incentive with a $5 incentive for refusal conversion
Next best approach was $2 initial incentive with a $2 conversion incentive
Both boosted first stage response rates over no incentives by approximately 5 percentage points
Tested use of priority and regular mail for initial and refusal conversion mailing - mailing did not have an independent effect beyond incentives for the first stage of data collection
Study of Potential Bias and Multimode Approach
Conducted an extensive bias test in 2007
Key components included:1. In person follow-up for telephone nonrespondents –
used to study potential nonresponse bias2. In person contact for households not in the RDD
frame – important to understand potential coverage bias
Multimode Work
Results not released yet but some results are generally clear
• If budget permits, in-person follow up can significantly increase response rates if used to help with telephone refusal conversion
• If not corrected, surveys based on RDD samples of land line phones alone can incure coverage bias in studies of prekindergarten through high school students and their families
Frame Adjustments
NHIS indicates that in 2007 16% of households were cell only• Approximately 14% of children live cell phone only households• Approximately 15% of adults live in cell phone only households
Can correct for some of coverage bias through weighting
Preferable to incorporate cell phones into frames though
• This may reduce overall response rates• Challenging for calculating proper weights
http://www.cdc.gov/nchs/data/nhis/earlyrelease/wireless200805.pdf