PROBABILITY-BASED SAMPLING USING Split-Frames with Listed Households

Transcription

1 PROBABILITY-BASED SAMPLING USING Split-Frames with Listed Households Mary E. Losch Mansour Fahimi University of Northern Iowa Marketing Systems Group Center for Social & Behavioral Research

2 Presentation Outline Need for Creative Sampling Methods Split-Frame RDD Sampling: Sampling considerations Weighting considerations Case Study: Employed sampling design Employed weighting methodology Selected results Concluding Remarks 2

3 Creative Sampling Methods EPSEM sampling is ideal for overall estimates, but: May not address domain estimates efficiently Requires heavy screening for rare subgroup surveys Need creative sampling methods to: Reduce cost of and time needed for data collection (screening) Maintain statistical integrity of survey sampling 3

4 Split-Frame RDD Sampling Sampling Considerations A representative sample mirrors population: Geodemographic characteristics: State, counties, etc. Gender, race, age, education, etc. Behavioral characteristics: telephone status RDD sampling alternatives: Landline RDD alone: not representative samples Dual-frame RDD alone: representative but expensive for rare subgroups 4

5 Split-Frame RDD Sampling Sampling Considerations Split-frame RDD to the rescue: Landline stratum: Targeted (listed landline) households: Listed Unlisted All remaining landline households Cellular stratum Optimal sample allocation to strata: Minimize cost of data collection Subject to meeting precision/analytical requirements 5

6 Split-Frame RDD Sampling Weighting Considerations Weights are needed to compensate for: Differential selection probabilities Design Weights Differential nonresponse Poststratified/Raked Weights Cost of weighting: Oversampling by Design Effective Sample Size Differential Nonresponse Effective Sample Size 6

7 Case Study Study of Women in Iowa Reproductive health study targeting women Conducted July November ,391 completed interviews AAPOR RR3 Landline Listed-Targeted 23% Landline Listed-Untargeted 14% Landline Unlisted 20% Cell 39% 7

8 Case Study Employed Sampling Design Strata Sample Interview Listed households with females , Remaining listed households 26, Unlisted households 7,000 Cellular 17, Select counties with listed households 7, Total 70,700 1,391 8

9 Case Study Employed Weighting Methodology) Missing data for weighting variables were imputed using a hot-deck process Design weights were computed to reflect selection probabilities Proc WgtAdjust of SUDAAN was used to rake the design weights along: Race/ethnicity Age Education County size 9

10 Case Study Demographic Comparisons Across Sample Strata Category Census Estimates Listed Targeted n=606 Remaining Listed n=224 Cellular n=217 Age % 30.2% 16.9% 14.7% % 27.9% 34.4% 45.2% % 41.9% 48.7% 33.1% Race White 92.5% 96.5% 92.9% 91.2% Black 4.5% 1.3% 2.2% 5.5% All Other 4.4% 2.3% 3.1% 5.1% Ethnicity Hispanic 5.8% 3.0% 5.4% 3.2% Marital Status Married 27.1% 29.7% 43.2% 31.3% Divorced/Separated/Widowed 3.8% 2.1% 4.1% 1.8% Never married 67.7% 68.4% 52.7% 66.8% Education Less than high school 10.7% 5.0% 6.2% 5.1% High school/ged 23.1% 24.6% 23.2% 24.9% Some college/vocational 42.5% 40.6% 40.6% 42.4% Bachelors degree 20.1% 24.8% 22.8% 23.0% Some graduate or more 3.6% 5.1% 7.1% 4.6% 10

11 Case Study Select Variable Comparisons Across Sample Strata See or Hear BC Info in Last 6 Months? Maximum Difference Listed Targeted n=606 Remaining Listed n=224 Cellular n=217 % Yes 4.9% 71.3% 66.7% 71.6% See or Hear Info about Stork? % Yes 17.1% 77.2% 60.1% 75.1% Fear Having Unintended Pregnancy Mean Rating on 1-5 Agreement Scale Influence of Religious Beliefs on Birth Control Mean Rating on 1-5 Influence Scale Availability of Low Cost BC in Community % Excellent 5.5% 18.4% 22.9% 23.9% % Poor 4.5% 4.5% 9.0% 5.3% 11

12 Concluding Remarks Landline RDD alone is not representative Dual-frame RDD alone is representative but can be expensive when targeting subgroups Employing Split-frame RDD will: Reduce screening cost Increase sampling variability Need to: Include mutually exclusive and exhaustive sampling strata Determine an optimal sample allocation 12