Weighting Stratified Surveys Stratification Scenario n n n
- Slides: 9
Weighting Stratified Surveys
Stratification Scenario n n n Stratified cluster survey performed Each stratum has its own disease prevalence estimates and coverage estimates Want to combine stratum estimates into a single national estimate
Stratification example Stratified cluster survey – 3 strata Region Population distribution Urine samples tested distribution Low UI N % N % A 372, 978 62. 9 334 34. 5 112 33. 5 B 127, 841 21. 6 324 33. 5 185 57. 1 C 92, 117 15. 5 309 32. 0 119 38. 5 Total 592, 936 100 967 100 416 - What is the national average? Naïve approach: (33. 5 + 57. 1 + 38. 5)/3 = 43%
Stratification example Naïve approach assumes each stratum has equal weight (i. e. , 1/3 rd) Region Population distribution N % Urine samples tested distribution Low UI N % A 372, 978 62. 9 334 34. 5 112 33. 5 B 127, 841 21. 6 324 33. 5 185 57. 1 C 92, 117 15. 5 309 32. 0 119 38. 5 Total 592, 936 100 967 100 416 - The above table presents the population distribution and the sample distribution. Note how the distributions by population and sample differ. Should each stratum’s estimate be treated equally when estimating national average?
Stratification: weights There are many ways to create a weighted population estimate One approach is to: 1) For each stratum multiply % of the population in the stratum by the % of samples in the stratum with low UI 2) Sum this # across all stratums and divide by 100 Stratum weight = (% of population in stratum) x (% of population with low UI) Region Population distribution Urine samples tested distribution Low UI N % N % A 372, 978 62. 9 334 34. 5 112 33. 5 B 127, 841 21. 6 324 33. 5 185 57. 1 C 92, 117 15. 5 309 32. 0 119 38. 5 Total 592, 936 100 967 100 416 -
Weighted point estimate: [(62. 9% x 33. 5%) + (21. 6 x 57. 1%) + (15. 5% x 38. 5%)] / 100 = 39. 4% 2107. 1 + 1233. 4 + 59. 6 Interpretation: the prevalence is 39. 4% Naïve estimate was 43% (a biased population estimate)
Another example n view. Epi 10 - Complex Survey Data based on the Expanded Program for Immunization (EPI) method with 10 strata n The view. Epi 10 file is an example where an EPI survey was performed in each of its 10 provinces, a stratified cluster survey.
Another example
How to add weights to data? In addition to calculating weight by hand many programs, such as Epi Info, can also calculate weights First some calculations will need to be made by hand then the program can be written For example in region A the weight is 62. 9/ 34. 5 = 1. 823. Once all the weights are calculated they need to be added to the data program. Each individual in Region A will have the weight 1. 823