Weighting Your Data
Feb 01, 2016
Weighting Your Data
General Information
Allows you to generalize your results to the entire population
Necessary for all sampled based surveys Requirements
Scientifically selected sample Complete documentation High overall response rate
What does weighting do?
It accounts for the probability of selection within the population non-response distribution of the target population by age and sex
It allows one participant to represent many others
Probability of Selection from the Sample (Individual Weight)
Accounts for the variation in the selection probabilities in the sample
= 1/[(selection probability of cluster 1)* (selection probability of cluster 2)*(as many clusters until reach primary sampling unit)]
Example
There are 12 districts and 3 are randomly selected; probability of selection at district level =.25
District Towns selected
Towns in district
Probability of selection (town)
INDIVIDUAL WEIGHT
A 8 30 .27 1/(.25*.27)
B 3 10 .30 1/(.25*.30)
C 5 22 .23 1/(.25*.23)
Non-Response
Non-response can cause bias Information on response must be collected during
interviews Interview tracking form collects this information
Entered during data entry Automatically attached to dataset with Epi Info program
Calculating the Non-Response Weight
Calculate non-response for each age and sex stratum
Non-response weight =
1/(response rate for age-sex stratum)
Example
Males by age
Response Rate (RR)
Non-response weight = 1/RR
25–34 .94 1.064
35–44 .95 1.053
45–54 .87 1.149
55–64 .79 1.266
Non-response weights
would be calculated for
both men and women.
Population Distribution
Used to adjust the sample to the target population "Post-stratification adjustments"
Need population information stratified by age and sex to calculate
Calculating Population Weight
Population weight =
Proportion of population/Proportion of sample
Population weight =
(Age-Sex of population / Total population)
(Age-Sex of sample / Total Sample)
Example
Males by age
Pop N Proportion of population (A)
Sample N Proportion of sample (B)
Weight
= A/B
25–34 2,000 .200 30 .075 2.667
35–44 1,760 .176 40 .100 1.760
45–54 1,440 .144 50 .125 1.152
55–64 1,600 .160 60 .150 1.067
Total 10,000 400
Post stratification would be calculated for males and females
Overall Weight for Individual Steps
W1: Individual weight W2s1: Non-response weight W3: Population weight
WStep1= W1*W2s1*W3 WStep2= W1*W2s2*W3 WStep3= W1*W2s3*W3
Weighting and STEPS
Tools to help calculate weights STEPSsampling.xls Interview_Tracking_Form.xls
Weights
The Weights spreadsheet is used to document the sample selection and attach the weights to the dataset in Epi Info
Part of the spreadsheet is automated and parts need to be filled in by hand
Weights: Example
Automatically entered
Enter by hand by matching information from PSU and Clustering SSU
Information available from Clustering SSU
spreadsheet Weighting Info
Information available from Rand Hhold
spreadsheet Weighting Info
Attaching the Weights to Your Dataset
Once you have documented your weights using the STEPSsampling.xls Interviewtracking.xls
THEN …
You can use the generic analysis programs to automatically attach the weights to your dataset