class 4_repairs for undercoverage

Question 1

Link et. al (2008)- description

Answer 1

Compared RDD and Addressed based mail in 6 states
5/6- low response rate
Examined
-Response rate
-representativeness
-Costs

Question 2

Link et. al (2008)- results

Answer 2

post card boosted responses
Mail over rep whites and college ed.
Mail under rep- large HH and rural 
Mail cheaper
using ABS led to another source of error

Question 3

Post survey fixes

Answer 3

post stratification- prominent method
Raking
margins
propensity weighting

Question 4

Three forms of non observation error

Answer 4

Sampling/Selection bias
coverage
non response

Question 5

Post stratification

Answer 5

prominent method
reduce variance on sample est. by adjusting to known totals
an adjustment for coverage error
Adjust weights to pop figure- calibrate who responded back to target pop
ignores non response and undercoverage

Question 6

Raking

Answer 6

make sure surv pop match target pop on
fancier version of post strat.
Adjust weights for marginal on specific variable
lead to less extreme adj. generally

Question 7

Examples of Calibration weighting

Answer 7

Post strat
Raking
GREG

Question 8

Propensity Weighting

Answer 8

Adjust weights for selection bias, noncoverage, or non response
Comes from obs studies
Use covariates to predich which group case gets into

Question 9

Steps for propensity weighting

Answer 9

1. Fit model to predict who volunteers, covered, responds

2. use fitted probability to adjust weights

Question 10

Propensity weighting in Web Panels

Answer 10

calibrate web panel est. to another survey
Calculate prob of being in RDD sample
Use est prob. to adjust weight