Class 6_Stat methods for analyzing non response Flashcards
Lessler and Kalsbeek (1992)
stochastic view
deterministic view
-Measuring non response bias- no good RR
Compared R and NR on aux. variables available for full sample that can be correlated with survey variables- differ could suggest bias
use response propensity as indicator of likelihood to responsd and correlate it with key variables
Lessler and Kalsbeek (1992): Stochastic view
Decision to participate based on person specific prob of responding
can be impacted by various design decisions
Lessler and Kalsbeek (1992): Deterministic view
population includes two types of people- R and NR
Continuing of resistance idea
Do people who choose to participate later produce diff estimates
Ways to reduce non response bias
Extrapolation
Nonrespondent sub sampling
Weighting adj
Extrapolation
Use observable features of NR to predict what their survey responses would have been
Non respondent sub sampling
Subsample of NR selected
alt design strategies to increase response
Weighting adj
Use aux variable to form weighting classes(have to be related to key variables Y and response propensity)
Bethlehem (2002)
Expands stochastic view
-Bias arises from covariance bw response propensity and key survey variables
Est. NR bias based on diff bw respondents and the pop on aux. variables
GREG estimator requirement
requires good auxilliary info for the whole pop
O’Muircheartaigh and Campanelli (1999)
example of est. int variance in area prob sample
effects of int on non contact and refusal
O’Muircheartaigh and Campanelli (1999) Results
int vary of achieved RR
Urban areas have higher rates of NR
Int good at reducing whole HH R=reducing whole HH non contacts
between int variance larger than between area variance
Schouten et al (2011)
role of paradata in studing NR bias
“representatives” indicator
measure degree to which respondent rep full sample
Schouten et al (2011): Unconditional Partial R indicator
SD of the response propensities based on single variable Z- determine contribution of Z to the lack of rep
Schouten et al (2011): Conditional Partial R indicator
SD of the response propensities based on all other aux variables besides Z -measure contribution of Z to lack of rep. while controlling for other aux. variables