|
⇤ ← Revision 1 as of 2025-04-28 18:54:15
Size: 1616
Comment: Initial commit
|
← Revision 2 as of 2025-04-28 18:56:33 ⇥
Size: 1630
Comment: Correction
|
| Deletions are marked like this. | Additions are marked like this. |
| Line 3: | Line 3: |
| '''Improving on Probability Weighting for Household Size''' (DOI: https://doi.org/10.1086/297852) was written by Andrew Gelman in 1998. It was published in ''Public Opinion Quarterly'' (vol. 62, no. 3). | '''Improving on Probability Weighting for Household Size''' (DOI: https://doi.org/10.1086/297852) was written by Andrew Gelman and Thomas C. Little in 1998. It was published in ''Public Opinion Quarterly'' (vol. 62, no. 3). |
| Line 5: | Line 5: |
| The author explores a specific type of [[Statistics/SurveyWeights|inverse probability weighting]]: adjusting for the selection of an individual from a sampled household with multiple potential respondents. This adjustment introduces bias because this selection criteria is related to a pattern of [[Statistics/NonResponseBias|nonresponse]], i.e. multi-person households are more likely to have an available respondent. | The authors explore a specific type of [[Statistics/SurveyWeights|inverse probability weighting]]: adjusting for the selection of an individual from a sampled household with multiple potential respondents. This adjustment introduces bias because this selection criteria is related to a pattern of [[Statistics/NonResponseBias|nonresponse]], i.e. multi-person households are more likely to have an available respondent. |
| Line 9: | Line 9: |
| The author finds measurable nonresponse bias such that unweighted estimates are similar or better than weighted ones. The author also finds that the bias is best corrected by [[Statistics/PostStratification|poststratifying]] estimates on [[Statistics/Binning|bins]] of household size (i.e., 1, 2, 3, 4, 5+, etc.; collapsing the right tail as necessary). | The authors find measurable nonresponse bias such that unweighted estimates are similar or better than weighted ones. They also find that the bias is best corrected by [[Statistics/PostStratification|poststratifying]] estimates on [[Statistics/Binning|bins]] of household size (i.e., 1, 2, 3, 4, 5+, etc.; collapsing the right tail as necessary). |
Improving on Probability Weighting for Household Size
Improving on Probability Weighting for Household Size (DOI: https://doi.org/10.1086/297852) was written by Andrew Gelman and Thomas C. Little in 1998. It was published in Public Opinion Quarterly (vol. 62, no. 3).
The authors explore a specific type of inverse probability weighting: adjusting for the selection of an individual from a sampled household with multiple potential respondents. This adjustment introduces bias because this selection criteria is related to a pattern of nonresponse, i.e. multi-person households are more likely to have an available respondent.
This is tested with 1988 presidential race polling done by CBS News and the New York Times, and also with the NES 1988 time series.
The authors find measurable nonresponse bias such that unweighted estimates are similar or better than weighted ones. They also find that the bias is best corrected by poststratifying estimates on bins of household size (i.e., 1, 2, 3, 4, 5+, etc.; collapsing the right tail as necessary).
Reading notes
Alternatively, apply a nonresponse adjustment with respect to household size.
ANES has historically taken the view that their data is self-weighting. This analysis probably demonstrates the typical rationale behind such assertions; the weighting methods that were in use were not sophisticated enough to meaningfully correct for biases.