|
Size: 3688
Comment: Design
|
← Revision 7 as of 2026-02-06 20:44:55 ⇥
Size: 4003
Comment: R -> r
|
| Deletions are marked like this. | Additions are marked like this. |
| Line 24: | Line 24: |
| === Sampling === | |
| Line 26: | Line 25: |
| The studies have traditionally been administered through face-to-face interviews. Telephony (including CATI eventually) modes were added over time. In 2012, web surveys were added. | === Instrument === |
| Line 28: | Line 27: |
| Every time series incorporates a fresh cross-section sample. In addition, prior respondents are empaneled and recontacted. The 2000 study included an experiment to incorporate a new RDD sample. |
Current study design is: * study for every presidential election * two waves per study: * 'pre-election' survey fielded in the month before an election * 'post-election' recontact survey fielded in the month after an election |
| Line 34: | Line 35: |
| === Weighting === | === Sample === |
| Line 36: | Line 37: |
| Each time series study's fresh cross-section is designed to be 'self-weighting'. | The studies have traditionally been administered through face-to-face interviews. Telephony (including CATI eventually) modes were added over time. In 2012, web surveys were added. In 2020, video interviews were used to supplement some web interviews. In 2024, towards the end of the fielding period, an abbreviated paper-and-pencil (PAPI) survey was distributed to nonrespondents. |
| Line 38: | Line 39: |
| TODO: document weight design for 2012+ data files | Each time series incorporates a fresh cross-section sample. In addition, prior respondents are empaneled and recontacted. |
| Line 40: | Line 41: |
| The full time series data file, including fresh cross-section data and panel data, requires weighting. Base weights reflect unequal probabilities of household selection as the PSU. If the household contained more than one eligible adult, the weights are adjusted for unequal probability of selecting the respondent. | |
| Line 42: | Line 42: |
| Weights are adjusted for non-response bias. The panel sample is adjusted particularly for attrition.Methodology write-up suggests only geographic characteristics are available for this computation. | === Weights === Each time series' fresh cross-section sample is designed to be 'self-weighting'. Nonetheless, with the modal experiments/changes since 2012, multiple weights have been made available to study the samples separately or together. This also means that survey weights are necessary for analysis using an entire time series data file, as it includes the fresh cross-section sample, panel sample, and (potentially) supplemental sample. Base weights reflect unequal probabilities of household selection as the PSU. If the household contained more than one eligible adult, the weights are adjusted for unequal probability of selecting the respondent. Weights are adjusted for [[Statistics/NonresponseBias|nonresponse bias]]. Weights for the panel sample is adjusted for attrition. Note: methodology write-up suggests only geographic characteristics are available for this computation. |
| Line 46: | Line 53: |
| TODO: document when weights began to be re-scaled to have a mean at 1. | At some point, the weights began to be re-scaled to have a mean at 1. |
| Line 48: | Line 55: |
| So-called 'post-election weights' use the pre-election weight as the base weight. It is corrected for attrition between the waves. It is also re-raked ''unless'' there are significant discrepancies between the demographics of retained and attrited cases. Such re-raking always includes presidential vote choice (if in a presidential election year) (to official certified results) and voter turnout (to estimates by United States Elections Project). === Frequency === Until 2002, the time series studies were administered for both presidential and midterm federal elections (every 2 years). The presidential election studies had two-waves, corresponding to a ''pre-election'' interview fielded in the month leading up to election day, and a ''post-election'' interview fielded in the month following election day. The midterm election studies had just the post-election component. In 2002, the studies have been reoriented to always have two waves. The 2002 study is the last administration for a midterm federal election; since then all studies have been restricted to presidential elections. |
Post-election weights use the pre-election weight as the base weight. It is corrected for attrition between the waves. It is also re-raked ''unless'' there are significant discrepancies between the demographics of retained and attrited cases. Such re-raking always includes presidential vote choice (if in a presidential election year) (to official certified results) and voter turnout (to estimates by United States Elections Project). |
| Line 66: | Line 65: |
| Until 2002, studies were administered for ''both'' presidential and midterm federal elections (every 2 years). Note that only presidential elections had a two wave survey instrument; the midterm elections only used a post-election survey. Despite being a midterm election, 2002 used a two wave instrument. There have been no midterm election studies since 2002. |
American National Election Studies
The American National Election Studies (ANES) are surveys administered around every federal election in the United States.
Usage
See here for notes on the time series data files.
See here for supplemental notes on the cumulative data files built from those time series studies.
Design
Instrument
Current study design is:
- study for every presidential election
- two waves per study:
- 'pre-election' survey fielded in the month before an election
- 'post-election' recontact survey fielded in the month after an election
Sample
The studies have traditionally been administered through face-to-face interviews. Telephony (including CATI eventually) modes were added over time. In 2012, web surveys were added. In 2020, video interviews were used to supplement some web interviews. In 2024, towards the end of the fielding period, an abbreviated paper-and-pencil (PAPI) survey was distributed to nonrespondents.
Each time series incorporates a fresh cross-section sample. In addition, prior respondents are empaneled and recontacted.
Weights
Each time series' fresh cross-section sample is designed to be 'self-weighting'. Nonetheless, with the modal experiments/changes since 2012, multiple weights have been made available to study the samples separately or together.
This also means that survey weights are necessary for analysis using an entire time series data file, as it includes the fresh cross-section sample, panel sample, and (potentially) supplemental sample.
Base weights reflect unequal probabilities of household selection as the PSU. If the household contained more than one eligible adult, the weights are adjusted for unequal probability of selecting the respondent. Weights are adjusted for nonresponse bias. Weights for the panel sample is adjusted for attrition. Note: methodology write-up suggests only geographic characteristics are available for this computation.
Weights are post-stratified to CPS targets. Post-stratification is considered necessary when the discrepancy between sample and known population characteristics is larger than 5%, and not necessary when smaller than 2%. Methodology write-up suggests that only one- and two-factor dimensions are used. Weights are trimmed at ~5 times the average.
At some point, the weights began to be re-scaled to have a mean at 1.
Post-election weights use the pre-election weight as the base weight. It is corrected for attrition between the waves. It is also re-raked unless there are significant discrepancies between the demographics of retained and attrited cases. Such re-raking always includes presidential vote choice (if in a presidential election year) (to official certified results) and voter turnout (to estimates by United States Elections Project).
History
The project was launched by the University of Michigan as the Michigan Election Studies. The National Science Foundation began funding it in 1977, at which point they were renamed to the National Election Studies.
Until 2002, studies were administered for both presidential and midterm federal elections (every 2 years). Note that only presidential elections had a two wave survey instrument; the midterm elections only used a post-election survey. Despite being a midterm election, 2002 used a two wave instrument. There have been no midterm election studies since 2002.
To distinguish the studies from similar projects based in other nations, they became known as the American National Election Studies in 2005.
The studies are still led by UMich but are now in collaboration with Duke, University of Texas at Austin, and Stanford.