ANES Cumulative
The fresh cross-section samples of ANES time series studies are aggregated into cumulative data files.
Contents
Weights
From the data center: "Over the years, the most common ANES study design has been a cross-section, equal probability, sample. These designs are typically "self-weighting"...".
In recent distributions of the cumulative data file, there are several weights. Most weight factors are simply "1.0" per the stated design. Principally, there are 9 weights based on a cross-selection of two analysis choices:
For 2012 and 2016, should all data be used (z), or only data from face-to-face interviews (x), or only data from web interviews (y)?
For 1970, should data collected from the most common version of the questionnaire ("Type 0", 0009) be used, or from the "Type 1" questionnaire (0010), or from the "Type 2" questionnaire (0011)?
Apart from these studies, the weights are identical across the variables.
Note that mode of interview is captured in VCF0017, but this is not the same as (1.). The weights are partitioned for sample design, i.e. treatment groups.
In summary, this will probably suffice:
svyset [pweight=VCF0009z], vce(linearized)
If working with only face-to-face interview data, try using VCF0009x instead.
There is also a post-election weight (VCF9999) if working exclusively with respondents to the post-election interviews.
Data
Variable |
Description |
Notes |
VCF0004 |
Year of study |
|
VCF0016 |
Sample indicator (panel "recontact", "fresh", etc.) |
|
VCF0017 |
Mode of interview |
|
VCF0803 |
Liberal-Conservative self-identifiation, 7-point scale |
|
VCF0824 |
Liberal-Conservative compelled choice, 3-point scale |
Follow-up if VCF0803 was refused or if VCF0803="Moderate" |
VCF0808 |
Should guarantee jobs and living standards |
|
VCF0809 |
Should guarantee jobs and living standards, 7-point scale |
|
VCF0830 |
"Aid to blacks" 7-point scale |
|
VCF0843 |
Should increase defense spending, 7-point scale |
|
VCF0006 is a unique ID within each study (as indicated by VCF0004). VCF0006a is unique across all studies.
Quality
A quick exploration of the data reveals possibly unintended interview patterns, and illogical combinations.
VCF0824 is a follow-up question; it should only be asked if VCF0803 was refused or if a response of "Moderate" was given (in a subset of years). The intention is to compel respondents to 'pick a side'.
. svy: prop VCF0803 VCF0824, percent
(running proportion on estimation sample)
Survey: Percent estimation
Number of strata = 1 Number of obs = 37,572
Number of PSUs = 32,951 Population size = 37,502.3
Design df = 32,950
---------------------------------------------------------------------------------------------------
| Linearized Logit
| Percent std. err. [95% conf. interval]
--------------------------------------------------+------------------------------------------------
VCF0803 |
0. NA; no Post IW; form III,IV (1972); R not | 4.66 0.13 4.42 4.92
1. Extremely liberal | 2.20 0.09 2.02 2.39
2. Liberal | 8.44 0.17 8.11 8.79
3. Slightly liberal | 8.48 0.17 8.16 8.82
4. Moderate, middle of the road | 26.05 0.28 25.51 26.60
5. Slightly conservative | 11.25 0.19 10.88 11.62
6. Conservative | 13.92 0.22 13.50 14.36
7. Extremely conservative | 2.97 0.11 2.77 3.18
9. DK; haven't thought much about it | 22.03 0.27 21.51 22.56
|
VCF0824 |
0. R has made choice on 7pt scale VCF0803 | 55.93 0.31 55.32 56.54
1. Liberal | 11.71 0.21 11.31 12.13
3. Moderate ('middle of the road') | 9.74 0.19 9.38 10.11
5. Conservative | 16.69 0.24 16.22 17.17
7. R refuses to choose | 1.81 0.08 1.65 1.98
8. DK | 3.45 0.10 3.25 3.66
9. NA in either question; technical error (2020) | 0.67 0.05 0.58 0.77
---------------------------------------------------------------------------------------------------The natural expectation, then, is that a valid response to the follow-up should require non-response or a "Moderate" response to the lead. This is not the case:
. tab VCF0803 VCF0824 if inlist(VCF0803,1,2,3,5,6,7) & inlist(VCF0824,1,3,5)
| If Compelled to Choose Liberal
Liberal-Conservative | or Conservative
Scale | 1. Libera 3. Modera 5. Conser | Total
----------------------+---------------------------------+----------
1. Extremely liberal | 4 3 2 | 9
2. Liberal | 21 24 7 | 52
3. Slightly liberal | 40 90 15 | 145
5. Slightly conservat | 12 79 63 | 154
6. Conservative | 6 29 44 | 79
7. Extremely conserva | 1 3 12 | 16
----------------------+---------------------------------+----------
Total | 84 228 143 | 455 The following tabulations seem to be of interest, but are reproduced here without a thoughtful explanation.
. generate ctd = (inlist(VCF0803,5,6,7) & inlist(VCF0824,1)) | (inlist(VCF0803,1,2,3) & inlist(VCF0824,5))
. keep if inlist(VCF0803,1,2,3,5,6,7) & inlist(VCF0824,1,3,5)
(67,769 observations deleted)
. tab VCF0017 ctd
| ctd
Mode of Interview | 0 1 | Total
----------------------+----------------------+----------
0. All personal | 95 15 | 110
4. All internet | 317 28 | 345
----------------------+----------------------+----------
Total | 412 43 | 455
. tab VCF0016 ctd
Cross-section |
composition: fresh | ctd
Cross or panel case | 0 1 | Total
----------------------+----------------------+----------
0. Fresh Cross case | 412 43 | 455
----------------------+----------------------+----------
Total | 412 43 | 455
. tab VCF0004 ctd
Year of | ctd
Study | 0 1 | Total
-----------+----------------------+----------
2016 | 412 43 | 455
-----------+----------------------+----------