|
?:comment
|
-
Election Study Notes on Module 2 (ZA5180)
(en)
-
FOR COUNTRY SPECIFIC INFORMATION
see Election Study Notes
(en)
-
|
Sample Weight |
Demographic Political |
Political Weight |
No Weight Provided |
ALBANIA (2005) |
|
|
|
X |
AUSTRALIA (2004) |
|
|
|
X |
BELGIUM (2003) |
|
X |
X |
|
BRAZIL (2002) |
X |
X |
|
|
BULGARIA (2001) |
|
X |
|
|
CANADA (2004) |
X |
|
|
|
CHILE (2005) |
|
X |
|
|
CZECH REPUBLIC (2002) |
|
|
|
X |
DENMARK (2001) |
|
|
|
X |
FINLAND (2003) |
|
|
|
X |
FRANCE (2002) |
|
X |
|
|
GERMANY (2002 Mail-Back) |
X |
X |
|
|
GERMANY (2002 Telephone) |
X |
X |
|
|
GREAT BRITAIN (2005) |
X |
X |
|
|
HONG KONG (2004) |
|
|
|
X |
HUNGARY (2002) |
|
X |
|
|
ICELAND (2003) |
|
|
|
X |
IRELAND (2002) |
X |
X |
|
|
ISRAEL (2003) |
|
|
|
X |
ITALY (2006) |
|
X |
|
|
JAPAN (2004) |
X |
|
|
|
KOREA (2004) |
|
|
|
X |
KYRGYZSTAN (2005) |
|
|
|
X |
MEXICO (2003) |
X |
X |
X |
|
NETHERLANDS (2002) |
|
X |
X |
|
NEW ZEALAND (2002) |
X |
X |
X |
|
NORWAY (2001) |
|
|
|
X |
PERU (2006) |
|
X |
X |
|
PHILIPPINES (2004) |
|
|
X |
|
POLAND (2001) |
X |
X |
|
|
PORTUGAL (2002) |
|
|
X |
|
PORTUGAL (2005) |
|
X |
X |
|
ROMANIA (2004) |
|
|
|
X |
RUSSIA (2004) |
X |
|
|
|
SLOVENIA (2004) |
|
|
|
X |
SPAIN (2004) |
|
X |
|
|
SWEDEN (2002) |
|
|
|
X |
SWITZERLAND (2003) |
X |
|
|
|
TAIWAN (2001) |
|
|
|
X |
TAIWAN (2004) |
|
X |
|
|
UNITED STATES (2004) |
X |
X |
|
|
(en)
-
Seven election studies have provided a single weight that is a combination of two or more of the three weight categories (sample, demographic, and political). For these election studies, the weight is duplicated in the two or more appropriate weight variables. Thus, analysts attempting to use two or more categories of the weights simultaneously will need to account for this duplication. There are duplicate weights in the CSES Module 2 Full Release for these seven studies:
- Brazil (2002): sample, demographic
- Germany (2002 Mail-Back): sample, demographic
- Germany (2002 Telephone): sample, demographic
- Great Britain (2005): sample, demographic
- Ireland (2002): sample, demographic
- Poland (2001): sample, demographic
- United States (2004): sample, demographic
There were also five election studies that provided separate weights in two or more categories (sample, demographic, political):
- Belgium (2003): demographic, political
- Mexico (2003): sample, demographic, political
- Netherlands (2002): demographic, political
- New Zealand (2002): sample, demographic, political
- Peru (2006): demographic, political
- Portugal (2005): demographic, political
Sometimes weights build on each other. For example, a demographic weight might take an existing sample weight and supplement it with additional demographic corrections. Because CSES does not create the original weight variables (rather, the variables are provided by collaborators), we do not know firsthand if the multiple weights for an election study are independent, or if they build on each other. In the absence of documentation from the collaborator, one piece of evidence that can be used to guess at this is to look at the correlations between the different weights. The correlations for the seven aforementioned election studies are as follows:
- 0.8544: New Zealand (2002): demographic-political
- 0.6827: New Zealand (2002): sample-demographic
- 0.5600: New Zealand (2002): sample-political
- 0.5575: Belgium (2003): demographic-political
- 0.3137: Netherlands (2002): demographic-political
- 0.1351: Mexico (2003): sample-political
- -0.0157: Portugal (2005): demographic-political
- -0.0091: Mexico (2003): sample-demographic
- -0.0051: Mexico (2003): demographic-political
- -0.0011: Peru (2006): demographic-political
Pairs of weights with higher correlations are perhaps more likely to be a situation where one weight builds on another, or at least they share some of the same types of corrections. As confirmation of the value of this test, we see in the election study note for the New Zealand (2002) weights that the weights in New Zealand are cumulative, which is hinted at by the high correlations for New Zealand in the above list. We also know from the Belgium (2003) election study note that their political weight takes into account demographic corrections as well. This is confirmed by the relatively high correlation (0.5575) of the Belgian demographic and political weights in the table above.
C. OVERSAMPLES
Analysts can use the election study notes for the weight variables to identify design issues such as oversamples, which are usually accounted for in the development of sample weights. For example, we know from the election study weights that the designs for these four studies intentionally oversample:
- Canada (2004) oversamples the eight smaller provinces.
- Germany (2002 Telephone) and Germany (2002 Mail-Back) both oversample the East German population.
- New Zealand (2002) oversamples the Maori population.
Typically a sample weight makes corrections to account for any oversample and makes the election study representative again of the general population. For countries with oversamples that provide multiple weights (sample, demographic, political) some hope to see that the demographic and political weights are cumulative, carrying forward any oversample corrections from the sample weight. In CSES Module 2, this seems to be the case. Of the four aforementioned election studies, only the Germany (2002 Telephone) and New Zealand election studies provided multiple weights. We know that the New Zealand weights (sample, demographic and political) are cumulative because this is mentioned in the election study notes. This is also supported by the high correlations between the New Zealand weight variables. Both election studies from Germany provided a single weight each that makes both sample and demographic corrections, including an adjustment for the oversample of the East German population.
D. OTHER NOTES ABOUT THE WEIGHTS
Analysts will want to keep in mind that the CSES weights are prepared to be election study weights, not country weights. To convert the weights to country weights requires an adjustment for those countries for which one or more polities or election studies appear in the dataset.
Where a weight of a particular type is not available, these variables are coded "1" for each respondent.
(en)
-
NOTES: B1010
These variables report the original ("raw") weights provided with the respective deposited data files.
Selected documentation from the election studies about their weights is provided in the election study notes that accompany these original weight variables (B1010).
A. TYPES OF WEIGHTS AVAILABLE FOR USE
Election studies are allowed to provide weights to CSES in as many as three categories:
1. "Sample" weights include those intended to correct for unequal selection probabilities resulting from "booster" samples, oversampling of a particular population, procedures for selection within the household, as well as other features of the sample design.
2. "Demographic" weights adjust sample distributions of socio-demographic characteristics to more closely resemble the characteristics of the population.
3. "Political" weights reconcile discrepancies in the reported electoral behavior of the survey respondents from the official vote counts.
Use of weights is at the discretion of the analyst based upon the considerations of her/his individual research question. We recommend that analysts familiarize themselves with the weights, their components, and their methods of creation before applying them.
Analysts can choose to make use of this set of original weights (B1010), make use of either set of derivative weights created by CSES (B1012 or B1014), create their own weights, or supplement the original weights. And of course, depending on the analysis, analysts sometimes choose not to use weights at all.
- The derivative "Polity Weight" (B1012) has been created so that for each weight (sample, demographic, political), each respondent within the election study has a mean weight of "1". If you are running a frequency, for instance, this weight will work so that the N in your frequency table comes out to approximately the same as the number of interviews in the study. This derivative weight is created by dividing the individual weight for each respondent within an election study by the mean for that weight for all respondents in that election study.
- The derivative "Dataset Weight" (B1014) has been created so that each election study in the dataset will contribute equally to analyses of respondents, regardless of the number of interviews in each election study.
B. ELECTION STUDIES WITH MORE THAN ONE WEIGHT
Most election studies either provide a single weight in one of the three categories (sample, demographic, political), or they do not provide a weight at all. Twelve election studies, however, provided weights for more than one category.
Table: Summary of Election Study Weights, by Type of Weight.
(en)
-
Original Weight: Political
(en)
|
![[This page as RDF]](https://data.gesis.org/gesiskg/static/rdf-icon.gif){kind=icon}