The Correlation between Wikimedians of Mainland China Membership and Voting Pattern in 2017 on Chinese Wikipedia
All documentations are released under Creative Commons Attribution-ShareAlike 3.0 Unported. All programs are released under GNU General Public License 3.0.
We investigated the correlation between Wikimedians of Mainland China (WMC) membership and voting pattern.
We sampled all community elections in 2017 on Chinese Wikipedia. Community election is defined as an election where the community decides whether the nominee should receive an administrative privilege including administrator, bureaucrat, checkuser, or oversighters. Any election that was closed before the two week election period is not counted in the sample. We used the public list of WMC members to compile a list of WMC members.
Every user that had casted a valid vote is counted as an instance. Each support vote is coded as 1. Each oppose vote is coded as -1. If a user was absent or voted other than support or oppose, their vote is coded 0.
We removed one invalid vote due to error in the sampling process. We reattributed 16 votes as the users that casted them had either changed their signatures or their usernames. A 23-by-217 matrix is made from the sample.
In order to obtain a lower dimension representation of the data, we used the
pca function provided in scikit-learn to perform principle component
analysis to 2 dimensions. We obtain first two principle component, PC1 and PC2.
We performed logistic regression in the log-likelihood of being a WMC member in response of changes in PC1 and PC2.
Logistic Regression Model
lrm(formula = Membership ~ PC1 + PC2, data = sample)
Model Likelihood Discrimination Rank Discrim.
Ratio Test Indexes Indexes
Obs 217 LR chi2 107.73 R2 0.589 C 0.918
FALSE 166 d.f. 2 g 2.928 Dxy 0.836
TRUE 51 Pr(> chi2) <0.0001 gr 18.684 gamma 0.841
max |deriv| 2e-09 gp 0.298 tau-a 0.302
Brier 0.096
Coef S.E. Wald Z Pr(>|Z|)
Intercept -2.0626 0.3001 -6.87 <0.0001
PC1 1.2297 0.2791 4.41 <0.0001
PC2 -3.7190 0.6354 -5.85 <0.0001We simulated random voting by drawing from the uniform distribution {-1, 0, 1}
for comparison. We repeated the PCA for non-WMC users and the simulated data.
