Fig 1.
Number of edits and quality by category granularity.
The noteworthy pattern is visible in the left-hand side of the diagram (articles in coarse categories). The 20% of articles with the coarsest categories receive above-average numbers of edits but their quality evaluations deteriorates compared to those with a medium category granularity.
Fig 2.
Average probability to be featured for articles in 10 edit classes.
Fig 3.
Average return for the ten granularity classes.
Darker bars represent coarser articles.
Fig 4.
Probability of being top-importance for the ten granularity classes.
Fig 5.
Return by category granularity separately for 10 different edit classes restricted to the 10% or articles receiving the highest number of edits.
The bar chart for each edit class can be read in the same way as the single bar chart in Fig 3.
Table 1.
Linear regression for the logarithm of the number of edits.
Table 2.
Logistic regression for FA-probability.
Table 3.
Logistic regression for FA-probability.
Fig 6.
Mean number of edits and effect of granularity on the number of edits, separately for each TLC.
Mean number of edits is displayed in the x-axis. The linear regression coefficient α1 of the granularity variable explaining the number of edits (compare Eq 1) is displayed in the y-axis. Area of points is proportional to the number of articles in the respective top-level category. All parameter estimates are significantly different from zero (p < 0.001).
Fig 7.
Average quality and coefficient of granularity explaining quality, separately for each TLC.
The baseline probability of featured articles in the respective TLC is displayed in the x-axis. The logistic regression coefficient of the granularity variable, when controlling for the number of edits (parameter θ2 in Eq 2), is displayed in the y-axis. Coefficients that are significant (insignificant) at the 5% level are displayed as red (gray) dots.