One of the acknowledged and shared long-term objectives of CV and of other .SE Q&A sites appears to be the creation of an on-line high-quality "Repository of Knowledge" (RoK).

While "quality" can be assessed and judged on its own, it is always useful to have some feedback on the matter. Specifically, on whether the CV-RoK is actually used as such.

I think that we could have an indication of whether this happens or not, by examining the time-distance between upvotes and answers.
While upvoting fresh questions is subject to a "go vote" bias (i.e. vote to show your appreciation for the quality of the answer, irrespective of whether the answer actually was immediately useful to the voter), when upvoting older answers it is more likely that the vote sends the message "this answer helped me".

If we denote $T_{v_j}$ the calendar timestamp of an upvote to answer $A_j$ and $T_{A_j}$ the calendar timestamp of this answer, then we could create the empirical frequency of the variable $TD = T_{v_j} - T_{A_j}$, measured in days.

We could take, say, the upvoting activity of one month, and form the empirical frequency distribution to a few intervals, say

% of Votes casted on January 2018 to answers that are no more than 30 days old
% of Votes casted on January 2018 to answers that are 31-90 days old
% of Votes casted on January 2018 to answers that are 91-120 days old

etc The meaningful length and number of such intervals would be decided by trial and error I guess.

Also one could consider forming this distribution not by considering the upvoting activity of one month but the upvoting activity of a quarter, etc. (this increases the sample size, but allows for fewer sequential distributions to be compared)

I don't expect that the proportions of votes to older questions will be large. But in any cases it is an indication about how much the CV-RoK is used. Also tracking this percentage over time, i.e by seeing whether it increases, decreases, or stays the same, would also give a sense of what happens with all this material.

Certainly this will be just an indication. People that found an old question useful didn't necessarily upvoted it (they may not even be CV users). Also, since an answer can be edited, there is an issue as to what timestamp should be used for the answer - the original submission date or the date of the latest edition?

But the main practical issue here is: are the timestamps required to perform this exercise available? Is there a record somewhere that provides the information "this upvote was cast on January 15th 2018 to an answer that was uploaded on July 10th 2012"? I am pretty certain that they are available to the SE company, but is it part of the user- or mod-accessible stats?

  • 3
    $\begingroup$ Have you tried data.stackexchange.com ? $\endgroup$
    – Tim Mod
    Aug 20, 2018 at 9:36
  • $\begingroup$ @Tim I am aware of this site but I am not very familiar with its workings. Let's see if someone that knows it better can tell whether the required data are available. $\endgroup$ Aug 20, 2018 at 9:46
  • 3
    $\begingroup$ I think there's several other indicators of 'repository-ness': 1. the number of searches that hit pages here (which I think only people at the highest rep level and moderators can see, but which you might perhaps be able to get by SEDE query, I don't know), and 2. The number of questions closed as duplicates. I know we do close a fair few as duplicates (though perhaps only a small proportion of the total questions) $\endgroup$
    – Glen_b
    Aug 20, 2018 at 9:56
  • $\begingroup$ @Glen_b Indeed, these are also indicators that can be used on the matter. $\endgroup$ Aug 20, 2018 at 10:22


You must log in to answer this question.

Browse other questions tagged .