Prioritising newer user is Pareto superior to prioritising more active ones

In my second estimation, a surprising result appeared. When the community manager is capacity-constrained, an allocation of capacity that prioritises newer users reduces inequalities. This is intuitive in the sense of the Kims. But an allocation that prioritises more active users does not increase either membership strength (mobilisation) nor total number of comments (activity). The capacity allocation rule "focus on new members" is Pareto-superior to "focus on the most active members".

Meanwhile, I noticed that the capacity constraint (100 comments per period) was only biting in relatively few cases. When the online community is still small, the community manager can easily engage with everybody who is active. It is only over the 400 members or so (and then only if the global chattiness parameter is high) that the constraint "bites". So, maybe the lax constraint eliminates the tradeoff between activity and inclusivity?

I launched a small scale simulation to explore this. I chose high values for global chattiness and intimacy strength, so as to get newer members to be much less active than older ones. Half of the model runs used the same capacity constraint of 100 comments per period; the other half had a much tighter capacity of 10 comments per period.

The dummy variable d_priority_newer takes value 1 if the allocation rule is "prioritise newer members", 0 if it is the alternative "prioritise more active members".

Results:

• corr d_priority_newer dropouts = -0.5106 No surprise: prioritising newer members in community management fights the exclusionary action of a high intimacy strength. The presence of the onboarding policy tends to smooth out the difference in performance between our two capacity allocation criteria: corr d_priority_newer dropouts if onboard == "false" = -0.6010

• corr d_priority_newer totalcomments = -0.0235 Surprise. This correlation is negative, as expected, but very weak. T-tests on the null that totalcomments | (priority = "newer") = totalcomments | (priority = "more active") does not reject the null hypothesis (N = 64). The means are almost identical.

• Even more surprising: when we discard the model runs in which the onboarding policy is active, the correlation becomes positive (but still small: 0.0236).

• T-tests confirm the intuition from the correlation coefficient.

  H0: mean (dropouts | priority == "newer") == mean (dropouts | priority == "more active") => REJECT
  H0: mean (totalcomments | priority == "newer") == mean (dropouts | priority == "more active") =>  DO NOT REJECT
  

• This last result is robust to different specifications (with and without onboarding, with and without randomised chattiness, at low and high capacity constraints, and the various combinations of these). The only exception is that H0 on dropouts is not rejected when (1) onboarding is present and (2) the capacity constraint is lax (100 comments per period). Only in that case, the presence of the onboarding policy does most of the heavy lifting in reducing the number of dropouts.

This page is based on these data.