Countering underproduction of peer produced goods

TL;DR

This study addresses underproduction in commons-based peer production by analyzing the English Wikipedia revision history to test how contributor experience and account use relate to task selection. It uses two samples—a cross-sectional revision sample (192,672 revisions) and a within-person history sample (42,602,912 revisions)—and measures underproduction with the factor $f = -\log(quality\_rank / popularity\_rank)$, alongside ORES quality and monthly views, with experience captured by revision count and a binary IP-based indicator for non-account edits. The results show that greater experience is associated with edits to underproduced content, displaying a $U$-shaped pattern with a notable inflection around $150$ revisions, while anonymous contributions do not consistently outpace account-based edits; within-person analyses confirm a shift toward underproduced goods as editors gain experience, consistent with a technosocial learning mechanism. The findings underscore the value of retaining contributors, including those who edit without accounts, as a lever against underproduction, but also highlight tradeoffs related to privacy, equity, and governance in retention strategies.

Abstract

Peer produced goods such as online knowledge bases and free/libre open source software rely on contributors who often choose their tasks regardless of consumer needs. These goods are susceptible to underproduction: when popular goods are relatively low quality. Although underproduction is a common feature of peer production, very little is known about how to counteract it. We use a detailed longitudinal dataset from English Wikipedia to show that more experienced contributors -- including those who contribute without an account -- tend to contribute to underproduced goods. A within-person analysis shows that contributors' efforts shift toward underproduced goods over time. These findings illustrate the value of retaining contributors in peer production, including those contributing without accounts, as a means to counter underproduction.

Figures (3)

• Figure 1: The marginal effect of having higher experience on the average alignment of an article selected for editing. Increasing values indicate increased levels of underproduction, i.e. low-quality but highly-viewed topics.
• Figure 2: The left pane shows a GAM smoothed line fit to a 10% sample of the data from the random sample with the result of the linear model superimposed as a dashed line; the right pane shows the same GAM line with the result of the polynomial model superimposed as a dotted line. Note the log-scaled $x$-axis.
• Figure 3: Marginal effect of increased experience on contributor task selection from our within-person sample. We use median individual-level fixed effects. Dashed lines are predicted values using a linear model, while dotted lines use a quadratic model, see Table \ref{['tab:withinAlign']}.