Research:Exploring Wikimedia Donation Patterns

From Meta, a Wikimedia project coordination wiki
16:05, 2 June 2016 (UTC)
David McDonald
Duration:  2016-June – 2018-April
This page documents a completed research project.

Every year, Wikimedia Foundation relies on fundraising campaigns to help maintain the services it provides to millions of users worldwide. However, despite many individuals who donate through these campaigns, these donors represent only a small percentage of Wikimedia users. Improving our understanding of donors and their behaviors too may result in more effective fundraising campaigns and limiting the burden of these campaigns on Wikipedia users.

Research Showcase Presentation Slides

Recent research suggests that different types of people may be drawn to different articles due to their personal reading interests. In other words, the articles themselves may provide some indicators about whom the users are and their interests. It may be possible, then, that (1) that fundraising campaigns on certain pages may be more successful (i.e., higher donation rate) due to the readers of the page and that (2) certain page/article properties can be used to predict the donation rate on the page. In this work, we, therefore, seek to explore the following research question:

  • RQ: Do donations across Wikipedia pages vary in some systematic ways?

In relation to this question, we also investigate specific hypotheses based on measurable features of the pages:

  • H1: Pages on more task-oriented topics attract more donations.
  • H2: Pages on which users spent more time attract more donations.
  • H3: Pages of higher quality attract more donations.

Insights gained can allow us to better anticipate donation behaviors on individual pages, and offer more interesting and effective banner messages.


Our goal is to explore whether and how certain features of the Wikipedia page, such as the topics, page quality, and the time users spend on the page, predict the donation rates. Using page-level aggregated donation data from the 2015 fundraising campaign for the French and English Wikipedias, we first collected the article content and meta-information for each page. To be able to model the impact of contents, we then categorized the articles into task-oriented and non-task oriented topics. We also collected the quality category for each article using the Objective Revision Evaluation Service (ORES). Finally, we collected the median and average time users spent on each page as a measure of engagement with the contents. Based on these article features, we developed a number of models to predict the rates of donation on those pages.


  • June 2016: Collect French Wikipedia content
  • March 2017: Collect English Wikipedia content
  • June 2017: Collect additional features - quality ratings, topic categories
  • September 2017: Explore regression models with collected articles features to predict donation rates
  • April 2018: Summarize findings in a report/publication

Policy, Ethics and Human Subjects Research[edit]

In our analysis, we have access to and use only the donation data aggregated per page, hence preserving the anonymity of the individual Wiki users.


Our results suggest the existence of reciprocity mechanism in donations, which leads users to be more likely to donate on pages that provide more value to them. This is indicated by users being more likely to donate on pages on task-oriented topics, pages with higher quality, and pages on which they spend more time. Our findings lead to practical design implications, offering design interventions that can potentially increase donation rates. One such intervention is reinforcing indebtedness by highlighting the potential of non-task oriented pages to provide task-oriented utility, or by highlighting quality on task-oriented pages. Another potential design intervention amounts to triggering anticipated reciprocity by highlighting the future potential of currently low quality, underdeveloped articles.

Detailed Results Presentation: Reciprocity & Donation - Research Showcase Video


  • Gary Hsieh, Jilin Chen, Jalal U. Mahmud, and Jeffrey Nichols. 2014. You read what you value: understanding personal values and reading interests. In Proceedings of the SIGCHI Conference on Human Factors in Computing Systems (CHI '14). ACM, New York, NY, USA, 983-986. DOI=
  • Rafal Kocienik, Os Keyes, Jonathan T. Morgan, Dario Taraborelli, David W. McDonald, and Gary Hsieh. 2018. Reciprocity and Donation: How Article Topic, Quality and Dwell Time Predicts Donation on Wikipedia. In Proceedings of the ACM on Human-Computer Interaction, Vol. 2, CSCW, Article 091 (November 2018). ACM, New York, NY. 19 pages.