Research:Optimizely Fundraiser Analysis
This page documents a planned research project.
Information may be incomplete and change before the project starts.
Key Personnel
- Ryan Faulkner
- Dan Siroker
Project Summary
Using Optimizely software to analyze WMF Fundraiser data. Dan Siroker expressed some interest to analyze WMF fundraiser data. This could include de-deuping the data and performing A/B testing of visitors falling into different sub-groups (e.g. number of banners seen, types of banners seen etc.)
Methods
Data is to be prepared for general consumption by researchers with IP information anonymized (implementation is being discussed with Roan Kattouw). The WMF will publish this data on the data dumps server (dumps.wikimedia.org). The data will be sourced from 1) WMF squid request logs stored and processed on locke.wikimedia.org and 2) donations data retained in our CiviCRM and drupal databases on db1025.eqiad.wmnet. This data is to be processed into the following three forms based on table schemas that define the anonymized data:
Banner Impressions - ip_hash, banner, article_hash, browser, country, language, request_time
Landing Page Impressions - ip_hash, banner, utm_campaign, utm_medium, landing_page, article_hash, browser, country, language, project, request_time
Donations - ip_hash, banner, utm_campaign, utm_medium, landing_page, donated amount, timestamp
Specific methods to be used by Dan Siroker have not yet been determined however, the software that may be used to analyze WMF Fundraiser data may be found at the Optimizely Homepage.
Dissemination
Wikimedia Policies, Ethics, and Human Subjects Protection
All data published to dumps.wikimedia.org will be stripped of any fields that may be used in personally identifying any donors or readers that are the source point of the server impressions and donation records held by the WMF.
Benefits for the Wikimedia community
Potential Analysis that will aid in making the WMF annual fundraiser more efficient. This could include:
- Improvement of A/B testing methods to increase efficiency of campaign testing
- Sensitivity of conversion metrics to duplicate requests for visitors
- Learning about donor and reader behaviours based on user-experience (banner views, types of articles visited etc.)
Time Line
Funding
This is a volunteer project. Resources for pre-processing the data for public consumption will be provided internally by WMF staff accounted for under the 2010/2011 annual budget.
References
External links
Contacts
Dan Siroker - dan@optimizely.com
Ryan Faulkner - rfaulkner@wikimedia.org