I. Research AbstractThis report discusses the findings of an experiment designed to test which actions trigger a "view" on eight popular video sharing websites.
II. MethodologyFive scenarios were tested in this experiment:
1. Full View – watching a video from start to finish multiple times
2. Half View – watching a video more than half way (but not to completion) multiple times
3. Refresh – refreshing the browser after the video has begun playing
4. Embedded – watching a video multiple times through a player embedded on another site
5. Embedded Autoplay – watching a video that is coded to start automatically when the page loads through a player embedded in another site
All tests were run from a static IP address on a single computer between 9/12/10 and 9/17/10 from an account different from the video author’s for each site. When available, privacy settings where enabled to prevent third parties from viewing the video and skewing the data. The video was a brief clip of a wall. All titles, descriptions, and tags were labeled with a nonsense word – “fldsfkj” – to prevent accidental viewing through a search.
The methodology employed in this study is a starting point and by no means an exhaustive list of testable scenarios.
III. Summary of Findings
|Site||Full View||>1/2 View||Refresh||Embedded||Embedded Autoplay|
Site Specificsblip.tv – Still the most stringent sites for counting views, blip.tv limits view counts to one per IP address per session. Views for videos embedded into another site (i.e.: Embed (>1/2 view) and Autoplay) were also limited to one per session. If videos were watched again after leaving the site and some time had elapsed, it would count as an additional view.
Dailymotion – Video was marked “private” except when in use for the study. The privacy boxes were unchecked for “comments”, “group”, and “feature on profile” to reduce the chance of someone stumbling on the video by accident. Views in all categories were counted. In the last study, the site did not acknowledge refreshes or embedded videos in view counts.
Metacafe – The video was live during the entire course of this experiment, as no option existed to make it private. Metacafe explained that it is a video entertainment site meant to showcase videos with mass appeal to broader audiences, not just personal sharing with a few people, therefore privacy settings are not available. However, no third party influence was apparent in the results. Despite their formerly restrictive standards (in the last study, only one count/IP address), all views in the full, half, refresh, embed, and autoplay categories were counted, including multiple viewings. No views were counted for cases where the page loaded but not the video.
MySpace – Video was marked “private” when not actively in use to prevent outsider viewer influence. Myspace counts all views except in the case where the page loads but not the video. While views on the videos embedded into another site were counted, it was seemingly erratic at times. Holding the length of time the video was viewed constant, the count fluctuated between accurate and less than the actual views. There was no discernable pattern as to how MySpace counted a view for these types of videos.
Viddler – The permission to view and embed the video was marked “just me” except during active use for this experiment. All views in the full, half, and refresh categories were counted and displayed as “views” on the site. Viddler also tracks cases where the page a video is embedded on is loaded, but the user doesn't click to play the video itself ("impressions"). This figure was kept separate from views performed on the Viddler site, and revealed in the “stats” tab under the video itself.
Vimeo – This video was marked private when not in use for the experiment. Permission to link to or embed the video was also off during this time. Full, half, and refresh categories were limited to one count per IP address. Although what counted as a view on the Vimeo site was restricted, it did not apply to views of the videos embedded on another site. Even multiple viewings were included, although this number was separated into an “embedded stats” tab. Also under that tab were links to the sites which the videos were embedded.
Yahoo! Video – No option was given to mark the video “private”, but no apparent outside viewers influenced the experiment. Yahoo! counted all views except in the case where the page loads but not the video.
YouTube – Video was marked “private” except during use for the experiment where it was marked “unlisted”, a feature that only allows users with the URL address access to the video. All categories except for autoplay were counted. Repeated views were counted in those scenarios.
Most Stringent Sites for Counting Views
|Yahoo! Video||less stringent|
AnalysisThe overall trend appears to be loosening formerly-restrictive standards on what counts as a "view."
YouTube is the only site to not count embedded auto-play views, a restriction they put in place back in 2008 after a group of Avril Lavigne fans tried to use embedded auto-plays to propel the artist to YouTube's all time most viewed video. Surprisingly, no other video site followed suit after the prominent incident.
Metacafe once used an IP-based restriction, but multiple views from the same computer now qualify as discrete counts. In the case of blip.tv, IP-based restrictions have been standardized to include embedded videos. Dailymotion, which did not previously count embedded video views, now includes both embedded videos and refreshes in view numbers.
In the interests of full disclosure, TubeMogul powers portions of the video analytics available to blip, Dailymotion and Viddler users. However, each of these sites uses their own methodology for what counts as a "view" -- Tubemogul's numbers tend to be used for engagement metrics like viewer attention span.
ConclusionFor better or worse, publishers and advertisers often measure success of online video initiatives in terms of "views." While the industry has moved closer toward a standardized metric of what a "view" actually constitutes, it remains relatively unclear.
Similar to how "time-spent on site" can be a more useful metric than just "visits" in Web traffic measurement, more transparency and detail in video metrics would likely give publishers and advertisers greater opportunities for growth. Luckily, the technology to do this is readily available, and rapidly expanding in terms of adoption.