Methodology
Our sample includes data for views by day for 10,916 videos over a 90 day time period. In order to exclude casual creators of online video (i.e. "Mikey's Birthday"), each video in the sample achieved a minimum of 1,000 cumulative views over the 90 day time period. A statistician fit a curve to the data and projected out views by day for a full year. All data is from videos deployed by TubeMogul.Results
For the most part, the data speaks for itself, with the graphs clearly skewing right by all measures.- Below: a breakdown of views by day as a percentage of total views over 90 days. Significantly, 50% of all views occur in the first two weeks, peaking at day three, which constitutes 11% of all views.
- A scatterplot of total views by day for all 10,916 videos:
Average views per video per day over a 90 day time period:
One Year Forecasts
- Our approximated curve, compared to known data points (note day 36, in which one video "went viral," pushing up the data point):
A breakdown of views by day as a percentage of total views, with a full year forecasted:
A density function representing percentage of annual views by day:
Conclusions
On average, videos are time-sensitive. Trends pointed out elsewhere, such as "evergreen" (non-time sensitive) content always fetching views or videos randomly "going viral," seem more of a rarity than an underlying trend in the data.However, since we only projected out to one year, we effectively capped a long tail that over time might add up to a significant percentage of overall views. Even in our one-year forecast, it is interesting that after 154 days, a typical video still has 25% of its annual views left--hardly a "flash in a pan," although the long tail is declining in potency over time as the function approaches the x axis.
Works Cited/Thanks:
- Aaron Growitz, Statistician.
- TubeMogul's proprietary data and our partners.
