The social skunkworks

Blog Comments (0)

As the social web continues to expand in reach and influence, we expect social metrics to have an ever-bigger role in allowing us to anticipate – and, by extension, influence – audience tastes and expectations.

By systematising the forces existing around the US box office, for example, can we begin to better define and quantify this role?

This is the question we set out to try and answer back in early 2011.

We collected assorted data ahead of the domestic openings of 135 movies – representing a total of over $3bn dollars in opening weekend ticket sales – and used an algorithm we’ve developed to project each movie’s 3-day weekend.

The graph below (click to enlarge) visualises the results. It shows actual performance (blue bars) against our projections (red bars) across all 52 weeks and 135 movies.

In order to benchmark our results we collected weekly predictions from four industry sources:, Box Office Mojo, ScreenCrave and Box Office Guru.

(We describe our algorithm’s output as ‘projections’ because they are generated using entirely quantitative methods, as opposed to being what we would term a ‘prediction’, which we understand as having a qualitative or anecdotal aspect.)

The results of our benchmarking were as follows:

Title Sample size* Avg. deviation 91 $4.7m
Box Office Mojo 104 $5.8m
ScreenCrave 45 $6.4m
Box Office Guru 31 $6.9m
* On some occasion predictions were unavailable for certain sources, resulting in incomplete sample groups.

At first glance $7.7m may seem like a large average deviation, and $3m a substantial difference in relation to the’s $4.7m average.

A quick glance at the data, however, indicates that our average deviation is substantially affected by a small number of aberrations.

What if we remove the ten movies for which our projection was most inaccurate?

The average deviation drops to just $5.4m, bringing us much more into line with the anecdotal averages.

We contend that the key to developing a system capable of projecting box office more accurately than industry analysts are able to qualitatively predict it lies in addressing specific considerations raised by this handful of movies.

The aim of this post is to invite anybody with an interest in being part of the next phase of this project to get in touch. We’re particularly interested in hearing from:

  • Partners studio-side and across the wider industry who may be able to provide insight into what’s happening ‘above the line’.
  • Other data partners, in particular owners of social platforms and/or analytics tools with a potential applicability to gathering social data around movie releases.
  • Experts in data analysis, specifically regression analysis, able to help us take a closer look at our extended data-set.

If you are any of the above – or have an interest of your own – please get in touch.

» Blog » The social skunkworks
On June 20, 2012

Comments are closed.

« »