Data Science Proves We Love to Hate-Watch TV

gameofthrones - ramsay bolton
Courtesy of HBO

Study by startup Canvs finds 'hate' is the emotion most strongly associated with TV tune-in

One of the true pleasures of HBO’s “Game of Thrones” the last few seasons has been seething at the outrages committed by oleaginous psychopath Ramsay Bolton. And now, thanks to Big Data, it’s clear that TV viewers are highly motivated to watch a show if they hate something about it.

For dramas and reality shows, various expressions of “hate” by viewers were the strongest emotional indicator that TV show viewership will increase for the following episode, according to a study by data-analytics startup Canvs. The company looked at nearly 6,000 episodes of recent cable and broadcast shows, cross-referencing Nielsen ratings with Twitter comments associated with each show categorized into one of 56 emotional buckets (like “love,” “hate,” “annoying,” “beautiful” and “boring”).

A key finding: For every percentage point increase in “hate” responses, there’s a 0.7% increase in viewership for the next segment of a show in reality and drama genres. That’s more than double the power of “love,” for which a 1% increase corresponded to a 0.3% viewership boost for dramas and 0.2% uptick for reality shows.


The Walking Dead Season 6 Finale

‘The Walking Dead’ Season 6 Finale Draws Social-Media Hate

“If you can’t stand a Kardashian, you are more likely to watch the show next week,” said Canvs founder and CEO Jared Feldman. “‘Hate’ is largely seen as negative — every network wants to reduce their negative scores on social media — but this study flips that logic on its head. You actually need ‘hate.'”

Of course, this isn’t really a shock: Villains have been integral to human storytelling for thousands of years. But the study shows the specific magnitude that feelings of hate have in drawing an audience into a plot.

For comedies, rather than “funny,” Canvs found “love” and “beautiful” to be the biggest emotional indicators of whether viewership will increase for the next episode. For every 1% increase in “love,” there’s a 0.1% increase in viewership the next episode — but a 1% increase in “beautiful,” yields a 0.3% increase in viewership the next episode.

“On comedies we were very surprised ‘funny’ wasn’t there. But it’s par for the course. It’s part of every comedy,” Feldman said. Interestingly, the Canvs study found a 1% rise in “funny” reactions for drama series actually produced a 0.3% viewing increase.

Feldman suggested that TV producers will be able to use the emotional-response data to glean insights into which storylines and characters achieve maximum viewer lift, or let networks identify elements to feature in on-air promos and social-media tune-in campaigns.

For the study, Canvs analyzed 5,709 episodes of 431 comedy, reality and drama series that aired between January 2014 and June 2015. Of those, 29% were new shows and 71% were returning shows. Broadcast networks comprised 40% of the episodes studied, while basic cable made up 55% and premium cablers like HBO and Showtime represented 5%. Because the study relied on Nielsen data, it excluded shows on Internet platforms like Netflix, Hulu and Amazon Prime Video (which are not rated by Nielsen).

The predictions aren’t guarantees that TV viewership will go up or down — Feldman compares the data analysis to a weather forecast. But Canvs’ predictive methodology was accurate to within three percentage points over the time period surveyed, according to the company. Some in the industry may question the accuracy of the underlying Nielsen TV panel-based ratings (which have been beset by glitches in the past) but whatever their shortcomings they remain the principal measure of television audiences today. Asked about the statistical validity of the study, Canvs chief scientist Sam Hui said in a statement, “The sampling error for Nielsen is non-systematic and random — they are already incorporated in the error terms in our model, while the effects that we identified are systematic and predictable.”


TV digital advertising spending

TV Shows That Pack Emotional Punch Have Higher Ad Recall: Twitter Study

Canvs uses thousands of different words, phrases, emoji and slang terms to identify emotional responses to TV on social media; in fact, 65% of those are not words found in a dictionary.

Using the same methodology, the company is launching a product called Canvs Viewership Probability, which will provide scores estimating whether viewership will go up or down for a given show’s next episode. Initially, the company will make the CVP research available to TV execs for their own networks, but it plans to roll out syndicated research in the next few months (which Feldman said will be more of interest to ad agencies). In addition, Canvs is working with Facebook directly to compile research for TV viewing based on its data.

Feldman founded Canvs using social-media analytics research from Hui, previously a professor at NYU’s Stern School of Business and currently an associate professor at the University of Houston’s Bauer College of Business. The New York-based company recently hired Justin Fromm, previously with Hulu and ABC, as director of research.

Canvs customers include Viacom, Sony Pictures Television, NBCUniversal, CAA and UTA. Investors include KEC Ventures, Rubicon Ventures, BRaVe Ventures, Social Starts and Milestone Venture Partners.

Filed Under:

Want to read more articles like this one? SUBSCRIBE TO VARIETY TODAY.
Post A Comment 4

Leave a Reply


Comments are moderated. They may be edited for clarity and reprinting in whole or in part in Variety publications.

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

  1. >Feldman suggested that TV producers will be able to use the emotional-response data to glean insights into which storylines and characters achieve maximum viewer lift, or let networks identify elements to feature in on-air promos and social-media tune-in campaigns.

    God, this is depressing…I think that writers and creators already have enough of an uphill battle trying to make compelling shows with characters people want to watch each week, because of the constant meddling from network execs who seem more concerned with ratings and ad dollars than on producing groundbreaking television with quality content each week. Giving them (the network suits) this kind of analytical tool is only going to make it harder for show creators and their writers to get or keep anything on the air that could be seen at all as potentially odd or challenging to viewers, which means that in the future, we can likely look forward to more cookie cutter show clones that cater to improving their “Viewer Probability” ratings, rather than aiming for creativity, originality, character development and an interesting story arch. Who has time for logical plot development in this day and age?!

    Imagine someone trying to pitch something like Twin Peaks, Six Feet Under or even Seinfeld if these kinds of tools existed or held as much weight when those shows were on as they do today! I doubt any one of them would ever have made it past the pilot stage, let alone been given enough time to become critical faves or the water-cooler shows of their day, if they had been expected to shape their scripts and cast roles based on feedback from analytical reports quantifying the influence of seemingly random details of a show on an audience’s viewing habits. Or having that show’s future hang in the balance because of a low “viewership probability” feedback report that could, possibly, under the right set of circumstances, potentially predict, one day, down the road, in the near or far or unimaginable future, the chance or maybe even the likelihood, that a viewer may watch or stop watching a television show forever because data from the pilot revealed that women between 18 and 45 don’t like men named David to be played by actors named John who star in shows airing on Wednesdays in the fall. Even though that data might be as reliable as the weather channel.

    I could just see Larry David being told by the NBC execs that there needs to be more “love” between George and Elaine and that Elaine should be referred to by Jerry as “beautiful” in a script at least three times per episode because studies show that viewers are more likely to watch during the commercial breaks and not go take the dump they were thinking about taking if a show’s leading man states that he thinks the female lead is “beautiful” every 7.6 minutes per episode. The funny thing is, Seinfeld’s show kind of introduced the dawn of the meddling network execs, but even though they were portrayed as Clueless Corporate Meddlers, at least the PTB let Seinfeld have enough time to find its audience. These days, if a show doesn’t immediately attract viewers, it’s pulled from the schedule faster than you can say “Newman!”

    Is it any wonder that there is rarely a time anywhere in the world where you can’t find an episode of Law and Order on t.v. these days? How many sick variations of that fucking show do we seriously need? I’m so tired of that by-the-numbers procedural police franchise crap hogging all my channels.

    You know what could be interesting, though? How about CSI: Minnesota-/Emotional Indicator Data-Analytics Emergency Division? Or CSI: MEIDEA Division.

    Ooooh, look….ME IDEA!

    Or if you drop the word “division” and just use the D, it’s CSI: ME I DEAD!!

    I have a feeling the data will predict that audiences will prefer that last version. Me Ideas are SOOOO last century.

    • Brick Tamland says:

      Well, the end product doesn’t instill much faith if the company is comparing the reliability of the results to a weather forecast.

      Seems like non-quants are preying on fellow non-quants under the guise of insights. In this case, people love to hate watch. But who knows if that’s actually true — the technology can’t say that confidently even though it’s stat sig. (So what’s p value or CI look like?)

      Not sure how this is remotely actionable or valuable to anyone, but I guess I’ll pack an umbrella in case it rains ‘funny’…

  2. Healthy Skepticism says:

    How stat sig are the results? Those percentages are so small one could argue randomness here. Disappointed that “this” is being shrouded as data science….

    • Todd Spangler says:

      Canvs says its data set is large enough to be statistically significant – as noted in the story, the company compares these predictions to weather forecasts.

More Digital News from Variety