Social Spikes at the Oscars: Placing Our Bets
Oh boy, it’s Oscar time! For me that means seeing every movie that I can before the big show, filling out my picks to win each category, and looking forward to watching social media react to the winners and losers during Sunday’s show. People around the world love to talk about all things about award shows - from pre-show through Monday morning quarterbacking for all the winners.
— Jonah Hill (@JonahHill) January 13, 2014
But today I’m not really going to talk about the films or the red carpet or the speeches.
Today, I want to talk about spikes.
A great thing about working at Mass Relevance is we have something going on with every large event out there. From the Super Bowl to the GRAMMYS to the Oscars and beyond, we’re powering social experiences for someone. And with large crowds participating, we always see huge social spikes when defining moments happens.
What can we learn about past patterns to help us predict what these spikes will look like at future events?
Really, It’s An Honor Just to Be Spiked
First, let’s define a spike. Twitter spikes occur when individuals across the world express their excitement, anger, or surprise to big events. They manifest themselves as sudden influxes in social media activity, typically triggered by:
- On-air calls-to-action (“vote now!”, etc.)
- Large, dramatic moments in events (Best Picture award announced, etc.)
- Surprises (touchdown in the big game, etc.)
Bad snap. False start. Penalty declined. SAFETY! Seahawks lead 2-0 after the first play from scrimmage. #SB48
— Seattle Seahawks (@Seahawks) February 2, 2014
In the past, we’ve analyzed spikes resulting from on-air CTAs at last year’s MTV Movie Awards, showing a 640% spike in social reaction when the audience was prompted to vote. We saw huge spikes as we powered American Idol’s #idolAgree effort. And of course, Beyonce.
But today we’re talking about the Oscars. Looking at the patterns of Twitter conversation levels during the 2013 Oscars, we see that not all spikes are created equal.
Each spike corresponds to the actor/actress/film winning their Oscar. As you can see, Argo's Best Picture spike was the one that ruled them all. Anne Hathaway’s spike just absolutely dominated Christoph Waltz’s social bump. And Daniel Day-Lewis, who can play just about anyone, didn’t even get close to the reaction that Jennifer Lawrence received.
So why are all these spikes different?
Let’s find out. We'll apply what we know about social spikes and use it to help predict spikes for this year’s Oscars winners. We’ll break it down by first picking apart the 2013 Oscar winner spikes, and then applying what we’ve learned to this year’s awards show. Who’s with me?
And the Most Influential Spike Factors in a Supporting Role Are...
There are a lot of elements that impact the size of a social spike. A few that we’ll investigate today are:
- Popularity: the more popular someone is on a social networks and larger their resulting spike. This makes sense: if more people talk about Jennifer Lawrence vs Daniel Day-Lewis on Twitter, then more of those people will Tweet “OMG” when she wins vs. when he wins.
- Surprise: larger social spikes happen when the audience doesn’t see the event coming. This would be more like “OMG!” vs “OMG” but you get the point.
So let’s figure out how to best account for popularity on Twitter. We could look at Twitter followers for each 2013 Oscar winner, but unfortunately not all of the nominees are on Twitter (and that approach wouldn’t work for Best Picture, which wouldn’t have a Twitter handle of it’s own).
Instead, let’s look at how often their names were mentioned the month before the Oscars to establish a baseline for level of conversation around each nominee on Twitter.
Now that we have a sense of “normal” for each term, we can compare the Twitter spikes when they received their awards and show it as a percent of normal mentions, like so:
Wow - roll out the red carpet for Ang Lee! His spike, which seemed small when just glancing at the raw numbers, is actually worthy of a standing ovation. Similar for Christoph Waltz.
<STATS> One method of figuring out if these two metrics move together is to calculate a correlation between the two data sets. The correlation coefficient measures the degree to which two sets of numbers move together. The highest correlation coefficient possible is 1, which would mean that two sets of numbers move together in lock-step. When comparing the average daily mentions vs. Oscar spikes for each winner above, the correlation between the two is .95, which is encouraging sign that these form a linear relationship.</STATS>
Ladies and Gentlemen: Place Your Social Bets
Now we need to factor in the element of surprise. The assumption here is that if someone won who wasn’t supposed to win, they’d get a larger social bump in mentions. But how do we measure...surprise? That’s tricky.
For that, we’ll need to bring in more data. And to get that data, it’s time to head to the place where I do my best thinking: Vegas, baby.
There’s two things everyone knows about Vegas:
- Doug is on the roof - just look on the roof.
- Vegas lets you bet on just about anything.
Every year, Vegas lays down odds for who they expect to win each Oscar. So what better way to gauge surprise than to use the same numbers as crazy people that put real money down? Let’s add some Vegas betting lines to our table and roll the dice:
So how do we read this? The lower the Vegas payout, the more they think that person will win the Oscar. Anne Hathaway and Daniel Day-Lewis were both front-runners to win the Oscar last year, so betting $1 on them would only return you your original bet plus 2 cents. That’s a lame return, but low risk. And when they won the Oscar, no one in Las Vegas was surprised.
Ang Lee was widely thought to lose to Steven Spielberg, so he was the most “surprising” win in the category and would have returned a cool $2.50 plus your original $1. Vegas is not big on surprises, and didn’t like this one bit.
So how do the Vegas odds correlate with last year's Twitter Oscar spikes?
The Vegas odds and Twitter Spike Ratio correlate with a .95 correlation coefficient. That’s pretty good, but remember this is a tiny data set, so it’s easily skewed by just one number (especially by Ang Lee’s big upset.) And of course, correlation does not equal causation, but we’ve got a basic level of validation for a linear relationship. We can also measure how closely the data fits to the line by using R-squared, which in this case is a respectable .91)
Just to test out if this thing works, let’s add in some more data. The same Twitter Spike Ratio and Vegas odds data, but we'll add data from this year’s Golden Globes.
Hmm, not as good. Adding in this year’s Golden Globe winners, we reduce the correlation to .60 (w R-Squared falling as well.)
What happened? Well, maybe the Golden Globes are different. Maybe a few outliers like Matthew McConaughey (who, if you remove from the data, brings the correlation back up above .70), skewed the data because his name is hard to spell when tweeting. Or maybe this theory is all junk. Who knows. But we’ve come this far, let’s take our .60 correlation and keep going.
To the Projection Booth
So based on what we’ve built so far, we should be able to predict Twitter spikes for Sunday’s big show based on 1) nominees Twitter chatter over the past month and 2) their Vegas odds for winning.
Predicting every single category would give us a huge list, so I’ll just pick a few. We’ll do Best Actor, Best Actress, and Best Supporting Actress.
The first thing I quickly realized is that the Oscar odds are pretty crazy for some actors and actresses. The largest payout winner we’ve seen in our data so far has been McConaughey, at 2.75 odds. Christian Bale is at 100/1 odds to win. So, we’ll need to scale the model a bit to adjust for the extreme surprise chances here. I won’t get into the details (it includes logarithms and today’s a Friday so I'm pretty sure you're not in the mood for that.)
Here’s the goods:
The big takeaway here is that if the Oscars want a lot of Tweets on Sunday night, they should hand the trophies over to Jennifer Lawrence, Leo, and Julia Roberts. All three are underdogs (which, based on this model, would help amplify the Twitter reaction) and all three get a lot of daily social conversation. Based on my numbers, it’s a perfect recipe for a spike.
On the low side, if Bruce Dern, Judi Dench, and Lupita Nyongo take the Oscars in the three categories I analyzed, the Twitter spikes should be less dramatic.
So be sure to watch the show sunday night and follow me on Twitter (at @chriskerns) where I’ll do my best to live-tweet the actual spikes vs. my predicted ones. I’m hoping the methodology we walked through here helped you understand a few methods of forecasting via data and maybe I’ll even call a few spikes correctly. But who knows what's going to happen -if someone shows up in a swan dress, we can all throw math right out the window