Did you know that Led Zeppelin has never won a Grammy (the Hall of Fame “participation” trophies they received notwithstanding)? Neither have legendary artists like The Who, The Doors, the Grateful Dead, or Bob Marley.
Or, did you know that in the 1988 Grammy Awards, classic rock band Jethro Tull beat out Metallica for “Heavy Metal Album of the Year.” The only thing “metal” about Jethro Tull was its flute.
In 1999, Shania Twain had the only country album deemed worthy of a nomination for Album of the Year. Did you know she still lost the Grammy for Country Album of the Year to the Dixie Chicks?
If anyone tells you they can predict the Grammy winners with supreme confidence, they’re frauds. The way the Grammy winners are selected — by a group of biased and often-irrational music industry insiders — makes predicting them something of a fool’s errand.
So, when Mark Cuban asked us to participate in a “Grammy Prediction Special” on his AXS TV network, pitting our polling technique against music industry insiders, not everyone on our team thought it was a great idea. Our head data scientist Ross wanted to kill me on more than one occasion for agreeing to do it. But, when a dude like Mark Cuban comes to you with an opportunity like that, you don’t say no.
Our technology is good at predicting “empirical” things like movie box office numbers, retail trends, or even the Michigan Consumer Sentiment Index. But a subjective event like the Grammy’s is a whole different level of difficulty.
Yes, the model we developed did pretty well last year (In five out of six Grammy categories to be exact). But that was our first and only try. It took us dozens of attempts to get movies right and almost two years to consistently nail the Michigan Index. What if our Grammy success last year, which spawned the model we used this year, was just a fluke?
We were also asked to predict a few categories that we didn’t study last year, like Best Rap Album and Best Group/Pop Duo. Our models in those categories are completely untested. And, thanks to Murphy’s Law, they also happen to be the two most competitive categories this year.
Predicting the votes of music industry veterans — people with grudges, friendships, financial ties, and social agendas — is effing hard. Who the voters are and how they base their decisions is a mystery. As I sat on the AXS TV panel , my counterparts, famed music critic, Bob Lefesetz, and renowned DJ, DJ SKEE, forgot more about the music industry during the first commercial break than I’ve learned in my entire life. Bob probably knows all of the Recording Academy voters personally — his experience and track record is unmatched. There’s every reason to believe that predicting the Grammy winners is one of those few remaining frontiers where inside-knowledge and instinct will trump data science.
That makes the challenge all the more fun. After all, we’re not doing pediatric oncology around here. If we trip over our “instruments” Sunday night, we’ll refine our models and try again next year.
HOW DID WE DO IT?
Before getting into our predictions, let me explain how we formed them. If you know our business, then you know that we are able to poll people inside the content of hundreds of web and mobile sites across the country every day. Peppered among the thousands of different poll questions we ask were questions about the nine (*) Grammy categories we studied. Respondents would randomly encounter one of those questions during their normal online travels.
(*) The Grammy Prediction Special only included 8 of our 9 predictions for time reasons. We share the ninth prediction below.
People were not asked who they wanted to win the categories; rather, they were asked to predict who they expected to win. That’s an important, if seemingly-nuanced point. We are asking people to evaluate all of the things they hear, read, AND feel in formulating their vote, not just their personal preference.
For the Grammys, we also created a stand-alone web page, where respondents could answer all nine questions in one sitting. We drove traffic to that site through prompts at the end of our third-party polling widgets, as well as through social media and PR activities. In all, from the day the Grammy nominees were announced to the date the show aired on AXS TV, we tallied exactly 187,345 votes. From past polls voters had answered, we also knew a lot about their demographics and other characteristics.
If this was a conventional public opinion poll, we would have simply tallied up those 187k votes, reweighted them to reflect the US adult population, and published the numbers. But the Grammys are not a popularity contest. More importantly, not every poll respondent is equally likely to predict the winners in the various categories. Unlike a political election, when it comes to predicting something like the Grammys, all votes are NOT created equal.
To figure out which votes (and voters) should be trusted more than others, we went back to last year’s Grammys, where we collected over 50,000 responses to questions about six Grammy Categories. Now knowing the winners, we could examine which among those 50,000 votes were right and which ones were wrong. We also put more value on some past predictions than others. For example, someone who predicted the artist Fun. to win Best New Artist last year was a valuable voter because that was such a divided category. Meanwhile, someone who picked Adele for Best Pop Solo Performance didn’t score as highly in our models because a large majority of respondents predicted she’d win.
We found that there were clear commonalities among the accurate predictors in each category. Naturally, an avid follower of music news and trends performed better than someone who wasn’t. In most categories, respondents under age 35 did much better than people over 35. In other categories, people under 30 were the only ones that mattered. In Rock-related categories, the good predictors were more likely to be men. In pop categories, they were women. You get the idea. We also examined metadata for each of the questions, such as when and where they were answered.
As we compiled this year’s predictions, we valued votes differently based on what we knew about the voter. In some cases, we may even have encountered the very same person who voted accurately last year and came back again. In the rock categories we had one ideal age and gender distribution. In the Best Pop Group/Duo category we had another. For the Rap category, we tweaked the distribution of votes by race. In the Country Album category we saw correlations with the regions where voters lived. The end goal was to find a manicured group from among those 187,000+ votes that gave us our best chance of predicting a specific category.
So, what did we conclude?
ALBUM OF THE YEAR
Prediction: The Heist by Macklemore and Ryan Lewis
Confidence Level: High
Why It Should Win: The most astute music fans and last year’s good predictors, especially those under age 30, sided with Macklemore and Ryan Lewis overwhelmingly. If we trust our models, this one should be a slam dunk.
Why We Could Miss It: There is diverse pool of nominees in this category. You could see Macklemore/Ryan Lewis splitting votes with someone like Kendrick Lamar and Daft Punk and leaving a broadly popular artist like Taylor Swift standing.
SONG OF THE YEAR
Prediction: “Royals” by Lorde
Confidence Level: Fairly High
Why It Should Win: Lorde’s support across the board was very high among music enthusiasts and particularly everyone under age 35. At face value, this one should be a runaway winner.
Why We Could Miss It: The Recording Academy has a history of using the awards to make political statements. And there is no hotter topic among the left-leaning types prevalent in the music industry than LGBT rights, the subject of Macklemore and Ryan Lewis’s “Same Love.” Among our voters, people who support the legalization of gay marriage were 148% more likely to pick the Macklemore/Ryan Lewis tune. Politics is a wild card in this one.
RECORD OF THE YEAR
Prediction: “Royals” by Lorde
Confidence Level: Confused
Why It Should Win: Similar to the Song of the Year category, Lorde draws a lot of support from predictors across the demographic spectrum. Moreover, it is extremely uncommon (and illogical) for a Song of the Year winner to lose Record of the Year honors to a nominee NOT even nominated for Song of the Year. That would be the case if other category favorites “Get Lucky” by Daft Punk or “Blurred Lines” by Robin Thicke take the prize.
Why We Could Miss It: No category had our data science team pulling out their hair like this one. It’s extremely competitive and close. The Vegas odds seem to favor Daft Punk, yet, no matter how we sliced and diced our numbers, we didn’t show Daft Punk near the top. If we miss this one, we have some work to do for next year.
BEST NEW ARTIST
Prediction: Macklemore and Ryan Lewis
Confidence Level: The Highest
Why They Should Win: This one of the biggest blow-out categories this year. Macklemore and Ryan Lewis win across the board and nobody else is even close.
Why We Could Miss It: If our performance last year was a complete aberration and our models are based on nothing but anomaly, then maybe someone else steals this award. Otherwise, it’s the closest thing to a shoe-in we found this year.
BEST RAP ALBUM
Prediction: The Heist by Macklemore and Ryan Lewis
Confidence Level: Mixed
Why They Should Win: They win across all of our key predictor segments, particularly the under-35 crowd. Moreover, if we are confident they will win Album of the Year (and we are), how could they not win their own genre category? In all eight previous occasions where a rap album was nominated for Album of the Year, it took home the Best Rap Album award that year. Anything to the contrary would simply expose a fundamental flaw in the process.
Why We Could Miss It: We’ve never studied the Best Rap Album category, so we have nothing to test our model against. We did see higher levels of support for Kendrick Lamar among African-American predictors and for Jay-Z among the general population. Still, it was not enough to overcome Macklemore/Lewis.
BEST COUNTRY ALBUM
Prediction: Red by Taylor Swift
Confidence Level: Very High
Why They Should Win: This category rivals Best New Artist for “runaway” status. Swift wins among every group we tested, with a few geographic exceptions. Blake Shelton did well in a few states in the south and west. Again, this is an example of an Album of the Year candidate being the only nominee in its genre category. How does that not make Red a lock unless the process is messed up?
Why We Could Miss It: Blake Shelton performed fairly well among the most avid country music fans in our predictor group. Conceivably, given that Swift has drifted more into the pop/mainstream, the Grammy voters could make this a protest vote in favor of an artist deemed more authentically “country.”
BEST ROCK ALBUM
Prediction: Mechanical Bull by Kings of Leon
Confidence Level: Low
Why They Should Win: This category is split between two contemporary artists and four classic rock artists. Our numbers show support among older predictors split fairly evenly among David Bowie, Neil Young, Black Sabbath, and Led Zeppelin. Of the two contemporary bands, Kings of Leon out-performs Queens of the Stone Age handily, leaving them with the largest percentage of the votes in the category.
Why We Could Miss It: This is the one category we missed last year, so perhaps we just haven’t figured it out yet. It’s also easy to find precedent where the Academy voters choose a nostalgic nominee, like Led Zeppelin who has never won a Grammy, as something of a Lifetime Achievement Award.
BEST POP DUO/GROUP PERFORMANCE
Prediction: Robin Thicke, TI, & Pharrell Williams
Confidence Level: The Lowest
Why They Should Win: Because somebody has to. In the best model we could develop for this category, Thicke & Associates won by less than ½ of a percentage point. We’re little more than lucky if we get this one right.
Why We Could Miss It: Ugh. It doesn’t help that we never studied this category before. It’s worse that there are so many viable candidates in the mix. Jay-Z and Justin Timberlake are about as popular as it gets. Daft Punk is dead-even with Thicke & Company among the coveted Under-30 music enthusiasts. Flip a coin (preferably a three-sided one).
BEST ROCK PERFORMANCE
Prediction: “Radioactive” by Imagine Dragons
Confidence Level: High
Why They Should Win: This category was left off the AXS TV show both because of time but it looks like a pretty safe bet. Imagine Dragons wins among just about every group “imaginable.”
Why We Could Miss It: This might be Led Zeppelin’s best chance for a pseudo-Lifetime Achievement Award nod. A significantly older Recording Academy voting group could throw these numbers out of whack. But, it’s unlikely.
Well, there you have it. Who knows how the unpredictable and oft-irrational group of Grammy voters will surprise us on Sunday night? All we’re saying is that we have a pretty good idea of who should win, if logic, historical data, and thousands of informed and insightful music predictors are any indication.
Enjoy the big night.