Friday, April 19, 2013

2013 NBA Playoffs Predictions

It's that time of year again! It is what an entire season has lead to, the famous "win or go home" period.
We all just love the Playoffs, when superstars are made and forgotten, when every single match-up has its own little history behind it, when teams without any hard feelings against each other a Game 1 tip-off end up hating each other by Game 3. Unparalleled intensity, game after game!

This year's Playoffs trigger mixed feelings, as it seems the Champions have already been elected. Can anybody beat Miami? Can anybody make it at least somewhat of a challenge for them?

But even so, there are bound to be plenty of upsets we thrive on,  and of course we will all have eyes on the Howard-led Lakers. They can't afford to go through the Playoffs the same way they roller-coasted through the season, but if they continue on their current trend they can be a potential threat.

So what are my predictions? Well, based on season performance and home/away results, as well as taking each team's specific homecourt advantage into account, here's what I obtained for the most likely outcome at each phase with associated probability:

So it does seem that Miami has the clearest path to the trophy, but there no guarantees. And if we look at each team probability of wining the championship, we obtain the following table:

NBA team Champion Probability
Miami Heat 22.8%
Oklahoma City Thunder 13.1%
San Antonio Spurs 11.3%
New York Knicks 9.5%
Denver Nuggets 7.9%
Los Angeles Clippers 7.2%
Memphis Grizzlies 6.5%
Brooklyn Nets 4.8%
Indiana Pacers 4.8%
Chicago Bulls 2.8%
Golden State Warriors 2.5%
Atlanta Hawks 2.4%
Los Angeles Lakers 1.5%
Houston Rockets 1.3%
Boston Celtics 1.0%
Milwaukee Bucks 0.8%

While the table confirms that Miami has the greatest probability of winning the trophy, the probability is still less than 25%, offering challengers a great opportunity. The probability is still incredibly high considering it is almost as big as the sum of probabilities for the two next contenders (Oklahoma and San Antonio). This is in part due to Miami definitely enjoying a huge Conference advantage battling weaker foes, and if it falls it will most likely be in the Finals. An extra dimension favoring the Heat to consider and not (currently) taken into account in my model is wear-and-tear: the Heat will be more rested and less likely to be injured when they meet the Western Conference finalist...

I will regularly post updates as more games get played.

Who's searching for Michael Jordan?

In a previous post we analyzed the trend and seasonality in Google searches for Kobe Bryant. We then looked at the evolution of queries for Lebron James, Derrick Rose, Steve Nash. But what about the greatest NBA legend?

Here's the raw data from Google Trends for Michael Jordan, strating in 2004 despite his last game being in 2003:

Using the same decomposition analysis detailed in the Kobe Bryant post, we get the following results:

For a retired NBA player, Michael isn't doing too badly!

From the seasonality it would seem that he continues to have the offseason / regular season / Playoff pattern, but a closer look reveals the spikes are in February! The only explanation I have is that it is his birthday and being Michael Jordan people tend to hear about it and follow-up on a search, perhaps to refresh his memory at how good a player he was.

Since 2009, searches for him have actually steadily increased, which basically correspond to the big spike when he was inducted in the Hall of Fame (August 2009). The second huge recent spike is the renewed interest in him when he celebrated his 50th birthday. It was also rumored that he was still in such good shape that he could return to play a few games in the NBA! The big dip after the spike is probably due to people that would have normally searched for him in March had searched for him already the previous month!

Thursday, April 18, 2013

Who's searching for Steve Nash?

In a previous post we analyzed the trend and seasonality in Google searches for Kobe Bryant. We looked at another famously injured NBA star, Derrick Rose, as well as Lebron James. Let's turn our attention to another star who made lots of headlines for switching teams: Steve Nash.

Here's the raw data from Google Trends:

Using the same decomposition analysis detailed in the Kobe Bryant post, we get the following results:

We're starting to become experts analyzing these decompositions now.
One interesting insight is that, similarly to Derrick Rose and unlike Kobe and Lebron, Steve's peaks are indeed during the Playoffs, but in May as opposed to June, indicating that while he often makes the Playoffs he's never played a game in the Finals...

Of course we see a giant spike in the 2012 offseason, right after the 2012 Playoffs. Well Steve decided to take his talent from Phoenix to LA. We all know since then that the Lakers have not really impressed us this last season, but at the time the Steve-Kobe sent shockwaves worldwide!

Wednesday, April 17, 2013

Who's searching for Derrick Rose?

In a previous post we analyzed the trend and seasonality in Google searches for Kobe Bryant. We then looked at the evolution of Lebron James' queries. But what about another famously injured player whose return is eagerly awaited?

Here's the raw data from Google Trends for Derrick Rose, the 2011 MVP:

Using the same decomposition analysis detailed in the Kobe Bryant post, we get the following results:

What can be said from the decomposition?

Although he started playing or the Chicago Bulls in 2008, his popularity only increased (rather significantly) during the 2010 offseason. Since then his popularity has maintained high levels.

The seasonality is harder to rely on given the small number of seasons he has played, but we do again see the offseason/regular season/Playoffs effect, but unlike Kobe and Lebron that usually reach the later stages of the Playoffs, Derrick's peaks are in April/May for the beginning of the Playoffs only. Could Google search peaks be a criteria for distinguishing the future stars from the current stars?

As far as outliers are concerned, the main one is at the end of the regular season in 2011, and guess who won the MVP at that time? Despite his popularity, he was still not in the same league as the Kobes and Lebrons, so this MVP title did have many people searching to learn more about him.

One surprising element is the absence of a big residual peak the following year when he injured himself at the beginning of the Playoffs. I guess a lot of the additional queries were picked up by the Playoff seasonality thus explaining the absence of a bigger outlier in April 2012. We are seeing a big dip in queries right now simply because this is the beginning of the Playoffs and he has yet to return to the Bulls' lineup...

Tuesday, April 16, 2013

Who's searching for LeBron James?

In the previous post we analyzed the trend and seasonality in Google searches for Kobe Bryant. But how do these compare to the other most talked about superstar of the moment, Lebron James?

Here's the raw data from Google Trends:

Using the same decomposition analysis detailed in the Kobe Bryant post, we get the following results:

So what can we observe?

Well, the overall trend has been an increase since 2008 although there was a significant uplift in mid-2010 followed by a drop early 2011. Things have pretty much flattened since.

Similarly to Kobe, their is a strong seasonality in the searches, the amplitude of which has steadily increased over time. But unlike Kobe for whom the pattern was rather complex with a clear distinction between off-season, regular season and playoffs, Lebron's searches are not very different during the off-season and regular season, they only take off at the end of the season and during the playoffs.

But the residuals are really interesting here: look at that massive spike in July 2010. The playoffs were over by then, so what happened?

Well, Lebron became one of the most hated athletes (except in Florida perhaps) as he announced in a big show called "The Decision" that he would leave for the Miami Heat in the following season.

The other two spikes are in June, the first one when the Miami Heat lost to Dallas in the Finals, the second when Miami beat Oklahoma in the Finals.

However, if queries continue to spike regularly in June, the decomposition will incorporate those spikes as part of the seasonality, and the only big outliers will be major events like "The Decision" (there are rumors of Lebron returning to Cleveland, I anticipate an even bigger spike if that happens one day).

Monday, April 15, 2013

Who's searching for Kobe Bryant?

Unless you live on another planet without any WiFi connection to Facebook or Twitter, it was pretty hard to miss out on Kobe Bryant's injury in a crucial game against the Golden State Warriors last week. A torn Achilles has just ended his season (although to be fair I am not sure he will be missing that many games, even if Lakers do make the Playoffs...).

But this got me thinking: how many people searched for "Kobe Bryant" on the day and next of his injury? Could a huge spike be seen in Google Trends?

The above chart is at the monthly level so we only have partial April data. Nonetheless, Google Trends predicts that April 2013 will be Kobe's all star spike in searches, his MVP (most valuable peak) if you will.

But we also seem to notice some older spikes in the data, and a closer look reveals that these tend to regularly occur in June. What type of seasonal pattern lurks in the data?

In order to investigate seasonality, and also to tease out the underlying trend in searches for Kobe, I processed the data using R's stl() function (seasonal decomposition of a time series with loess) to split a time series into its trend, seasonal component and residuals.

Here's what I obtained:

We can't see the April spike yet because the downloaded data doesn't have Google Trends' estimates for the month, but we can predict a big spike given that halfway through April we already have almost as many queries as in a full month.

Going back to the figure, the top panel has the raw monthly data since 2004.

The next panel displays the smoothed loess trend. After a dip throughout 2004, Kobe has slowly but surely gained momentum since.

The next panel shows the seasonal component. We easily identify a strong pattern with a low volume in the off-season, an uplift when the season starts which remains quite constant during the regular season, another uplift during the playoffs with the maximum attained in June as we had observed previously.
It's also interesting to see how the seasonality's amplitude has steadily increased over the years.

Finally, the last panel shows the residuals, what cannot be explained by the trend and the seasonality. The cluster of yellow bars in the summer of 2004 could be due to the fact that Shaquille O'Neal was traded away from Los Angeles and there were many rumors around Kobe's future with the Lakers.

But how big the spike will be in April 2013 is anybody's guess!

Wednesday, April 10, 2013

Drama in Lakerland continues

Edit as of 4/10/2013 2:40pm
The drama continues, my model had the tiebreaker backwards, giving LA the advantage of the tie. Corrected values are at the bottom of the post.

The Lakers making the Playoffs or not has without a doubt been the most talked about conversation of the 2012-2013 season. And you would think that with just a handful of games to be played in the regular season we would finally have at definite answer or at least a pretty good idea.

But we would be forgetting this is LA/Hollywood we're talking about (I just know a movie is going to be made one day around this whole story).

So the Lakers currently have the 8th and final spot in the West (with a Laker win/Utah loss combination yesterday), half a game in front of Utah, but Utah has the tie-breaker in case the two teams end the season with same number of wins.

So what does the rest of the season look like? The experts have weighed in on's blog but what do the stats say?

Running some new simulations based on the latest results, factoring the strength of the opponents and home game advantage for the last games, we have the following results:

Position Lakers Probability Mavericks Probability Jazz Probability
7 0.4 NA NA
8 73.7 6.5 19.4
9 20.3 17 62.7
10 5.6 76.5 17.9

Because I am more of a data-driven guy, I hereby declare the Lakers IN THE PLAYOFFS!!

4/10/2013 2:40pm
The drama continues, my model had the tiebreaker backwards, giving LA the advantage of the tie. When corrected, the Lakers playoffs drop to 50.1% !!
As a statistician I cannot pronounce myself with this value....

Here are the details:
Position Lakers Probability Mavericks Probability Jazz Probability
7 0.6 NA NA
8 50.1 7.1 42.2
9 39.3 18.1 42.6
10 10 74.8 15.2