Twitter College Football Review: A Tale of Two Heisman Winners

College football is a business and college football fans are a vocal group of customers.  Like many businesses, college football has a new opportunity to track customer opinion: The Twitterverse.  For a bit of a of “Holiday diversion” we are going to take a look at several college football stories from the past season using Twitter as the data.Manziel vs Winston

The first story is a look at the past two Heisman trophy winners.  The chart shows something call Twitter sentiment for Johnny Manziel and Jameis Winston.  Sentiment analysis is basically the ratio of positive to negative Tweets. The higher the score the more positive the Twitterverse is about a subject.

What we have done is track sentiment for a one year period for each player. The blue line shows the weekly sentiment for Johnny Manziel.  This starts off very high following his historic Heisman victory.  But there are some ups or downs in the off season.  The downs can be linked to some of the news stories about Manziel gambling in casinos or showing up at high profile events.

The big dip for Manziel occurred in June.  Following his comments about being eager to leave College Station, his sentiment dropped from 93 to 20.  “Johnny Football” did recover much of the lost sentiment but then dropped to an even lower score of 18 following the mini-controversies surrounding his appearances at UT frat parties, and leaving the Manning passing academy early.  Interestingly, his autograph controversy in August did not generate an equal backlash as his sentiment bottomed out at around 30.

From a marketing perspective, this is a fascinating story.  The biggest damage to his brand occurred when he offended the locals by talking bad about College Station and partying in Austin.  The impact of breaking NCAA rules was not as large an issue.  It’s one thing to break a rule, but it’s much worse for a brand to insult it’s primary customers.

In contrast, there isn’t much to say about Winston during the off season.  While his sentiment also bounces around, this is more of a problem of small numbers (minimal data) rather than anything else.  Manziel’s Twitter traffic absolutely dwarfed Winston’s.  In marketing language we would say that Manziel had much higher brand awareness.  This is important in the context of the Heisman race since publicity matters quite a bit.

winston_manzielDuring the season, we see a steady climb for Winston and an up and down pattern for Manziel.  These patterns obviously have a lot to do with how the teams played.  For example, A&M’s late season losses dropped Manziel’s sentiment to the mid twenties.   Departing from college football for a moment, think about what this means for companies interested in tracking customer satisfaction!  The Twitter data almost tells us exactly what happened in games each week.   We say almost because one of Manziel’s biggest gaining weeks followed a close loss to Alabama.  In other words, Twitter gives us instant feedback about team quality.

Just like Manziel, Winston also had a huge late season drop in popularity.  He bottomed out with a sentiment score of 13 in the middle of November.  In Winston’s case, the drop can be attributed to legal issues.  The interesting thing about Winston’s drop is how quickly the public forgave.  The combination of “no charges” and the Heisman victory drove Winston’s sentiment to 97 in December.  Winston also finally passed Manziel in terms of total Twitter activity in December as well.

In terms of the final word – a look at these two brands / quarterbacks over time says something profound (and obvious) about sports brands: If you win then the public will be very forgiving.

Mike Lewis & Manish Tripathi, Emory University 2013.

NFL Fans at the “Twitter Water Cooler”

Note: This was originally published on September 4, 2013

The start of the NFL regular season is upon us.  In cities across America, NFL fans will engage in the practice of “Monday Morning Quarterbacking,” giving an analysis of their team’s performance on the previous day.  Some fans will of course be delighted after a team victory, while others will be dejected after a crushing defeat.  We decided it would be interesting to analyze how the thirty-two NFL fan bases felt the day(s) after their teams played in the regular season.  While we don’t have the ability to observe the millions of “water cooler” conversations that occur every week, we do have access to millions of Twitter conversations about NFL teams.

We used Twitter data to describe fan base reactions to team wins and losses during the sixteen-game 2012 NFL Regular season.  Our process for data collection can be illustrated with an example using the Buffalo Bills.  Imagine that the Bills played a game on a Sunday.  We recorded whether the Bills won or lost the game.  We then collected all tweets in the Buffalo area that mentioned the words “Buffalo Bills”, “Bills”, or other very frequent terms used to describe the team.  We collected the tweets from Monday (one day after the game), Tuesday (two days after the game), and Wednesday (three days after the game).  We then analyzed each tweet and characterized its content as positive or negative.  Next, we calculated the overall sentiment (roughly the indexed ratio of positive to negative tweets) of the Buffalo Bills related tweets for each of the three days.  We repeated this process for all thirty-two teams, and for all regular season games*.

The chart above displays the average sentiment of fans both after wins and losses.  The chart is based on data from the regular season for all thirty-two teams.  It is interesting to note that by three days after a win or loss, fans on average seem to either come down from their win “high” or recover significantly from their loss “low”.  While the chart above looks at all NFL fan bases in aggregate, we thought it would be interesting to classify each NFL fan base on the following dimensions**:

1)   Happiness After a Win (Highest ratio of positive to negative tweets after win)

2)   Sadness After a Loss (Lowest ratio of positive to negative tweets after loss)

3)   Stability (Least difference in positive to negative tweet ratio between after wins and after losses)

1) Happiness After a Win

The New Orleans Saints’ fans seemed to have just over a 9:1 positive to negative tweet ratio in the two days after the team won a game during 2012 regular season.  We believe that rankings on any of these dimensions are most likely driven by fan expectations (which is in part a function of past and current performance) and by the “expressiveness” of fans.  Since we are presenting descriptive statistics, and not explicitly modeling these drivers, it is tough to make a definitive statement as to why we see this particular order of teams.  Although, is anyone really surprised to see Cleveland or Oakland in the top 5?

2) Sadness After a Loss

The Pittsburgh Steelers’ fans seemed to take losing really badly in the 2012 season.  This could be because of fan expectations.  The Steelers finished 12-4 in 2011, but failed to make the post-season in 2012.  Early losses to the Raiders and Titans produced especially negative Twitter reaction, as did late season losses to the Browns and Bengals.

3) Stability

We measured “stability” by looking at the difference between average sentiment after wins and average sentiment after losses.  Dallas Cowboys’ fans seemed to never get too negative after losses, nor were they tremendously positive after wins.  Colts’ fans were even more understanding after a loss, but more positive on average than Cowboys’ fans after a win.  This could be due to the Colts being a young team that did not have high expectations going into the 2012 season.    The Atlanta Falcons only lost three times during the regular season, and the last loss was meaningless, as the Falcons had already secured home-field advantage throughout the playoffs.  Thus, there was very little negative reaction to the last loss.  The Philadelphia Eagles’ fan base is an interesting story.  It may be surprising to many to see them in the list of “stable” fans.   A better moniker for these fans in 2012 might be “resigned”.  It seems that as Philly began to lose more games, fans started to look forward to the next season, and a new head coach.  A majority of the fan tweets after a game were about changes for the next season, and not about the most recent loss.

The Oakland Raiders’ fan base best resembled “Dr. Jekyll and Mr. Hyde” during the 2012 season.  Fans were extremely happy when the team won, and terribly negative after a loss.  The Raiders were the only team in the top 5 of the “Happy” and “Sad” fan rankings.

 

Mike Lewis & Manish Tripathi, Emory University 2013.

*There are, of course, several caveats regarding this study.  First, while we only used tweets from the team’s geographic market, there could always be fans of other teams who may have tweeted about the local team.  Similarly, there are fans of the team that do not live in the local market, whose tweets would have been excluded.  Second, though we used terms that were associated/descriptive of the team, there are tweets related to the team that we undoubtedly excluded because they did not mention the terms we were looking for.  Third, the volume of tweets is not the same for each team.  We are confident, however, that a minimum threshold was met for each day, such that the sentiment score was not heavily influenced by a small number of tweets.  Fourth, this study is only over one year; it would be beneficial to perform a multi-year study.  Finally, there was one game in the 2012 season between the San Francisco 49ers and St Louis Rams that ended in a tie.  We have excluded that game from this analysis.  The Twitter data was collected using Topsy Pro Analytics.

**We computed each of these metrics using one day after, an average of one and two days after, and an average of the first three days after.  Since the rankings were fairly robust across these specifications, we only report the average of one and two days after the game.

What if the Heisman Trophy Really Was A Popularity Contest?

Ballots for the Heisman Trophy were due yesterday.  Ostensibly, the Heisman Trophy “annually recognizes the outstanding college football player whose performance best exhibits the pursuit of excellence with integrity”.  However, many have argued throughout the years that the Heisman is essentially a large popularity contest.  This view is supported by the millions of dollars annually spent by universities on publicity campaigns for their Heisman candidates.  There are 928 voters for the Heisman Trophy.  This includes members of the media, former winners, and 1 “fan vote” that represents the public at large.  We were curious to see what would happen if the general public was completely responsible for determining the winner of the Heisman Trophy.  As with past studies, we decided to use Twitter as a proxy for the views of the public.  Below, we present our methodology and results.

The first thing to consider is how does one define “Popularity” on Twitter.  Often, studies will use the volume/number of mentions on Twitter as a proxy for popularity.  However, this measure does not account for sentiment (positive, negative, or neutral), which could be important in the decision to vote for someone.  So, we constructed a daily “popularity” measure that is the product of the volume of tweets mentioning a candidate and the average sentiment of those tweets (Note: we tried several specifications of the “popularity” measure, but the rankings were robust).

Once we had a method for determining popularity, we decided to look at the six Heisman Trophy finalists: Johnny Manziel, Jameis Winston, A.J. McCarron, Tre Mason, Jordan Lynch, and Andre Williams.  The pie chart on the left looks at the sum of the popularity measure for each candidate over the entire season (mid-August to Dec 9th).  Johnny Manziel is by far the leader of the pack.  This could potentially be attributed to the stellar start of his season, as well as his huge following.  Heisman-favorite Jameis Winston is in second place, and A.J. McCarron is third.  It’s incredible that Manziel leads Winston by more than a 2:1 margin.  We realize that Heisman voters mark their ballots for 1st, 2nd, and 3rd place, and we are simply looking at most “popular”.

We performed a similar analysis, looking at only the last month and looking at only the last week.   It’s remarkable to see the variation in “popularity” over time.  Tre Mason had a relative 5% popularity if you look at the full season, but 11% over the last month, and 24% over the last week.  In the analysis of popularity over the last week, James Winston barely edges out Manziel for 1st place.  To better understand the factors behind these movements in popularity, we would have to perform content analysis on the tweets to determine what topics were being discussed with respect to these athletes; that is left for a future study.

It is interesting to note that in the their final straw poll, Heisman Pundit has the following ranking: 1) Winston, 2) Lynch, 3) Manziel, 4) Mason, and 5) McCarron.  The “popularity” measure for over the last week gives the ranking: 1) Winston, 2) Manziel, 3) Mason, 4) McCarron, and 5) Lynch.  Jordan Lynch is the only player of these top 5 that plays for a “Non-AQ” school (Northern Illinois).  Perhaps Lynch in second place is evidence that voters look at performance on the field, and not just popularity, however if Heisman Pundit’s straw poll is correct, it seems a lot can be explained by recent popularity.

Mike Lewis & Manish Tripathi, Emory University 2013.

 

 

Twitter Analysis: The Seat is Warm for Philbin & Shanahan

During the NFL season, columnists & “insiders” provide their speculation on coaches that are on the proverbial “hot seat”.  It seems like coaches can be on the “hot seat” before the season even starts, and they can jump on and off the seat on a weekly basis.  We assume that the columnists & “insiders” are basing their speculation on institutional knowledge.  While Emory Sports Marketing Analytics does not have access to NFL team management, we do have the ability to gauge fan/customer opinion through Twitter.  We would like to present the NFL Coaching Hot Seat from the fan perspective.

The methodology for creating our “hot seat” is straightforward.  Using Topsy Pro, we collected all tweets from the last thirty days that mention a head coach.  Each tweet was then characterized as having positive, negative, or neutral sentiment based on its content.  We computed the ratio of negative to positive tweets for each coach (e.g. If the ratio is 2, the coach has twice as many negative tweets as positive).  We believe that this ratio can serve as a proxy for public sentiment towards a coach.

A quick scan of today’s sports news shows that most columnists & “insiders” believe that Greg Schiano (Tampa Bay) and Leslie Frazier (Minnesota) are receiving the most heat from their coaching seats.  Our analysis of public sentiment over the last thirty days shows that Joe Philbin (Miami) has the highest negative to positive tweet ratio (2.47).  At a distant second is Mike Shanahan (Washington), and Mike Tomlin (Pittsburgh) is in third.  Thus, it seems that if the public were making coaching decisions,  Philbin and Shanahan would be on the hottest seats!  We realize that much of the negative tweeting about Coach Philbin is probably connected to the Martin-Incognito incident, but ultimately this does reflect on Philbin’s job status.

It is interesting to note that in terms of sheer quantity of  tweets (number of mentions), Rex Ryan is the leader in the NFL over the last thirty days.  Schiano  and Philbin are second and third, respectively.

Mike Lewis & Manish Tripathi, Emory 2013

Ranking the Most “Volatile” Fans in the SEC: LSU, Ole Miss, & UGA Lead the Way

Last weekend, Georgia beat LSU in a highly entertaining, closely contested football game.  After the game, fans were undoubtedly sad in Baton Rouge and elated in Athens.  These emotions were manifested through the tweeting activity of fans in both cities.  Using data from Topsy Pro, we were able to collect football-related tweets originating from Athens and Baton Rouge after the game.  There were almost twice as many tweets originating from Athens, and the ratio of positive to negative tweets was 9:1 in Athens, whereas the ratio was 1:9 in Baton Rouge.  As transplants who have lived in Atlanta for a few years now, we can attest to the overwhelming passion towards SEC football in the South.  Recently, we used data from Twitter to describe the emotions of NFL football fan bases during the 2012 regular season.  We decided that performing a similar analysis on the SEC football fan bases would be an interesting study.  We decided to empirically determine which SEC football fan bases really “live & die” by the performance of their teams.

The methodology for our study was straightforward.  We considered all of the regular season games from 2012 and the first five weeks of the 2013 season.  For each game, we recorded who won the game, and we collected football-related tweets from all of the SEC college towns for one, two, and three days after the game.  It would be reasonable to ask why we didn’t collect tweets from Atlanta for a UGA game or from all of Kentucky for a UK game.  We were trying to isolate tweets primarily from fans of the SEC team, and we believe that the college town is the best proxy for mainly fans of the college.  Atlanta is full of UGA fans, but there are also Alabama fans, Auburn fans, Florida fans, and pretty much fans of all SEC teams.  We wanted reactions of UGA fans to the UGA games, not the reactions of Auburn fans to the UGA games.  By football-related tweets, we mean tweets that mentioned any words that were commonly related to the particular college football team.  The tweets were coded as positive, negative, or neutral.  We were able to determine the “sentiment” of the collection of tweets as a rough index (1-100) of the ratio of positive to negative tweets.

Thus after each game, we were able to calculate the sentiment of the fan base.  We determined on average how positive a fan base was after a win, and how negative they were after a loss.  To understand the “volatility” of a fan base, we looked at the delta between the average sentiment after a win and the average sentiment after a loss.  In other words, how big is the difference in a fan base’s “high” after a win and “low” after a loss.  We believe that this metric best captures “living & dying” by the performance of your team.  After computing this metric for each fan base, we determined that LSU has the most “volatile” fans in the SEC.

The chart on the left gives the full rankings for the SEC.  It should be noted that these rankings were robust to whether we looked at how fans felt one, two, or three days after a game.  We believe that volatility is in part driven by 1) the expectations of the fan base and 2) the expressiveness of the fan base.  The top three schools in our rankings seem to get to the top for different reasons. The volatility of LSU & UGA fans is driven more by extreme negativity after losses, whereas the volatility of Ole Miss fans is a function of high levels of happiness after wins. This could, of course, in part be due to expectations.  UGA & LSU fans may have higher expectations than Ole Miss fans.  An examination of the data reveals that LSU fans had an extremely negative reaction to the Alabama loss last year and the Georgia loss this year.  These fans even had an overall negative reaction to a close WIN over Auburn last year!  UGA fans spewed a lot of vitriol on Twitter after the loss to Clemson this year.  Ole Miss fans, on the other hand, did not have overly negative reactions to losses, and were very positive after wins (e.g. the win over Texas this year).   It is interesting to note that the Alabama fan base is at the bottom of the volatility list.  Alabama only lost one game during the period of this study (a good reason for publishing this list again next year when we have more data).  But, even after wins, the Alabama fan base is not very positive on Twitter.  There are several tweets that are critical about the margin of victory.  If Alabama does ever go on some type of losing streak in the future (as unlikely as that seems), it will be fascinating to observe the reaction on Twitter.

Mike Lewis & Manish Tripathi, Emory University 2013.

 

 

 

 

Twitter Analysis: College Station Buzzing About Alabama

It’s amazing what a difference a year (or ten months) can make.  Last November, Johnny Manziel was a redshirt freshman leading a two-touchdown underdog team into the hostile environment of Bryant-Denny Stadium.  This weekend, the Heisman-Trophy winning, John Hancock-machine leads the sixth-ranked Aggies into a matchup with Alabama that is undoubtedly one of the most anticipated games of the college football season.  ESPN College GameDay will be in College Station, even though the game is on CBS.

We decided to use Twitter to study (1) how much more chatter is there about the game this year versus last year and (2) how is the chatter different between the two campuses?  Our key findings: (1) The pre-game chatter has increased over 600% in College Station and 350% in Tuscaloosa as compared to last year and (2) The level of pre-game chatter was over 3oo% greater in College Station versus Tuscaloosa in 2012 and over 400% greater in 2013.

The methodology for our study is quite straightforward.  As in our Michigan-Notre Dame “rivalry” study, we used Twitter data from Topsy Pro Analytics.  We essentially collected all of the tweets originating from College Station, TX and Tuscaloosa, AL in the Sunday-Wednesday period before the game in 2012 and before the upcoming game.  In the pool of tweets from College Station, we counted how many of them mentioned a term that was related to Alabama (e.g. “Alabama”, “Bama”, “Tide”, and “Saban”).  We divided this number of tweets by the total number of tweets collected from College Station.  We performed this analysis separately for 2012 and 2013.  This gave us the Twitter Share of Voice for the match-up in 2012 and 2013 in College Station.  We did the exact same thing for the pool of tweets from Tuscaloosa in 2012 and 2013, but we looked for Texas A&M related terms (e.g. “Aggies”, “TAMU”, “Manziel”, and “Johnny Football”).  We believe that the Twitter Share of Voice metric is a good proxy for the level of game related chatter in the two markets.

The results indicate that while both communities seem to care a lot more about the game this year than they did last year, the Texas A&M community cares a lot more about the match-up than the people in Tuscaloosa.  We look forward to performing a similar analysis when Alabama plays Auburn later this year.

Mike Lewis & Manish Tripathi, Emory University, 2013.

Twitter Analysis: Michigan Cares More about Notre Dame “Rivalry”

Last week, Notre Dame Coach Brian Kelly described the Michigan-Notre Dame game as not “one of those historic, traditional Notre Dame rivalries.”  These comments helped invigorate discussions, newspaper columns, and College GameDay signs debating the magnitude of the Michigan & Notre Dame rivalry.

Rather than listen to “experts” tell us about the significance of the game (or fabricate memories of the game), we decided to use Twitter to study how much people cared about the game in South Bend, IN and Ann Arbor, MI.  The setup for our study was fairly simple.  Using data from Topsy Pro Analytics, we were able to examine tweets originating from South Bend and Ann Arbor.  We compiled a list of words that could be used to describe Notre Dame (e.g. “Notre Dame”, “ND”, “Fighting Irish”) and a list of words that could be used to describe Michigan (e.g. “Michigan”, “UM”, “MICH”, “UMICH”).  We then collected all tweets that mentioned any of the Notre Dame related words and originated from Ann Arbor.  We also collected all tweets that mentioned any of the Michigan related words and originated from South Bend.  We believe that these tweets are capturing the level of “rivalry” that each campus has toward the other campus*.

For the game played on September 7, 2013 in Ann Arbor, we looked at tweets on September 5th and 6th (pre-game).  We also examined tweets on September 8th (post-game).   We computed the Twitter Share of Voice for tweets about Notre Dame in Ann Arbor and for tweets about Michigan in South Bend for both Pre and Post-game.  As an illustration, to compute Twitter Share of Voice for Notre Dame related tweets in Ann Arbor, you simply divide the number of tweets that mention Notre Dame in Ann Arbor by the total number of tweets in Ann Arbor.  We believe that this Share of Voice metric helps control for the relative sizes of Twitter bases in the two cities.

The results from the 2013 game are very interesting.  Pre-game, the Twitter Share of Voice in Ann Arbor for Notre Dame related tweets was 60% higher than Twitter Share of Voice in South Bend for Michigan related tweets.  This implies that people in Ann Arbor cared more about the game (at least on Twitter) than people in South Bend.  Post-game, the Twitter Share of Voice went up by 57% in Ann Arbor.  The sentiment (ratio of positive to negative tweets) of the post-game tweets also rose by 40%, whereas there was no change in sentiment in South Bend (Michigan won the game).  We could interpret this as Notre Dame fans were relatively unaffected by the loss.

Perceptive Michigan and Notre Dame fans could argue that these results are skewed because the game was played in Ann Arbor.  We have excluded tweets from the day of the game to try to correct for any game site effects.  However, to get a better understanding of the “rivalry”, we performed a similar study of the 2012 game which was played in South Bend.  Even though the game was played in South Bend, the pre-game Twitter share of voice was 18% higher in Ann Arbor. The Notre Dame victory only created a 14% increase in the post-game share of voice in South Bend, and a 23% increase in tweet sentiment.  Thus, looking at data from the past two years, there seems to be an asymmetry in this “rivalry”.  That is, it seems Michigan cares a lot more than Notre Dame.

Mike Lewis & Manish Tripathi, Emory University 2013.

*Obviously, both of these universities have alumni all over the world.  We are limiting our study to South Bend & Ann Arbor because we believe this (1) captures current students and (2) is the cleanest way to separate out the two fan bases.

Instant Twitter Analysis: USC angrier but Texas Cares More

When teams lose fans get angry and coaches get fired.  Twitter now allows us to get an instant picture of fan anger.  Over the first week and day of the college football season, two coaches have emerged as mostly likely to be run out of town.  According to Topsy, Mack Brown has been the subject of the most negative tweets (2,550) but Lane Kiffen has the highest rate of negative to positive tweets (2 times as many negative as positive tweets).

USC fans are angrier but Texas fans are more involved.

Tebow Fatigue?

In the past, we’ve discussed Tim Tebow in the context of the brand equity he created for the University of Florida.  With his recent departure from the New England Patriots, we thought it would be interesting to see how fans reacted to his being cut this time around (as compared to in April from the NY Jets).  The chart below simply illustrates the ratio of positive to negative tweets that mentioned “Tebow” on April 29, 2013 [when Tebow was cut from the Jets] and on August 31, 2013 [when Tebow was cut from the Pats].  The ratio dropped from 1.55 to 1.05.  Thus, while overall there were still more positive than negative tweets when Tebow was cut from the Pats, the ratio has declined dramatically from the first cut by the Jets.  There were also fewer mentions of Tebow overall.  Does this signal Tebow fatigue?

Mike Lewis & Manish Tripathi, Emory University 2013.