Super Bowl Party or Super Spreader Event?

TLDR: All the medical professionals have been urging people to avoid large crowds but would Americans abide by the rules and avoid Super Bowl parties?

I calculate the size of the average Super Bowl party by dividing the estimated number of viewers by the estimated number of TVs watching the Big Game. The results show that the average party was the lowest it has been since 2007, reversing a trend of growing party sizes.

Although it is still likely higher than the medical professionals would have liked and large parties likely still occurred, there are signs that some individuals did heed some warning and avoided the large crowds.

---

Everyone's aunt's favourite heart-throb, Dr. Fauci, and medical professionals alike, have asked everyone to stay home, wear masks, avoid crowds, etc., etc. – I think we all know and are near our wit's end with the terms 'social distancing,' 'unprecedented,' and 'normalcy' (technically, unless we are taking about geometry, normalcy is not the correct word... it should be normality). 

But despite all these warnings, would Americans forgo gathering on what some see as the holiest of days of the year - Super Bowl Sunday?

The Nielsen Company, famous for its research into the television viewing habits of Americans, actual can provide insight into this question. Every year, Nielsen provides the ratings—the estimated number of televisions watching the Super Bowl—and the total number of viewers. Nielsen is a trusted source on this topic and I have talked about them ad nauseum on this blog... (like here for football, or baseball, or even basketball).

As a quick summary, a television program's ratings are calculated as follows: Rating = 100 × (# TVs tuned to the Program) / (# of TVs) For example, when Nielsen says a television program had a rating of 5, that means 5% of all TVs in the designated marketing area were watching that program. The remaining 95% could be watching a different program or turned off entirely.

While the typical Super Bowl attracts both large ratings (+40% of all TVs in recent years), it also attracts large crowds at Super Bowl parties. Aside from the gut-wrenching series finale of M*A*S*H, the Super Bowl makes up the top 30 most-watched American television programs. Nielsen also provides an estimate on how many people are watching the game. If we multiply the number of people watching by the number of TVs, we can get the average number of people watching the same TV which we can think of as the average Super Bowl party size. It can be calculated as follows:

Super Bowl Party Size = Estimated Audience / [Rating × (# of TVs)]

So, using the formula above how did the US do? Below, I plot the average Super Bowl party sizes for the last 15 years:

What we see is a stark reversal in an otherwise increasing trend of larger Super Bowl parties. For the first time since 2007, less than two individuals gathered to watch the Super Bowl. (For those of you who think an average Super Bowl party of 2.25 people is lower than you anticipated, consider that for every massive party there are likely numerous households with one individual watching plus all the people following Shania Twain's rule of having a party for two).

While there were lots of concern that Super Bowl parties would turn into super spreader events, it appears some have heeded the advice of the experts and opted for a smaller crowd or avoided parties altogether. While this likely meant that there were still the same crowded parties that were held 15 years ago, it also means there is still some hope that people are listening to the rules!

Good job America!

So long XFL, we hardly knew you...

So long XFL, we hardly knew you...

And perhaps you hardly knew yourself.

Because if you did, you would know that before the Coronavirus-induced suspension of play, the laying-off of all employees, and the Chapter 11 bankruptcy, there were some positive signs of life. And, yes, the viewership was consistently decreasing week-over-week (a pattern not unique to the XFL, mind you), but the league was at or near its floor in viewership - the solid base of fans it could count on. And the magnitude of this fan base would put it in some very good company - namely the other major, non-NFL, sports leagues.

---

There was never really high expectations for this year's Xtreme Football League (XFL) given that it was the second incarnation, the original of which had folded after just one season back in 2001. That original XFL was also hardly the first American football league with the lofty goal of becoming a legitimate rival to the mighty National Football League (NFL). Yet, insofar as these attempts to develop a viable competitor, none have found success.*

Yet despite the poor track record of all the previous endeavors, the 2020 version of the XFL saw an opportunity to not just bring the more of the same American football, but to deliver on a unique brand of football: kick-offs were modified, PAT kicks were eliminated, and it seemed that nearly everyone in the stadium would be mic'd up.

Thus, on 8 February 2020, the XFL was (re)born. This eight-team league kicked off a ten-week schedule just seven days after the NFL Super Bowl. Each member team was slated to be in action each week and all games broadcast nationally on ABC, ESPN, FOX, or a counterpart thereof. The  first games were generally well-received by the 3.3M fans that tuned in and the style of play in was described as "quirky" but "a lot to like." While these television audience sizes were less than half of what the 2001 predecessor attained during its debut, this year's viewership was certainly enough to label the XFL 2.0 a viable product - if it could only keep it up.

Below I plot the number of viewers for each XFL game and add a simple logarithmic trend line before a Coronavirus-induced hiatus was announced. By week three, we can see why, after a tepid, yet acceptable start, people predicted this league was doomed: the average viewership had dropped by 50% compared to its debut. However, it should be noted that this pattern has been seen time and time again, such as NFL's Thursday Night Football: the large hype for the first event of the season does not last long.

Alas, the XFL viewership declined in a somewhat predictable way, but this decline began to settle around week four. If we think that the XFL had in fact bottomed out, the average viewership for the remainder of the season would have been around 1.25M. When comparing this estimated average viewership to other major sports leagues, this is a very reasonable level (below I use the most recent full season of other leagues to compare against):

Perhaps the XFL hoped to have garnered interest enough to rival that of the NFL, but with a 2019 average of 15M viewers per game, the NFL remains way out of the XFL's league - pun intended. The XFL also fell just short of keeping pace with NCAA college football which averaged 1.7M viewers on ESPN in 2019.

However, if we move the goalposts just a bit, two places where the XFL stood out from crowd was against the 243K viewers of ArenaBowl XXXII (the Arena Football League's version of the Super Bowl) and the 200K viewers of the 107th Grey Cup (the slightly more polite version from the Canadian Football League - a game played on a oft-snowy field in November in Canada where Shania Twain once rode a dog sled with a Mountie for the half-time show).

The point is that XFL wasn't really a strong competitor of the NFL. But as a league of second-tier talent and an informal moniker of 'minor-league' football, the XFL had outstanding success. Perhaps they fell victim to their own hubris and lofty expectations, or perhaps other, unknown issues befell the league but sadly, XFL, you were gone too soon.

RIP XFL - you will be missed ... ?

---

* The exception being the American Football League whose outstanding strategy of having teams based in then-untapped markets led to a merger with the NFL.

Ben Roethlisberger is the most owned Steeler QB in Fantasy Football

Despite his season-ending elbow injury in a week 2 game, 13.1% of fantasy football leagues still have Ben Roethlisberger on their roster.

Check out the fantasy ownership of Sunday, Monday, and Thursday Night Football matchups and their television ratings here!

Update!

Big Ben is once again the most owned Steeler quarterback in fantasy football!

Keep your options open: Change in MLB viewership when substitutes are available

TLDR: Every television program on the air compete for the same sets of eyes. To understand the preferences of consumers, we can use television data to observe the changes in viewership when multiple programs air simultaneously.

Results suggest that 10% of would-be MLB viewers instead watch their local NBA team in a playoff game and only 3.6% watch the local NHL playoff game. In cities with two MLB teams and a NHL team, the MLB team A steals away more fans from MLB team B than the local NHL team playing playoff hockey!

---

Every television program on the air - and increasingly, television programs on demand - compete for the same sets of eyes (my beautiful optometrist girlfriend assures me eyes usually come in sets).

Sporting events are no exception: looking way back to my second ever blog post, I discussed how at least some of the decline in the National Football League's Sunday Night Football broadcasts was due to the timing of the games and other programs like the Cubs-Cleveland World Series game 5 or the 2nd Presidential Debate.

To understand the preferences of consumers, we can use Nielsen television data to observe the changes in viewership when multiple programs air simultaneously. Nielsen is the media company that provides estimates on the number of people who tune in to a given television program.

For this calculation, I focus on Major League Baseball for two reasons: a) the data is readily available, and; b) there are plenty of games with and without overlap from similar television programs. I define a similar television program as a broadcast of other local teams of the four major North American sports leagues (where local is roughly teams in the same city or marketing area). Narrowing the definition to local teams ensures that we capture viewers who actually have a choice between the several options. Therefore, a the local MLB regular-season game can potentially overlap with:

• the local National Football League team's pre-season game;
• the local National Basketball League team's playoff game;
• the local National Hockey League team's playoff game, or;
• the other local Major League Baseball team's regular-season game in the same marketing area, with a focus on when they are not playing each other.
• e.g. New York, New York (not Sinatra but the Mets and Yankees).

I run an estimation to predict the total number of viewers for each MLB game, adding controls for the team, opponent, day of the week, the quality of the teams, and, finally, the availability of alternative sporting events. If the number of viewers declines in response to the availability of one of the aforementioned alternative sporting events, we would say the event and MLB are substitutes for each other. If we assume the 'cost' of watching a given sporting event stays relatively constant over time, the magnitude of the decline becomes an indicator of how strong of a substitute each event is for MLB.

Below is a graphical depiction of the results:

As expected, the number of viewers of the local MLB declines when an alternative sporting program is airing simultaneously. While it is no secret that on a per-game basis NFL is the most watched sport in North America, it is still quite astonishing that 30% of would-be MLB viewers turn off their local baseball team in favour of their local NFL team participating in a pre-season game! Some may argue that NFL pre-season overlaps the MLB regular-season at a time when the post-season fate of many MLB teams has already been decided and there is very little excitement in watching a losing MLB team finish out a season. But the estimation described above does in fact control for the quality of the MLB team meaning that on average, whether the team is leading the division or a perennial basement dweller, 30% of their fans rather watch a 'meaningless' pre-season NFL game.

Furthermore, 10% of would-be MLB viewers instead watch their local NBA team in a playoff game and only 3.6% watch the local NHL playoff game: in cities with two MLB teams and a NHL team, MLB team A steals away more fans from MLB team B than the NHL team playing intense playoff hockey takes from MLB team B!

While I would not necessarily prefer to watch the local NBA playoff game over the local NHL game (and likely never watch a pre-season NFL game), what do you guys think? Do the results seem reasonable? If all were playing at the same time, which would you choose to watch?

Let me know in the comments!

---

If I had a dollar for every reader ...

Turns out I actually can (or at least fractions of a dollar).

If you are enjoying what you are reading and want to support Sports and Economics (in that order), please consider turning off any ad-blocker and clicking some ads and some ad revenue will be sent my way!

Predicting the number of viewers of an NFL game: a Fantastic new approach!

TLDR: What can you get with $700,000? How about a 30 second commercial during Sunday Night Football. With such a large price tag, it is imperative that businesses accurately estimate how many eyeballs are expected to see their pricey Sunday night advertisement.

I find the best method of forecasting NFL viewers is not by using a measure of the featured teams' quality. Instead, I find that the share of Fantasy Football ownership can help to explain around 73% of the variation in NFL viewership: 12.5% better than team quality!

And now that I have begun to provide readers with the NFL nationally broadcast game viewership data, you too can experiment with these types of analyses!

$700,000 can buy you a lot of things: a house, a used 100-foot yacht, or a 30 second commercial during Sunday Night Football. And while it is nowhere near the $5 million price tag for a Super Bowl commercial, I imagine most businesses find it important to accurately estimate how many eyeballs are expected to see their pricey Sunday night advertisement.

So how would a business go about predicting the number of viewers for a National League Football (NFL) broadcast? They would likely need to begin with some background information.

First, the NFL regular-season consists of 17 weeks. There are three three nationally broadcast games in a typical NFL week: the aptly name Thursday, Sunday, and Monday Night Football based on the weekday the game is played, abbreviated to TNF, SNF, and MNF, respectively.

Second, Nielsen, a media company, provides data on the number of people who tune in to each of the three nationally broadcast games each week. Many readers have likely heard the term ratings, which is the number of televisions tuned to a specific broadcast divided by the total number of televisions. Instead, a business may want to focus on the audience, or the total number of people exposed to their advertisement.

Nielsen estimates for the size of the audience and/or ratings are then used by television stations to set the price of advertising. Since SNF usually has the most viewers, it is the most costly of the three nationally broadcast games, but TNF and MNF are not much cheaper and will cost around 385,000 to 500,000 for 30 seconds.

With all of this information in mind, the next step is to predict the number of viewers using a multivariate regression model. I use the Nielsen estimated audience size as the independent variable and consider many different explanatory variables in attempt to explain the variation in the audience size. The most logical suggestion for an explanatory variable would be a measure of the quality of each team featured in the match, such as FiveThirtyEight's ELO. While this is a good start, is team quality the best explanatory variable of viewership?

Instead, I suggest looking at data from an industry that has seen a growth of +370% from 2005 to 2017: Fantasy Football! For those who are not familiar, Fantasy Football is a game played among viewers wherein each individual becomes a manager of a hypothetical football team in a league made up of their friends, colleagues, co-workers, etc. At the beginning of each actual NFL season, Fantasy Football managers create their own teams wherein each actual NFL player can be owned by no more than one Fantasy Football manager within each Fantasy Football league. The performance of the actual NFL players contribute to the points to the Fantasy Football team that own said player. And while the advent of technology has made tracking the points of each Fantasy Football team much easier, watching the players owned by one's Fantasy Football team adds to the excitement of the NFL viewing experience.

And surprisingly, Fantasy Football also adds significant value to a model for predicting the number of viewers of a NFL game!

I have considered several multivariate models in order to predict NFL viewers, each varying by only a single regressor. First, I use the aforementioned ELO, or the quality of each team featured in the broadcast. I then turn to the Fantasy Football ownership of the NFL players that are featured in the game-broadcast; Quarterback (QB); Running Back (RB); Tight End (TE); Wide Receiver (WR), and; overall. Ownership is defined as the percent of Fantasy leagues where the player is owned and is provided by ESPN. I then sum the two values for ownership of Team A's player and Team B's player when Team A is playing Team B on a nationally broadcast NFL game.

Below is a graphical depiction of the results:

At 64%, ELO (team quality) does quite a great job at explaining the variation of the viewership (i.e., a relatively high R-squared). However, it does not outperform other measures...

Instead, turning to the Fantasy Football ownership, the ownership of Wide Receivers helps explain 73% of the variation in the viewership. Trailing just slightly, overall ownership of all the positions between the two teams explains 72%. Quarterback ownership does about as good of a job as ELO and Tight End ownership does not quite outperform ELO. Running Back ownership does little in terms of explaining the audience at all: I suspect this is because Fantasy Football managers are required to have at least two Running Backs on their team and the ownership of first-string Running Backs in the NFL is above 90% for 30 of the NFL's 32 teams.

While this is an interesting find, it also has quite practical purposes: from fans to advertisers to league organisers, it is in everyone's best interest to offer the most exciting and interesting games from a viewers perspective. And now that I have begun to provide readers with the NFL nationally broadcast game viewership data, you too can experiment with these types of analyses!

Viewers Aren't Having a Knee Jerk Reaction to NFL Player Protests

---

TLDR: There has been "a 17% decline in the television audience of [...] NFL games."

"The decline in these measurements of NFL viewership is not entirely due to kneeling protesters. Nor is it due to the increased awareness of head injuries."

This decline can be seen as "a shift in how fans consume NFL products not a shift in how much fans consume NFL products."

---

The National Football League (NFL) has seen no end of issues as of late: deflated balls, alleged-doping scandals, concussion lawsuits, and now, the twitter account of the president of the United States:

Marshawn Lynch of the NFL’s Oakland Raiders stands for the Mexican Anthem and sits down to boos for our National Anthem. Great disrespect! Next time NFL should suspend him for remainder of season. Attendance and ratings way down.

— Donald J. Trump (@realDonaldTrump) November 20, 2017

Instead of debating the issue, I want to evaluate and contextualise this final sentence, "attendance and ratings way down," which we may have missed if not for the additional 140 character allowance.

While Mr. Trump is mostly true, I find that the decline in these measurements of NFL viewership is not entirely due to kneeling protesters. Nor is it due to the increased awareness of head injuries. In fact, we probably don't really know what is causing the majority of this shift in demand away from live NFL television.

What is true is that Nielsen has measured a 17% (unconditional) decline in the television audience of nationally broadcast NFL games in 2017 over 2015. Attendance, on the other hand, is a bit trickier to measure....

Because of the difference between paid attendance and turnstile attendance, we will only know how many tickets are sold to each game instead of how many fans show up. Therefore we cannot know the impact of events such as anthem protests have on the number of people attending live football games. In fact, the average attendance as reported by teams is on par with past seasons. Ultimately, without insider access to each NFL team's in-seat attendance data, we cannot fully evaluate the statement that attendance is down because:

1. Season ticket holders and individuals who purchased tickets in advance of the game are always included in the paid attendance figures. Therefore these individuals cannot elicit a response to new information, such as inclement weather which may cause some to not attend, and;
2. Reported attendance is typically at or near capacity for the eight home games of most teams. With few observations and such little variation in a dependent variable, we are unlikely to pick up an effect of an explanatory variable.

Instead we can focus on the television ratings to demonstrate what is the cause for the modest viewership of the three nationally broadcast games the NFL airs each week. These three broadcasts are aptly named Monday, Thursday, and Sunday Night Football, with the latter having the largest viewing audience. Note that for exhibition, I focus on Sunday Night Football, however these results can be generalised to each of the three prime time broadcasts.

I begin by collecting the Nielsen-estimated number of viewers of each of the prime-time games since 2014, as summarised by Sports Media Watch. I then collect information on the quality of each team using FiveThirtyEight's ELO.

I also want to consider other factors that may be affecting the viewership such as the focus of head injuries in contact sports or the aforementioned anthem protests. For head injuries, I use information from Google Trends for the weekly number of searches for "concussions" in the United States. In the case of anthem protests, ESPN publishes a weekly list of players who have made some sort of protest-like gesture during the Star-Spangled Banner before their respective game. I define a anthem protest as a player kneeling or sitting during the anthem (therefore I do not consider the numerous players who raise a fist or choose to remain in the locker room during the anthem).

Next, I run a simple regression to predict the natural logarithm of viewers using weekly or game-specific data on the games, the teams, and the number of protesters. A graphical depiction of the results is below. To interpret the graph, consider the following example: if the quality of the teams featured on Sunday Night Football in 2017 was equivalent to that of which we saw in 2015, we would expect to see the number of viewers increase by the area of the white bar.

So, what do we learn from this exercise?

The first thing to note is that the vast majority of people are still watching Sunday Night Football through traditional methods (i.e., live television). Additionally, there has been a slight increase to (licit) online viewing of live NFL.

Second, the largest share of the explained loss in viewership can be mitigated through something I have talked about once before - better matchups. The NFL does have the power, especially in the latter half of the season, to make use of Flex Scheduling and change which teams are featured in the Sunday night broadcast. If done correctly, this could strongly increase the overall viewership for the remainder of the season: I have previously suggested the NFL could have used a better strategy brought in an additional 7-16 million viewers in the 2016 season alone.

Third, kneeling protests have only a small negative effect on viewership (only 200k viewers!). This effect is really no bigger than concussions which is another issue I have talked about that has been plaguing the NFL.

Instead, I cannot definitively say what is it that has been driving the majority of viewers away from Sunday Night Football. The red area of 'unexplained' decline in viewership may point to larger trends, such as cord-cutting wherein homes are selecting to forgo cable in favour of on-demand services such as Netflix.

Finally, some context: a simple comparison of other measures of demand for football point to a less worrying trend for the NFL. First, the unexplained decline in viewership of Sunday Night Football is not all that far off from the decline in viewership of arguably the closest substitute, NCAA Division I Football Bowl Subdivision (FBS).

Now compare the NFL viewership with the increase in the Fantasy Football memberships and the non-change in NFL ticket sales (keeping in mind the limitations of reported attendance numbers) and the NFL does not look to be in such bad shape. (Not to mention the $7B the league will receive each year for broadcast rights until 2021)

What I would interpret this as is a shift in how fans consume NFL products not a shift in how much fans consume NFL products: Nielsen measure neither mobile viewership nor viewership of a recorded program more than a day later, not to mention illicit online viewership.

And while Mr. Trump may be correct in saying NFL viewership is down, it may be a little overstated to think that it is all driven by player protests.

NFL television ratings, or the lack thereof

The viewership of the 2016 NFL regular season was lackluster to say the least.  The television ratings seemed to stumble out of the gates in September.  This prompted several public responses from commissioner Roger Goodell, attempting to explain away the downturn in viewership - he pointed the finger at everything from baseball playoffs to the extended coverage of the presidential election.

I want to review some facts of television and some of the metrics the industry uses to quantify their success.  I will show that the NFL, and similar media, have an upward battle to gain back the lost viewers.  They are "swimming against the stream" - figuratively and literally.  The first part will discuss what really happened to the NFL in 2016.

Some background information you will need:

• television ratings = (# of TVs watching program X) / (# of TVs)
• viewers = (# of TVs watching program X) x (# of people watching per television)
• the NFL has three prime time broadcasts three each week of its season (17 weeks per season):
• Sunday Night Football (SNF),
• Monday Night Football (MNF), and;
• Thursday Night Football (TNF).

I have collected data on the ratings and viewers of each of the prime time NFL broadcasts for the 2014, 2015, and 2016 seasons from Sports Media Watch.  For purposes of exhibition, I focus on SNF, which is the most watched of the three prime time games with an average rating of around 12 (MNF and TNF both have an average rating of around 8).  SNF has had the greatest ratings of any Fall show since 2010 (as per TV by the Numbers).

I want to see how different the viewership was in 2016 when compared to 2015, if at all.  I begin by plotting a three-week moving average of the number of viewers of each SNF game in these two years, as shown below.  Focusing on the solid black line, the 2015 season started with an average of around 25 million viewers for the first three weeks and finished with an average of 20 million viewers in the final three weeks.  By contrast, the 2016 season, the solid yellow line, started with an average of 22 million viewers and ended with an average of 24 million for the first and last three weeks respectively.  Up until week eight, the 2016 season's viewership looked bleak.

Now consider a similar broadcast: The Walking Dead (TWD).  TWD also airs during Sunday prime time and has 16 episodes to each season, comparable to the 17 SNF broadcasts per season. Also noteworthy is that TWD is the top-rated, non-sports broadcast.  The dashed black and yellow lines in the figure above illustrate the similarity in the trends of viewership for TWD and SNF (TWD data also comes from TV by the Numbers).  This exercise is the first indication that the 2016 downward trend in viewership may not be unique to the NFL.

Returning our attention to the NFL, a superficial explanation may suggest the ratings of the 2016 season are a result of poor scheduling choices or uneventful play.  In order to test this theory, I turn to a measurement of the relative competition in each match-up: FiveThirtyEight's Elo.  This gives a relative value of the skill level of each team prior to a particular game, where higher values indicate higher skilled teams.  I run a simple regression of viewership as follows:

(1) viewers_i,j,e= f(Elo_i + Elo_j, event_e)

where Elo_i + Elo_j is the sum of the Elo scores of each team and event_e is an indicator of which prime time game is being played: SNF, MNF, or TNF (note that I have chosen SNF to be the base).

I use the 2014 and 2015 NFL seasons to predict viewers for the 2016 season.  I then plot the residuals by week as shown below.  The black crosses show how the 2016 viewership fared against the model: points above the zero line indicate that the number of viewers was higher than predicted value given the level of competition of the match.  As the narrative suggests, many of the 2016 points lie below the zero line suggesting that the actual viewership did not live up to expectations.

I have highlighted two weeks in particular: five and eight.  These two SNF games were broadcast under exceptional circumstances: during the 2nd presidential debate and a Major League Baseball World Series (playoff) game.  These two games were aired simultaneously with strong substitutes, yet other weeks during the 2016 NFL season equally under-performed their respective expected viewership.  In other words, baseball and politics do not fully explain away the low NFL viewers.

I rerun equation (1) and add in season fixed effects and then, separately, control for team fixed effects.  The team effects are a set of dummies, one for each of the 32 teams, that take the value of one if that team is participating in the match.  These results are shown below.

Variable	Model 1	Model 2
Sum of Elo	7.50***	1.03***
	(5.20)	(5.09)
SNF	-base-	-base-
MNF	-7.51***	-6.55***
	(-14.34)	(-13.14)
TNF	-12.12***	-10.26***
	(-20.00)	(-12.60)
Year is 2014	-base-	-base-
Year is 2015	0.23	0.25
	(0.57)	(0.79)
Year is 2016	-1.25***	-1.38***
	(-2.64)	(-3.53)
Team FE	No	Yes
Constant	-2.45	-10.53*
	(-0.53)	(-1.71)
Adjusted R2	0.83	0.90
Observations	143	143

The results above suggest that in 2016 there was an average of 1.25 million to 1.38 million less viewers per game after controlling for the level of play.  Thus, the 2016 NFL regular season ended with 70 million less viewers than 2015 for it's prime time broadcasts, equating to a 10% decline in total viewership.  However, one thing we are failing to address is if the NFL could have chosen a better slate of games to feature in its prime time broadcasts.  What was the NFL's next-best option?

The 2016 NFL season did have one of the highest rated regular season games of all times between the New York Giants and the Dallas Cowboys.  The success of the aforementioned record-setting game prompted the NFL to change the schedule of the games for the next week to feature the Cowboys in a prime time broadcast again.  The use of the "flexible scheduling" is stated to "[ensure] quality matchups on Sunday night [...] and [give] surprise teams a change to play their way onto primetime.''  The NFL can, in theory, pick which match-ups to feature in nine of the 17 SNF games.

I would like to illustrate how the NFL could have gained from better scheduling.  First, here are the rules of flexible scheduling:

• In effect between weeks 5 - 17.
• Max 2 games can be flexed between weeks 5 - 10.

And my assumptions:

• NFL is risk neutral (prefers to maximize predicted ratings rather than  some sort of trade-off between magnitude of the estimated ratings and the magnitude of the estimates standard error).
• The is no cost to changing the schedule.
• Other networks cannot reserve games from being flexed (e.g. FOX wants to air the game on their network instead of going to NBC which hosts SNF).

I run the same models as detailed in the table above and predict the millions of viewers for each of the games that had could have been scheduled to play on SNF  Then I have the model choose the games with the highest predicted ratings to be featured on SNF for that week.  I limit the number of times a new game can be flexed in during the weeks 5 through 10 as per the rules of flexible scheduling.

Using Model 1, we see there would have been seven out of a possible nine scheduling changes.  The results suggest that the NFL could have added an additional 7.1 million viewers by using this more effective schedule.  Note that had the NFL based it's decision making process off of this model, they would never have aired the record-breaking game between the Cowboys and Giants in week 14.  Instead they would have opted for slightly higher expected viewership in the Seattle Seahawks and Green Bay Packers game.

Using Model 2, only five games would have been flexed into SNF which would increase the aggregate number of viewers by 16.2 million.  Recall that Model 2 uses team fixed-effects - this model highly favours teams like the Dallas Cowboys, New England Patriots, Denver Broncos, and New York Giants.  Model 2 suggests that SNF should always feature at least one of the aforementioned teams.

 

Models 1 and 2 rarely agree.  Although both would suggest the NFL made scheduling errors in weeks 6, 11, 13, and 17, the two models offer conflicting solutions to the problem.  In reality, the cost of changing the schedule may outweigh the benefit of the additional expected viewers.  Ultimately, flexing in more popular games would not have fully reversed the downturn in NFL viewers.

In summary:

• the 2016 NFL regular season fared poorly in viewership.
• -70 Million or -10% compared to 2015.
• this occurred despite the strength of available substitutes airing contemporaneously.
• there were gains to be made by rescheduling SNF games.
• 7.1 to 16.2 million viewers in expectation.
• scheduling would not have completely solved the issue of lost viewers. 
• 2016 had a unique effect on viewership.
• similar television programs saw a decline in viewership

Thank you for reading.  Please join me in the next post about television ratings where I will discuss the last point in bold above.  I also promise not to talk about the NFL.  Please add any comments or questions below.

the cost of a concussion

In case you have not turned on a National Football League (NFL) broadcast in the past five years, concussions are becoming an ever-increasing hot button topic.  From 2011 to 2015, there were more than 1,200 formally diagnosed concussions during games and practices.  Yet, with all the talk about the epidemic of head injuries, there is still questions to be answered: the most pertinent to the economist being the effects of the injury on the wage of players.

My goal is to predict the effect of a concussion on future salary.  Before explaining the details of the estimation, here are some pieces of background information that are important to note:

- All players must be under contract before they play in the NFL.
- After four years of NFL service, players are free to negotiate with any of the 32 NFL teams for a contract (known as free agency).
- Anything else?

I compiled data from several sources to get a dataset of salaries from the 2013 to 2016 NFL free agency periods and then performance and concussion events from the 2012 to 2015 NFL seasons. There were 109 quarterback contracts signed in the 2013 to 2016 NFL free agency periods. Since concussion information was available only for the 2012 to 2015 NFL seasons, I considered only the concussion events that occur in a quarterback's final year of his contract. There were nine such occurrences.

I considered only quarterbacks for two reasons. First, there was a relative balance between the number of quarterback hires and the number of teams of the NFL in the sample period: with few substitutes, neither the buyer nor the seller of can have substantial bargaining power over the other. Second, there is a clear measure of productivity for the quarterback position: other positions have a wide menu of duties and/or their contribution is not easily quantifiable.

Consider the individual agent bargaining for his salary with a general manager. These negotiations are bounded by the individual's reservation wage (the minimum value he is willing to take in order to show up for work) and his marginal revenue product (MRP - the player's contribution to the employer's revenue). Denote this boundary as follows:

(1) r_i,t ≤ w_i,t ≤ MRP_i,t

Since the NFL requires players to be under contract before they are eligible to play, the player and the team must negotiate a wage for period t+1 in period t, therefore the agreed upon wage is:

(2) w_i,t+1 = MRP_i,t+1 + rent_i,t+1

where the objective function of both parties is to maximize their respective rents. Note that at the time of the negotiation, MRP and rent are both unknown and therefore some sort of estimated value of future MRP is required, such as:

(3) E(MRP_i,t+1) ≅ E(Performance_i,t+1) = f(Performance_i,t,Age_i,t+1)

where future MRP is approximated by some performance measure(s). Predicting future performance is commonly referred to as estimating the aging curve.

To determine the wage of a player in the next period, I use information on the player at the time of contract negotiations. Combining the concepts of (2) and (3), I run the follow specification using OLS:

(4) ln(wage_i,t+1) = f(Concussion_i,t,ln(wage_i,t),Performance_i,t,Age_i,t+1)

where ln(wage_i,t+1) is the natural logarithm of the average annual salary of the player: the agreed upon wage. The quarterback's win percentage and the passer rating are used as measures of performance in period t. Age is the player's age at the time of signing the new contract and ln(wage_i,t) is the natural logarithm of his current salary. I also consider the year in which the player signed the new contract.

A graphical depiction of the estimation is below. Here I have not yet separately controlled for the concussions in the regression. I plot the fitted values against the actual values. The hollow black circles represent a given quarterback contract. The red circles represent contracts signed following a season in which the quarterback had a concussion. Points above the black 45 degree line indicate the player signed a contract that was below expectations given his observable productivity. It becomes clear that the red dots appear only on or above the black line.

I then estimated the model described in (4). The results of this regression are shown below:

Variable	Coefficient
Concussion	-0.5385***
	(-2.18)
Win%	1.7096***
	(4.25)
Passer Rating	0.0053***
	(2.71)
Previous Salary	0.5182***
	(5.10)
Age	-0.0447*
	(-1.85)
Year FE	Yes
Observations	109
Adjusted R2	0.685
Estimated Concussion Effect	-39.8%
95% Confidence Interval	(-69.8%,-9.8%)

The results are rather shocking - the implied effect of a concussion is found to have a -40% impact on the next salary of the player. However, recall that the above equation (3), and therefore equation (4), would require that the negotiated salary is always at least as great as the player's reservation wage and no greater than his MRP: alternatively stated as equation (1) always holds.

Now consider that there is no feasible bargaining range to satisfy (1). There may be a reason that some players were not offered contracts in the next period and we should be wary of a selection bias. In our sample period, there were 109 quarterback contracts signed but 152 player-contracts had expired. These 152-109 = 43 censored observations were used in addition to the 109 non-censored observations in a two-step Heckman model to remove the potential selection bias. In the first step, I estimate a probit model which predicts if a player will get a new contract using only performance and age. The results of this probit model are used in the second step OLS regression to correct for any potential selection bias. You can read up on this estimation technique here for more information.

Variable	Coefficient
Concussion	-0.5344***
	(-2.15)
Win%	1.7075***
	(4.25)
Passer Rating	0.0060*
	(1.73)
Previous Salary	0.4034
	(5.10)
Age	-0.0263
	(-0.38)
Year FE	Yes
Observations	152
Adjusted R2	0.689
Estimated Concussion Effect	-39.5%
95% Confidence Interval	(-69.5%,-9.6%)

The results of the Heckman model do not change the output therefore the OLS regression likely did not suffer from selection bias. Again, a concussion is found to correlate to a 40% lower salary than would have otherwise been observed but-for the head injury. This is a huge impact but it leads me to several conclusions.

First, although the effect of a concussion is found to be negative and statistically significant, the range of the 95% confidence interval is massive. It is hard to understand what the true effect may be if the range is 60 percentage points. This issue should be corrected with more concussion data becoming available.

Secondly, this may be the first insight that the NFL managers are cautious of concussions. There may be some sort of stigma around a diagnosed concussion wherein a player will miss increasingly more time with each subsequent concussion. This is despite the evidence that individual performance is no different post-concussion compared to pre-concussion.

Lastly, money lost from concussions and injuries may be part of a larger zero-sum game. The teams of the NFL are subject to both salary floors and ceilings - there is a minimum and a maximum amount of money each team must allocate to it's players each year. If one player is paid below their estimated MRP, another player may be given more.

Thoughts? Comments? Thanks for reading.