One Last Time with Team Interceptions, Part 4

 One Last Time with Team Interceptions, Part 4

SOM PRO FOOTBALL LINKS

So, the question came up- why not generic team charts, like/ similar to batting cards?

Let’s look first at the problem batting cards were trying to solve.  Take a Sixties baseball team and the pitchers (in sum) will bat about as much as 2/3rds of a regular, and mostly not very well.  But you get exceptions.  When you try to cut a team’s total pitcher batting into the individuals you eventually get a handful of relief pitchers who banged a triple in five at bats, or a homer in ten at bats, and unless you were Earl Wilson in 1968 a personal card reflecting this this is somewhat unrealistic. 

Most pitchers stink when batting so (other than personal cards) there are two ways to solve this – you can aggregate the whole team (a method I use in Statis-Pro Baseball, which is going to have the effect of grouping a pitcher with no power like Mickey Lolich in with Earl Wilson) BUT it is just easier to manage and works for probably 25 out of 28 teams.   Or you can group the pitchers by type.  This is what Strat-O-Matic did, first with four pitcher’s batter’s cards, and then with eight during the baseball card game’s Golden Age.   

They chose five cards for pitcher with no power, and three for pitchers with power:

IF NO HR'S OR AB/HR>=68 

Bavg           Card 

.000 - .134      1 

.135 - .179      2 

.180 - .224      3 

.225 - .269      4 

.270 +             5

 

 

IF AB/HR < 68 

Bavg           Card 

.000 - .199      6 

.200 - .249      7 

.250 +             8 

 

One look and you can see the first five cards are probably reasonably spaced; this is probably not a bad way to model pitchers with no power. No, it won’t accommodate pitchers with a lot of walks or extra base hits past a certain point but those are rare anyway.  It is the last three cards that bother me, that range for Card 6 is from truly lousy pitcher’s hitters who catch hold of one (here’s looking at you, Bartolo Colon) to pretty good hitters who might have reasonable power, like Dave McNally or 1972 Bob Gibson.  That range is too broad for Card 6.  Now- does it matter? 

Well SOM baseball fan Flying Pickle actually put all pitcher’s batting cards in a spreadsheet about fifteen years ago, and here were the results:

 

Figure 1: Pitchers Batting Cards, 1901-2008

There aren’t that many Card 6 guys anyway, and what would they get? For a number one starter maybe 40-50 AB?  Cards 6-8 are less than seven percent of all pitchers who bat.  The question of whether there ought to be ten or twelve pitcher’s batting cards (more than eight) is mooted by the fact these players are peripheral to their team’s overall offensive experience.   Most players will want the simpler answer or the most complex, i.e.  individual cards.

And that brings us to Team Interception Returns.  I’m pleased that I finally got the last two teams from 1980 to finish my Wideboys collection; this final event was over a year in the making.  I might be poorer, but I sure have a smile on my face, because I’ve been horsing around now for a day or two with this season’s cards in my off time.  One might say I have 1980 on my mind. Two of my favorite teams are 1980 LA and 1980 Dallas, the Wild Card runts of the 1980 NFC despite good records.  1980 LA is an offense first team with one of the team’s weaker run defenses of the last decade, they finished with a pretty good turnover plus minus but with terrible kicking.  They still got 405 points even with Haden in for fifty attempts, more than ten percent.  The Cowboys led the 1980 NFL in scoring in spite of having the ninth best offense in yards. Dallas had Tony Dorsett push for 11 TDs and 1185 yards on the ground while five other runners chipped in 991 yards and another ten TDs. Dallas scored like this year’s Detroit Lions, but in one less game, and the Rams were not far behind, even though neither team made a lot of kicks. What was setting them up?

Aha!  The 1980 Season’s interception returns helped to define the 1980 season!

Figure 2: A sampling of team interception returns by team from the 1980 season:

The World Champions and two of the ten playoff teams are in the top yardage tier, teams that averaged more than thirty yards of interception returns a game.  The next tier is limited more by opportunities, as Denver would have led the league in yardage had they matched the Oakland interception total.  Still, we see another three playoff teams there, and only one team, Philadelphia in the bottom tier.  This is not uncommon.  LA and Dallas might have been 10-6 type teams had their returns not elevated them to 11-4 (with Ferragamo) for LA and 12-4 for Dallas. Their team returns were not peripheral to their success, they were the key to it. Another way to look at it is Atlanta, also 12-4, was plus sixteen in turnovers and had great special teams, a plus kicker in Mazzetti, a good punter in James, good coverages, and a top three punt return unit.  All that offset an iffy pass defense. Why shouldn’t Dallas and LA get the same boost from their secondaries?

One last point I’ll make is the seasonal record even at first blush for each team holds a lot of information.  Let’s look at 1980 Dallas- the first rung of data that you see in PFR:

Figure 3:  1980 Dallas Return Results First Pass from PFR: (Click to make bigger)

So, we have the longests.  And if a guy has two returns with a longest of 2 yds that means the other return is a zero.  If a guy has 3 returns for 56 yards with a longest of 56 yards the other 2 are zeros.  Just at first glance you have half of the returns and you can guess they might be “chunky”.

Looking up the weekly records by player for Thurman, Waters, Breunig, and Wilson finishes the record.  Thurman’s best day saw him get two for 42 yards with a longest of 35.  That means the other is seven yards.  Here’s how it fills out:

Figure 4: 1980 Dallas Final Record

And this is the Team Card from the Game:

Figure 5: 1980 Dallas Returns

I could see maybe adding in that 56 yarder in the twelve spot and cutting the +12 yard result in the seven spot to +9 but that increases the fractional TD pct by a few points, so I didn’t do that.  You can see the chunky nature of the team’s returns in the nine 30 (or so) yard chances they have out of 36.   In the final analysis once you get this close it’s not going to matter, Mike might do better with his reckoning, but it’s a pretty good chart. Taking into account returns “not returned using the rule for that” for 0 yards this chart is manifestly better at describing 1980 Dallas than the stock chart.

We’ll put one last chart up there, 1980 Seattle:

Figure 6: 1980 Seattle Seahawks Team Card (23/95 yards, 0 TDs, 4.1 yds per return)

I’ll let gamers be the judge as to whether or not a generic chart is a better fit for this team’s efforts.  In my opinion, though, team interceptions are more like running cards then baseball’s pitcher batting cards, where there can be many ways to get to the final rushing stats line. If there weren’t we could just use the extra runners from the set of extra player cards and change a few results to make a season.  And, of course, we don’t do that. We can do better than that.

 

Fred Bobberts

Initial Date of Publication: 2/10/2024

PS - I'm not going to spend one more minute of my time on this.  If you don't like the feature, turn it off.  I'm done feeding the trolls.

The (Second to the) Last of Three (no, Four) parts on the Team Interception Feature in Strat-O-Matic Pro Football.

 The second to the Last of Three oops four parts on the Team Interception Feature in Strat-O-Matic Pro Football.

 

 The real-life 1970 Lions scored 4 TDs on 28 interceptions.  Is this better or worse than expected?

SOM PRO FOOTBALL LINKS

“Sometimes, the season itself is the outlier”

 –Scott Everngam

 In the previous installments, we explored the design intent of the feature, and the challenges in evaluating the raw interception return data a season might produce. We’ve shown the feature is pretty good at distributing the most return yards to the most threatening teams; in as few as six replays real trends start to appear that mirror real life team results.  We’ve show it is an improvement over the stock interception chart.  But we haven’t tackled the final objection- why is it a season like 1975, which had 533 interceptions but only 25 interceptions returned for TD, or 1977, which tallied 562 interceptions but only 28 returned for TD will come out consistently high when using this feature? 

Well to look at this, I created a graph of every season’s Real Data from 1956 to 1981, effectively bracketing these years, plotting the season’s overall interception return yardage average and pct of overall interceptions returned for TDs.  And while the fit is not perfect, it is not exactly a random walk, either.  At the season level, the best fit for TD pct vs return average over the range of data we see is a second order power law, implying a fairly strong push upward in “house pct” as the average yardage per return increases.  And here it is:

Figure 1: Season Level Real Life Data for Average vs TD PCT, 1956-1981

This is all real NFL Seasonal data compiled from each team’s data underneath it, except for the 1975 replay data point the original chart data points in red.  The original chart was developed in 1968 for the game’s initial release and very much is a product of its time; it reflects the five seasons in the mid-Sixties that surround it.   1975 and 1977 in real life were below the curve of expected TDs their yardage would have normally produced, with 1975 in particular about 1.8 pct lower than predicted for its 14.5 yards per return.   Over 533 interceptions this would produce a deviation of 8-10 interceptions versus a typical such season over NFL history.  The effect in 1977 is less pronounced but probably still about 7 interceptions low.  The Fifties teams and quite a few mid Seventies teams are low outliers, while the mid to Late Sixties trend high.  Sometimes the real data from the season itself is the outlier, as Scott observed once while creating CMs.

It’s a pat answer but I haven’t explained why this is true. It’s a heck of a lot of data, if this were a normal distribution 26 seasons and 13,095 interceptions should produce a pretty nice chart, not the messy correlation we see above.  But the overall data set is neither continuous nor normal; the y axis response is low frequency binomial (yes/no) data, and each season is a collection of subsets of team data that lie beneath its sums.  If you look at the team data over the same period, you see it has some very interesting properties:  

Figure 2: Team Level Data, Aggregated by Range

There are no seasons that sum out as low as the lowest team ranges or as high as the highest team ranges.  Seasons are populations of teams. At the team level both the TD percentage and the percentage of 0 TD returns in the sample range are strongly correlated to the average yards per return. The latter case is in the inverse, as teams with interception return averages less than 12.0 will fail to get even one TD more than half the time.  It’s important to note that even real-life teams that had very high return averages might fail to tally even one TD in one out of eight (or so) cases.

Another way to look at it is 18 pct of teams had 17 yds or greater per return; these teams accounted for 25 percent of the total yards and 30 percent of the total touchdowns during this interval.  Six out of seven of them with have at least one TD, as opposed to two thirds of the total.

I keep harping on these high average teams because they help to explain why certain higher average return seasons have such high variability.  Season such as 1975, which averaged in the mid 14 yards per return range do not have every team as a high performer.  Instead, a higher seasonal average is more an indicator that there are also some high potential teams mixed in with the rest of a typical population, rather than a guarantee of high return touchdown percentages.  If we return to the original chart, we see three teams had similar overall return avg stats but different TD percentages (again real-life data):

Figure 3: Team Level Data for three NFL seasons in the mid 14s for average yard per return

  

That’s quite a range of TDs and House Pcts from a narrower range of inputs, but actually all three are possible based on the underlying data, which is decidedly not continuous nor normally distributed.  There is no guarantee that a team with a high return average will meet its potential and no guarantee that a team won't uncork a TD when the rest of the real-life teams with similar averages mostly didn't. 

Looking at 1975 we see that other than Baltimore that key team population at/or/over 17 yards a return is not carrying its weight in TDs, with four teams out of the top eight not getting any at all.  That number should be one out of eight when considering NFL History, so this type of distribution will happen about 1/80 times. Unlikely, but not impossible.

Figure 4: Team Level Data for 1975

Now we look at the middle season, 1971.  Led by Houston the top teams get 12 return TDs in 115 returns.  Washington, which is in the next range subpopulation down, also chips in a nice year:

Figure 5: Team Level Data for 1971 

And, finally, the high field, 1966.  Here the top teams are robust and are also supported by good seasons from Miami, Buffalo, and LA.  This outcome is also merely unlikely, but not impossible:

Figure 6: Team Level Data for 1966  

In summation, some corrections on team charts can be made to try to tame high observed interception return TDs. But there are limits on how far you can go, and to how much these adjustments will translate in seasons with higher return averages. Throughout NFL history, it can be shown that there is a strong correlation between team interception return average and the percent of these interceptions returned for touchdowns.  And the goal of the feature is to closely match relative return yardages within a season, if fed the same number of interceptions per team. 

It is also possible that the season itself is an outlier when its combined team population's results are compared to all teams in NFL History.  One must consider what the predicted outcome is likely to be, not just a deterministic idea of what a statistic should be based only on a value, and not within its context.

Note: the original questions on the Forum concerned 1977, which is why I mentioned it.  But, unfortunately, I don’t own 1977 to test in detail, so I chose 1975 to illustrate the point.  Teams like 1975, 1977, 1979 and 1981 are all somewhat analogous, while the mid to late Sixties are also analogous on the high side.

Fred Bobberts

Original Publication Date: 2/4/2024