Fantasy: Using PFF Premium Stats to Win Your League, Part 1

We pour over cheat sheets excavated from deepest recesses of the internet. We buy magazines filled with expert opinion that cost significant portions of our league entry fees. We readily play armchair orthopedist, scrutinizing every pixel of tweeted images of MRI’d hamstrings. Indeed, the diehard fantasy player’s thirst for information is never slaked. Fortunately, the hard-working folks here at Pro Football Focus take great pride in providing you, the diehards (and, hopefully, faithful subscribers), with a Vegas-buffet-sized spread of decadent informational goodness. Yes, several weary souls have actually taken the time to meticulously grade every player on every play of every game,^[1] and as a result of their heroic efforts, we have the Premium Stats section of the website and the spreadsheeted treasure trove of statistical information contained therein.

Of course, as noted fantasy football theorist Albert Einstein once astutely pointed out, information is not knowledge. Simply knowing that Dez Bryant scored a +12.5 pass rating and a -2.1 blocking rating in 2011 is one thing; actually applying that information toward your goal of winning your fantasy league’s 2012 championship is another matter. In this series of articles, I aim to synthesize the myriad team and individual statistics provided by Pro Football Focus (from here on abbreviated, “PFF”) into reliable, actionable, bite-sized chunks of knowledge.

The broad question at the center of each article will be: how do the various premium stats provided by PFF relate to the relevant bottom-line for fantasy players—i.e., fantasy points? We’ll begin answering this question by looking at team-level statistics and team-level fantasy points. Specifically, in this article I’ll be investigating the relationship between PFF’s cumulative by-team offensive stats^[2] and cumulative by-team offensive fantasy points. Or, to rephrase the question: given what PFF is telling us about a team’s offensive performance in the areas of passing, rushing, pass and run blocking, and penalty avoidance, how can we predict future fantasy performance for players on that team?

Important Terminology

Before I begin detailing my methodology and results, I should first define two major analytical concepts that will be making regular appearances throughout these articles: convergent validity and predictive validity. Convergent validity is, to put it simply, the degree to which a measure relates to something it should theoretically be related to. For example, we might have a theory that PFF’s overall team offense rating should be related to that team’s overall offensive fantasy points. If we find there’s indeed a strong statistical relationship between those two measures, we’d say that the PFF overall team offense rating displays high convergent validity. This would indicate that the play-by-play observations made by PFF’s raters (i.e., those who compile the Premium Stats) are picking up on something meaningfully related to the fantasy points bottom-line, and therefore it’s a stat that demands our close attention as fantasy players.

Measures that display high predictive validity are potentially even more useful. Predictive validity refers to the ability of a measure to forecast a future outcome. Going back to the above example, if the 2010 PFF overall team offense rating can reliably predict the team’s 2011 offensive fantasy totals, we’d say the rating has high predictive validity. Simple enough, right? This concludes today’s psychometrics crash course. Now that your eyes are satisfactorily glazed-over, we’ll dive headfirst into the deep end of our data pool.

Method & Results

Pro Football Focus began their current method of statistical tracking back in 2008, which means we now have four seasons worth of fastidiously collected data at our disposal. As mentioned above, the current work will be looking specifically at cumulative team offensive stats across all four seasons. There are six categories of offensive stats in total: Overall, Pass, Rush, Pass Block, Run Block, Penalty. All of these should be fairly self-explanatory (but I’ll refer you here for more details). Although the numerical range varies for the different stats, generally, more extreme negative numbers indicate the team performed poorly in that particular facet of the game while more extreme positive numbers indicate they performed well. I took these stats and compared them to the team’s total fantasy points accumulated through Passing, Rushing, and Receiving.^[3] To calculate a team’s total fantasy points in each of the three categories, I simply took a total of the fantasy points made by any player who passed, ran, or caught the football, respectively, at any point for their team over the course of the season. For example, the Minnesota Vikings 2011 Fantasy Passing Total was 176 thanks to the contributions of Christian Ponder (1853 yards passing, 13 TDs, 13 INTs, 100 points), Donovan McNabb (1026 yards passing, 4 TDs, 2 INTs, 53 points) and Joe Webb (376 yards passing, 3 TDs, 2 INTs, 23 points). The 2011 Carolina Panthers Fantasy Rushing Total of 396.7 came from a combination of Cam Newton (155 points on the ground), DeAngelo Williams (126), Jonathan Stewart (100), Richie Brockel (7.2), Steve Smith (5.6), Josh Vaughan (2.4), and Armanti Edwards (0.5). I could go on, but I’m sure you get the idea.

This leaves us with nine variables per case (the nine variables are comprised of the six stats from Pro Football Focus listed above plus the three categorically separated fantasy totals, and a case is a particular team in a particular season) and 128 cases in total (32 NFL teams × 4 seasons). It’s a data set that’s certainly large enough to identify statistically reliable relationships between variables. The question, of course, is what, if any, relationships exist? To get a big-picture view of this question, I conducted bivariate correlations between the six PFF variables and the three categories of fantasy totals. The results are summarized in the table below.

	Pro Football Focus Cumulative Team Offense Stats (2008-2011)
Team Fantasy Points	Overall	Pass	Rush	Pass Block	Run Block	Penalty
Passing	.548**	.830**	.129	.092	.133	.241**
Rushing	.484**	.196*	.506**	.246**	.418**	.181*
Receiving	.492**	.801**	.108	.074	.077	.222*

If the numbers in the table mean nothing to you, that’s okay—the most important information is contained in the asterisks. If a number has a single asterisk next to it, the asterisk means there’s a statistically reliable relationship between the two variables referenced in the intersecting row and column (for the statisticians out there, it means the correlation is significant at the p < .05 level); and very reliable relationships are indicated by a second asterisk (p < .01). Additionally, generally speaking, the closer a number in the table is to 1 (and the farther it is from 0), the stronger the relationship between the two variables. By examining these numbers, we can start to make some inferences about the convergent validity of PFF’s statistics as they relate to fantasy points.

The Overall stat shows a very reliable positive relationship with all fantasy point categories—as the Overall rating increases, the team generally scores more fantasy points. (If this wasn't true, we’d have a major problem.) The Pass statistic is strongly correlated with both passing and receiving fantasy points,^[4] but it also has a weak, albeit reliable, relation to rushing fantasy points. (Perhaps this lends credence to the offensive philosophy that a team should use effective passing to open up the run.) Rush has a clear correlation with rushing fantasy points but not the other two categories, suggesting the convergent validity for the Rush statistic is particularly strong.

Maybe the most interesting of the findings depicted in the above table are those regarding the Pass Block and Run Block stats. As reasonable intuition would posit, Run Block is strongly associated with fantasy points achieved on the ground, but not air-related points. However, we have a counter-intuitive finding that the pattern is the same for the Pass Block measure—i.e., it’s positively associated with rushing fantasy totals, but not passing/receiving totals. So why might this be the case? Well, perhaps unsurprisingly, Pass Block and Run Block are strongly correlated,^[5] so if a team does one type of blocking well, chances are they do the other pretty well, too. It’s possible that general good blocking is just more important to the running game than it is to the passing game. We’ll return to this notion in a moment.

Now that we’ve established that PFF’s team-level stats are meaningfully associated with on-field fantasy performances (antithetical term alert!), we’ll next explore how we can use these stats to our advantage when gauging player value for the following season. In more precise terms, we’ll see how well stats from a season predict fantasy points the following season. The table below shows correlations between a season’s PFF team stats and the following season’s fantasy totals. This time around, the sample is smaller—by 25%, to be exact—since we have 32 teams and 3 comparisons for each: 2008 to 2009, 2009 to 2010, and 2010 to 2011. Still, 96 cases represent a relatively large data set for this sort of analysis.

	Pro Football Focus Cumulative Team Offense Stats (2008-2010)
Team Fantasy Points the Following Season (2009-2011)	Overall	Pass	Rush	Pass Block	Run Block	Penalty
Passing	.303**	.425**	.043	-.014	.170	.202*
Rushing	.248*	.059	.076	.259*	.213*	.107
Receiving	.245*	.420*	.038	-.072	.114	.146

A few things immediately stand out when looking at these numbers. First, the correlations are generally much lower than they were when we previously looked at the relationship between PFF stats and fantasy performance within a season. This is to be expected—the PFF measures are designed to capture events happening within seasons, not across them. Second, Pass is the strongest cross-season correlate. Considering quarterback is probably one of the more stable positions in the NFL, this also makes perfect sense. We expect relatively consistent performances from the likes of Tom Brady, Drew Brees, and Aaron Rodgers.

Finally, the most surprising and potentially informative finding revolves around the predictors (and non-predictors) of fantasy rushing totals. I was certainly quite surprised to find that the Rush stat, despite being strongly associated with fantasy rushing within a season, turns out to be a poor predictor for fantasy rushing the following season. This could be due to a number of reasons. Like all the team-level stats, these numbers are blind to the players they’re built upon, so they aren’t sensitive to season-to-season changes in personnel—something that could easily affect a team’s running game given the short shelf-life of NFL runners. Moreover, running back performance is (anecdotally) more volatile compared to other positions, and season-to-season performance just may be harder to predict than, say, quarterback or wide receiver.^[6]

On the other hand, the fact that both Pass Block and Run Block are predictive of rushing performance the following season is extremely educational. For starters, it adds support to our working theory that pass blocking and run blocking are indicative of a more general “goodness of blocking” factor, and that this factor is important to a team’s ability to run the football. To further test this idea, I averaged each team’s Pass Block and Run Block stats and used that number to predict fantasy rushing totals the next season. While either blocking measure can reliably predict rushing performance on its own, using this Pass/Run Block Average (PRBA) metric as the predictor moves the correlation between blocking and rushing points into very reliable territory.^[7] This relationship between PRBA and the following season’s fantasy rushing totals is shown in the graph below.

PRBA as a Predictor of Rushing Points

The red line going through the data points represents the line of best fit through the data (i.e., the regression line).^[8] In other words, it shows what our “best guess” would be for a team’s rushing points for the following season given its PRBA from its current season. So, for example, if a team averaged around zero in its blocking performance, we’d expect, on average, for them to score about 267 fantasy points rushing. As you can see, there’s considerable variation around that line; it's by no means a perfect predictor. That being said, there’s also clearly a positive linear relationship between the predictor and the outcome. In fact, there's a 99.4% chance that the linear relationship we're observing between the previous season's PRBA and rushing points is meaningful (and not random). For you stat nerds out there, the r² value for the line of best fit is .076. This means that 7.6% of the variation in rushing points is accounted for by our regression line.

You might look at that number and say, “Okay, so what? A predictor that accounts for 7.6% of the variation doesn't seem that useful.” You would perhaps have a valid point until you consider the following: (1) previous year's OL performance, as described by PRBA, is one of any infinite number of factors that can affect rushing points (other factors may include, e.g., the skill of the team's RBs, personnel changes, off-season playbook changes, the time of the game, the left guard's birthday, what the team's quarterbacks coach had for breakfast that morning, etc.), and even if we don't account for another one of those infinite other factors, our single PRBA number still explains 7.6% of the rushing output the following season; and (2) the previous season's PRBA is our best predictor among other plausible team-level predictors by a fair margin. For example, a team's Overall rating, a global measure of how good the offense is, only accounts for 6.2% of the variation in rushing points the following season. The Rush stat, a global measure of a team's rushing ability and a stat we might reasonably expect to predict the next season's rushing points, is a terrible predictor, only accounting for 0.5% of rushing-point variance. In other words, PRBA, a stat based on performances that are (mostly) independent of the players racking up the rushing totals, is fifteen-times better at predicting rushing points than Rush, a stat specifically tracking rushing performance. That's pretty incredible, right? Even if PRBA isn't a great predictor of rushing points in absolute terms, it's still clearly a number that demands our attention. The takeaway is this: as a rule of thumb,^[9] good blocking performance in a given season leads to a good fantasy rushing output the following season.

Discussion

So what does this mean to us as we prepare for the upcoming season? My recommendation is that you add PRBA to the list of factors to consider when making roster decisions about the running back position. I'm not suggesting you give it top priority by any means, but if you’re deciding whether to take Running Back A or Running Back B with a critical draft pick, then, all other things being relatively equal, take the guy whose team Pro Football Focus rated as better blockers the previous year. To give a concrete illustration, let’s say you had a late first round pick in last season’s draft and were trying to decide between Rashard Mendenhall and Maurice Jones-Drew as the cornerstone of your team.^[10] The younger Mendenhall was coming off a career season in 2010; the veteran Jones-Drew was coming off a bit of a down one (by his high standards). If you looked at the team’s respective PRBAs for 2010, -70.5 for the Steelers and -10.3 for the Jags, the decision would have been (a) easier (Jones-Drew's blockers look considerably better), and (b) the right one (Jones-Drew outscored Mendenhall, 209 to 147, on the ground in 2011).

This knowledge could also be used to determine which breakout performances by runners have the best chance of being replicated the following season. Heading into the 2011 season, many experts projected LaGarrette Blount as a solid #2 RB in standard leagues^[11] after he came on strong at the end of 2010, finishing with over 1000 yards rushing and 18^th overall in ground production amongst RBs. Those high expectations may have been tempered after considering the 2010 Bucs finished with an awful PRBA of -68.35 in 2010. (Blount went on to finish 25^th in ground production amongst running backs in 2011 despite having practically no competition for playing time.)

In addition to helping us gauge the repeatability of strong performances, team blocking performance may assist in identifying potential diamonds in the rough to scoop up with your (frequently league-swinging) middle-round picks. The 2008 Baltimore Ravens, for example, failed to produce a 1000-yard rusher, but their PRBA (20.85) ranked in the league’s top ten for the season. Owners who selected Ray Rice (a second-year player who was a relatively unknown quantity at the time after a mostly quiet rookie season) in the middle rounds of their 2009 drafts likely made out very well that year after Rice finished with 1339 rushing yards and 7 touchdowns.

Summary

Pro Football Focus proudly provide a glut of information to satisfy even the geekiest, stat-hungriest of their readers. The point of this series of articles is to begin distilling this information into nuggets of wisdom that can be used to your advantage in drafting and managing your fantasy teams. The take-home point from today’s study: when targeting a running back for drafting, pre-season trading, keeping/throwing back, etc., pay careful attention to his team’s blocking performance—not just in the running game, but pass blocking as well. In the next article in this series, I’ll apply these same team-level analyses to identifying trends in IDP performance. Future installments will look at individual-level statistics as predictors of individual and team fantasy performance, so stay tuned. In the meantime, I’ll leave you with this food for thought: Marshawn Lynch ranked fifth in points on the ground as Seattle’s workhorse last season. So where did his team finish in PRBA in 2011? Fifth-worst (-56.15). If you’re in my league, nothing would make me happier than watching you spend that second-round pick on him. I’ll gladly take Mark Ingram and his #1-in-PRBA Saints teammates five rounds later.

[1] It’s not just an empty motto, people. The details of the PFF methodology can be found here. Go back.

[2] If you’re not yet a subscriber, you can check out the 2008 sample stats for free by registering here. Go back.

[3] These totals are based on standard scoring (i.e., Passing = 1 pt. per 25 yds. passing + 4 pts. per TD – 2 pts. per INT; Rushing = 1 pt. per 10 yds. rushing + 6 pts. per TD; Receiving = 1 pt. per 10 yds. receiving + 6 pts. per TD). No points were subtracted for fumbles because (a) it couldn’t be determined whether the fumble occurred on a rush or pass play and (b) fumbles recovered by a player’s own team weren’t tracked in the data set used. Go back.

[4] Passing and receiving fantasy points would naturally be highly correlated with each other as well—a pass that nets fantasy points for a passer typically nets fantasy points for a receiver. Go back.

[5] Correlation coefficient = .331**. Go back.

[6] Look for future installments of this series to address this issue in deeper detail. Go back.

[7] Correlation coefficient = .276**, r²= .076. Go back.

[8] For those interested, the exact equation for the line is: Rushing Points Expected = 266.975 + .553(PRBA). Go back.

[9] That is, a decision-assisting heuristic, not to be confused with a hard-and-fast Law of Fantasy Football like… well… are there any? Go back.

[10] Considering many fantasy football publications had these players closely ranked going into the 2011 season, I feel this is an appropriate example. Pro Football Focus, for instance, gave Mendenhall and Jones-Drew preseason RB rankings of #6 and #7, respectively. Go back.

[11] Pro Football Focus pegged Blount as the #15 preseason RB in 2011. Go back.