Projecting Upside: The Predictability of Receiving Touchdowns

Fantasy Football Pocket Drafting

If targets are the lifeblood of fantasy scoring and the addition of air yards helps us explain 80 percent of a wide receiver’s receiving yards, then volume can be viewed as the force that drives opportunity to collect fantasy points. But touchdowns act as the fantasy point multiplier. Therefore, it’s valuable to determine what is and what isn’t predictive of touchdown scoring.

Methodology

Wide receivers from 2008 to 2017 with at least 50 targets in both Year N and Year N+1 are included in this study.1 Data was collected using the RotoViz Screener.

How Predictive Are Receiving Touchdowns Year-to-Year?

Most of us intuitively know that TDs aren’t predictive year-to-year. On average, there are 1.54 passing TDs per team per game2 and only a handful of players even reach double-digit scores. The small sample size makes it difficult to justify using TD output as a reliable stat for predicting future TDs. And the data supports this logic:

WR Receiving Touchdowns Year-to-Year Correlations

Variable Year N Variable Year N+1 R-squared
Total reTDS Total reTDS 0.079
Red-Zone reTDS Total reTDS 0.029
10-Zone reTDS Total reTDS 0.018

The R-squared values are all close to zero, indicating that total, red zone, and 10 zone receiving TDs do not correlate with next-season TDs.3 Because touchdowns are not predictive year-to-year, we need to find stats that correlate with next-season touchdowns.

Explaining In-Season Receiving Touchdowns

To determine what stats will be useful for predicting next-season touchdowns, first we have to determine what stats correlate with in-season receiving touchdowns.

WR In-Season Receiving Touchdown Correlations

Explanator VariableResponse VariableR-squared
Red Zone reTDSTotal reTDS0.674
Total reTDRTTotal reTDS0.657
10 Zone reTDSTotal reTDS0.468
Red Zone reYDSTotal reTDS0.463
Total reYDSTotal reTDS0.421
Red Zone reTRGSTotal reTDS0.391
10 Zone reYDSTotal reTDS0.352
10 Zone reTRGSTotal reTDS0.322
Total reTRGSTotal reTDS0.295
Red Zone reTRGMSTotal reTDS0.25
Total reTRGMSTotal reTDS0.246
10 Zone reTRGMSTotal reTDS0.202
Red Zone reSRTotal reTDS0.105
Total reSRTotal reTDS0.081
Red Zone reCRTotal reTDS0.064
10 Zone reSRTotal reTDS0.052
10 Zone reCRTotal reTDS0.032
Total reCRTotal reTDS0.031

Findings:

  • Moderate Correlation:
    • Total reTDRT: Receiving TD rate is just shy of being statistically strong for correlating with in-season touchdowns.4
    • Red-Zone reYDS: This one isn’t surprising because the majority of TDs are scored in the red zone, and scoring a TD generates the maximum receiving yards on that given yard line.
    • Total reYDS: Total receiving yards explain 42.1 percent of the variability in TDs.
    • Total/Red-Zone/10-Zone reTRGS & Total/Red-Zone reTRGMS: Volume is king in fantasy football. And volume also helps drive TDs (moderately).
  • Not Useful
    • Red-Zone/10-Zone reTDS: While 67.4 percent of in-sample WR touchdowns are scored in the red zone, earlier we already saw that touchdowns don’t correlate year-over-year.
    • 10-Zone reYDS and 10-Zone reTRGMS: Red-zone reYDS and reTRGMS provide greater strength.
    • reSR and reCR: It was worth a shot to look at, but … no dice.

How Sticky Are These Stats?

Finding a stat that correlates with same-season touchdowns is useful for predicting next-season touchdowns if that stat is also sticky year-to-year.

Proxy Stat Year-to-Year Correlations

Variable Year NVariable Year N+1R-squared
Total reTRGSTotal reTRGS0.296
Total reTRGMSTotal reTRGMS0.29
Total reYDSTotal reYDS0.245
Total reCRTotal reCR0.178
Red Zone reTRGMSRed Zone reTRGMS0.153
Red Zone reTRGSRed Zone reTRGS0.146
Total reSRTotal reSR0.097
10 Zone reTRGMS10 Zone reTRGMS0.086
Red Zone reYDSRed Zone reYDS0.075
10 Zone reTRGS10 Zone reTRGS0.075
Total reTDRTTotal reTDRT0.013
Red Zone reCRRed Zone reCR0.011
Red Zone reSRRed Zone reSR0.004
10 Zone reYDS10 Zone reYDS0.004
10 Zone reTDRT10 Zone reTDRT0.004
10 Zone reSR10 Zone reSR-0.001
10 Zone reCR10 Zone reCR-0.001
Red Zone reTDRTRed Zone reTDRT-0.002

Findings:

  • Sticky: None
  • Moderately Sticky
    • Total reTRGS/reTRGMS/reYDS: Not restricting the field position provides a larger sample size, and stickier stats.
  • Somewhat Sticky
    • Total reCR: Unfortunately, we already know that catch rate doesn’t correlate with TDs.5
    • Red-Zone reTRGS/reTRGMS: Considering 67.4 percent of receiving TDs came in the red zone and volume is king in fantasy football, the fairly weak red zone target stability is disappointing.
      • For those who think elite WRs have stickier red-zone opportunity: After restricting the data to only include WRs with 100 total targets and 10 red-zone targets year-to-year, the red-zone target year-to-year correlation was 0.046.
  • Not Sticky (Bouncy?): Theme – small sample sizes aren’t sticky.
    • Total/Red-Zone/10-Zone reTDRT: While touchdown rates strongly correlate with touchdowns, they have a near zero year-to-year correlation.6
    • 10-Zone reTRGMS/reTRGS: Not surprising given the small sample size in the 10 zone.
    • Red-Zone/10 Zone reYDS: Red zone yards had a promising in-season correlation with TDs, but the lack of year-to-year stability will likely render it useless for predicting next-season TDs.
    • Red-Zone/10-Zone reCR & Total/Red-Zone/10-Zone reSR: Still meaningless for predicting TDs.

What Stats Predict Touchdowns Year-to-Year?

We want to be predictive, and not just descriptive, so the hypothesis now becomes: if X correlates with in-season touchdowns AND X is sticky year-to-year, then X should predict next-season receiving touchdowns.

Variable Year NVariable Year N+1R-squared
Total reYDSTotal reTDS0.109
Total reTRGSTotal reTDS0.091
Total reTRGMSTotal reTDS0.072
Red Zone reTRGSTotal reTDS0.063
10 Zone reTRGSTotal reTDS0.04
Red Zone reYDSTotal reTDS0.033
Red Zone reTRGMSTotal reTDS0.025
Total reTDRTTotal reTDS0.019
10 Zone reTRGMSTotal reTDS0.018
10 Zone reYDSTotal reTDS0.014
Total reSRTotal reTDS0.007
Red Zone reTDRTTotal reTDS0
Total reCRTotal reTDS-0.001
Red Zone reSRTotal reTDS-0.002
Red Zone reCRTotal reTDS-0.002
10 Zone reTDRTTotal reTDS-0.002
10 Zone reSRTotal reTDS-0.002
10 Zone reCRTotal reTDS-0.002

Findings:

  • No individual stat is a strong predictor of next-season TDs.
  • Knowing that targets are the lifeblood of fantasy scoring and that red zone targets explain 39.1 percent of in-season TDs, it’s logical to think red-zone targets can be used to predict next-season TDs. However, a 0.063 R-squared value indicates red-zone targets are, on average, not good at predicting next-season TDs.

Using multiple linear regression to determine if any combination of these stats are more predictive than using them individually, two combinations showed slight improvement:

Variable Year N Variable Year N+1 Adj. R-squared
Total reTDS + Total reYDS Total reTDS 0.115
Total reTDS + Total reTRGS Total reTDS 0.11

Even with the relative increase in correlation strength, the adjusted R-squared values indicate the combinations are only weakly predictive of next-season TDs.

There was no significant difference when I only included WRs who remained on the same team year-to-year.

Conclusion: Touchdown Regression

Before digging into this data, I thought some combination of total and red-zone opportunities could improve receiving TD projections. However, it’s evident that it’s really difficult to reliably predict next-season touchdowns for WRs.7 Therefore, we can:

a) Improve touchdown projections on a case-by-case basis using information that’s difficult to control for in a study like this, and/or

b) Embrace receiving touchdown variance and pinpoint players who are due for touchdown regression.

In order to find touchdown regression candidates, we can use total reTRGS and total reYDS because these stats are moderately sticky year-to-year and correlate with same-season touchdowns (explaining 42.6 percent of TDs).

Screen Shot 2018-05-18 at 11.02.47 AM
Expected reTDS = 0.139 + (-0.015 * Total reRTGS) + (0.008 * Total reYDS)

Notice that the estimate for targets is negative. In other words, if two receivers have an equal number of receiving yards, but one has more targets, the receiver with fewer targets should actually be expected to have more touchdowns. This is somewhat surprising, but also intuitive, because, as mentioned above, a touchdown represents the maximum yardage that can be gained from each yard line.  This means that, all things equal, WRs with higher yards per target on average score more touchdowns than WRs with lower yards per target.8 Essentially, yards are more indicative of expected touchdowns than targets. Therefore:

  • For positive touchdown regression: look for WRs with a lot of yards, but few touchdowns.
  • For negative touchdown regression: look for WRs with more touchdowns than their total receiving yards would predict (even if they have a lot of targets).

It’s important to remember that regression doesn’t mean a WR will overcompensate for their previous efficiency (or lack thereof), but rather that they will likely return to (near) league average. Also predicting regression is only useful if we can expect that WR to sustain the same pace of yards and targets, which means things that this model can’t account for like depth chart or coaching changes need to be taken into consideration.

  1. Fifty targets are an arbitrary number I chose with the hope of only including fantasy relevant players with sufficient sample size. A total of 530 WRs met this criteria. I also ran the data for the last five seasons to account for potential shifts in offensive philosophy, but 1) I didn’t see a significant difference in the result, and 2) the sample size was cut in half.  (back)
  2. Over the last three years, there have been 2,369 passing TDs, which is 24.7 per team or 1.54 per team per game.  (back)
  3. I tested changing the inclusion criteria from 50 targets to a more stringent 100 targets and 10 red zone targets year-to-year in order to include only top-end WRs, but I found that all correlations were even weaker.  (back)
  4. Keep in mind that we’ve controlled for volume by requiring a target threshold. TDs are simply the product of targets and TD rate, so it makes sense that TD rate would correlate with TDs at a certain minimum volume level.  (back)
  5. Interesting aside: catch rate becomes stickier the more you increase the target threshold.  (back)
  6. I also manipulated the sample to only include players with 10 as well as 15 red zone targets year-to-year, and the r-squared values were near zero.  (back)
  7. There are still variables that should be tested, such as career quarterback touchdown rate, among others.  (back)
  8. Even though yards per target itself isn’t especially sticky (R-squared of 0.061 in this data set), both yards and targets are.  (back)