The general take-home message of the wide receiver (WR) reliability analyses I’ve reported so far on I//R is that per-route stats are more reliable indicators of a WR’s “true” ability than are per-target stats. For specifics, below is a table summarizing my findings and translating them into regression to the mean (RTM):
|Stat||Unit of Obs||Stabilization Point||% RTM after 200 Obs|
|YPT||Targets||205 (1,027 Rtes)||50.6% (83.7%)|
It’s pretty straightforward, but read the table like so: Targets per Route Run (TPRR) is observed on a per-route basis, and it takes 184 observed routes for a WR’s current TPRR to represent 50% “true” ability and 50% randomness. For the purposes of application, the last column tells us that, if a WR’s run exactly 200 routes, we need to regress his current TPRR 47.9% of the way towards the league average.1
Elsewhere in the table, we see the main evidence supporting my earlier conclusion: Yards per Target (YPT) takes nearly thrice as long to stabilize as does Yards per Route Run (YPRR). And perhaps the biggest indictment of YPT is that, despite touchdowns being highly susceptible to randomness, we learn about a WR’s “true” Touchdowns per Route Run (TDPRR) more quickly than we learn about his “true” YPT.
Given that tight ends (TEs) are also an integral part of a team’s receiving corps, the logical next step is to see whether or not they produce the same pattern of results exhibited by WRs. Therefore, today I’ll answer the question, “When do YPRR, TPRR, and YPT stabilize for TEs?”
Aside from changing positions, my methods for this TE reliability analysis were exactly the same as what I did for WRs. If you’re new to I//R and want step-by-step details, click here. Also for space considerations, I’m not going to explain how to read the results tables, which should be familiar to regular readers by now. Again, new readers should click the just-one-sentence-ago link if the tables below — or my comments about them — make no sense.
Let’s start with YPRR:
|Games||n||r||R2 = 0.50||Avg YPRR||Obs 1.75 YPRR|
For WRs, I previously found that YPRR stabilizes at 14 games, which translates to 351 routes run given that an average WR runs 26.1 routes run per game. From the table above, we see that YPRR stabilizes at 17 games for TEs. Although that’s slightly longer in terms of games, the fact that an average TE runs only 18.2 routes per game means it actually takes slightly fewer routes run (318) for a TE’s YPRR to stabilize than it does for a WR’s. Either way, though, the results are similar by and large.
Next up is moment of truth for YPT:
|Games||n||r||R2 = 0.50||Avg YPT||Obs 8.50 YPT|
BAH GAWD! YPT stabilizes at 52 games for TEs, which is 13 more games than it takes for WRs. However, because the average TE runs fewer routes per game than the average WR, as well as being targeted fewer times per game (3.3 vs. 5.2), the per-target (172) and per-route (943) conversions slightly favor TEs. In terms of application, the general idea here is that a receiving TE in a pass-happy offense (e.g., Jimmy Graham, 28.9 routes run per game in New Orleans) will see his YPT stabilize much faster than a blocking TE in a run-happy offense (e.g., Luke Willson, 11.8 routes run per game in Seattle).
That said, there’s a statistical red flag in the YPT results table that I have yet to mention in these reliability articles because (a) I haven’t come across it until now, and (b) it’s a really esoteric point. Namely, the split-half correlations that form the basis of “stabilization points” in this method are all over the place. In contrast to most every other reliability results table I’ve posted so far, the “r” column in the above table doesn’t follow a steady linear progression. It’s essentially unchanged from the initial 8+ games group (i.e., Row “4” in the table) all the way through the 32+ games group (i.e., Row “16”).
More to the point, even after things start to resemble the typical pattern in the 40+ games group (i.e., Row “20”), I found wild variations between the 25 random iterations that the method requires. For instance, in the 56+ games group (i.e., Row “28”), the lowest split-half correlation in any iteration was .086, whereas the highest was .730. To put that in context, the same group had an “r” range from .648 to .860 for YPRR, and even the range for TDPRR was narrower than that of YPT (.178 to .648). All of this suggests that YPT for TEs is exceptionally volatile — more volatile than any metric I’ve looked at so far for members of a team’s receiving corps.
|Games||n||r||R2 = 0.50||Avg TPRR||Obs 20.0% TPRR|
I previously found that TPRR stabilizes at 7 games for WRs; here TPRR stabilizes at 13 games for TEs. Unlike YPRR and YPT, our conclusion about TPRR from games played is consistent with our conclusion from routes run: It takes more routes run for a TE’s TPRR to stabilize (228) than it does for a WR’s TPRR to stabilize (188).
DT : IR :: TL : DR
Overall, as was the case with WRs, per-route metrics for TEs seem to be more reliable than per-target metrics. Furthermore, two of my findings about reliability — and therefore RTM — in the context of TEs vis-à-vis WRs are particularly interesting: (1) Because TEs run fewer routes per game than WRs, the difference in stabilization points for the two positions is minimal; and (2) the lack of confidence I have in YPT for WRs dwarfs the lack of confidence I have in YPT for TEs.
The math here is 184 / (200 + 184) = .479, or 47.9%. ↩