clock menu more-arrow no yes

Filed under:

The Myth of The Streaky Shooter

New, comments

The Sixers Science Team examined if there’s any truth to the claim that Robert Covington is a streaky shooter.

NBA: Detroit Pistons at Philadelphia 76ers John Geliebter-USA TODAY Sports

He’s Streaky, Duh!

Ben Detrick (The Ringer/NY Times): Covington is a streaky shooter, but the defense is real and it’s spectacular.”

Andrew Porter (WIP Sports Editor): “Covington is who he is — a streaky shooter who plays solid D.”

Marshall Harris (NBC Philly): “Covington’s streaky shooting has been an issue to say the least.”

UPDATE: We created an interactive shooting log for Covington. Click HERE if you want to play around with it.

All of the verified Twitter personas above share the opinion that Robert Covington is a streaky shooter. Additionally, 92% of the voters in our poll agreed with this sentiment.

Referring to Covington as “streaky” is like referring to 90-degree ankle dislocations as “gruesome”; it’s just the word that everyone uses by default.

Anyone can see that what happened to Gordon Hayward was gruesome, but when it comes to a phenomena like streaky shooting — we humans are not especially capable at the task of perfectly assimilating months worth of data through sheer observational power, so we decided to investigate the answer ourselves.

We tested Robert Covington’s three-point shooting against that of similar-quality shooters. The evidence of Covington’s streakiness does not leap out of the data in the way the basketball pundits suggest.

By the end of this article, you will have three things:

  1. A thorough description of our studies and their findings.
  2. Potential explanations into why there is such a large gap between our results and the perception of Covington’s play.
  3. Notes on the exact methodology.

A quick Q-and-A:

Q: “How did you avoid bias when devising the study?”

A: Before the study, the team had conflicting hypotheses.

Three quick notes on the study:

First, on bias and the scientific method: If researchers have an agenda, there is risk that they will study an issue multiple ways until they get the result they were hoping for. This was not a problem in this study as, first, neither of us has a particularly strong desire to prove whether Robert Covington was streaky or not streaky. To the extent unconscious confirmation bias can result from trying to get the outcome you expected, we were immunized from this effect by the fact that the two of us went into the study with differing beliefs on the question at hand.

Second, on the value of consistency: in this study we look to see how Covington compares to his peers on various measures of “streakiness” or “consistency.” Left aside is the question of how much consistency matters. Say you have two players, Player A ,who goes 3-for-8 from three virtually every night, and Player B, who swings wildly, going 7-for-8 from deep one night and following that with an “0-for.” If they each make 37.5% of their threes for the season as a whole, how much more valuable is Player A? This warrants an entire study to itself and goes beyond our scope in this paper.

Third, we define streakiness here with respect to a norm, typically the norm of what other NBA players do. Of course we could say that anyone who ever makes or misses three in a row is streaky, and by that definition all players are streaky players. Frankly, that doesn’t seem very interesting. For purposes of this discussion, a streaky player is one who is significantly streakier than the norm.

NBA: Philadelphia 76ers at Denver Nuggets Isaiah J. Downing-USA TODAY Sports

Our friend and colleague Mike O’Connor of The Athletic & Sixers Science recently wrote an article discussing the issue of Covington and his streakiness. He showed some evidence for the phenomenon, but his study was centered on examining Covington’s shooting technique and shot quality by position — which is something you definitely should read.

The Findings

Our study considered three kinds of streakiness:

  1. Within Games
  2. Across Years
  3. Across months

Within Games

An important note, examining streakiness within a game allows you to explain if game-to-game streakiness is a real phenomenon. First, we selected any full time 2017-18 player who has shot threes at approximately the same frequency and accuracy as Covington. Our exact method of selection will be at the bottom.

The Comps

As you can see, these players are not Covington clones; no one would mistake Damian Lillard for Covington. But as you can also see, “The Beard” shoots about as many threes as Covington and with about the same accuracy, so he is a good comp as far as this particular study goes.

For each player, we selected all the games in which they took at least five threes. Our view was that going 2-for-2 or 0-for-3 wouldn’t qualify as a “hot” or “cold” night, whereas going 5-for-7 or 1-for-9 would.

This gives us our final sample, approximately 50 games this season for each player.

Cold: 0-25% from three

Medium: Between 25 and 50% from three

Hot: 50% or above from three

For the avoidance of doubt, 2-for-8 was low, 2-for-7 medium, 2-for-5 medium, 3-for-6 high, etc.

Here’s what we found:

As you can see, Covington is around the middle of the pack, registering about as many medium games as the others. The only really streaky player appears to be JJ Barea, who does have far fewer medium games than others.

Covington does have more cold games than the average player in the group, but the difference is minor; no one would look at this list and say Covington is the super-streaky guy here.

As many of you know, the evidence for the existence of a “hot hand” in the NBA is vanishingly small.

For now let us just say this: Most studies historically have found no evidence whatsoever for the notion that players who have hit three in a row are any more likely to hit their next shot. This is fairly shocking as, even if such common tropes as “he was feeling it out there’” and “the basket looked as big as the ocean” were total nonsense, you’d think the data would show a hot hand anyway.

After all, some nights you are slightly injured, or facing a good defender or good defensive team, or you didn’t sleep well or are ill or hungover, or your superstar teammate is sitting so you are guarded more tightly. Yet all that stuff combined appears to add up to approximately nothing; or it adds up to something, but is fully counteracted by defensive adjustments or shooter overconfidence when feeling hot, or by other effects.

NBA: Los Angeles Clippers at Philadelphia 76ers Bill Streicher-USA TODAY Sports

Lately there have been glimmers of evidence that maybe there is a teeny-tiny hot hand effect -- one you can see if you use an electron microscope. Which, again, would be easily explained by nagging injuries and other variables -- even in the absence of “feeling it.”

So given that there’s been near-zero evidence of within-game streakiness at all among NBA players, it shouldn’t shock us that we don’t see it with Covington in particular.


But what about over longer periods of time?

An easy thing to check is annual streakiness -- does Covington have unusually good and bad years? One might think so, after all he shot a mere 33% last season, hardly the hallmark of the elite shooter.

Here are the 3P% marks for Covington’s four seasons in Philly:

As you can see, there’s a significant spread there. But is it abnormal? We found all the players with 3PA/36 and 3P% similar to Covington’s over the past four seasons combined. We only kept those who played all four years. For each player we computed the standard deviation of their annual 3P%’s That is, all these guys averaged about the same as Rock, and the question is, were most of them more consistent?

Here they are:

Voila! Covington is far from streaky by this measure, indeed he is among the most consistent players on the list. As a reminder, we chose ALL the players who shot as often and as well as Cov over the full four years, there was no cherry-picking. It’s just entirely normal for a 36% 3-point shooter to have a 33% year or a 39% year, even if he shoots for high volume.

For those who don’t love standard deviation, we added a column showing the gap between each player’s best and worst 3P% in these four years.

People like to quote figures as to how many threes it takes for random variation to even out; you’ll hear numbers like 750 shots and such. Of course no sample size fully eliminates randomness, but here’s a rule of thumb: if a player takes 400 threes a year, then it will be a common occurrence for them to be off from their true level by 2.5% or more; indeed that will occur about one year in three.

NBA: Miami Heat at Philadelphia 76ers Bill Streicher-USA TODAY Sports

In fact, one year in 20, just by luck, they will be off from their actual level by 5% or more.

This 2.5% standard deviation changes with the square root of the number of shots, so to get it down to 1.25% the player needs to take a full 1600 threes, several years worth for most gunners.

OK, we know the Covington-is-streaky forces are saying, “Covington doesn’t look streaky game-by-game or year-by-year, but it’s the months! After all, wasn’t he close to 50% at one point this year? And hasn’t he been in the twenties lately? If that isn’t streaky, what is?!”


Monthly

Again, we could do an analysis of statistical distributions to show how common such events would be in the absence of streakiness.

We think it’s more interesting to look at other shooters and see how they compare. We use the same sample from the annual test.

For each player we took every month in which the player took at least 50 threes -- Covington took 100 in November, less than 50 means the player was not in many games, perhaps because of injury or because it was a short NBA month like April.

We compute the standard deviation of their monthly 3P% below:

Once again Covington does not seem especially streaky.

Here we do it by best month vs. worst for those who don’t love standard deviation:

Once more, if anything Covington is more consistent than most. Yes, it seems like a 42% month and a 26% month from the same player is proof of streakiness. But it just isn’t so!

If I do a word-association test with you and I say James Harden, there are lots of things you might say. ”Beard.” “MVP.” “Buckets.” “Fouls.” “Flops.” But “Inconsistent”? I don’t think so! Yet he shoots the same number of threes as Covington, with similar accuracy, and has somewhat larger month-to-month accuracy spread than Covington.

If one looked hard enough, one could probably find a way of cutting the data that leads to a different conclusion, but we believe that a majority of these studies will reach a similar conclusion to ours: Covington is not an especially streaky shooter.

That still leaves us with a puzzle: why does he SEEM so streaky?

NBA: Utah Jazz at Philadelphia 76ers Bill Streicher-USA TODAY Sports

We have two theories:

First, it’s well-documented in psychology research that humans tend to observe patterns even when the reality is random.

Here’s a bet you can win with your friends!

First, have them write a long sequence of H’s and T’s that are as random as they can possibly make them — say 50 letters long. Then, have them toss a coin 50 times and write down the sequence that results using H for Heads and T for Tails, HHTHTHTT etc.

Here’s where the juicy party is — you bet them that you can guess which one they made up and which they got from tossing coins. You can even offer them better odds to make it worth their while.

Here’s how you win: almost for sure, the one with the longest consecutive sequence of H’s or T’s is the truly random one, — the one that came from tossing the coin. In a sequence of 50 tosses you will most likely have at least one run of 5 or 6 straight times with the same result.

Conversely, nobody would ever put 5 straight heads or tails in a sequence when they’re intentionally trying to make it look random. We think truly random sequences have more alternations than they do. Try it, unless the other person knows this bet, you will win virtually every time.

The application to basketball is clear: when someone sees five straight makes or misses, they think something must be happening -- the player is “hot” or “cold.” It just doesn’t seem possible that five straight is the result of randomness. The truth is, lady luck works in mysterious ways -- producing five straight bricks is not outside her powers.

Here it behooves us to include a quote from recent Nobel Laureate Richard Thaler, who once told us, “When someone makes three in a row and then hits another, they say he was “hot.” When he misses three in a row and then hits one, they say he was “due.” Between “hot” and “due” you can explain a wide range of phenomena.”

To play devil’s advocate, there’s certainly some evidence to suggest the hot-hand exists:

So, that may explain why some people think all players are streakier than they are. When someone misses his first couple threes, we often here calls to sit him down or bar him from shooting again that night. This is probably bad strategy, but it is an entirely natural reaction for us as humans, we simply see those two misses and imagine a pattern where in reality randomness reigns.

But while we have certainly seen many players referred to as streaky, we have little doubt that it is Covington, more than any other Sixer, who is seen that way. Why Covington?

A small part of the answer may come from the fact that many players who shoot as many threes as Covington are truly exceptional distance shooters, 40+% shooters like Steph Curry, Klay Thompson, or Philly’s own -- JJ Redick.

A quick glance at the data suggests to us that Covington isn’t any streakier than those sharpshooters. He seems less consistent because he’s just less good at shooting.

The reason this can’t be the whole story, though, is that as we’ve seen above, there are many fine players who shoot the three about as often and as well as Covington; it’s not our impression that everyone thinks all those guys are super-streaky.

Something special is happening with our perception of Robert Covington. What is it?


An Easy Target

Here’s our best theory: Covington plays the right way.

By “the right way,” we mean that, as has been oft-repeated in recent years, the best shots in basketball are dunks, layups, free throws, and three-pointers.

Some go so far as to say players should almost never take any other kind of shot. Basketball is a game of equilibrium, and if the defense wants to guard every three up close and give up easy 10-footers, those 10-footers may become efficient shots.

But we aren’t there yet -- at this point, it really is the case that most players should not be taking many long 2s, or even mid-range 2s. Robert Covington knows this; not only does he know it, he implements it.

This year 66% of Covington’s jump shots have come from beyond- the-arc. Compare, for example, JJ Redick at 53% or Dario Saric at 44%.

Why does this make Covington appear streaky?

If Covington goes 1-for-8 from three, he’s doomed to a bad shooting night. Most likely, he will have a low shooting percentage and number of points that night.

Whereas other players, ones who have not embraced the mantra of efficiency, will salt their evening with some relatively easy two-point attempts.

Consider a night where we have:

Covington: 2-for-9 from 3, 0-for-0 from 2

Saric: 0-for-5 from 3, 3-for-5 from 2

Neither shot great! But in this example, Covington will have had a more efficient shooting night (6 points on 9 shots vs. 6 points on 10).

The problem is that it will feel as though he shot worse, because his shooting percentage was 8 points lower than Dario’s. As a fan it’s going to seem like both shot poorly, but Dario was merely “bad” while Covington was “pathetic.” This despite the fact that we intentionally cooked the numbers so Covington had the better shooting night, properly measured.

NBA: Philadelphia 76ers at Miami Heat Steve Mitchell-USA TODAY Sports

We suggest that NBA fans in the aggregate are not yet fully used to how much more valuable three-pointers are than twos. And neither are players! How often do we see Redick pass up a “somewhat-guarded three” — with say around a 32% chance of success — for a rhythm dribble into a long two-pointer that he was 45% to make? That’s usually a bad trade!

32% * 3 gives you an expected value of .96 points versus 45% * 2 giving you an expected value of .90 points.

But as we say, the right way, the winning way, is to prefer the 32% three to the 45% two. That’s what Covington does, night after night. This is one of the reasons why the Sixers are vastly better with him on the court than they are without him; the on-off split has been greater with Covington this year than for any player -- save Joel Embiid.

But this article isn’t about Covington’s level of quality (see: Kevin Love’s terrific piece). This article is about streakiness.


Conclusion

Covington doesn’t have bad streaks from three any more than other shooters of his quality. Yes, he has more than Curry, Redick, or Thompson, but this is attributable to them being superior long-range shooters.

When we compare him to shooters who match his efficiency and volume, guys like Lou Williams, Damian Lillard, or Allen Crabbe — we see that he’s not an especially streaky shooter.

Many of the other players at his level of shooting ability still favor inefficient shots. Since these things are mostly random, on a particular night when the threes don’t fall, they hit a couple of those long twos, making their night seem more productive.

If you want to say the hypothetical Dario night above is less “cold” than the hypothetical Covington night, then that definition of hot and cold will enable the words to match the eye test. But if one cares instead about winning basketball games, the best choice is to encourage the efficient style of play employed by Covington, rather than denigrate it by calling it “streaky” or “inconsistent.”

NBA: Philadelphia 76ers at Charlotte Hornets Sam Sharpe-USA TODAY Sports

Methodology

Game By Game Qualifications

  • 3P% [+/- 1.2 from Covington (34.4-37.8%)]
  • 3PA/36 [+/- 1.2 from Covington per 36 (6.5-8.9 3PA)]
  • >1000 MP this year

Monthly/Yearly Qualifications

  • 3P% [+/- 1% from Covington over 4 year sample (35.7%)]
  • 3PA/36 [+/- 1 from Covington per 36 over 4 year sample (8.0 3PA/36)]
  • >3500 MP over 4 seasons