I'm not a thrill-seeker. In fact, I'm very much the opposite. I once screamed so loud on a carnival ride that the attendant stopped it just so I could get off.

The biggest rush I can stomach in life is being underweight on a popular play in DFS. That player can underperform for 58 minutes. But if they bust off a long touchdown late in the game, your fade is ruined, and your bankroll is bruised. You're basically holding your breath for three straight hours.

This was not a good year to get your adrenaline fix in this fashion.

I've been looking at the production of players who were heavily rostered for each of the past five seasons. In 2020, the public hit across the board at rates higher than they had the previous half decade. If you tried to be contrarian, you had a lot of long Sundays.

This -- by itself -- matters. If popular plays pan out more often, there's less incentive to deviate in NFL DFS. It's something we have to account for within our process, so that alone is noteworthy.

But there were still some inefficiencies in our collective decision-making. We've got to pounce all over those while we still can.

Today, we're going to dive into how players performed when they were popular on FanDuel for 2020. We'll look at their performance relative to past years, see in which circumstances swallowing the chalk was advisable, and -- importantly -- where the popular plays still managed to fail, giving us lingering windows to deviate.

Lucky for those who value life and don't want to jump out of an airplane for funsies, we can still get our heart thumping from the safety of our couches on Sundays this fall.

Defining Success

Before we actually look at what went down, we have to set some guidelines for the study.

First, our "popular" players will be those on the most rosters each week in the FanDuel Sunday Million. At quarterback, tight end, and defense, this will be the top three players in roster rate each week. At running back and wide receiver, we'll run through the top seven options, accounting for the additional roster slots (including the flex) for those positions. This winds up being roughly equivalent to the players on at least 10% of rosters each week.

In order to determine whether a popular player "hit," we need a baseline of success. This will vary by position. So, to get a baseline, I took the average point-per-thousand-dollar output (or value) for all players at each position who were on at least 1% of rosters in the FanDuel Sunday Million. This way, we're not comparing the popular players to those who were completely off the radar and didn't have a role.

Here's the baseline value for each position. In this instance, a quarterback's baseline value of 2.55 means that the average quarterback on at least 1% of rosters produced 2.55 FanDuel points for every $1,000 worth of salary.

These bars for success are low. Getting 1.89-times value out of a running back with a $7,000 salary means they're scoring 13.2 FanDuel points. You need more than that if you're trying to take down a tourney. So keep in mind that these baselines are low, but even with similar baselines in years past, players had trouble clearing them.

That was less of an issue in 2020.

Position-Wide Hit Rates

Just a couple of years ago, even low baselines couldn't make the chalk palatable. In 2016, only 39.2% of popular tight ends hit their baseline value. Wide receivers hit at a 47.1% clip as recently as last year.

That changed in 2020.

Every single position hit more than half the time when they were popular. That might seem like a product of the low baselines for success, but this was a serious deviation from what we saw in previous seasons.

Quarterbacks, receivers, and defenses all either set or tied their highest hit rate in the five-year sample. Running backs hit at their second-highest mark. Even when we account for the fact that the baseline was low, popular plays were popping this year.

Truthfully, this shouldn't be a surprise. With daily fantasy analysis and projections being more readily available than ever, bad chalk is more likely to get weeded out throughout the week. We should expect this up-tick to carry over into future years, and it should make us less fervent in our commitment to avoiding the popular plays.

But as mentioned, that doesn't mean we were perfect. Not all of the misses this year can be blamed on variance. After accepting the fact that the chalk is likely to hit at a higher rate going forward, it's our job to try to find spots where we can still be different. One mistake that doesn't seem to be going anywhere is point-chasing.

Point-Chasing

Whenever a player goes nuts one week, we're going to get itchy. Their salary likely won't budge too much, so we'll think they're a screaming value the next week, as well.

That's not how things play out in reality, though.

The table below looks at how well players do the week they're popular versus what they did the previous game. We'll exclude Week 1 from this and any players coming off of injuries. The right-hand column shows what percentage of popular plays scored more points the week they were popular than they did the previous week.

More than 60% of heavily rostered receivers scored fewer points when popular than they did the previous week, and the position-wide average went down almost three points. The only position to score more points when popular than the previous week was defense and special teams.

This doesn't mean we should avoid players coming off of a blow-up as a rule. Dalvin Cook followed up his 47.6-point outing by dropping 38.2 when popular in Week 9, and we're marking that as a win every time. It's moreso to say two other things:

1. You should expect players with big performances in Week X to be popular in Week X+1, and

2. You should expect those players to score fewer points when they're popular than they did the previous week.

Basically, it should raise a flag in your mind when someone goes bonkers and projects to be heavily rostered the following week. You should be hyper vigilant of warning signs they could bust.

It would help to know what those warning signs could be. Let's get into that next, going position-by-position and seeing which conditions often led to busts and where our decision-making was still lacking.

Quarterbacks

As you saw in the opening section, we were good at identifying high-quality plays at quarterback this year. They hit almost two-thirds of the time, even at a respectable 2.55-times value. We were especially good at identifying upside at the position.

Of our 51 popular quarterbacks this year, 17.6% of them managed to provide at least four-times value. That was nearly double the position-wide rate among quarterbacks on at least 1% of rosters.

Of the 29 quarterbacks in the Sunday Million player pool who hit at least four-times value, eight were among the most popular options at the position that week. Two of three quarterbacks to hit five-times value were popular. So when it came to capturing the best plays, the public did so fairly successfully.

That's the plus side. It means in general, we shouldn't be too wary of quarterbacks who project to be popular. You also don't see too many hyper-chalky options here as the max roster rate for any single quarterback was Patrick Mahomes at 24.7% in Week 16. Thus, this position is pretty safe.

There were still some stumbles in our decision-making, though. A big one was weather.

Talking about wind on Twitter will often lead to blow-back (pun intended), saying that we overreact to it and avoid good plays as a result. The data refutes that. There were 11 heavily rostered quarterbacks in wind speeds of 10 miles per hour higher this year, and only 5 of them hit the baseline value.

That's compared to 23 of 30 popular quarterbacks who hit their baseline value when wind speeds were less than 5 miles per hour. As the wind speed increased, so did the bust rate.

One part of the wind discussion that is true is that wind speeds do lead to reduced popularity. Whereas 29.3% of all NFL games had wind speeds of 10 miles per hour or higher, only 21.6% of the popular quarterbacks came from that bracket. That's the direction you want to be in. But with the elevated bust rates associated with higher winds, it's clear we still have an opening to be underweight on quarterbacks playing in elevated winds.

The second inefficiency is one that was spoiled last week in our analysis of quarterbacks who made perfect lineups: we under-invest in spots where quarterbacks are underdogs entering the game.

This is a spot where we can't look at hit rates for quarterbacks at various spreads. The reason is that only 4 of our 51 popular quarterbacks for the entire season were underdogs. There were more quarterbacks who were underdogs in perfect lineups (six) even though there were just 17 total perfect quarterbacks.

The table below compares the decision-making (the percentage of popular quarterbacks in each bracket) to the results (the percentage of quarterbacks in perfect lineups in each bracket). To provide additional context, the "Overall" column shows what percentage of teams broadly fell into each spread bracket during the regular season.

A higher percentage of quarterbacks made the perfect lineup while slight underdogs than there were in the league as a whole. Yet they still rarely wound up being popular options.

And those four quarterbacks who were popular while underdogs? They all scored at least 31 FanDuel points, and three of them were in the perfect lineup.

Looking into roster rates isn't just about identifying which types of players busted and which hit. It's also about seeing in which types of quarterbacks we are over- and under-investing. At quarterback, it seems obvious we aren't as enthusiastic about slight underdogs as we should be, and we should try to exploit this edge early and often in 2021.

Outside of the underdog quarterbacks, the split that hit highest was those in games with a total of 53.5 or higher. Those 19 quarterbacks hit 78.9% of the time compared to 56.3% for quarterbacks in lower-totaled games. Things went well when we used quarterbacks in projected shootouts.

We do see a bit of a gap here in terms of decision-making. Although we did a decent job of avoiding quarterbacks in games with a low total, we should have been even more strict about it.

The quarterbacks in the lower-total range didn't bust often. But they also rarely flashed upside with none of the popular quarterbacks in a game with a total under 48 exceeding 3.52-times value. The other 41 popular quarterbacks topped that mark 12 times. With how important it is to capture upside, we can afford to be highly skeptical of projected popular quarterbacks in games with lower totals.

Overall, we can feel comfortable using a quarterback even when they project to be popular. The penalties here are lower than most other positions. Even with that in mind, we should look to be underweight on popular quarterbacks playing in wind or games with low totals, and we should actively seek out quarterbacks who might be undervalued with their teams coming in as slight underdogs.

Running Backs

The overall hit rate at running back -- 58.0% -- was high. It's the second-highest mark for the position in the past five years, and it ranked third out of the five positions. But there were more leaks in the process here than at quarterback.

Specifically, there wasn't as big of a production gap between the popular backs and the full grouping as we saw at quarterback.

We were good at identifying running backs with a high floor. The ceiling equation was different, though, as only 34.8% of the backs to hit four-times value were among the seven most popular options for the week. The main culprit here seemed to be overconfidence in heavily favored backs.

As you'll recall from our look at perfect lineups, only 4.8% of backs in perfect lineups were on teams favored by double-digit points. But 13.4% of our popular backs came from that bracket.

That alone should tell you that we're too high on backs whose teams are heavily favored. But those backs were also far more likely to bust than backs in other scripts.

The two charts above tell us two things.

First, our brains are wired to target backs on teams that are heavily favored. We should expect increased popularity for backs tied to heavily favored teams.

Second, those backs are more likely to bust than other options.

If we can find spots like this where we know a player is both more likely to be popular and to fall short of expectations, that's a delectable formula for being underweight relative to the field. This is far from being a universal creed, but things aren't going to align this well at other positions. We should take advantage of the opportunities presented to us.

Things were far prettier when the backs were in projected tighter scripts. When the spread was fewer than five points in either direction, backs hit their baseline value 64.4% of the time. Those in games projected to be less competitive hit just 51.7% of the time.

We did see a more minor split based on the total, as well. Popular running backs hit 61.2% of the time in games with a total of 48.5 or higher compared to a 53.8% hit rate when it was projected to be lower-scoring.

Combining these two points, there were 10 running backs who were on teams favored by double digits in a game with a total of 48 or lower. Only 4 of those 10 hit. We should be wary of backs in this bucket in 2021.

That's the negative. The positive is that we were solid at picking quality value backs to roster.

Of our 119 popular running backs, 22 had a salary of $5,900 or lower. That group hit at a higher rate than any other salary range.

It's true that a lower salary requires fewer points to hit baseline value, and that's one of the big issues with viewing things through the lens of value. But even from a raw points perspective, 18.2% hit 20 FanDuel points, 31.8% scored 18 points, and 54.5% scored at least 13 points. That's output we can accept at these salaries, even if the lower end isn't going to help you take down a tournament.

What we saw in the perfect lineups is that the lower-salaried backs made the cut with a disproportionate chunk of their scoring coming via receptions. Some of the names who thrived at a lower salary when popular were Tony Pollard, Giovani Bernard, Leonard Fournette, Jamaal Williams, and Jonathan Taylor, all of whom provided juice as pass-catchers. If you see a guy in that mold at a lower salary, you can comfortably roster them even under the assumption their popularity will be high.

As with quarterback, we've got a solid set of guidelines at running back. We should be skeptical of projected popular backs when their teams are heavily favored and when they're in a game with a lower total. We should be more welcoming of the chalk when it comes in a tighter script or if it's a lower-salaried back who catches passes stepping into a larger role. This should position us to avoid the mistakes of the public without ignoring spots conducive to upside.

Wide Receivers

At wide receiver, there wasn't as much inefficient decision-making as we saw at quarterback and running back. There were still some signals of what chalk was good and which was underwhelming, though.

As mentioned earlier, wide receivers came through at a higher rate this year than they have any of the previous seasons. That's a plus. We were specifically good at pinpointing quality floor plays, though, with the ceiling analysis being similar to what we saw at running back.

We correctly pinpointed 28.6% of the wide receivers who hit four-times value compared to 34.8% for running backs. So although things were good, they weren't quite perfect.

A lot of the inefficiencies at wide receiver were super process-oriented and logical. There was no one split that we had to avoid, but steadily picking away at weak spots could better position us for handling the chalk.

A good overview of this is looking at the profile of the receivers who hit baseline versus those who didn't. This chart divides them into those two camps. The differences in the two groups can tell us what kinds of receivers are more likely to shine.

The receivers who hit were in higher-totaled games with tighter spreads and lower winds. They were also less likely to be coming off a monster game the week before. Seems pretty cut and dry.

In digging into these splits individually, you'll see slight edges. Receivers hit 63.1% of the time in minimal wind compared to 57.1% of the time when speeds were 5 mph or higher. The hit rate for receivers in games with a total of 48.5 or higher was 63.2% compared to 56.3% when it was lower. This is in line with what you'd expect, which is reassuring. It means we can take a process-based approach to the position.

If a receiver who projects to be popular is coming off a big game, playing in a sub-optimal game environment, and dealing with some wind, we have plenty of incentive to jump ship. That's especially true with our ability to pinpoint eruption games being just mediocre.

But we can feel decently comfortable if a receiver who checks all the process boxes projects to be popular. You still have the ability to deviate given what we saw with upside at the position, but they're far from being a must-avoid as long as they fit what we typically want out of a wide receiver.

Tight Ends

Tight end was almost the opposite of wide receiver. Although the hit rate was low -- meaning we lagged in identifying floor -- we had the ceilings plays on lock. All three of the highest point-per-dollar outputs at tight end this year were from guys who were popular in that week's Sunday Million.

Sure, one of those games was when Taysom Hill was the New Orleans Saints' starting quarterback. But the other two performances in the top three at the position were by Darren Waller and Jonnu Smith, so they weren't all gimmes.

Given the shakiness in the floor, you have leeway to duck out on tight ends who project to be popular. The odds they bust are still high. But if the stars align for a big output, we also have justification for going with the flow.

That's the difference between receiver and tight end. There were also some similarities, though. Both positions were more likely to hit when the total was higher and when the wind was lower, and point-chasing was an issue. The difference, though, is that tight ends weren't as dependent on a tight script.

Just in general, it's likely wise to be underweight on anyone tied to the passing game when the wind speeds go up. But with the spreads, we saw something similar here in the perfect lineups.

The average spread for tight ends in perfect lineups was higher than for any other position. We can explain this pretty easily. Tight end is a position heavily dependent on touchdowns, and teams that are favored are far more likely to generate those. At receiver, we can rely more on yardage, which is often a product of a back-and-forth affair.

This leads to a divergence in our decision-making between the two positions. Let's pretend the Kansas City Chiefs are 10-point favorites in a game with a total of 53 and the winds are low. The total says we can go at the pass-catchers. But the spread is one that fits better with flocking to Travis Kelce than Tyreek Hill. As such, if both project to be popular (which they likely would), we should be more willing to pony up with Kelce and be underweight on Hill. The Chiefs were favored by an average of 11.2 points the three times Kelce made a perfect lineup compared to 6.8 for Hill, so this does seem to track in practice. It's an exercise we can deploy in similar situations with other teams in 2021.

Defense and Special Teams

At defense, we did a good job of not point-chasing and identifying quality floors. As a result, the hit rate here was high.

We were not as good at identifying ceiling performances.

This isn't a surprise. To generate upside at defense, you need tuddies, and those are harder to predict here than any other position. So, this isn't all our fault. There are some tweaks we can make, though, to better position ourselves to capture upside.

Namely, we're too afraid to deviate from the baseline process. If you're looking for floor, you want to find a team that's heavily favored in a game with a low total. That's why we had success in identifying floor, but that's not the formula to use when looking for upside.

Here's a comparison of decisions versus results based on the spread at defense.

And here's the same for totals.

In both instances, we over-invested in the traditional process-based areas. We weren't willing to go at teams in a tight script or projected higher-scoring games, and the data says we should have been. The inefficiency in decision-making here was even greater than what we saw with quarterbacks and running backs.

That -- combined with our inability to pinpoint upside -- should influence how we handle defenses that project to be popular. Even though popular defenses had a high floor and a high hit rate this year, we should be willing to deviate from the chalk.

The whole point of looking into roster rates is to determine tournament strategy. In a tournament, our goal is to identify upside so that we can find difference-making totals. We didn't do that well with defenses this year, and we know there are large flaws in the decision-making process at the position. When both of those are true, we have little incentive to conform with the crowd.

If there's a defense that checks our traditional boxes but doesn't project to be popular, go for it. The popular defenses that hit had a higher average spread than those that flopped, so there are advantages to attacking a positive script. But if that defense is tracking to be popular, this isn't a position where we need to have a big fear of missing out.

Conclusions

Yes, the public as a whole did a good job this year in identifying safe chalk. And yes, we should expect that to continue. That does not mean we should suddenly be willing to follow the crowd no matter what.

We saw big gaps between our decision-making and what indicated a possible ceiling performance at quarterback, running back, and defense. At quarterback, we should have been more willing to roster slight underdogs. At running back, the public got overzealous with heavily favored backs. And at defense, we ignored potentially good process plays because they weren't heavily favored or in a game with a low total. The hit rates at all of those positions was high, but we can still find outlets for being different.

At wide receiver and tight end, we needed to be more strict in sticking to our process. The bust rates for receivers and tight ends in high winds and games with low totals makes it easy to justify deviating if a player projects to be popular in those conditions. We also saw increased bust rates out of receivers in non-competitive scripts and tight ends whose teams weren't projected to rack up points.

Let's pretend that from now on, all popular players hit at a 60% rate. That still leaves 40% of the time where they'll fall short, and even some of the 60% hits will be just barely clearing the baseline. We'll always have leeway to deviate when we see fit, and based on some of the data here, we know what those windows to fade look like.

So, if you're like me and can't work up the courage to get your Dominic Toretto on, fear not. We've still got justification for fading the public when the circumstances are right, and we can go right back to that well in 2021 despite what we saw this past year.