David’s Week 8 FanDuel Lineups: Modeling Defense

October 31, 2015 - by David Hess

Week 7 was yet another overall negative week for my FanDuel entries, though there was one silver lining — my tournament stacks ended up slightly positive for the week.

When reviewing why I lost, I noticed something surprising (to me): a high percentage of my head-to-head opponents made the same lineup choices at running back (Freeman & Gurley were both picked by 92%) and defense (74% picked St. Louis), and those choices turned out well.

This prompted me to take a look at the data to see if I also should have predicted those choices and/or performances. As it turns out, a simple defensive model I just put together did indeed predict the high St. Louis ownership. More on that below.

Insanity Is Doing The Same Thing Over And Over And Expecting A Different Result

At first I thought a few losing weeks to start the season was just variance rearing its ugly head. But we’re almost halfway through the season, and my cash game (i.e. head-to-head & 50/50) lineups have had only one winning week.

More worryingly, the losing weeks haven’t been just slightly down; they’ve been losing big. In 5 of 7 weeks, my cash game lineups have had ROI’s of -28% or worse, including worse than -63% three times.

This is almost the direct opposite of last season, where my cash game lineups were a net positive in 7 of 8 weeks, and had an ROI better than +69% four times.

On the other hand, my tournament lineups have continued to keep pace with last year. I’m up slightly for the season in my tournament plays, and that’s including the poor tournament performance of my cash game lineups, which I have been throwing into one or two tournaments each week. So far I’ve been positive in 3 of 7 weeks, compared to 4 of 8 last year. Given that the general strategy in big tournaments is to mix in a few big wins along with some expected small losses, this is acceptable.

As a result, I’m making a couple changes:

I’m going to cut down on the volume of my head-to-head plays
I’m going to change my lineup creation strategy some, and rely a bit more on my own projections rather than just trusting RotoWire and numberFire (for some positions, at least)

Projecting Team Defense Scores

As mentioned in the intro, 74% of my head-to-head opponents last week chose the St. Louis defense. This seemed like a crazy high pick rate to me, and it made me suspect that a lot of the “sharp” DFS players are using similar models or heuristics to pick their defense.

If that’s the case — that the vast majority of sharp players are coming up with the same defensive pick — then I figured the models they’re using can’t be that complicated. If they were, there would have been less consensus.

So, I rolled my own simple model last night. I did this quickly, and took a lot of shortcuts. It’s definitely got some methodological issues, but it’s “good enough for government work,” as they say, and the outputs pass the smell test. Last week, it showed St. Louis as the clear best pick for defense.

Because I don’t really think it’s a great model, I’m not going to get into the nitty gritty details, but I will say that the variables I used were (from most to least important):

Vegas predicted opponent score
Opponent’s average FanDuel points “allowed” to other team defenses
The defense’s average FanDuel points scored in previous games
Vegas predicted score of the defense’s team
Game location

This week I’m going to base some of my defensive choices on the outputs of the model, with some manual adjustments for injuries.

Week 7 Results Recap

Initial 2015 NFL Bankroll: $1,000.00
Bankroll Entering Week 7: $819.20 (-18.1%)

Head to Head & 50/50 Entry Fees: $120 (14.6% of bankroll)
Head to Head & 50/50 Net Winnings: -$94.80 (-79% ROI)
Tournament Entry Fees: $60 (7.3% of bankroll)
Tournament Net Winnings: +$2 (+3% ROI)

Total Entry Fees: $180 (22.0% of bankroll)
Total Net Winnings: -$92.80 (-52% ROI / -11.3% of bankroll)

Bankroll Entering Week 8: $726.40 (-27.4% for the season)

Week 8 Lineups

Head-to-Head & 50/50s

After looking into my opponents’ picks last week, I’m changing up my strategy a bit this week. At positions where I think some specific pick will be very popular, I’m going to default towards making the same pick, if I don’t have a strong reason to think that player or defense will underperform.

Let’s look at St. Louis last week as an example of why this strategy makes sense.

There was very heavy usage for a few key players last week, including the St. Louis defense. I didn’t use St. Louis, they scored 25 points, and I was in a big hole right off the bat in 74% of my head-to-head matchups.

Picking a team other than St. Louis last week was analogous to picking a big underdog to win a game in an NFL pick’em pool. If the favorite wins you lose ground to most of the pool, but if the underdog wins you give yourself a leg up on everyone. It’s a high-risk, high-reward strategy. On the other hand, if you pick the favorite, and the underdog wins, you’re not in terrible shape since very few opponents probably gained ground on you. It’s a low-risk, low-reward strategy.

In other words, without a very strong reason to believe that St. Louis wasn’t going to do well, picking the “favorite” was probably the best move last week in head-to-head matchups. (In tournaments, on the other hand, fading the popular team can make more sense.)

In order to project pick%, I’ll mainly be relying on this Reddit thread, where people share pick% info from their Thursday tournaments.

As a quick reminder, the original strategy here was to be willing to pay for volume points at QB & RB, even at the expense of a little bit of value. I’m still going to do that; this increased attention to pick% is just an addition to my already existing strategy guidelines.

Here’s my primary lineup for Week 8:

Some notes on this lineup:

Philip Rivers was the projected best value at QB according to both RotoWire and numberFire. And with good reason — he’s averaged more than 55 pass attempts per game over the past three games, and this week faces a Baltimore defense that has allowed the most QB FanDuel points in the NFL.
Based on Thursday ownership percentages, I expect Todd Gurley and Justin Forsett to be the most popular two running back in cash games this weekend. Both are projected as solid plays, so I see no reason to go against the “chalk” here.
Stefon Diggs was the most popular cash game wide receiver in Thursday contests, so he’s another chalk play.
For my other two wide receivers, I tried to find players who were moderately popular, had solid projections at both sites, and have shown some consistency this season.
Tyler Eifert is behind only Antonia Gates and Austin Seferian-Jenkins in average FanDuel points this year, and he’s facing a Pittsburgh defense that has given up the third-most TE FanDuel points.
Chris Boswell was among the most popular kickers in Thursday contests, and is projected as a good value pick by a kicker scoring model I whipped up last night (very similar to my defensive model). Plus, he costs the minimum, allowing me to splurge elsewhere.
The Rams are the whole reason I built my simple defense model last night. Lo and behold, the model projects them as the best value this week, even before any subjective adjustments for San Francisco’s injury woes. Plus, the Rams were the “chalk” defense in Thursday contests.

Here’s my secondary lineup:

Notes on this lineup:

The choice for my second QB came down to Cam Newton or Andy Dalton. It seemed like a toss up to me based on the projections, so I ended up going with Newton simply because he was slightly more popular on Thursday.
Todd Gurley is the only player duplicated in this second lineup. He’s been a points monster recently, and the betting lines seem to imply that the Rams will have a relatively easy time of it against San Francisco, with could mean lots of clock-killing carries for Gurley. It seems relatively safe to use him in both lineups, especially considering his projected popularity.
Devonta Freeman is this year’s Le’Veon Bell, in that he’s projected with monster points every week. And per my strategy guidelines, I don’t mind paying for running back points.
Antonio Brown should benefit greatly from Ben Roethlisberger’s return, plus he’s facing a Cincy defense that is 29th in DVOA against #1 receivers.
Steve Smith is facing a San Diego defense that is 30th in DVOA against #1 receivers, and that just allowed touchdowns to both Amari Cooper and Michael Crabtree. Plus, Smith is the Ravens’ top red zone target, and acts as a bit of a hedge against having Forsett in my primary lineup.
Just like last week, John Brown is questionable for Arizona, and if he misses the game, Michael Floyd seems like a good candidate for increased targets. Even with Brown active, Floyd has scored the lat two weeks.
Antonio Gates is likely to miss this weekend’s game, making Ladarius Green a great bargain at tight end.
My simple defensive model projected Houston as the third best value, after St. Louis and Seattle. I decided to go with Houston rather than Seattle mainly to save $500, which helped me afford Devonta Freeman.
Matt Bryant was chosen because we had the highest consensus rank on FantasyPros among the kickers I could afford.

Tournament Lineups

As is the case most weeks, my first step this week was to come up with some single-team stacks that seem to have the potential for a big day.

Some of these stacks were based on fantasy-points-allowed to each position from this season (using RotoGuru as my source), and others were keyed on a particular player or two who were well-projected by consensus sources, or who looked like a potential great targets-per-dollar value.

In addition, I’ve done some research recently into what characteristics are common to “monster” QB/WR games. Preliminary research seems to show that moderate home favorites (roughly 5- to 9-point faves) in games with moderately high totals lines (over 45) are good candidates for stacks. This week, that includes the Falcons and Panthers.

Here’s what I ended up with:

Atlanta (vs. TB) — Matt Ryan (QB), Devonta Freeman (RB), Julio Jones (WR). The Falcons fit my “monster games” profile, plus the Bucs have allowed the 6th-highest FanDuel points per game to wide receivers (and highest over the last 3 weeks). Freeman doesn’t fit as nicely into the template, but … he’s Devonta Freeman.
Carolina (vs. IND) — Cam Newton (QB), Greg Olsen (TE). Carolina also fits the “monster games” profile, but the Panthers don’t have a star wide receiver for Newton to throw to. Instead, TE Greg Olsen leads the team in targets. I may enter an alternate version of this lineup that includes Carolina’s Jonathan Stewart. That would nearly corner the market on Carolina red zone touches.
Houston (vs. TEN) — Brian Hoyer (QB), DeAndre Hopkins (WR), Texans (DEF). Hopkins has the highest number of targets per dollar in the NFL over the past 4 weeks, which gives him the potential to have a 30+ point game in any given week. I may enter an alternate version of this lineup that includes Texans WR Nate Washington, who put up nearly 30 FanDuel points last week.
Arizona (@ CLE) — Carson Palmer (QB), Chris Johnson (RB), Michael Floyd (WR). The Cardinals have several skill players who seem to have the potential for a big game (Larry Fitzgerald, John Brown, Michael Floyd, Chris Johnson). Here, I went with a coupe of the less popular options, in the hopes a big day would give me a boost that nobody else benefits from
Cincinnati (@ PIT) — Andy Dalton (QB), Marvin Jones (WR), Tyler Eifert (TE). Pittsburgh has given up the third-highest average FanDuel points to tight ends, and Eifert is already one of the top scoring TEs heading into this matchup. And at wide receiver, A.J. Green may be the big name, but Marvin Jones has seen more targets recently (both overall, and in the red zone).

Filling in around the stacks with solid consensus values, with some emphasis on trying to find players with high upside, here’s the result (click to enlarge):

I also couldn’t help myself, and had to enter one mostly model-based lineup. For this lineup, I made simple projection models for every position except wide receiver. I then narrowed my choices down to the top few highest-value picks at each position (according to the models), and let RotoWire optimize a lineup using only those players. At wide receiver, I simply tried to use players that were consensus solid values at both RotoWire and numberFire. Here’s what I ended up with.

The above lineup feels more to me like a 50/50 lineup than a tourney lineup. But I’ll enter it in both types of contests, and see how it does.

For the stacks above that lineup, I’ll be entering them each in a FanDuel Sunday Million tournament, as well as a smaller 100-person or 250-person tournament. The idea is that if they do astronomically well, I’ll be rewarded for that in the Sunday Million, but if they just score in the top 1% or so, I’ll get a better return in the 100-person tournament.

The most likely outcome for any specific tournament lineup is that I win nothing, but with multiple stacks, hopefully at least one will still do well enough to recoup my buy-ins, and let me tread water while I wait for a big payout.