The Data Behind the Voodoo: Goalie Selection
Ask most NHL DFS players and they will tell you that goalie selection can be the most tilting part of the lineup selection process. I mean it’s easy though right? Just take a guy who’s favored to win and voila. Goalies typically perform in a specific point range and you want a winner at a reasonable cost; and don’t even try to predict shutouts! Goalie selection is so important that we wrote a strategy round-table piece on it earlier this year.
Using our projections from this season; we made an attempt to validate one form of goalie selection over another. Which goalie selection method leads to the best results? Which one has the highest upside? How much should you rely on Vegas? In total I’ve evaluated seven different methods for choosing goaltenders, including my own goalie selections; which is mostly science, with a splash of art. The conclusions I am able to draw are intriguing.
With any data set, I wanted to determine exactly what might be influencing the information; which could skew the results. When looking at Vegas lines, I used the goalie who was the heaviest favorite and the cheapest favorite at puck drop; not when lines are originally posted in Vegas. Therefore the accuracy of the goalie selections allowed for line movement and news that came out during the day.
The other issue I’ve bandied about is how and when to account for seasonal changes for teams. To project goalies, an analysis of team level data is needed, goals allowed at home and on the road in all score states and strength states are needed, but the issue is, how much do you weigh this year’s results to last years? A rolling 30 game average (home or away) may be the best method, but it becomes very cumbersome to handle the data to make those projections. Team personnel changes every year, sometimes significantly; and nevermind coaching changes. Take Bruce Boudreau out of Anaheim and insert Randy Carlyle; that’s a different team on day one; just as an example.
What about home/road splits this far into the season? Before Saturday’s games the Nashville Predators had given up just 13 goals in ten home games. They gave up nine goals over the weekend in consecutive losses to New Jersey and Philadelphia. So which trend is to believed? The 13 goals in ten game start of the disaster over the weekend? What it really did was jostle their early season success with some regression back toward the mean. 22 goals allowed in 12 games is more about what you’d expect, even if we considered the Preds an elite defensive team at home. Do we?
Site selection matters as well. DraftKings has tightened up goalie pricing this year, while there are still some anomalies on FanDuel. For the of this article I am using DraftKings pricing, scoring and totals.
Let’s look at what the seven methods of goalie selection are. I’ve included my own picks from the last 32 nights, because that’s as far as I could go back on DraftKings. I’ll examine why the results might be as they are, and also break some myths about goalie selection. On any given night some goalies will be the choices for more than one category, so there is also some overlap and not always the same overlap every night.
Here is the chart with the all the data broken down by results. If you see the key on the right the bars from right to left are as follows: (You’ll probably be need to expand the chart)
Light Blue – Any goalie result of under 2.0 points scored. – Total Disasters
Orange – From 2.0 points to 4.9 points. Not the result you’d want but not a Total Disaster
Gray – From 5.0 to 7.9 points. The most common set of results, and where we’d at least like to do when selecting a goalie.
Yellow – From 8.0 to 9.9 – Now were getting into the great results
Dark Blue – Anything from 10.0 and above. These are separating performances!
DFS Goalie Outcomes By Goalie Selection Method (Link)
The Same Data in Table Form (Link)
Heaviest Favorite (HF) 31-21, 2.21 GAA, 4SO (4.94 DK Pts)
Most DFS players presume the biggest key to getting points is finding the goalie who will win. Looking at Vegas odds before puck drop, we can presume that no goalie will have a better chance of winning than the heaviest favorite of the night. Size matters of course, the biggest favorite on a three game night is much different than the biggest favorite on a 12 game night. The heaviest favorite of the night has a record 31-21 for a 59% winning percentage.
Looking at the HF bar on the bottom you will see that choosing the heaviest favorite for the most part eliminated your “total disaster” outcomes of less than 2.0 points scored. Through 52 games, that’s nine results that had the potential to sink your entire lineup. In 45.5% of the games played so far, selecting the heaviest favorite of the night scored under 5.0 points. Not a number you’d like to have to anchor your lineups. What is interesting is the lack of 10+ point performances. Usually the heaviest favorite of the night is not going to be facing an up-tempo team that can generate shots at an elite level. Your ceiling is limited, because even with a dominating team performance the other team won’t likely generate many chances.
The heaviest favorite averaged 4.94 points scored, good for 3rd among the scientific models (not including my own choices). The heaviest favorite of the night has averaged 25.77 saves, which is the lowest of all methods. These goalies have earned four shutouts to date.
Cheapest Favorite (CF) 32-20, 2.29 GAA, 3 SO (5.26 DK Pts)
Now compare the cheapest favorite of the night with the heaviest favorite of the night. The first thing that should stand out; is there is no discernible advantage in taking the heaviest favorite of the night over the cheapest. The only outcome range where the heaviest favorite performed better was in “Total Disaster” results (11/52). If you’re playing cash games there is a little more risk, but once we get beyond that “under 2.0” level the cheapest favorite was on par or better with the heaviest favorite of the night.
If you haven’t already, perhaps this is evidence you should almost always forego the heaviest favorite for the cheapest if relying solely on Vegas. On the whole the cheapest favorite has a higher average total than the heaviest favorite, you just have to have the stomach for the under 2.0 risk. The risks of that sub 2.0 number will give you more salary for the rest of your lineup.
Lowest Expected Goals Allowed (LxG) 35-17, 2.15 GAA, 3 SO (5.35 DK Pts)
The first two categories were Vegas based. The next few are all based off of our projections. The best scientific goalie selection method is the Lowest Expected Goals selection. We get to this method by weighing goals for and goals against for each goalie against their opponent. Often times the highest rated goalie in this category will have an expected goals allowed of between 1.90 and 2.20. The drawback of selecting a goalie from this category is they are often among the most expensive goalies on the schedule.
The results show that these selections are the ones most likely to produce a desirable result. The Lowest Expected Goals goalie has the lowest chance for a subpar performance. Just 38.46% or 20/52 nights did this goalie score under 5.0 points. On the flip side these were also the goalies mostly likely to return a performance that would net eight points or more, one of only two selection methods that will produce an 8+ performance more than 20% of the time, currently sitting at just over 28% through the first 52 dates on the schedule. Somewhat ironically this is not the category of goalie with the most shutouts.
Top Value (TV) 26-26, 2.71 GAA, 1 SO (4.45 DK Pts)
I’m a little disappointed in this selection method. I was hoping it would return even more top level results. It is the top performing scientific method for finding the 10+ results, with five so far. Selecting this goalie for GPP’s will get you an elite performance one in every ten days. I just assumed it might actually be a bit higher on average.
For a goalie to be the top value of the night, they need to be cheap, and face alot of shots. This isn’t the recipe for cash game success, but the ceiling of the goalies in this category is rather high. This category has the highest GAA at 2.71, so there is considerable risk here.
This looks like a feast or famine category, with only 23% of the results being between the 5-8 point categories. If you don’t have the stomach for a poor performance (almost 60% of the time) it’s not worth going here. This is a category for MME players, but not for cash game players.
Top Value Favorite (TVF) 28-24, 2.63 GAA 4 SO (4.63 DF Pts)
The top value favorite means the highest projected value goalie that also happens to be a favorite on the schedule. Goalies in this category are usually very small favorites, just a few percentage points better than a 50-50 prop to win the game, the 28-24 record reflects that.
Looking at the overall chart however, I’m dismayed at the 50% of results being under 5.0 points, and if you look at the difference between the top value play and the top value favorite play is about a decrease by 7% of the poor results (less than 5.0) and a more or less equal (8.0) or better results. These goalies are slightly more likely to win, presuming they aren’t the same guy.
On average these goalies make 26.6 saves, which is on the lower end of the scientific methods, less of course than the Top Values, simply because they are likely to see more shots, thus will be more of a value. The Top Value Favorite play can never be expected to see a higher shot total than the Top Value play.
Highest Projected (HP) 23-29, 2.69 GAA, 3 SO (4.29 DK Pts)
The highest projected plays are almost always under dogs, because they are facing fast-paced teams. There have only been a game or two this year where I can remember the highest projected play being a favorite, Carey Price against the Maple Leafs about two weeks ago was one but it is rare. Holtby at home against the Pens in that 7-1 game might have been the other. If you have a great goalie (like Price or Lundqvist) facing a fast paced team, that’s about the only time a favorite is also the highest projected.
The Highest Projected and Top Value plays are tied for having the highest negative performances (57% of the time they score under 5.0) points. You can expect the highest projected plays to have a low average because of the fact they are often either on bad teams or facing one of the league’s best teams.
The upper echelon results of over 8.0 points, we’re only seeing scores fall in that range just over 17% of the time. Similar to the Top Value plays, I expected to see a few more positive outliers, which just aren’t there. In only 9 of 52 nights has this goalie scored 8+
Highest Projected Favorite (HPF) 28-24, 2.60 GAA, 4 SO (4.48 DK Pts)
A twist to the highest projected play, in the same line of thinking of the top value favorite, to be the goalie who fits in this category on any given night is expected to win but is also expected to see a heavy workload. Devan Dubnyk is the only goalie who has appeared on this list more than three times. Dubs was as favorite last night in Edmonton against the Monster; a late line shift once Gus was given the start.
Many times this is a goalie who’s facing one of the faster road teams. Outside of the Lowest Expected Goals scientific model, this is the category of goalies with the most positive results of 8.0 points or higher. 21.15% of all results fall into these categories. Chad Johnson has a few of them.
Mine (M) 20-13, 2.06 GAA, 4 SO (5.35 DK Pts)
I’ve gone through how I select a goalie in the round table article, so I wont reiterate those points again here, but the results I’ve have, have been very good for the 32 nights I was able to track them. I’ve had a 8+ result on 8 of 32 nights. What I didn’t like seeing though was the 45% of results below 5.0 points, which is on par with Vegas totals, I would like to see these numbers migrate into upper levels as the year wears on.
What I attribute some of the poor results to is that I’ve perhaps too many times taken goalies who I expect to win, but are facing overly poor teams and thus my goalie wont see many shots. Comparing these numbers to the scientific methods, the save totals fall directly in the middle. The GAA of my goalie selections is 2.06, which is better than the scientific methods.
Everyone who plays DFS is going to select goalies on different parameters.
- For me I’ve tried to strike a balance between a guy I think will win with expected saves.
- There is no need to leap for the heaviest Vegas favorite when the cheapest will do
- Lowest expected goals will get you the most positive results but it will often come at a cost
- The Top Value goalie has returned the most 10+ games
- The value plays and top projected plays are goalies who are playing good teams, which in turns leads to the opportunity for more goals to be scored against
- The favorite value plays and projected plays still happen to be playing good teams, their teams are simply expected to win.
- All of the non-Vegas choices are likely to be facing good teams
- You can beat Vegas numbers rather handily with a combination of the lowest expected goals method
- Combine science with a flare of art for best results each night
- The projections are a tool that will help guide your art
- I will compile this data for the rest of the season, and perhaps we can make some definitive conclusions with hundreds of data points rather than around 50 for each.