2016-17 WCHA Week 5 BELOW Rating

Week 5 BELOW for the 2016-17 WCHA Regular Season

2016-17 BELOW, Week 5

TeamWk5 BELOW+/-Wk2 EWP
3-BSU1644+1969.61%
7-MTU1644+3369.61%
8-MSU1626-1967.38%
4-BGSU153054.31%
6-LSSU1517-652.44%
2-UAF1489+648.42%
5-FSU14701945.69%
9-NMU1450-3342.85%
0-UAH138333.77%
1-UAA1367-1931.74%

Huskies Closing in on Beavers

A strong weekend by Michigan Tech against a middle-of-the-pack Northern Michigan squad saw the Huskies surge into a tie in BELOW with Bemidji State.  The Beavers lead the league standings with a perfect 6-0-0 mark that in some way validates that BELOW thought that they were better last year than their sixth-place standings finish.

Your BELOW rating always goes up when you win, but when you’re expected to win as the Beavers were — it’s rare that ABOVE picks one team sweeping as a higher probability than a split; Alaska-Anchorage was only picked to out-point the home side 18% of the time — you don’t get as much of a bump from those wins.  That’s how Tech gains 33 BELOW points while Bemidji gets just 19.

Split in Mankato

You’d be forgiven for thinking that Minnesota State would sweep Ferris State.  The teams were 194 points apart in BELOW, and Ferris State was only 19% likely to do better than  split.  However, Mankato didn’t finish out a sweep on the road at Alaska, and they didn’t finish at home with Ferris.  This tells us — and BELOW reflects it — that the Mavs may not as elite as they were last season.

Logjam at the top

You could still throw a blanket over Bemidji, Tech, and Mankato.  The only direct interaction between the teams is the Mavs’ home sweep of the Huskies.  Bemidji State plays both sides twice this season, and MSU will travel to Houghton.  That A tiebreaker could well come into play this season, as the three teams that very likely finish at the top all play each other four times.

Nanooks Blemish the Lakers

Lake Superior’s 50th anniversary season finally saw the Lakers take a poor mark, as Fairbanks pulled scored a real-overtime game-winner on Saturday to pick up the split.  A Laker win would’ve put them in second place early in the season.

Why are the Lakers so much farther behind the Bemidji-Mankato-Tech triumvirate despite being right behind them in the standings and being the last perfect team out West?  Simply put, the Lakers weren’t that good last year, and last year is generally a good predictor of this year, at least in the early going.  Also, the Lakers’ two opponents are Alabama-Huntsville and Fairbanks, and neither team is any great shakes in BELOW’s eyes.  Even if the Lakers had blown the Nanooks out last night, their BELOW would have maxed out at 1565.

Chargers and Falcons Don’t Much Help the WCHA’s PWR

Sorry, WCHA — Huntsville and Bowling Green didn’t do a whole lot for our Pairwise Ranking this weekend.  After netting just over two goals a game, the Falcons did go out and drop four on Miami on Saturday night to break into the win column.  UAH had St. Cloud tied through two on Friday and down by one on Saturday, but a bad third the first night and a bad second on Saturday night ended up in a sweep of the Chargers.

Looking Ahead

We have four conference matchups coming this weekend, with Huntsville off to Anchorage, Bemidji off to Fairbanks, Ferris on the bus to BG, and Mankato making the trip to the Soo.

To date, UAH has played FSU, MTU, and LSSU.  FSU has also played NMU and MSU.  MTU has also played MSU and NMU.  LSSU has also played UAF.  NMU has also played BSU.  BSU has also played BGSU and UAA, and that closes the loop.  After three more weeks, every team will have played someone from each of the three tiers of the league (BSU-MTU-MSU, BGSU-LSSU-UAF-FSU-NMU, UAH-UAA) save MSU, UAF, and NMU, who won’t have played a bottom-two team.  (Yes, the Nanooks and Seawolves face off in non-conference play.)

ABOVE 1.3 Goals

I will  daisy-chain Weeks 6 and 7 together — and maybe 8 and 9 if that isn’t too time-consuming.  The goal would to not only calculate BELOWs to the end of November but to give probabilities for how the standings will look then.  I think that’s more valuable than simple calculation of a number that has intuitive to value to me but maybe not you.  From there, we’re not all that far away from routine projections for the end of the year, followed hopefully by what-if scenarios for you to try.

I also want to revisit goal differential with an eye to FiveThirtyEight’s point differential coefficient for NFL calculations.  When I’ve looked at the goal differential bonus before, I’ve noted that it’s generally the big teams who win out big.  Why reward them more for doing their job than, say, an upset win by a lower-ranked team?  I’m not sure that their methodology is the correct one, and I’m sure that those coefficients are correlated to a model that I don’t quite have built yet.

If I apply this, I’ll take it back to the beginning of the season, which will change the BELOW ratings that I’ve published to-date.  That’s fine — I’ll just re-calculate and start over from Week 5 forward.

Thanks as always for reading along.

2016-17 WCHA Week 5 Predictions (ABOVE 1.2)

2016-17 WCHA Week 4 Predictions

2016-17 Week 5: Alaska-Anchorage at Bemidji State

ResultProbability
BSU sweeps32.880%
Splits30.756%
BSU 5 points (W, OTW)9.112%
BSU 4 points (W, OTL or OTW, OTW)8.291%
UAA Sweeps8.185%
UAA 4 points (W, OTL or OTW, OTW)4.963%
UAA 5 points (W, OTW)4.459%

2016-17 Week 5: Alaska at Lake Superior

ResultProbability
Splits35.481%
LSSU sweeps21.464%
UAF Sweeps16.265%
LSSU 5 points (W, OTW)7.257%
UAA 4 points (W, OTL or OTW, OTW)6.845%
UAA 5 points (W, OTW)6.374%
LSSU 4 points (W, OTL or OTW, OTW)6.314%

2016-17 Week 5: Ferris State @ Minnesota State

ResultProbability
Splits33.022%
MSU sweeps30.582%
FSU Sweeps9.652%
MSU 5 points (W, OTW)8.722%
MSU 4 points (W, OTL or OTW, OTW)7.902%
FSU 4 points (W, OTL or OTW, OTW)5.284%
FSU 5 points (W, OTW)4.835%

2016-17 Week 5: Northern Michigan at/vs Michigan Tech

ResultProbability
Splits34.293%
MTU Sweeps26.784%
NMU sweeps12.191%
MTU 5 points (W, OTW)8.196%
MTU 4 points (W, OTL or OTW, OTW)7.247%
NMU 4 points (W, OTL or OTW, OTW)5.848%
NMU 5 points (W, OTW)5.441%

Welcome to ABOVE 1.2

This week, I’ve pulled in two- and three-goal wins as promised.  I haven’t reflected this yet in terms of probabilities of, say, a sweep with two three-goal margins — hey there, Lake Superior — but ABOVE does make the calculation for me, which is a stepping-stone to the next version.  I haven’t decided what to tackle yet, but soon it will be that I have to put these runs into a database.  When I get to that point, the output of these runs will go up at wchaplayoffs.com.  I’m not sure how I’ll output it just yet, but it’ll be something.

The value of the two- and three-goal wins comes when you start daisy-chaining weeks through each other.  These are the match-ups for Weeks 6 and 7:

  • Week 6: UAH @ UAA, BSU @ UAF, FSU @ BGSU, MSU @ LSSU.  OFF: NMU, MTU
  • Week 7: UAF @ UAH, LSSU @ MTU, BSU @ MSU, BGSU @ NMU.  OFF: UAA, FSU

A two-week ABOVE calculation would have the output from Week 5 as a starting point.  As an example, we’d know the results of Alaska-Anchorage’s trip to Bemidji State, so their matchup with Alabama-Huntsville in Week 6 would give Huntsville a set of BELOW outcomes that would range from around 1350 to 1450, depending on how week 6 goes.

In much the same manner, we’d have a range of possible results for Alaska and Bemidji State.  So let’s say that it’s 1440-1540 for Fairbanks depending on what they do in Bemidji.

As such, it could be that UAH would have a Week 5 BELOW of 1450 while UAF’s if 1440.  The ABOVE calculations for that matchup would look pretty even — something like the estimates in Week 4 for Ferris State-Northern Michigan.  But if it’s something like 1350-1540, and your estimates are far more like Minnesota State-UAF that week.

“But wait,” you say.  “The MSU-UAF gap was really big and yet the result was a split.  What gives?  Your system sucks!”  I hear you, but now hear me out.

Each one of these runs of predictions is done in a programming loop: four series are done at the same time, with the results of the first game feeding the second one.  Once you simulate that a certain number of times — 100,000 is something that you can run in about a minute — you have a pretty good idea of what the week will be like.

Right now, these series are independent of each other — that is, those results tables up there don’t care if UAA sweeps BSU with two three-goal wins while UAF creams LSSU.  But if you take Week 6’s results and then use them as inputs for Week 7, you’ve suddenly got a wider range of results, and the aggregate probabilities of any week take on greater meeting.  Suddenly, it does matter that the Seawolves routed the Beavers while the Nanooks blitzed the Lakers, because we’re dealing with the concept of conditional probability.

The goal is going to be to simulate the rest of the season given the estimated results of a series of simulations.  Remember when UAH looked hot early for sweeping FSU on the road?  That doesn’t look that smart now.  I could have created an estimate of how the league would look after that sweep, and there would be possible outcomes that look like what we have in reality.  Each game’s result not only gets us closer to the end of the season, but it gives us more uncertainty in the estimate.

So perhaps ABOVE 1.3 will be estimating Week 6 and 7 together and then ABOVE 1.4 be estimating Weeks 7, 8, and 9 together and out-putting estimate for league standings.  By that point, UAA will have played eight games and UAH, BSU, and MTU will be at twelve, and that’s a good point to evaluate the strength of the model’s assumptions (how much goal differentials should matter, how volatile the model should be overall, etc.).

The end goal will be that I’ll output a weekly estimate — and maybe one on Saturday — of where teams are when it comes to the playoffs.  This is the kind of things that you see with the 2016-17 NBA Predictions and 2016 NFL Predictions at FiveThirtyEight.  The FAQ-style description of the NFL predictions is relevant and probably less dense than the above 500 words.

See you on Monday with results, new BELOWs, and some thoughts on where I’m going with ABOVE.

2016-17 WCHA WEEK 4 BELOW Rating

Week 4 BELOW for the 2016-17 WCHA Regular Season

2016-17 BELOW, Week 4

TeamWk4 BELOW+/-Wk2 EWP
8-MSU1645-1669.73%
3-BSU1625067.25%
7-MTU1611065.45%
4-BGSU1530054.31%
6-LSSU1523+4853.31%
2-UAF1483+1647.56%
9-NMU1483+1847.56%
5-FSU1451-1842.99%
1-UAA1386034.16%
0-UAH1383-4833.77%

The Lakers Impress

While Alabama-Huntsville seems to be in a free-fall after a road sweep at Ferris State — they’re 3rd in league points but have already played 21% of their schedule — Lake Superior went down to Dixie and took care of business, dropping two 3-goal elbows on the Chargers.  The Lakers were 56% (Friday) and 63% (Saturday) picks to win, and they did so handily each time.

For this week’s predictions, I’ll look to break out 2- and 3- goal wins.  These do make a difference, and with some separation at the top, it makes sense to consider it.  I noted back last January that the big wins in the season to that point were generally in situations where there was a big gap in BELOW; that the gap for the Lakers was 44 and then 96 points is more in the realm of what we saw last year when UAH started fast with a road sweep in the Soo that finished with a 5-0 drubbing — the high-water mark for the Chargers’ 2015-16 season.

What does all this mean for the Lakers?  Well, BELOW — being reflexive and moderately transitive — is sorting the middle of the league by saying that they’re just about even with Bowling Green, which continued to struggle mightily in being swept by Ohio State.  Because UAH beat Ferris and NMU did, too, BELOW thinks that the Lakers are above the Chargers and Bulldogs and maybe better than the Wildcats.  LSSU does have just the two games under their belt, but it’s a promising start.

NMU Pulls Ahead of FSU

BELOW doesn’t care much for the shutout — the adjustment bonus is just 10 points for the shootout because, you know, it just doesn’t stinking matter that much.  I almost calculated it as a tie and went on with my life.  The “win” didn’t move the needle much, but a solid win on Friday pushes Northern Michigan back closer to 1500 / .500, their standard place.

Mankato Stumbles in Fairbanks

I said last week that I thought that Minnesota State would sweep Alaska, and the numbers agreed with that, too.  A resounding 7-1 trouncing seemed to bear that out, although the change in BELOW was just 15 points each direction because the Mavericks were expected to get positive results (wins in regulation or overtime) 75% of the time.  That number jumped to 78% for Saturday’s matchup, but the homestanding Nanooks delighted the fans at the Patty with a two-goal win.  Because the teams had a big gulf, and because order matters in BELOW,  a split leaves Fairbanks +16 on the weekend.

Looking ahead

There are four conference tilts this weekend:

  1. Alaska-Anchorage joins the 2016-17 WCHA party with a trip to Bemidji State.  This looks pretty lop-sided (1625-1386, or 80% positive results for the Beavers), but it just comes down to the fact that we have no real data on the Seawolves just yet.  This is one of those need-to-win series for BSU, being at home against a (presumably) weaker team.  Wins from UAA would close this gap pretty significantly.
  2. Alaska traves to the UP to face Lake Superior.  This will give us more information in answering whether either team is for real.
  3. Ferris State travels to Minnesota State.  There’s a little doubt if the Mavericks are the tops of the league after a rough night in Alaska; there’s a lot of doubt in Ferris fans’ heads if the Bulldogs will stay in the top half of the conference.  This looks to be an easy series for the Mavs, but so did the Fairbanks trip.
  4. Northern Michigan and Michigan Tech face off in a classic UP rivalry.  Will the Wildcats have Shane Sooth and Dominik Shine at their disposal?  Can Michigan Tech find a goaltender to be their rock?  This one should be interesting and ought to go a long way in determining whether the carryover leg up that the Huskies have in BELOW is worth keeping.

Thanks as always for reading.  I’m hoping to have a simulator later this season to help you noodle around with ABOVE and BELOW calculations to give you a sense of what’s going on here.  Unless you follow the math, I imagine that this can be a bit impenetrable at times.  I’ll keep trying to make it make sense!

2016-17 WCHA Week 4 Predictions (ABOVE 1.1)

2016-17 WCHA Week 4 Predictions

ResultProbability
Splits31.579%
LSSU sweeps19.629%
UAH sweeps14.747%
LSSU 4 points (W, OTL or OTW, OTW)9.517%
LSSU 5 points (W, OTW)8.755%
UAH 4 points (W, OTL or OTW, OTW)8.246%
UAH 5 points (W, OTW)7.527%
ResultProbability
Splits29.323%
MSU sweeps28.221%
MSU 4 points (W, OTL or OTW, OTW)11.366%
MSU 5 points (W, OTW)10.588%
UAF sweeps8.424%
UAF 4 points (W, OTL or OTW, OTW)6.439%
UAF 5 points (W, OTW)5.638%
ResultProbability
Splits31.723%
FSU sweeps17.288%
NMU sweeps16.877%
FSU 4 points (W, OTL or OTW, OTW)8.905%
NMU 4 points (W, OTL or OTW, OTW)8.853%
FSU 5 points (W, OTW)8.240%
NMU 5 points (W, OTW)8.113%

Welcome to Above 1.1

I mentioned last week that I was then using what I was calling the 1.0 algorithm — the same one I used last year that calculated wins, ties, and losses — as a tried, true, tested system.  It got me some pretty decent results: Tech and Bemidji sweeps were the highest-rated outcomes, and the results were there.

One thing that the model doesn’t take into account yet is the goal differential.  Currently, ABOVE 1.1 does the following:

  1. Calculates an expected winning percentage, which the model uses as a center point.
  2. Sets a 20% band around the center point.
  3. Four bands exist: below the 20% band, a win for one team; an overtime win between that band line and the center point; above the center point, it’s an overtime win and win for the other team.
  4. The dice roll generates a random number between 0 and 1 and slots itself into the band.
  5. ABOVE then re-calculates BELOW based on the nature of the win.

The key feature here is that 20% value.  Last season, just about one-in-five WCHA regular season contests went the full 65 minutes.  Will the new overtime rules change that calculus?  Huntsville and Michigan Tech played the first league game to go past 65 minutes in the league’s 10th contest.  We simply don’t have enough data to know when teams are going to push past 65 minutes to guarantee themselves a point.  Will it be something close to 20%?  We don’t know.

What I’m looking to do in the future is a self-correcting system that calculates the number of overtime games and then uses that to determine if the league is drawing down on the number of 65-minute games.  If they are, the percentages will move more toward pure wins/losses and away from this non-hockey stuff.  [I’m with you, Walt.]

ABOVE 1.2

The goal for next week is to account for the wider variances in goal differential.  I’ll create more bands — one-goal regulation wins, two-goal regulation wins, and three-goal regulation wins — and then use those bands and the dice roll to help create a stronger signal for the rise of better teams.  Will this make a huge change?  Probably not, but it’ll have some effect.

A note on the lack of BELOW calculations in this week’s tables

It’s important to note that there are multiple ways for a series to go; order matters.  Right now, you get a spread of 16 BELOW values per run: a first-game win against a second game win/OT win/OT loss/loss, then a first-game OT win against the same, etc.  If you add in two-goal and three-goal wins, it’ll get even wider.  Also, since order matters, the outcome is dependent upon that.

As such, I can’t reasonably provide you a BELOW output for each of the seven scenarios in the above tables at this point — there are four alone for the splits!  When I have this data in a database — hoping for the end of the year — this data will be more readily available.

My (gut feeling) picks

  1. LSSU – UAH split
  2. MSU sweeps
  3. FSU 4 points

2016-17 WCHA Week 3 BELOW Rating

Week 3 BELOW for the 2016-17 WCHA Regular Season

2016-17 BELOW, Week 3

TeamWk3 BELOW+/-Wk2 EWP
8-MSU1661071.64%
3-BSU1625+3567.25%
7-MTU1611+2565.45%
4-BGSU1530054.31%
6-LSSU1475046.41%
5-FSU1469045.55%
2-UAF1467045.27%
9-NMU1465-3544.98%
0-UAH1431-2540.20%
1-UAA1386034.16%

Hail the Beavers

BELOW thinks that there’s a new third member of the WCHA’s Big Three — and it’s Bemidji State.  While the season is young — we’ve played just 7% of the schedule — the Beavers have impressed, scoring 10 goals in league play while allowing just two (both to Bowling Green).

Whither the Falcons

We may wish to temper our enthusiasm for Bemidji, as out-of-conference play with Western Michigan would seem to indicate that the Falcons are struggling now, coming away with a 4-4 tie and an 8-2 loss to a squad that was 8-25-3 last season.

Mankato Beat Tech, and Tech beat UAH, …

BELOW’s reflexive nature means that you can apply the transitive property a bit.  Minnesota State crunched Michigan Tech last weekend, so the Mavericks are ranked higher.  Bemidji State splits the two, and from there, it’s a dog’s breakfast of half the league huddled together in the middle.

Roll Chargers

Despite being second in league points — a fact that still seems weird to me as an alumnus and someone who’s watched this team struggle for a decade — BELOW says that Alabama-Huntsville just isn’t that good based on past seasons and this weekend’s effort.  “Show me,” the algorithm says.  UAH’s efforts this weekend in Houghton did finally garner them a point — their first one in 12 games against Tech — but they lost ground.

Bitzer tames the Wildcats

Northern Michigan drops in the rankings because they couldn’t score this weekend after dropping eight goals on Wisconsin the weekend prior.  Atte Tolvanen can’t do it all, and Dominik Shine has yet to play a game for Walt Kyle’s squad.

Looking ahead

  • Lake Superior travels to Huntsville, a series that will determine if UAH is good — they swept another team in that rough range of BELOW — or if the first weekend was a fluke.  It’s fitting that the Chargers face the Lakers here, as UAH’s peak in 2015-16 was a sweep in the Soo.
  • Mankato travels to Alaska, a contest that they should handle easily (75% expected winning percentage).  If the Mavericks falter, it won’t just be the standings that reflect it.
  • Northern travels to Ferris State in a matchup of teams that were swept at home by lower-half squads of last season.  The two schools have hovered around 1500 since I started calculating BELOW.

 

2016-17 Week 3: ABOVE v1.0 Predictions

Not far ABOVE you

When I say v1.0 of ABOVE, my adjustment algorithm for re-rating BELOW, I’m talking about a straight probability-based system for calculating probabilities of the weekend’s contests that does not take into account the new overtime win/loss situation or goal margin.  We’re just going to do straight predictions.

I will talk more about ABOVE soon — as soon as I know why my Python script isn’t giving me reasonable data!

Here goes!  I’ll leave a discussion of the math for later down; if you’re not interested, stop with the tables.

Alabama-Huntsville at Michigan Tech

ResultProbabilityRoad BELOW (Change)Home BELOW (Change)
Tech sweeps49.18%1420 (-36)1622 (+36)
Tech wins, UAH wins18.70%1480 (+24)1562 (-24)
UAH wins, Tech wins18.26%1471 (+15)1571 (-15)
UAH sweeps13.86%1531 (+75)1511 (-75)

Bemidji State at Northern Michigan

ResultProbabilityRoad BELOW (Change)Home BELOW (Change)
Bemidji sweeps42.85%1631 (+41)1459 (-41)
Bemidji wins, Northern wins19.82%1571 (-19)1519 (+19)
Northern wins, Bemidji wins19.42%1581 (-9)1509 (+9)
NMU sweeps17.91%1521 (-69)1569 (+69)

The paradox of order

Have you noticed that the weaker team (Huntsville, Northern) benefits in terms of BELOW by losing on Friday and winning on Saturday?  Doesn’t this seem weird?  Let’s consider this from a mathematical and hockey perspective.

The better team is generally going to win any game that’s played — that’s probability, that’s hockey, that’s life.  There’s a reason that UAH is 0-10-0 at Tech all-time — MTU has been very good and UAH has been very bad.  If the Huskies beat the Chargers tonight — and it’s about 68% likely that they will — then BELOW’s rating system will say, “Yeah, we thought that UAH was better after their wins at Ferris State, but they’re just not better than MTU!” and rate them accordingly.  Conversely, if UAH breaks the streak, BELOW will say, “How about them Chargers!  They’re 3-0-0!”

But if Tech wins, the gap between the Huskies and Chargers widens, and MTU goes from 68% likely to win to 72% likely.  As such, an upset win on Saturday says, “Wow, these Alabama boys finally figured out a way to win in their 12th try!” and give them just a little more credit for winning that second game — after all, BELOW thought that the gap was wider.  Also, hockey fans are likelier to expect the sweep after the better team wins the first game.

Conditional probability

This is all a factor of conditional probability: if Tech (or Bemidji) wins, they’re heavier favorites to win the next time around.  In any ABOVE-iterated matches of BELOW-rated teams, the team with the higher BELOW is expected to sweep — and as  you can see above, the bigger the gap, the higher the likelihood of the sweep.

Furthermore, consider the fact that the probability of a Tech sweep is 49%, a good bit higher of the likelihood of a split (37%).  Compare that to the likelihood of Bemidji sweep (43%) and a split (39%).  If the matchup were teams closer in BELOW, a split would be likelier than a sweep either way, and as the teams close to part — like Ferris (1468) and Alaska (1467) are right now — the probability will come out to about 25% for either team sweeping and 50% of a split.

See you on Monday

I’ll re-calculate BELOW over the weekend and have a new post available on Monday.  I’ll probably write about goal differentials, and given my luck, a game will go past the first overtime and I’ll have to deal with that.

2016-17 WCHA Week 2 BELOW Rating

Week 2 BELOW for the 2016-17 WCHA Regular Season

2016-17 BELOW, Week 2

TeamWk2 BELOW+/-Wk2 EWP
8-MSU1661+5771.64%
3-BSU1590+5362.67%
7-MTU1586-5762.13%
4-BGSU1530-5354.31%
9-NMU1500050.00%
6-LSSU1475046.41%
5-FSU1469045.55%
2-UAF1467045.27%
0-UAH1456043.70%
1-UAA1386034.16%

With convincing weekends, the two Minnesota-based teams rose sharply in this week’s BELOW rating, with Minnesota State rocketing to the top and Bemidji State — well-rated down the stretch last year — jumping behind them into 2nd place, albeit just by a nose over Michigan Tech.

Implications

BELOW thinks that Bemidji and Tech are pretty much equal so far, and given that Tech is 0-4-0 and hasn’t looked very good in any of those losses, this is a reasonable assumption.  Remember that BELOW is looking back to last year as well as this year, so presuming that Tech is still pretty good — they’d be expected to beat Northern Michigan three games in five — isn’t a big stretch.

Looking forward

There are two WCHA series on the docket for this weekend: Tech hosts Alabama-Huntsville (MTU EWP 68%), while Northern hosts Bemidji (BSU EWP 63%).   Look for predictions on Friday (no, really).

2016-17 Week 1 BELOW Rating

Week 1 BELOW for the 2016-17 WCHA Regular Season

2016-17 BELOW, Week 1

TeamWk0Wk1+/-Wk1 EWPRank
7-MTU16431643069.49%2(t) - 0
8-MSU16041604064.54%2(t) - 0
4-BGSU15831583061.72%2(t) - 0
3-BSU15371537055.30%2(t) - 0
9-NMU15001500050.00%2(t) - 0
6-LSSU14751475046.41%2(t) - 0
5-FSU15181469-4945.55%10 - 0
2-UAF14671467045.27%2(t) - 0
0-UAH14071456+4943.70%1 - 6
1-UAA13861386034.16%2(t) - 0

The Results

Well, sometimes that 9.2% shot happens.  The news of the WCHA this week — the only news, but big news nonetheless — is that the Alabama-Huntsville Chargers, who’ve never finished higher than 8th in the league in their three seasons in the WCHA, went out and swept USCHO #20 Ferris State 2-1, 4-3 on Saturday and Sunday.  UAH hadn’t won their first two games of the season since 2001-02 (my senior year at UAH).  It was the Chargers’ second WCHA sweep and first-ever sweep of a ranked opponent.

The BELOW Rating Impact

Yes, yes, I hold an engineering degree from UAH, so this is special to me.  But we also see that it’s big for BELOW, as the pre-season gap in BELOW — 111 points — had every expectation that the Bulldogs would win at least one, and likely both, games this weekend, with an expected winning percentage of 64.5% in their first matchup.  That just didn’t happen.

As a result, the Chargers pick up 49 points in BELOW and now just barely trail the Bulldogs, 1469-1456, in that rating.  UAH jumps from a distant 9th to a close one and is within striking distance of 5th with a good weekend against Michigan Tech.

It’s important to note that this change would still mean that UAH would only be expected to win ~44% of its matches against a league-average (BELOW 1500) opponent, compared to ~37% with last week’s rating.  BELOW now thinks that UAH is merely mediocre and not bad.

Coming Up

This week’s matches should really move the needle, as Bowling Green travels to Bemidji State and Michigan Tech travels to Minnesota State.  I will have those projections later this week.

2016-17 Week 1 ABOVE Predictions

It’s October!  It’s my birthday!  It’s time for hockey!  IT’S TIME FOR NUMBERS!

I won’t bury the lede here: in a 1,000,000-trial Monte Carlo simulation, FSU sweeps about 23.5% of the time, UAH sweeps about 11.6% of the time, and the rest falls in between.  My model is still a work in progress (mainly in the data-management side; the calculations are easy), but it’s pretty simple.

Event% Chances
FSU Sweeps16.5%
Split54.3%
UAH Sweeps9.2%
New OT Involved20%

How I get to these numbers – it starts with BELOW

BELOW stands for Bringing Elo to the WCHA.   Elo-style rating systems work on a simple principle: you can go into a match with an estimate for the likelihood that contestant A will win, known as an expected value.  If contestant A wins, they are rewarded with a jump in their rating commensurate with the expected value.  If contestant A had a high expected value, they won’t receive much of a change.

Conversely, contestant B could pull off the upset, and an Elo rating system will reward them for that.  I explained this moderately well in December, and I’ll be refining that soon.  If you’re new here, this is where you should go.

How we got here – re-calculating BELOW

Ferris State, despite all the wonder that it did through the postseason, starts the 2016-17 season with a BELOW rating of 1518, or just a hair above an average baseline of 1500.  Alabama-Huntsville comes in at 1407.  Both teams have been regressed to the mean. by one-third; UAH, since they were farther from 1500 (1361) than the Bulldogs (1527), moved closer.

Introducing ABOVE

ABOVE is Adusting BELOW through Operative Value Experiments, the model that uses that Elo-style rating and iterates, time after time.  This used to be pretty easy, actually: in the regular season, you would win, lose, or tie; now you can win, win in 5×5 overtime, win in 3×3 overtime, or win in a shootout.  I used to assign a different value adjustment based on whether a non-tie was settled in OT, but I didn’t predict those on a macro level.

I actually am predicting 16 outcomes per weekend: two combinations of four results: win, overtime win, overtime loss, and loss by team A.  I could spit that data out at you, because to ABOVE, it actually matters whether Game #001 is a UAH win even if it’s a given that Game #002 is an FSU win.  It doesn’t matter to you.

A brief sidebar as to why it does matter to ABOVE

Think about it, though: if a team with a 1400 ranking beats a team with a 1500 ranking — pretty close to where Ferris-Huntsville is — a split will move the teams closer, but order matters.  To our example: a 100-point differential means that team A is expected to win 64% of the time.  An upset means that team B get 0.64 x 40 = 25.6 points, and the spread is now 1474 – 1426; this ELO Difference Calculator will show that the expected win for team A is now just 57%.  But if the reverse happens, and FSU wins first, the gulf is wider between the two teams, and the value of an upset is greater.  We saw this last season also with Ferris State opening.

In short, a split does not mean that ABOVE will calculate the same BELOW regardless of order.

Say, when do we get that model?

Well, there are two things at work:

  1. I’m flying back today after 12 nights in Iceland.  I haven’t had a chance to work on it while I’ve been here.
  2. I’m not sure yet how I want to model the new overtime.

New overtime

Now, the new OT is weird.  How do you model 3×3?  You can do it two ways:

  1. It’s a crapshoot, so model it as 50-50 and go on with life.
  2. Do some guessing based on GF/GA differential or something silly like that.

There’s also the question of how often teams will push to go into the extended overtime.  My current assumption in ABOVE is, “It’ll be like last year,” but I kinda doubt that.  A weaker team is unlikely to have the skaters to win in 3×3 or in a shootout, and they may see the prospect of getting three points greater than playing for just one.  We don’t know yet!  So I’m trying to go with what makes sense for now, and that’s just … not yet done.

Questions?

Please leave a comment below or hit me on Twitter @wchaplayoffs.  Today is a travel day for me — KEF to BOS — but I’ll see stuff as I have time.