<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Neil Pierre-Louis on Medium]]></title>
        <description><![CDATA[Stories by Neil Pierre-Louis on Medium]]></description>
        <link>https://medium.com/@neilpierre24?source=rss-a5bd1d2b057f------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*etT7OgjMjFq_SjYYkbk2dg.jpeg</url>
            <title>Stories by Neil Pierre-Louis on Medium</title>
            <link>https://medium.com/@neilpierre24?source=rss-a5bd1d2b057f------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sat, 21 Nov 2026 14:08:55 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@neilpierre24/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[2026 NHL Playoff Predictions]]></title>
            <link>https://medium.com/@neilpierre24/2026-nhl-playoff-predictions-9bfec43c34c5?source=rss-a5bd1d2b057f------2</link>
            <guid isPermaLink="false">https://medium.com/p/9bfec43c34c5</guid>
            <category><![CDATA[hockey]]></category>
            <category><![CDATA[sports-analytics]]></category>
            <category><![CDATA[nhl]]></category>
            <dc:creator><![CDATA[Neil Pierre-Louis]]></dc:creator>
            <pubDate>Sat, 18 Apr 2026 17:46:12 GMT</pubDate>
            <atom:updated>2026-04-18T18:16:51.577Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/750/0*R0LnZciugf_8A29W.jpg" /><figcaption>Joe Corvo after beating Buffalo in Game 2 of the 2007 ECF. Photo by Mike Lynaugh: <a href="http://www.mikelynaugh.com/2007/pages/IMG_0964.htm">http://www.mikelynaugh.com/2007/pages/IMG_0964.htm</a></figcaption></figure><p>The 2025–26 NHL season has <a href="https://x.com/pierreanalytics/status/1978284060163072338?s=20">lived up to the hype</a> this season, and I could not be more excited to share my postseason predictions for a field that has some newer blood in it. Let’s dive in.</p><h4><strong>The Model</strong></h4><p><a href="https://medium.com/@neilpierre24/2025-nhl-playoff-predictions-ef1c3dae092a">Similar to last season</a>, I am extremely lazy (busy with work and life) and did not create a game prediction model for the regular season, but I was able to spin up a last minute version for the playoffs. This model is trained on shot and game data starting from 2013 on. For the results below, I ran 10,000 simulations of each series based on home and away probabilities drawn from this XGBoost odds model which uses the following predictors:</p><ul><li>Home and Away ELO (A model made by <a href="https://x.com/ryanabest">Ryan Best</a> and <a href="https://x.com/Neil_Paine">Neil Paine</a> when they were at <a href="https://fivethirtyeight.com/methodology/how-our-nhl-predictions-work/">538</a>, great article)</li><li>Home and Away rolling EV xGF% (Even Strength Expected Goals For %) over ten games</li><li>Home and Away rolling GSAx (Goals Saved Above Expected) over ten games</li><li>Home and Away skater gamescore impact for each forward and defenseman slot (NEW)</li></ul><p>The way I calculated this gamescore impact was by using a very <a href="https://x.com/pierreanalytics/status/1822753236060713421?s=20">trivial version</a> of <a href="https://x.com/domluszczyszyn?lang=en">Dom Luszczyszyn</a>’s <a href="https://t.co/2KIWzRjbNf">gamescore model</a>. I calculated per second gamescore averages over the last 40 games before a game, with the most recent games mattering the most, and then multiplying that by a projected TOI that was calculated on the past 40 games to get 12 home and away forward and 6 home and away defense predictors. This addition was prompted by my want to have individual players factor into the model with the combination of not having projected lineups available.</p><h4>Resources</h4><p>Before I dive in, I want to list out the resources I used for this article! Shoutout to <a href="https://x.com/BennettSummy">Bennett Summy</a>, who has been with the Penguins for a year but worked on these team cards from my website that you see below.</p><ul><li><a href="https://www.pierreanalytics.com/#/">https://www.pierreanalytics.com/#/</a></li><li><a href="https://hockeystats.com/">https://hockeystats.com/</a></li><li><a href="https://www.naturalstattrick.com/">https://www.naturalstattrick.com/</a></li></ul><h4>Metropolitan Division</h4><p>(1) <strong>Carolina Hurricanes (53–22–7) vs (6) Ottawa Senators (44–27–11)</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/990/1*udR7OCk3LRmiZ5kLWXocnw.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/977/1*03qw938HB0-Caf88ReW3rg.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*EW0jxnpa6GHCsUrx8nPpUg.png" /></figure><p>As a Hurricanes fan myself, I was dreading this matchup. The Ottawa Senators have been a ridiculously good play-driving team with elite defensive play that stems from a <a href="https://www.pierreanalytics.com/#/8481606/year/2026">deep blue line</a> and <a href="https://www.pierreanalytics.com/#/8481596/year/2026">responsible two-way forwards</a>. Their goaltending was unfathomably atrocious near the midway point of the season, but once that regressed, the Senators have not looked back and have been one of the best teams in the NHL since the Olympic break.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*15_fx2Be2pGEbhU5tR8gTg.png" /><figcaption>Cumulative GSAx via TopDownHockey and Jfresh’s hockeystats.com</figcaption></figure><p><strong>How CAR will win: </strong>Big contributions from the Stankoven line and winning special teams. I imagine, especially for the games in Ottawa, that the top line will be pretty well neutralized, but that could give players like <a href="https://www.pierreanalytics.com/#/8482809/year/2026">Jackson Blake</a> and <a href="https://www.pierreanalytics.com/#/8482702/year/2026">Logan Stankoven</a> a chance to shine.</p><p><strong>How OTT will win: </strong>Staying out of the box and getting Ullmark a solid foundation early in the first couple of games.</p><p><strong>Final Prediction: OTT in 6</strong></p><p><strong>(7) Pittsburgh Penguins (41–25–16) vs (8) Philadelphia Flyers (43–27–12)</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/988/1*GY1S89NlXToYDAgHbelpJA.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/963/1*FujaKZmbB_ZrA0ta4Yl1bw.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*-CkafiydcJDb8TasvigYYg.png" /></figure><p>Before you dive into anything related to this series, just do me a favor and watch game 3 of the 2012 Eastern Conference Quarterfinal matchup between these two squads. Absolute cinema. Four players remain from that series on these two teams and will be looking to make a big impact in this rivalry series.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/957/1*3hOEfAggcl1gWB58zLbcuw.png" /></figure><p><strong>How PIT will win: </strong>Big games from <a href="https://www.pierreanalytics.com/#/8482475/year/2026">under the radar off-season/in-season acquisitions</a>.</p><p><strong>How PHI will win: </strong>Stay out of the box.</p><p><strong>Final Prediction: PIT in 7.</strong></p><h4>Atlantic Division</h4><p><strong>(2) Buffalo Sabres (50–23–9) vs (5) Boston Bruins (45–27–20)</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/992/1*4ICDjgi7GJyNtu7pFSIntA.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/962/1*r14xevCB6OmsMjk_802Fvg.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*I-PqBYFISecNOd1VbAFPTQ.png" /></figure><p>You could argue both of these teams would be pleasantly surprised to be where they are at in this point of the year with both of their preseason expectations. The Sabres were most commonly projected to be right outside of the playoff mix; instead, they have been the best team in the NHL since the new year and have claimed their first division title since 2010. The Bruins have had an interesting retool/transition year, not a lot of good underlying data with their play, but they have struck gold where it matters at the end of the day: goaltending and finishing.</p><p><strong>How BUF will win: </strong>The task is much harder than said, but if you can get a line with <a href="https://www.pierreanalytics.com/#/8484145/year/2026">Zach Benson</a> and <a href="https://www.pierreanalytics.com/#/8482659/year/2026">Josh Doan</a> to do a good job of neutralizing <a href="https://www.pierreanalytics.com/#/8477956/year/2026">Pastrnak</a> and <a href="https://www.pierreanalytics.com/#/8479987/year/2026">Geekie</a>, I think Buffalo takes this pretty easily given their depth compared to the rest of the Bruins forwards.</p><p><strong>How BOS will win:</strong> Counterattacking play, a Swayman masterclass, and getting good matchups against <a href="https://www.pierreanalytics.com/#/8474568/year/2026">Luke Schenn</a> and <a href="https://www.pierreanalytics.com/#/8479378/year/2026">Logan Stanley.</a></p><p><strong>Final Prediction: BUF in 6.</strong></p><p><strong>(3) Tampa Bay Lightning (50–26–6) vs (4) Montreal Canadiens (48–24–10)</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/997/1*uHiIC4Fs0xwEokQBVsy4dw.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/928/1*D24vIDDQQ64y9PcFephxDQ.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*M9OdRlKtxEqS183GV2vbTQ.png" /></figure><p>There was a lot of debate about how Montreal would follow up their playoff drought ending season given some underlying concerning defensive issues, and they have followed it up well with an extremely strong performance from some franchise pillars with one who is emerging as a top option in the goalie crease. This season was business as usual for the Lightning. Despite a rough start, they sit at the top of the Eastern Conference in goal differential and look to get back out of the first round since 2022.</p><p><strong>How TBL will win: </strong><a href="https://www.pierreanalytics.com/#/8476883/year/2026">Vasilevskiy</a> regains playoff form.</p><p><strong>How MTL will win: </strong>Keep the 3rd defensive pairing sheltered and really on <a href="https://www.pierreanalytics.com/#/8482487/year/2026">Dobeš</a> to have a couple of games that resembled the form he ended the regular season in.</p><p><strong>Final Prediction: TBL in 5.</strong></p><h4>Central Division</h4><p><strong>(1) Colorado Avalanche (54–16–11) vs (8) Los Angeles Kings (35–26–20)</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*CV0kLqrTkPg-ROsJHO0DOA.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/985/1*6eZjs-30RYCZpjQrXgCvRQ.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*d5P7vlMJHMNVTlrFtHIxmg.png" /></figure><p>Colorado has been pretty much flawless for most of the season other than their powerplay. The best team in the league that is an offensive juggernaut will take on a stingy Los Angeles Kings team that won just 22 games in regulation this year.</p><p><strong>How COL will win: </strong>I do not think that there are many avenues where Colorado loses this series, but I think they can ensure a quick victory if (big if) they can get their powerplay together with how stifling the Kings are defensively at 5v5.</p><p><strong>How LAK will win: </strong>Leveraging their great defensive play and forcing some overtime hockey.</p><p><strong>Final Prediction: COL in 4.</strong></p><p><strong>(2) Dallas Stars (50–20–12) vs (3) Minnesota Wild (46–24–12)</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1007/1*pLLtYVn3omjCAiyZd7neqw.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/966/1*DJbmoFCKDsVJAfNXGGhXIQ.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*GobPkc-c5EkheJLpXGtuxA.png" /></figure><p>Perhaps the most unfair first round Stanley Cup Playoffs matchup in quite some time. These two were neck and neck in the top 4 for most of the season before Minnesota dropped off a bit. One of these two teams will be looking back at a severely disappointing season when presented with their regular season resume.</p><p><strong>How DAL will win: </strong>They’ve gotta generate more offense in the 5v5 game. Dallas is very much a quality over quantity team when it comes to chance generation. They come out well in the xG battle despite having the lowest Corsi For (CF/60) rate in the entire league (per <a href="https://www.naturalstattrick.com/">NaturalStatTrick</a>), but I think they are going to need to devise new ways to get scoring chances, especially if the playoff whistles arrive.</p><p><strong>How MIN will win: </strong><a href="https://www.pierreanalytics.com/#/8478864/year/2026">Kaprizov</a> has his moment. 15 goals in 25 playoff games despite one bad run and an injured one.</p><p><strong>Final Prediction: DAL in 7.</strong></p><h4>Pacific Division</h4><p><strong>(4) Vegas Golden Knights (39–26–17) vs (6) Utah Mammoth (43–33–6)</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/992/1*Na6ejIdgKrMIkhD57EXD3A.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/957/1*SM43Yf9ev0xQEt87sHJMXA.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*qKhVykKRyFF2Z2qfrurlxQ.png" /></figure><p>Very weird season for Vegas. A top 10 team in a bunch of play driving metrics, but was so bad when it came to finishing and goaltending that they let their Stanley Cup winning coach Bruce Cassidy go with just eight games left in the season. They take on a young Mammoth team that is stronger than appears in several aspects of their game except for special teams.</p><p><strong>How VGK will win: </strong>Experienced players will have to show up and win the depth battle with Utah, especially given their goaltending woes.</p><p><strong>How UTA will win: </strong>Vejmelka steals the show and wins two games in Vegas.</p><p><strong>Final Prediction: VGK in 7.</strong></p><p><strong>(5) Edmonton Oilers (41–30–11) vs (7) Anaheim Ducks (43–33–6)</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/970/1*w-I2H6kO9pT08tpVhxQKww.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/986/1*5jK369eRKeUR7uKJrWK3kw.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*LpzgnDAiufGTvTxOi0ND0A.png" /></figure><p>This Oilers squad had a lot of question marks heading into the Olympic break this season and there were and are still a lot of rumblings as to whether their best is behind them. They look to take on a young Anaheim Ducks squad who has a very brilliant future ahead but also the only team in the playoff field with a negative goal differential.</p><p><strong>How EDM will win:</strong> A lot of this contends on Leon Draisaitl’s health given just how poor the depth on this squad has performed.</p><p><strong>How ANA will win: </strong>Engaging the track meet. This will probably be my hottest take of this article, but I think Anaheim has the offensive talent to run and gun with Edmonton even with Draisaitl back. We’ve seen what has happened to Los Angeles when you try to park the bus against this team.</p><p><strong>Final Prediction: EDM in 5.</strong></p><p>If you have made it this far, thank you for reading and I hope your team makes you proud this spring.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*aYcPsGjCYj64AhF0ViDKVw.png" /></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9bfec43c34c5" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Sample Size: Evaluating Reliability Rates for NHL Metrics]]></title>
            <link>https://medium.com/@neilpierre24/sample-size-evaluating-reliability-rates-for-nhl-metrics-839465cd0ace?source=rss-a5bd1d2b057f------2</link>
            <guid isPermaLink="false">https://medium.com/p/839465cd0ace</guid>
            <category><![CDATA[statistics]]></category>
            <category><![CDATA[hockey-analytics]]></category>
            <category><![CDATA[sports-analytics]]></category>
            <category><![CDATA[nhl]]></category>
            <category><![CDATA[hockey]]></category>
            <dc:creator><![CDATA[Neil Pierre-Louis]]></dc:creator>
            <pubDate>Mon, 27 Oct 2025 16:08:25 GMT</pubDate>
            <atom:updated>2025-10-27T16:08:25.049Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*vJCeNTz_588h6--t" /><figcaption>Andrei Kuzmenko carrying the puck for the Vancouver Canucks</figcaption></figure><p>Hot and cold streaks will forever be a popular topic of sports, and hockey is certainly not immune to that. Whether it’s a goaltender saving a team’s season or a forward going on a shooting percentage bender, it is important that we take some time to have some context as to when some performances can no longer be attributed to random noise, especially early in a season. I did some research and saw that this has not been a topic of discussion in the hockey analytics sphere, and I thought this could be of use. Today, we will look at several statistics, some at a player level and some at a team level, to identify when we expect the noise to decrease.</p><p><strong>What is “reliability” anyway?</strong></p><p>Before we dive into the bulk of the information, we should first define what it means for a statistic to become “more reliable”. I chose to keep my research in the sports domain and chose baseball, with the work that has been done in that realm. I highly recommend you read <a href="https://blogs.fangraphs.com/a-new-way-to-look-at-sample-size/">this article</a> by Jonah Pemstein and Sean Dolinar, as we will be using the same methodology.</p><p>A metric “becoming stable” in sports is not really a thing, especially in hockey, where even more luck and randomness are involved more than in other sports. However, when we obtain more samples of the rate of an event, often times we get increased reliability of the metric in relation to the talent level of the player. Let’s take shooting percentage for a hypothetical example:</p><p>A player scoring three goals on their first six shots of a season is great and an exceptional feat. However, it is not very reliable for estimating the talent level of that player. A player having 20 goals on 100 shots is still not very reliable, but it is more reliable than the first example.</p><p>We can actually measure this idea of reliability using Cronbach’s Alpha, a coefficient that measures internal consistency, or reliability! Reliability tells us the extent to which we would expect the same results of the metric we are measuring if we <strong>reproduced the events under the same conditions¹</strong>. This is an important distinction to make: we are not discussing predictive power or what we would expect, and as we know, our “conditions” here change all the time in hockey. Guys get hurt, guys play with different forward or defense groups, and opponents adjust. However, we can get a sense of whether a player’s ability or <strong>talent level</strong> thus far is actually legitimate or if it is mostly early-season noise.</p><p>Cronbach’s Alpha is on a scale of 0.0 to 1.0, where items in question, if alpha is closer to 0, have little to no consistency, and if alpha is higher and closer to 1, the items have consistent values. In academic settings, a good value for alpha, although arbitrary, is 0.7. Each graph you will see will have a horizontal dotted line at 0.7, but this is a reminder that a metric never stabilizes, it gets more reliable, or in this context, we have more of an indication of the talent level of the player, line, or team. There is no magic number where we can say, “Nice, this player or team is who they are,” and we still have meaningful takeaways about someone’s play with alpha being lower than 0.7.</p><p><strong>Methodology</strong></p><p><em>(Feel free to skip this section if you hate math and Python)</em></p><p>Cronbach’s alpha is defined as:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/150/1*o1Yv9w2YL67g0J1AiLBoDQ.png" /><figcaption>Cronbach’s Alpha Formula[²]</figcaption></figure><p>Let’s use Alex Ovechkin’s 2024–2025 NHL season to define these variables.</p><ul><li>N represents the number of items, or in this case, the number of shots Ovechkin had, which was 237.</li><li>V represents the average variance of each shot.</li><li>C represents the average covariance between these shots.</li></ul><p>When C increases, or the shots have a relation or consistent trend to them, alpha approaches 1². This is compounded further when this happens AND the number of shots, or our sample size, increases.</p><p>Since we are looking at statistical reliability within a season, I decided to look at players, pairings, lines, and teams grouped across the past three seasons. All the metrics we are looking at are percentage-based (faceoff win %, shooting %, etc.). Doing some transformations, we can group our data by these options and season, and then pivot the table to give us a matrix that looks like this.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/458/0*q_N9Oy8SPSGxbgYk" /></figure><p>Each column represents the event that has happened, so in this case, column 3 is the 3rd Corsi event in this segment of sampled shots that a pairing or line is on the ice for. “1” represents the Corsi event being a shot attempt <strong>for</strong> that pairing, whereas “0” is against. We can then compute alpha for a sample of 10 shots, 20 shots, etc. I used a package called <a href="https://pingouin-stats.org/build/html/index.html">Pingouin</a> to compute this in Python. I highly recommend this one in general, not just for Cronbach’s Alpha. It is open source and was created by <a href="https://raphaelvallat.com/">Raphael Vallat</a>, who is a staff machine learning scientist. I have linked the Python calculation below.</p><pre># Pingouin Cronbach Alpha code<br>def cronbach_alpha(<br>    data=None, items=None, scores=None, subject=None, nan_policy=&quot;pairwise&quot;, ci=0.95<br>):<br>    # Safety check<br>    assert isinstance(data, pd.DataFrame), &quot;data must be a dataframe.&quot;<br>    assert nan_policy in [&quot;pairwise&quot;, &quot;listwise&quot;]<br><br>    if all([v is not None for v in [items, scores, subject]]):<br>        # Data in long-format: we first convert to a wide format<br>        data = data.pivot(index=subject, values=scores, columns=items)<br><br>    # From now we assume that data is in wide format<br>    n, k = data.shape<br>    assert k &gt;= 2, &quot;At least two items are required.&quot;<br>    assert n &gt;= 2, &quot;At least two raters/subjects are required.&quot;<br>    err = &quot;All columns must be numeric.&quot;<br>    assert all([data[c].dtype.kind in &quot;bfiu&quot; for c in data.columns]), err<br>    if data.isna().any().any() and nan_policy == &quot;listwise&quot;:<br>        # In R = psych:alpha(data, use=&quot;complete.obs&quot;)<br>        data = data.dropna(axis=0, how=&quot;any&quot;)<br><br>    # Compute covariance matrix and Cronbach&#39;s alpha<br>    C = data.cov(numeric_only=True)<br>    cronbach = (k / (k - 1)) * (1 - np.trace(C) / C.sum().sum())<br>    # which is equivalent to<br>    # v = np.diag(C).mean()<br>    # c = C.to_numpy()[np.tril_indices_from(C, k=-1)].mean()<br>    # cronbach = (k * c) / (v + (k - 1) * c)<br><br>    # Confidence intervals<br>    alpha = 1 - ci<br>    df1 = n - 1<br>    df2 = df1 * (k - 1)<br>    lower = 1 - (1 - cronbach) * f.isf(alpha / 2, df1, df2)<br>    upper = 1 - (1 - cronbach) * f.isf(1 - alpha / 2, df1, df2)<br>    return cronbach, np.round([lower, upper], 3)</pre><p>I chose to randomly sample the number of items in each segment instead of using the chronological order of shots, faceoffs, etc for each group. I explain more in the section below, but Cronbach’s Alpha has some assumptions that are severely violated if we do not take this approach.</p><p>One issue I ran into that Dolinar and Pemstein also noticed, was a shrinking player pool size as the samples grew. It makes sense; the players who play the most will take the most faceoffs, shots, and will be on the ice for more events. However, this led to some turbulent alpha values when the last samples had a tiny player population, so I followed suit to what they did and set (I know, gross) arbitrary maximums on sample sizes³.</p><p><strong>Drawbacks, Limits, and Considerations</strong></p><p>I come back to this section in the conclusion, but I thought it would be important to write a quick blurb on some of the drawbacks and limits of using Cronbach’s Alpha in this context.</p><p>Cronbach’s Alpha has three assumptions⁴. I will list them and explain how these can be violated:</p><ul><li><strong>Unidimensionality<br>- </strong>Our assumption is that our items measure one attribute, which, for all of our metrics rings true.</li><li><strong>Tau Equivalence<br>- </strong>The assumption here is that we assume our items’ values are calculated in the same way. We definitely see violations here; some shots are more dangerous than others, and opponent and teammate strength make every event different.</li><li><strong>Uncorrelated Errors<br></strong>- This is where choosing our segments via sampling vs chronological order helps to mitigate. A good example is zone starts and fatigue affecting a series of five or so shots. We still cannot avoid this entirely, especially in larger samples where injuries, strength of opponent, and other uncontrollables that matter exist.</li></ul><p>Like other real-world examples, a lot of the way we obtain and track these metrics will not follow these three assumptions as much as we hope. There are alternatives, such as McDonald’s Omega⁴, that are better suited for situations like these, and I hope either I or another curious person can test reliability in this sphere that way.</p><p><strong>Faceoff Win %</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/790/1*hTuXkHrzUmlAOptOE957qQ.png" /></figure><p>I was surprised by how many attempts it actually took here for alpha to rise. I think the reliability here is definitely understated a bit, but it’s no surprise that most of the noise is gone after a certain number of draws.</p><p><strong>Shooting %</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/790/1*BeAFiWkXGoXJ_hO3GBWnuA.png" /></figure><p>Unsurprisingly, we see a slow slope with shooting percentage here and just a lot of noise in general. You do not even need this graph to tell you that, outside of a select few players, you should never trust a guy to repeat a ridiculous career-high sniping performance.</p><p><strong>Corsi For % (Player Level)</strong></p><p>Corsi is simply a measure of shot attempts, which includes shots on goal, goals, missed shots, and blocked shots. We can calculate Corsi For% (CF%) by just dividing the number of Corsi For by the total Corsi (Corsi For + Corsi Against).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/185/1*2KaapfZk9j8dk90TgqoMmg.png" /><figcaption>Corsi Formula</figcaption></figure><p>We could evaluate CF% reliability on a singular player level, but it does not tell us much for deployment, since players are part of pairings, lines, or five-man units, so I ended up calculating it instead based on D-pairings and forward lines. I filtered down to pairings that had at least 200 Corsi events. This is, unfortunately, an extremely arbitrary number, but I wanted to filter to an amount I felt constituted a trial of a pairing or line for a stretch of five or so games.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/790/1*jiqqWanLzYFomOFpnb2wFQ.png" /></figure><p>Two interesting points to note here:</p><ul><li><strong>Forwards and defensemen reach a Cronbach’s alpha of 0.7 at around 200 Corsi events together, which I have roughly pieced to be about 7 or 8 games. </strong>Again, not suggesting that this is when the statistic stabilizes; it never stabilizes! However, this number of games goes well with our intuition; it does seem like a good trial number for a line or pairing.</li><li><strong>Both groups continue to gain reliability for their CF% the more they play after this threshold. </strong>There seems to be a plateau around alpha = 0.85, which makes sense just given the amount of variance surrounding each event.</li></ul><p><strong>Corsi For % (Team Level)</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/790/1*c7vaM0fwogw_VwQkerbXvw.png" /></figure><p>We start to get a good sense of how a team performs possession-wise for stretches of 500 Corsi events or more, which roughly grades out to about 4 or 5 games. <strong>Again, not a prediction or projection of a team, but it gives you an estimate of how reliable a team’s CF% in that stretch is not just noise.</strong></p><p><strong>Controlled Entry %</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/789/1*gq4G_BNYI-G_5R-ceJd-Ww.png" /></figure><p><strong>Zone Exit %</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/789/1*tNHRZujKsUTwSSimwEPN8Q.png" /></figure><p>Exits and entry success percentages I found tough to have much interpretability in regards to alpha. There are quite literally infinite avenues of performing a zone entry or exit, and they are not events that happen as often as the ones listed above. I would take these reliability numbers with a grain of salt.</p><p><strong>Further Work</strong></p><p>I would love to try to see reliability for these metrics:</p><ul><li>Special teams (PP% and PK%)</li><li>Expected Goals For (xGF%) at a team and player level</li><li>Successful Pass %</li><li>Per 60 Metrics</li><li>Career Shooting %</li></ul><p>I would also want to try different tests for reliability, whether it be split-half reliability or McDonald’s Omega. Cronbach’s Alpha is simple to compute and interpretable, but it has limitations that I listed out earlier.</p><p><strong>Data</strong></p><p>All play-by-play NHL API data was scraped and manipulated by me! <a href="https://www.pierreanalytics.com/#/">Check out my website</a> to see some of the visuals I have created.</p><p>The microstat data comes from <a href="https://x.com/ShutdownLine">Corey Sznajder’s</a> All Three Zones <a href="https://www.allthreezones.com/">project</a>. I highly recommend subscribing. Corey does incredible work tracking all of this data and has really great outputs as well that you won’t get anywhere else.</p><p><strong>References</strong></p><ol><li>Carleton, R. A. (2017, April 18). <em>Baseball therapy: reliably stable (You keep using that word)</em>. Baseball Prospectus. <a href="https://www.baseballprospectus.com/news/article/31626/baseball-therapy-reliably-stable-you-keep-using-that-word/">https://www.baseballprospectus.com/news/article/31626/baseball-therapy-reliably-stable-you-keep-using-that-word/</a></li><li>Frost, J. (n.d.). <em>Cronbach’s Alpha: Definition, Calculations &amp; Example</em>. Statistics by Jim. Retrieved October 22, 2025, from <a href="https://statisticsbyjim.com/basics/cronbachs-alpha/">https://statisticsbyjim.com/basics/cronbachs-alpha/</a></li><li>Dolinar, S., &amp; Dolinar, S. (2015, June 22). A new way to look at sample size: math supplement. <em>FanGraphs Baseball</em>. <a href="https://blogs.fangraphs.com/a-new-way-to-look-at-sample-size-math-supplement/">https://blogs.fangraphs.com/a-new-way-to-look-at-sample-size-math-supplement/</a></li><li>Flora, D. B. (2020). Your coefficient Alpha is probably wrong, but which coefficient Omega is right? A tutorial on using R to obtain better reliability estimates. <em>Advances in Methods and Practices in Psychological Science</em>, <em>3</em>(4), 484–501. <a href="https://doi.org/10.1177/2515245920951747">https://doi.org/10.1177/2515245920951747</a></li></ol><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=839465cd0ace" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[2025 NHL Playoff Predictions]]></title>
            <link>https://medium.com/@neilpierre24/2025-nhl-playoff-predictions-ef1c3dae092a?source=rss-a5bd1d2b057f------2</link>
            <guid isPermaLink="false">https://medium.com/p/ef1c3dae092a</guid>
            <category><![CDATA[nhl-playoffs]]></category>
            <category><![CDATA[hockey-stats]]></category>
            <category><![CDATA[nhl]]></category>
            <category><![CDATA[hockey]]></category>
            <dc:creator><![CDATA[Neil Pierre-Louis]]></dc:creator>
            <pubDate>Fri, 18 Apr 2025 16:16:55 GMT</pubDate>
            <atom:updated>2025-04-18T16:21:37.381Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*lDZoqu0_amWsfT8H7b0uQQ.jpeg" /></figure><p>We have finally reached the best time of the year. Whether you are a Canadiens fan who is just happy to be here or a Hurricanes fan who has a “cup or bust” mentality, this is an exciting time and we are about to witness some unforgettable moments. Let’s dive in.</p><h4>The Model</h4><p>Unlike last season, I did not create a game prediction model for the regular season, but I made sure to spin one up in time for the playoffs. It is trained on the 2022 and 2023 seasons and tested on the 2024 data. For the results below, I ran 10,000 simulations of each series based on home and away probabilities drawn from this XGBoost odds model which uses the following predictors:</p><ul><li>Home and Away ELO (A model made by <a href="https://x.com/ryanabest">Ryan Best</a> and <a href="https://x.com/Neil_Paine">Neil Paine</a> when they were at <a href="https://fivethirtyeight.com/methodology/how-our-nhl-predictions-work/">538</a>, great article)</li><li>Home and Away rolling EV xGF% (Even Strength Expected Goals For %) over ten games</li><li>Home and Away rolling GSAx (Goals Saved Above Expected) over ten games</li></ul><p>I initially had two other variables, rolling GSAx and EV xG% over five games, but noticed too much volatility based on a singular game. The amount of load management over the last week of the season was also another factor in this decision. If I had more time, I would incorporate a gamescore metric I created over the summer for my player cards and move to a format that is not roster agnostic (shoutout to <a href="https://x.com/shawnferris98">Shawn Ferris</a> last year for this idea).</p><p>I calculated this season’s ELO by running the last three season’s games. As mentioned in the article above, 538 uses a formula that resets the ELO for each season by taking 70% of the prior season and adding it to 30% of 1505. I decided to do 50% and 50% with 1500. I also increased the K factor of the model to 7, treated shootout and overtime wins as ties, and did not include any ELO boost for a team at home.</p><p>The model has a log loss of 0.673 and a test accuracy of 58.2% for the test season (2023–2024). It’s better than a coin flip, but nowhere close enough to compete with Vegas.</p><p>The other variables are derived from expected goal values from shot data that I have scraped and transformed into those probabilities with their game data.</p><p>The team pages you see below in the series previews were designed and made by <a href="https://x.com/BennettSummy">Bennett Summy</a> who did amazing work on <a href="https://www.pierreanalytics.com/#/">PierreAnalytics</a> and is now doing great hockey systems work with the Penguins!</p><h4>Metropolitan Division</h4><p><strong>(1) Washington Capitals (51–21–9) vs (8) Montréal Canadiens (40–31–11)</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*PA7qShPknn22QUSA-qbCHA.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*g-h7svW3iJQ2tS2frvqm0g.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*4vtCtwsUntX6VQN9IxSdPA.png" /></figure><p>These are two teams that at times went on PDO benders to solidify their current place in the standings. The major difference is the floor of these teams. Washington is still a great squad that controls play even when they are not getting the bounces. Montreal, on the other hand, has struggled mightily in own their end all season long and their goalie <a href="https://www.pierreanalytics.com/#/8478470/year/2025">Sam Montembeault</a> will have a heavy workload this series.</p><p><strong>How WSH will win: </strong>Dominating possession and keeping Hutson and company hemmed in their zone.</p><p><strong>How MTL will win:</strong> Ivan Demidov could be a serious X-factor from the small (but impressive) sample size we saw against Chicago and Carolina. Montembeault will need to be on his A-game.</p><p><strong>Final Prediction: WSH in 5</strong></p><p><strong>(4) Carolina Hurricanes (47–29–5) vs (7) New Jersey Devils (42–33–7)</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*m8YU4RJgJ1ecDV3kElLmkg.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*wYMo-EgMLvGbbpMdrRg_KA.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*EyycaCjoXs6CDt3a_2sLYg.png" /></figure><p>The injury of superstar Jack Hughes was a devastating blow to the New Jersey Devils, but do not count this team out. <a href="https://www.pierreanalytics.com/#/8480002/year/2025">Nico Hischier</a> has been a dominant top center in the league for a few seasons and will not be denied easily. The Hurricanes are the favorite in this series, having just wrapped another successful regular season dominated by Seth Jarvis, a player who should be a Selke finalist this season.</p><p><strong>How CAR will win: </strong>Playing their game and switching up the powerplay. As with every other year, Carolina will most likely dominate the 5v5 game impressively. However, their powerplay has been subpar this season. The Hurricanes have four powerplay goals and have conceded four shorthanded goals in the 138 minutes of PP ice time that Brent Burns has had according to <a href="https://www.naturalstattrick.com/playerteams.php?fromseason=20242025&amp;thruseason=20242025&amp;stype=2&amp;sit=pp&amp;score=all&amp;stdoi=oi&amp;rate=n&amp;team=CAR&amp;pos=D&amp;loc=B&amp;toi=0&amp;gpfilt=none&amp;fd=&amp;td=&amp;tgp=410&amp;lines=single&amp;draftteam=ALL">Natural Stat Trick</a>, a trend that could be crucial in a close series.</p><p><strong>How NJD will win: </strong>Keep Carolina to the perimeter. Much easier said than done, but the Rangers did a decent job of this last season and allowed Shesterkin to face a lot of low-quality chances.</p><p><strong>Final Prediction: CAR in 6.</strong></p><h4>Atlantic Division</h4><p><strong>(2) Toronto Maple Leafs (51–26–4) vs (6) Ottawa Senators (44–30–7)</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*saBADbiFvVAqDOJp5WlW-A.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*3B5MAFweMaBULnnSuBIMVw.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*CCClmdh1ZXLbKxpnKvZ72w.png" /></figure><p>These two teams have a right to have real excitement going into this series. The Leafs have finally won the Atlantic Division and have avoided Tampa and Florida, while the Senators have broken their seven-year playoff drought.</p><p>I felt pretty down on the Leafs late into this season. You could argue metrics-wise that this is one of their weakest teams at possession in the Matthews-Marner era, but they are hot going into the playoffs and their goalie Anthony Stolarz has been a big part of that.</p><p><strong>How TOR will win: </strong>The talent will shine through. If Stolarz can keep his torrid pace and the Leafs get contributions from their top forwards, which at times has been a big “if” at this time of the year, this series will end quickly.</p><p><strong>How OTT will win:</strong> Win the special teams game and keep Travis Hamonic off the ice. <a href="https://x.com/JFreshHockey/status/1910522421049721082">Seriously.</a></p><p><strong>Final Prediction: TOR in 5.</strong></p><p><strong>(3) Tampa Bay Lightning (47–26–8) vs (5) Florida Panthers (47–31–4)</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*TGOC60Mv87UcwmNg_X8l7A.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*37mChWkGLy-e4NRRfNrRbw.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*08lROWiW3A3cQbmrgDzsGg.png" /></figure><p>Without a doubt, this will be my favorite series to watch. This is the fourth time since 2021 that these Florida teams will face off, and the hatred will be brewing. Florida slid to end the year but will have Matthew Tkachuk and Aaron Ekblad return for the playoffs.</p><p><strong>How TBL will win: </strong>Staying focused on the prize. Florida has a lot of agitators and will be doing anything to get in the heads of anyone in a blue jersey, even Vasilevskiy. Discipline will be key for a team that I think will be going back to the Stanley Cup Finals.</p><p><strong>How FLA will win:</strong> Stay out of the box. Florida has the edge 5v5 and should not let Kucherov and company set up their lethal powerplay.</p><p><strong>Final Prediction: TBL in 7</strong></p><h4>Central Division</h4><p><strong>(1) Winnipeg Jets (56–22–4) vs (8) St. Louis Blues (44–30–8)</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*yNUqCa9r628w1msYyhrxRg.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Dt4z6MJATDGTdfBBysaD6g.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*b59Pfvw_ToL4d71vrkPDNw.png" /></figure><p>Winning the President’s Trophy was a strong statement from the Jets who had a disappointing first-round exit to the Colorado Avalanche last season. They rank in the top 3rd of the league in several metrics and have piled a talented finishing performance on top of another Vezina season for Connor Hellebuyck. Don’t count out St. Louis though, they’ve been one of the best teams in the second half of the season while having solid underlyings for most of the year (<a href="https://www.hockeydb.com/ihdb/stats/leagues/seasons/teams/0000362019.html">sound familiar?</a>).</p><p><strong>How WPG will win: </strong>Connor Hellebuyck finally continues his regular season form of being the best goalie in the world.</p><p><strong>How STL will win</strong>: Limit the Jets’ speed and stay out of the box against the top-rated PP in the league.</p><p><strong>Final Prediction: WPG in 6</strong></p><p><strong>(3) Dallas Stars (50–25–6) vs (5) Colorado Avalanche (49–29–4)</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*2wvpAQOd_ZdQ_9Nvntg5Ew.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*9ZlQMXkuPqwMt2sG2qIiVw.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*-boxKisjRNQd-UX3W8KVfA.png" /></figure><p>The Rantanen bowl! There is a lot of star power in this highly anticipated series that could feature the return of Colorado Avalanche’s captain Gabriel Landeskog. Dallas has lost their last seven games, but the warning signs were there long before. They will need to put up a better defensive performance against one of the best rush offenses in the league.</p><p><strong>How DAL will win: </strong>Either an early Heiskanen return or Oettinger plays well. The defense has been <a href="https://x.com/shaynagoldman_/status/1910536193491554399">shelled 5v5 since Heiskanen’s injury</a>, but the forward group is talented enough to potentially cover that up, especially if Colorado gets into penalty trouble.</p><p><strong>How COL will win: </strong>Line matching.<strong> </strong>Get MacKinnon and Makar on the ice every time the Stars’ bottom four is out there. They will put up a ridiculous shot share.</p><p><strong>Final Prediction: COL in 6.</strong></p><h4>Pacific Division</h4><p><strong>(2) Vegas Golden Knights (49–22–10) vs (7) Minnesota Wild (45–30–7)</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*jWxR-oaO9B7bt3cbAmkIqw.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ephjNTrL4mlwo30CixFQJA.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ypRo5Xpz6clooHRR4iX-hA.png" /></figure><p>It feels like Vegas quietly assembled this 110-point season after their first-round exit to Dallas last season. Minnesota, on the other hand, came in grave danger of missing the playoffs until <a href="https://www.youtube.com/watch?v=2FFlxCnUZ_M">Joel Eriksson-Ek saved the season with less than a minute left.</a> Kaprizov’s return is huge for a Wild team that has been putrid offensively for most of the year and Zeev Buium could be an x-factor. I don’t think it will be enough for this Vegas powerhouse though.</p><p><strong>How VGK will win: </strong>They need their goaltending to be average. If they get the stops that should be made, they will win, this team is complete.</p><p><strong>How MIN will win: </strong>Buium and Kaprizov turn around some special team struggles and the Wild steal a couple of games in overtime.</p><p><strong>Final Prediction: VGK in 5.</strong></p><p><strong>(4) Los Angeles Kings (48–24–9) vs (6) Edmonton Oilers (47–29–5)</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*lseJJnCSzVpYTW9qyW5WfQ.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*_1-f5RRCtrBC1Xb5XimuCg.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ML4t1sYEYqX2GQlqSmYRiw.png" /></figure><p>I seriously cannot believe this is happening for the fourth year in a row. Thanks, Bettman. I think this is the year LA finally gets past Edmonton, and they can thank Edmonton management for that. They replaced quick younger players (Foegele, McLeod, Holloway) with aging forwards (Henrique, Arvidsson, Skinner) who not had a sizeable impact on this team so far. Skinner looks to have <a href="https://x.com/NHL_Sid/status/1912878639235899708">gelled with McDavid </a>at the end of the season, something to keep an eye on.</p><p><strong>How LAK will win: </strong>The best defensive team in the league will have to neutralize Edmonton’s rush chances, something that is almost impossible when the best player in the world is on the opposing team.</p><p><strong>How EDM will win: </strong>Depth scoring and Stuart Skinner giving the team a quality start. The rest will take care of itself.</p><p><strong>Final Prediction: LAK in 7.</strong></p><p>That wraps up my predictions. I hope you enjoyed the analysis, if you have any questions or comments, feel free to leave them here in the comment section or reach out to me via Twitter or LinkedIn! Here is my bracket and my model odds.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/958/1*hWuFcT-xia0BCeW_Us9r8Q.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/896/1*zKdm8WGT46uTwUft54zFGA.png" /></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ef1c3dae092a" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[A Guide to Building Your Own Full-Stack Sports Analytics Website]]></title>
            <link>https://medium.com/@neilpierre24/a-guide-to-building-your-own-full-stack-sports-analytics-website-a9247cb3b99f?source=rss-a5bd1d2b057f------2</link>
            <guid isPermaLink="false">https://medium.com/p/a9247cb3b99f</guid>
            <category><![CDATA[hockey]]></category>
            <category><![CDATA[web-development]]></category>
            <category><![CDATA[sports-analytics]]></category>
            <dc:creator><![CDATA[Neil Pierre-Louis]]></dc:creator>
            <pubDate>Wed, 20 Nov 2024 23:01:48 GMT</pubDate>
            <atom:updated>2024-11-20T23:01:48.744Z</atom:updated>
            <content:encoded><![CDATA[<h3>Introduction</h3><p>This guide is meant to be a comprehensive resource for those who are interested in full-stack web development for sports analytics. It will be primarily based on my project, <a href="https://www.pierreanalytics.com/#/">PierreAnalytics</a>, a hockey analytics website I started at the end of my senior year of college that was inspired by <a href="https://baseballsavant.mlb.com/">Baseball Savant</a>. Whether you’re looking to build a platform to show off on your portfolio or for a side project, I think this type of project is a fantastic opportunity for several reasons.</p><ul><li><strong>Lots of room for creativity.</strong> This tool can look however you want it and there are many powerful packages to help you get there. You’ll see the GIF later in the article, but the website has evolved so much since its first deployment!</li><li><strong>Versatility.</strong> Before starting this project, I had yet to do much front-end development and felt that was a significant weakness in my skill set. Whether it’s that, back-end development, or the database layer, taking the reigns on it will make you a versatile developer.</li><li><strong>Teams will notice it.</strong> Personal projects are a big reason why I ended up with the Red Sox and now having been here for some time, it is definitely an item we look for in candidates, especially for systems roles.</li></ul><p>Before we dive in, I want to share some advice that others have told me I have found to be very accurate in this process. You will make mistakes along the way, and that is okay! It is a very difficult project, especially for first-timers, and will require lots of time and effort. The website’s current state results from almost two years of work from me, a year of phenomenal work from <a href="https://www.linkedin.com/in/bennett-summy">Bennett</a>, and two years of model building and data analysis before I even started working on this. Even with all of that, plenty of bugs, poor implementation, and very questionable data engineering decisions from me still plague this site! Software development is a field where you will always learn something new, your programming style will change, and you’ll pick up sound habits and hopefully drop the bad ones. I even look at some of the initial PRs I made for my job at the Red Sox, and I cringe, even though I have been here for a little over a year now. What’s important is getting those reps in and learning from those mistakes.</p><p>Throughout this guide, I will give some opinions and tips on the relevant section. Then, we will walk through an end-to-end example from my website, starting from the raw data all the way to the webpage, which in this case will be this player card page.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*0LD9n9BgRh3k8iUAKxuaKQ.png" /></figure><h3>Design</h3><p>I think a design document for a project of this size should not be skipped, but this is very optional. I did not make one for this website, but this guide will mimic one. Projects of large sizes can be intimidating, and it certainly helps to dive into one with some semblance of a plan. This is not to say that this document is not mutable, and it is good practice to edit even after you start implementation, but the majority of your initial ideas should come here. Here are some tips for creating one:</p><ul><li><strong>Diagrams and mockups are your friend. </strong>Sometimes, it’s hard to see the big picture or even the components in it; giving yourself a visual of it makes this easier.</li><li><strong>Always consider alternatives. </strong>Whether it’s a different framework or component idea, odds are relatively high that an idea you have will not work out the way you want it to.</li><li><strong>Break the project up into doable milestones/issues. </strong>Having realistic goals and timelines is a necessity for anything on a large scale.</li><li><strong>Try to save the code/pseudocode for the actual project. </strong>This is something I have recently flipped my stance on, mainly due to the amount of times these snippets don’t work out for me or keep me locked away from creativity. Be detailed, but stay high level and save the programming for implementation.</li></ul><p>If you’re having trouble outlining the document, I recommend having these sections as a starter.</p><ul><li><strong>Overview/Objective:</strong> Very small blurb (2–3 sentences) that describes the goal of the project and why you are creating it.</li><li><strong>Requirements: </strong>Hammer home some of the broad details here. What data source will we need to scrape? What tables will we need to create? What will I use for website hosting?</li><li><strong>UI/UX: </strong>Goes hand-in-hand with requirements, but pertains to the look of the application. How do we want player pages to appear and what do we want the user experience to be?</li><li><strong>Timeline/Milestones: </strong>I mentioned breaking up the project into feasible goals, and this section is where you make that decision. You can choose to stay broad (Ex. 2 months to complete the backend) or dive into fine detail (Ex. 1 week to create a REST API for player cards).</li><li><strong>Alternatives/Challenges: </strong>As I mentioned, things will go wrong, or different routes will be taken, prepare for that in this section.</li><li><strong>Testing: </strong>Devise a plan for accounting for things like data quality, mobile-friendliness, error identification, etc. I think unit tests may be a stretch goal for a project like this, but if you can plan and implement that, you are doing better than a lot of the sports world!</li></ul><h3>Data and Structure</h3><p>This might be the most important section in this guide, given how much it shapes up how your full-stack application operates. Let’s take a look at the flow of data for PierreAnalytics:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ZWui0tPL3X-mEZK4crc-XQ.png" /></figure><p>Let’s break this chart down! We first start with several different NHL API sources that all share raw data pertaining to the league that we can transform into actionable insights and visualizations. Some of you may already be familiar with these endpoints, but if not, <a href="https://github.com/Zmalski/NHL-API-Reference">there is a wonderful reference on GitHub to follow</a>. Max Tixador (<a href="https://x.com/woumaxx">@woumaxx</a>) also made a <a href="https://colab.research.google.com/drive/1D8hax2Bs-k-8HVwttJVLnlEqAzNhWOEM?usp=sharing">great resource</a> if you are struggling to scrape data from the NHL API or don’t know where to start.</p><p>There are several parts of the flow here that I could improve. In an ideal world, I am not manually running any of these notebooks and I have one centralized database. These are quality-of-life improvements that you will notice the more you practice but are not necessarily the most important to address at this stage of the project.</p><h4>Data Flow</h4><p>I have a Python scraper that grabs play-by-play data from the API from the night before every morning at 7am. This action is automated by a job that I scheduled using <a href="https://www.windowscentral.com/how-create-automated-task-using-task-scheduler-windows-10">Windows Task Scheduler</a>, not a necessary thing to have for projects of this size, but I think automating as much of the data updating and cleaning process as possible is good practice and will save you lots of time.</p><p>A lot of data manipulation and model running happens with the Jupyter notebooks, but I will save that for a different article. To store this data, I highly recommend using an SQL database. SQL is a skill that is so often overlooked in the sports analytics community, but it is a powerful tool that can process data quickly and efficiently. PierreAnalytics uses an <a href="https://www.sqlite.org/">SQLite</a> database for a few reasons: it is very portable, is embedded into your application, and is excellent for a website like mine where my user base does not usually exceed 20 people a day. Your application might have different use cases that will require maybe something like Postgres or MySQL, but using this will give you the option to query data and make transformations. No more CSVs!</p><p>Here are a few tips I have for storing your sports-related data:</p><ul><li><strong>ALWAYS use player IDs when storing player-related data. </strong><a href="https://www.pierreanalytics.com/#/8480222/year/2024">Some players share names</a> and that can result in data issues if you are differentiating off of that. This should be your primary key! Same applies for teams.</li><li><strong>Decide what your models should look like before you start creating tables. </strong>When I say model, I am referencing the structure of a chunk of data, not a statistical predictor. In this case, if I want to create player cards for NHL skaters, I first need to decide what are the properties that make up a skater (position, name, team, etc).</li><li><strong>Follow a naming convention for your columns and stick to it. </strong>May or may not have laughed as I write this due to my clear violation of this rule, but set it for yourself, it will save you plenty of headaches. I recommend <a href="https://developer.mozilla.org/en-US/docs/Glossary/Snake_case">snake case</a>.</li></ul><p>Let’s jump into the example. Right now, we have shot data (shooter, shot type, who was on the ice, time, period, previous event, etc.), and we want a summary table of skaters. Here’s the flow of how that works:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/960/1*HB5CE9LE4rWgJ9dzDu6JlA.png" /></figure><ul><li><strong>Preprocessing/Data Transformation: </strong>Morphing the raw data into a form that can be ingested by the xG model (creating calculated columns like the previous event, location, shot angle, etc).</li><li><strong>xG Model -&gt; Output: </strong>We run the transformed data through the model and then attach the output, which in this case is the probability of a shot being a goal, to the dataframe as a column.</li><li><strong>Further Data Transformation: </strong>Now, we can calculate several types of summary tables (gamescore, goalie statistics, skater statistics, etc.). In this specific case, we would calculate things like xGF, xGA, and finishing for a season for all eligible skaters based on some ice-time threshold.</li><li><strong>Send it to the database: </strong>Once this is done, we can dump this into our SQLite database. For any Pandas users, SQLite integrates very nicely.</li></ul><h3>Back-End</h3><p>For PierreAnalytics, I use <a href="https://expressjs.com/">Express.js</a> for my back end. It is very lightweight and easy to set up, so I highly recommend it for users who are not doing a lot of calculations and logic on the back end. Of course, your application may have very different needs, or your skillset could be better suited in a different language. I have made a table below on a few different options, their languages, and pros and cons. Many other choices are out there, but I thought I’d give a few.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/631/1*TjHOvjBx-zlh1JtwNOcqLQ.png" /></figure><p>Here and the SQL layer is where you should really define what your models, or how your data is structured, should look like. Let’s look at a PierreAnalytics example.</p><p>We have our skater data in an SQLite table; now it’s time to extract the columns we need, build a response and an API, and have that API send it to the front end. There are two main questions we should consider here that go hand in hand:</p><ul><li><strong>What does the request look like? </strong>What information does the front end user need to provide?</li><li><strong>What should the response look like? </strong>What information should we give the front-end user?</li></ul><p>In this case, the user will need to provide a player ID and a season in order to retrieve the data. We can grab these numbers from the request, make a query to our table that we populated earlier, and build a response from there.</p><pre>app.get(&quot;/players/card&quot;, (req, res, next) =&gt; {<br><br>// Grab parameters from response<br>const id = req.query.id;<br>const season = req.query.season;<br><br>// Make sure parameters are valid<br>if (!id || !season) {<br>  return res.status(400).json({error: &#39;Undefined parameters&#39;});<br>}<br><br>db.get(<br>    `SELECT * FROM skaters WHERE player_id = ? AND season = ?`,<br>    [id, season],<br>    (err, playerRow) =&gt; {<br>      if (err) {<br>        res.status(400).json({ error: err.message });<br>        return;<br>      }<br><br>      if (playerRow) {<br>        const response = {<br>          xGF: playerRow.xGF,<br>          xGA: playerRow.xGA,<br>          wpa: playerRow.wpa,<br>          goals: playerRow.goals_60,<br>          sh_xGA: playerRow.sh_xGA,<br>          primaryAssistsPP: playerRow.primary_assists_PP,<br>          pp: playerRow.pp_60,<br>          primaryAssistsEV: playerRow.primary_assists_EV,<br>          TOI: playerRow.EV_TOI,<br>          gsae: playerRow.gsae,<br>          shortHandedTOI: playerRow.SH_TOI,<br>          powerPlayTOI: playerRow.PP_TOI,<br>        };<br>        res.status(200).json(response);<br>      }<br>   }<br>}</pre><p>You might have noticed that we have not passed specific details like name, nationality, and other descriptive fields. These are saved for other API calls instead of passing repeated details.</p><h3>Front-End</h3><p>We’ve now built a significant portion of this application, and now it’s time for the icing on the cake. If you do not know any HTML, CSS, or JS, this may be the steepest learning curve of full-stack development, and I would highly recommend diving into a guided course or YouTube video where you can code alongside the instructor. PierreAnalytics uses a front-end framework called Angular, which helps me structure my project. It has a lot of prewritten code that can do powerful tasks like styling and state management. There are plenty of front-end frameworks out there, but I would highly recommend using <a href="https://react.dev/">React</a>, <a href="https://angular.dev/">Angular</a>, or <a href="https://vuejs.org/">Vue</a>. They are the most commonly used frameworks that SWE teams use, and this still rings true for the sports industry. You also can go the route of no framework, and this may be easier if you have little to no front-end development experience.</p><p>This section will be high-level and more conceptual. There are so many resources out there that can teach you Angular, HTML, CSS, JS, etc much better than I can in this article, some of which will be linked below.</p><h4>Sports Front-End Design Tips</h4><p>Here are some good practices and tips for designing your front end. Some of these are applicable to general data visualizations!</p><ul><li><strong>Don’t overlook accessibility: </strong>Avoid red to green color gradients. Make sure your text sizes are large enough. Make sure your color contrasts <a href="https://webaim.org/resources/contrastchecker/">follow convention</a>! Displaying data, especially in sports, is all about telling a story, and you want to make sure that all of your audience can understand it and that they are getting the correct story.</li><li><strong>Make your data easy for users to see and manipulate: </strong>If you have a table of goalie data, make it filterable and sortable. Have a search tab for player pages. Create a visible navigation bar for all of your pages. When considering your layout, quick access is important in the sports world! If a user wants to see a table of all of Sebastian Aho’s goals from the 2022 season, receiving that information in as few clicks as possible will prompt them to keep coming back to that site.</li><li><strong>Account for mobile devices: </strong>I still struggle with keeping my website mobile-friendly, but it is a necessity. 63% of my all-time users are mobile, so it is important to test out new features and make sure they are usable on a phone screen.</li></ul><h4>Layout</h4><p>A couple of months ago, I had a great discussion with my co-contributor Bennett about home pages. Designing the site, the home page is where I concentrated the least amount of effort, even though that is the only guaranteed page the user will see. Turning it into an aggregation of resources that the rest of the site possesses was huge for UI/UX improvement and user engagement as you can see in the GIF below.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/600/1*almFimy6jakFy7xkBj9w7g.gif" /><figcaption>PierreAnalytics, June 2023 and November 2024</figcaption></figure><p>A few items that are different:</p><ul><li><strong>Way less white/empty space: </strong>Much more visually appealing.</li><li><strong>Randomly generated lucky player:</strong> Pops out immediately and gives the user quick access to one of the main tools on the site.</li><li><strong>Scores: </strong>It does not have to be scores, but I think having some type of summary of the league state on the home page is great for users to see on the first click.</li><li><strong>Navigation Bar: </strong>Quick access to the work.</li><li><strong>Release Notes:</strong> New features are less likely to go unnoticed now! (Future goal is to have those link to said feature)</li></ul><h4>Bonus: Angular</h4><p>If you do choose to go the framework route, this section may interest you! Angular is the first framework I learned, and the biggest learning curve for me was figuring out what components should consist of. Take the player page for example.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*pR1VqwdmWstI_yQdEDsHKQ.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*IfC-3jZIOtdmfVOYqxq1Hg.png" /></figure><p>Going into Angular, a commonly placed mistake is to think in this scenario, “One page — one component,” when in reality, a page like the one above can contain several. Each component in your directory should serve a singular role or task. In this example:</p><ul><li><strong>Card component: </strong>Displays the actual page and the other components.</li><li><strong>Timeline component: </strong>Displays the colored statistical percentiles.</li><li><strong>Shots component: </strong>Displays the rink with shot locations.</li><li><strong>Gamescore component:</strong> Displays a gamescore graph.</li><li><strong>Table component: </strong>Displays a table with shot data.</li></ul><p>There’s technically no “right way” to do this. I could have stuffed all of this code into one component, but massive components have poor readability and I would not be able to reuse a lot of these other features like the tables or the rink.</p><h3>Git</h3><p>I cannot stress how important version control is! The amount of times I have introduced a bug on my site or some data issue has been far too many — and thanks to Git, it is easy for me to track down what I did wrong. Funny enough, last summer, my laptop died due to water damage — had I not used Git for this project, the source code for this site would have been gone for good.</p><p>With a lot of starter projects, the flow may look something like this:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/540/1*jCG7Vc39TGOCx-ZpOWhxZQ.png" /></figure><p>There are a few issues here:</p><ul><li>No track of previous states is kept somewhere.</li><li>Very difficult for collaboration with some amount of overwriting becoming inevitable.</li><li>Hard to switch between devices/potential loss of code if your computer is damaged.</li></ul><p>Now let’s look at what the flow would be when we use Git:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*7I5x10_mIf7ih4uDsz83SA.png" /></figure><p>We can keep our code for different issues and projects isolated by creating a branch for each task. Even if you are working solo and decide that the branch flow is too difficult to manage, committing directly to the master branch will at least add to a log of changes that you can reference. This log becomes super helpful when you want to roll back a change, look at previous versions, or identify an error or regression.</p><p>Another benefit of using Git is linking it to deployments on your website. I currently host mine on <a href="https://www.heroku.com/">Heroku</a>, although I would recommend using AWS or Google Cloud after a few security issues I have noticed with Heroku. These services give you the option to automatically redeploy on any change to the master branch, which is very convenient for a website like mine that updates player data every night of NHL games.</p><h3>Conclusion and Resources</h3><p>If you have made it here, thank you so much for reading! I hope this was helpful, please reach out to me if you have questions. There is a lot of detail I left out that also is covered in the resources below. A lot of these tools and technologies I learned in school, projects, and work. Regardless of the medium, working with the languages, doing exercises, and writing your own queries/functions/classes will help you the most.</p><ul><li><a href="https://angular.dev/style-guide"><strong>Angular Style Guide</strong></a></li><li><a href="https://github.com/Zmalski/NHL-API-Reference?tab=readme-ov-file#get-team-stats"><strong>NHL API Reference</strong></a></li><li><a href="https://colab.research.google.com/drive/1D8hax2Bs-k-8HVwttJVLnlEqAzNhWOEM?usp=sharing"><strong>Max’s NHL API Scraper</strong></a></li><li><a href="https://www.lucidchart.com/pages/"><strong>Diagram Creator — Lucidcharts</strong></a></li><li><a href="https://blog.postman.com/how-to-create-a-rest-api-with-node-js-and-express/"><strong>Creating a REST API with Express.js</strong></a></li><li><a href="https://learngitbranching.js.org/?locale=en_US"><strong>Learning Git Branching</strong></a></li><li><a href="https://code.visualstudio.com/docs/sourcecontrol/overview"><strong>Using Git Source Control in VS Code</strong></a></li><li><a href="https://developer.mozilla.org/en-US/docs/Learn/Getting_started_with_the_web"><strong>MDN: Learning HTML, CSS, and JS</strong></a></li><li><a href="https://sqlbolt.com/"><strong>Learn SQL</strong></a></li><li><a href="https://angular.love/all-you-need-to-know-to-jumpstart-with-ngxs"><strong>Intro to NGXS</strong></a></li><li><a href="https://www.geeksforgeeks.org/deploy-angular-app-in-aws/"><strong>Deploying an Angular Site to AWS</strong></a></li><li><a href="https://www.namecheap.com/hosting/how-to-set-up-hosting-account/"><strong>Hosting a Website on NameCheap</strong></a></li><li><a href="https://d3js.org/"><strong>D3.js: Powerful JavaScript Data Visualization Tool</strong></a></li></ul><p>As I have mentioned before, PierreAnalytics is open source. If starting your own full-stack web development sports project seems too daunting at first, <a href="https://github.com/neilpl24/pierreanalyticsource">contribute to the repo to get a good feel for the technologies</a>. There is a list of issues that you can choose from or you can create your own.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a9247cb3b99f" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Offensive Revolution Part 1: Rebounds]]></title>
            <link>https://medium.com/@neilpierre24/the-offensive-revolution-part-1-rebounds-8e37199c3164?source=rss-a5bd1d2b057f------2</link>
            <guid isPermaLink="false">https://medium.com/p/8e37199c3164</guid>
            <category><![CDATA[hockey-analytics]]></category>
            <category><![CDATA[analytics]]></category>
            <category><![CDATA[sports-analytics]]></category>
            <category><![CDATA[hockey]]></category>
            <dc:creator><![CDATA[Neil Pierre-Louis]]></dc:creator>
            <pubDate>Fri, 01 Nov 2024 18:47:20 GMT</pubDate>
            <atom:updated>2024-11-04T15:04:02.950Z</atom:updated>
            <content:encoded><![CDATA[<p>Back in fall 2023, I found myself discussing hockey with a couple of friends, specifically the Chicago Blackhawks powerplay early on in the season. I was losing my mind, at least to my eye, that Connor Bedard was not getting enough touches and that rebounds were the only way that he was getting action. My friend replied, “The rebound is a dying art unless you’re Brady Tkachuk.” Having the shot data already scraped on my computer, I set out to see the results for myself — and what I found was not what I expected.</p><h3>Diving In</h3><p>Let’s start here — what is a rebound and how can we quantify it? According to <a href="https://www.naturalstattrick.com/glossary.php?players">Natural Stat Trick</a>, a rebound is any attempt made within three seconds of another blocked, missed, or saved attempt without a stoppage in play in between. It’s a great definition, but I will slightly tweak it for what I think is a “true rebound”, an attempt made within three seconds after a shot on goal.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/731/1*Y-Dw7N50O-kGpYHEZQ8LoA.png" /><figcaption>Python Code of defining a rebound</figcaption></figure><p><a href="https://github.com/Zmalski/NHL-API-Reference?tab=readme-ov-file#get-play-by-play">The NHL has an API that provides relevant play-by-play data</a> for each event in every game since 2007. Having game IDs for each event and this newly added rebound column, we can quickly group the dataset by season, calculate the number of rebounds and number of games per season, and build a quick graph.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/554/1*64OLstZNm84ng_VLhQxE1A.png" /><figcaption>Rebounds Per Game By Season</figcaption></figure><p>In the 2023–2024 season, there were almost <strong>TWICE</strong> as many rebounds per game (6.17) than we did in the 2017–2018 season (3.53)! Even when you account for a potential SOG discrepancy by season, you still see this increase. So what changed?</p><h3>The Death of the Point Shot</h3><p><a href="https://x.com/jhanhky/status/1732603039427485836?s=46&amp;t=GeC0iTO9_KrKxDtqbRZAPA">I talked about this with hockey consultant Jack Han briefly on Twitter</a>, but teams are putting up fewer shots from the point each year (visualization below). What does this have to do with rebounds? The further the shot is taken from the net, the more time a goalie has to react to it — they have more time to freeze, catch, or trap for goalies than slot chances or chances closer to the net. These higher-quality shots in tight forces goalies to fight off shots, leading to more rebounds. We can also prove this point by looking at where on the ice players take shots that lead to these prime rebound chances.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/685/1*i4vBP24UcLuY7xkz4K7tlw.png" /><figcaption>Python Code for looking at shot-generating rebounds</figcaption></figure><p>Our data is now set to where we can plot these shots using the NHLRink package. (Massive shoutout to <a href="https://x.com/the_bucketless">TheBucketless on Twitter</a> for creating this package, he’s a beast!)</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/640/1*8LE7ddl3O43DZhKguq0eaQ.png" /></figure><p>Unsurprisingly, we see the majority of these shots occur in the slot area, which tracks with our intuition. If we also look at a general shot attempt heat map season over season, we’ll see the chances from the outer spaces and the point diverge into that slot area.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/640/1*_XabgXEYVnjtnPLTBfPH7g.gif" /></figure><p>In our most recent season, we actually see the heat map in the upper slot expand further into the east-west direction. This change is really where I wish we had passing and tracking data to see if teams are intentionally moving offensive schemes this way to <a href="https://players.brightcove.net/6415718365001/EXtG1xJ7H_default/index.html?videoId=6345016405112">exploit goalie movements</a> and weaknesses.</p><h3>What does this mean for actual gameplay?</h3><p>We have disproven my friend’s notion of the rebound being a dying art. What else can we draw from this data?</p><p>That will come up next in part two, where I discuss how we can evaluate goalie play through expected rebounds and a new topic involving on-ice finishing for skaters.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8e37199c3164" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>