VoxBlog

Editor’s Note: This story is included in The Athletic’s Best of 2022. View the full list.

I was obsessed with hockey growing up — expectedly so, considering the career path I chose. It’s been an intense lifelong love affair with the game and one of the earliest things I can recall sparking that was collecting hockey cards. 

Every year, McDonald’s would put out a hockey card collection that came with meals and every year I would beg my dad to bring me there as much as possible so I could complete my collection. I’m pretty sure there are still binders full of them in my parents’ basement somewhere, along with binders full of non-McDonald’s cards too. I was captivated by the design, the player photos, the players themselves and obviously the stats on the back. It’s how I learned as much as I could about key players in the pre-internet era, a phrase that instantly makes me feel ancient. 

Hockey cards have declined in popularity, but one of its core concepts — and why I was so drawn to them as a child — has been translated pretty well to the internet era of hockey discourse. I’m talking about player profile dashboards, quick at-a-glance summaries of relevant player stats — basic or advanced — in an easily digestible manner through data visualization. It’s the digital heir to the back of a hockey card, armed with more (and better) data at its disposal. More charts and colour-coding, too.

Advertisement

It’s that specific niche of analysis that has really taken over the forefront of hockey analytics, especially on Hockey Twitter where the content most shared and engaged with tends to be some type of player profile. And everyone has one, with versions from HockeyViz, Evolving Hockey and JFresh being the most prevalent, along with my own.

The issue with my player cards has always been availability. The cards themselves could only be generated on my own computer so anyone who was curious about how their favourite players stacked up was at the whim of whether I would share it or not. Ever since I started tweeting out some sort of visual player summary, the number one request I constantly receive is when they’ll become publicly available. 

Good news: That day is today.

It’s been a long time coming and the reason it’s taken so long is that I didn’t really have the resources at my disposal to pull it off with no automation to create something interactive. But the demand is likely high enough that it’s worth the trouble of doing it anyway without that power (and thankfully the addition of Shayna Goldman to the team adds to the manpower needed to accomplish it). It’s a proof of concept that this is the type of content fans want to see and use. I see it every day on Twitter with people referencing the various player cards already out there and so I wanted to make my own work in that domain significantly more accessible. That type of transparency is important to gaining a better understanding of how the underlying model works and trusting its inputs more. That, and seeing where there are still flaws and blind spots with certain players and player types — this model, like any other, isn’t perfect.

This project also offered an opportunity to reimagine my own player cards, and I was inspired to take that term literally. I wanted to create my own hockey cards for the digital age, one that borrowed from the feel of the cards I grew up with (while also sticking to our own brand’s guidelines). The goal was to not only summarize a player’s strengths and weaknesses in a clear and concise way but to also create a cool package to interact with. Luckily, thanks to working at The Athletic, I had access to a database of player photos via USA Today to capture that vision. 

Every team will have their own separate page showcasing cards for 12 forwards, six defenders, and yes, even two goalies as well. You can find a link to each page at the bottom of this article. That won’t cover every player, but it’ll cover all the starters. Here’s what they look like:

For those familiar with the previous version of my player cards, you’ll notice there are some obvious changes — namely, the lack of an age curve. That chart type isn’t going away, I just felt it wasn’t something that was necessarily appropriate year-round. It fits more when players sign new contracts, but not so much on a random Tuesday in January when you just want to know how good the player is right now. That needed to be the focus, especially if the design was to be influenced by real-life hockey cards. I also omitted penalty differential and usage from the visualization. Not that those two things don’t matter, it’s just that some sacrifices needed to be made for the sake of simplicity and to bring new concepts into the fold.

Advertisement

The main difference between this card and prior versions is that this card is not only showing what a player is projected to do but also what they’re actually doing this season (and have done in the past, via a three-year summary at the top). One of the main challenges with explaining projected values is that it’s sometimes difficult to separate them from what a player has already done. I see it all the time with questions like “Player X looks a lot better than your GSVA suggests” and the reason for that is his projected value takes into account his priors. Generally speaking, a person’s intuition is bang-on and their actual value for the season is closer in line with what they’re seeing.

If a player is on pace for 80 points but was previously closer to a 40-point scorer, how much can that sudden spike be trusted? What can we expect going forward? After regressing and accounting for sample size, a projected value suggests an answer, but that doesn’t mean what they’ve done so far isn’t also interesting. Having both concepts side-by-side adds valuable context to the type of season they’re having and its sustainability. That was the main goal of the redesign. 

It may be a bit confusing to see a player with a higher pace number than projected, but a shorter pace bar than projected. That’s due to the regression that’s happening with the projected values and the small sample sizes that the pace values are dealing with. It leads to a tight range of projected values, but a very wide range of pace values. An example: A player may be projected to be a 40-goal scorer, but is actually on pace for 45. That may seem like he’s overperforming — and he is — but relative to his peers, he may actually be lower. A 40-goal projection might have him in the 90th percentile, but a 45-goal pace might only land him in the 85th percentile. What that generally means is his goal-scoring is a lot more reliable and while he may regress closer to 40 goals, his peers who are pacing above him should fall even further.

That’s easily seen here, not only by the numbers on the side, but also by the bar lengths. If a projected bar is longer than a player’s pace bar, it means the model expects that player to improve based on his history. A longer pace bar compared to the projected bar means expect him to come down to earth a bit. Those bars are based on what percentile a player falls into at his position within that metric, a number that isn’t explicitly stated. It’s sometimes helpful to know that a player is in the 95th percentile exactly, but I didn’t want to imply false precision with that number, especially with how variable all the stats can be from month to month. When it comes to a player in the 50th and 60th percentile, odds are you’re mostly splitting hairs between the two.

This is a new data visualization so I can imagine it may be overwhelming to some, especially those less versed with analytics. For that reason, I’ve created a visual guide on how exactly to read it below (and I’ll be in the comments with any additional questions people may have):

The current goal is to update these cards as often as possible with once per week being the goal. We obviously want the data to be as up-to-date as possible for sharing purposes, though it would obviously be unrealistic for it to be updated daily. Each card has a serialized date number so you know exactly which version you’re looking at and how updated the data is.

Advertisement

As for where to actually find them, we’ve created a landing page for each team and you can find the links to each one below. Each landing page will have links to all other teams at the bottom so you can jump to and from different teams with some ease. It’s not as easy as having a single interactive page of course, but this is hopefully only the beginning.

I know this feature is something a lot of folks have asked for over the past few years. I hope it was worth the wait.

Team Player Card Pages

Atlantic: BOS, BUF, DET, FLA, MTL, OTT, TBL, TOR

Metropolitan: CAR, CBJ, NJD, NYI, NYR, PHI, PIT, WSH

Central: ARI, CHI, COL, DAL, MIN, NSH, STL, WPG

Pacific: ANA, CGY, EDM, LAK, SJS, SEA, VAN, VGK

ncG1vNJzZmismJqutbTLnquim16YvK57kmlsaXFkbnxzfJFrZmlrX2WEcLXNramonKWYtq%2BzjKisq2WenbluvMuasJ6qXZius7DSaA%3D%3D