Optimizing Gameplay Using Data: How to Win More in Magic: The Gathering Arena
Magic the Gathering is an incredibly complex card game. Building a good deck requires knowledge of the cards themselves (there are thousands) and an understanding of prevalent strategies. The draft format - where players take turns selecting cards from a pool and build a deck on the fly - has its own added complexities. Today we’ll focus on draft data from the Dominaria United set (provided by 17Lands) and examine how often players choose certain cards vs. how those cards actually perform in-game.
The Challenge of Optimizing the Draft
For the past 30 years, players have argued over many aspects of drafting, from the optimal number of “land” cards in a deck, to how many cards of each type a player should select (e.g. creatures, sorcery spells, enchantments), and other hard to answer questions.
Take the ever popular draft format. Eight players arrange themselves around a table (either physically or virtually) each with three 15 card packs of cards. Each player opens a pack, selects one card for themselves to “draft”, and passes 14 cards to the player to their left. They receive 14 cards from the player to their right, select one to keep, and pass the remaining 13 cards on. This goes on until the table runs out of cards. This happens two more times with new packs, with the drafting direction alternating each time.
From there, players build the best 40 card deck they can from the 45 cards they’ve selected and any number of basic land cards. Here’s an example of what a draft deck might look like, according to this template by Wizards of the Coast (the game’s creators).
While heuristics, templates, and expert opinions can help players reach reasonable conclusions, only analyzing the data itself can provide an unbiased look into the game. In years past, this has been difficult. As a primarily paper card game, accumulating enough detailed and accurate data for analysis has been nearly impossible.
MTGA and 17Lands
Enter Magic the Gathering Arena (MTGA) - a hugely popular online version of Magic - and associated data trackers like 17Lands. 17Lands tracks and provides MTGA draft data for thousands of drafts and millions of picks, and publishes a whole host of analytics: performances by card and color, deck type, and more. For deeper analysis, 17Lands allows third parties to access this data directly.
Using one of these datasets for the recent Dominaria United (DMU) set, we’ll tackle one of these questions: do players prioritize cards that help them win?
Card win rates
In order to assess player draft bias, we’ve imported a dataset of 1,800 DMU drafts into Datawisp, a no-code data analysis platform. We’ll first analyze win rates of each card in the set and look at how this metric might be used in practice. Then, we’ll compare win rate to average draft pick order and investigate the difference between the two and what might be causing it.
To begin, here’s a look at a snapshot of the draft data.
The data contains the following information:
Draft_ID: A unique identifier for each player’s draft
Draft_time: When the draft took place
Pick_number: The pick in question - pick 0 being the first pick in the pack
Pick: Name of the card chosen
Pick_maindeck_rate: Percentage of a player’s games the card was in their deck
Event_match_wins: The number of matches won by this drafter in this draft
Event_match_losses: The number of matches lost by this drafter in this draft
Rank: The rank a player started the draft at
For players opening their first draft pack, a reasonable question might be “which cards are even good?” One way to assess a card’s quality is by its win rate. Calculating this is fairly straightforward - we know the number of wins and losses for each draft and the cards selected in that draft. Note that decks usually do not contain all cards drafted; we can account for this by only calculating based on cards that were played more than 50% of the time.
Let’s take a look at the top (and bottom) card win rates in the set to get an idea of just how much card quality matters. Here, we see that, minus a clear outlier in Urza Assembles the Titans (with a 0% win rate), the worst cards win just over 40% of the time, whereas the best cards win over 60% of the time.
Top 10 cards by win rate:
Bottom 10 cards by win rate:
We can visually display the distribution using Datawisp:
Since each match has a winner and a loser, we’d typically expect the average card’s win rate to be near 50%. However, it’s worth noting that players who submit their data to 17Lands tend to be above average players who care about improving their results using data, so the average is actually closer to 57%.
Cards in Magic the Gathering have different levels of rarity - common, uncommon, rare, mythic rare - which are very roughly correlated to their power level (not always the case). A draft pack contains one rare or mythic rare, two uncommons, and 11 commons.
Here’s an example of what a DMU pack might look like, along with win rates for each card in the pack:
0.506 - Plaza of Heroes
0.554 - Baird, Argivian Recruiter
0.530 - Dragon Whelp
0.578 - Knight of Dusk's Shadow
0.585 - Meria's Outrider
0.561 - Coalition Warbrute
0.570 - Magnigoth Sentry
0.562 - Heroic Charge
0.582 - Radiant Grove
0.578 - Pixie Illusionist
0.562 - Sacred Peaks
0.568 - Flowstone Kavu
0.566 - Contaminated Aquifer
0.582 - Battle-Rage Blessing
Our data suggests that a player picking only on expected win rate should select Meria’s Outrider (common, 58.5%) over Battle-Rage Blessing (common, 58.2%), and avoid the incredibly unexciting rare, Plaza of Heroes (50.6%), for an extra 7.9% win percentage. While experienced players might know that Outrider is a more versatile and playable card than Plaza in draft, being able to quickly assess the top 2-3 cards in a pack is very difficult and often players default to picking the rarest card in the pack.
Draft Pick Order
After the first pick, there’s an additional element to consider - opportunity cost. Most of the cards not initially chosen won’t be seen again because they’ll be snapped up by other drafters. So when choosing between two cards, a drafter also needs to assess the likelihood that a card will be available for selection later on in the draft. To see which cards are generally picked first, we can look at avg_pick. Below are the top picked cards in DMU draft:
Liliana of the Veil is the least passed card: on average, there are 0.054 cards passed before Liliana is picked (she’s passed roughly once for every 20 packs she appears in). But her win rate is only 57.6% - not significantly higher than the mean. In comparison Shivan Devastator has an average pick of 0.128 but a win rate of 60.7%. These discrepancies continue throughout the dataset.
We would expect that cards drafted later would have a lower win rate, but that’s not entirely true, as shown in the following graph:
Rarity Draft Bias
This chart, showing average pick on the X axis vs win rate on the Y axis, and colored by card rarity, leads us to believe a few things:
Higher picks are associated with higher win rates. In general, the early picks seem to be correlated to higher win rate cards.
Higher rarity cards are almost always picked before lower rarity cards, even if lower rarity cards have higher win rates. This could be for a few reasons. Rares and Mythics are more exciting and splashy, but they’re also usually more sought after to have in your collection after the draft.
There are a few cards that perform very differently than their current pick order would suggest, something that clever drafters can take advantage of.
Some points of interest:
Underdrafted and Overperforming
Drafted after - 2.261 cards | Winrate - 0.638 | rare
Boasting the top win rate among all cards, but on average drafted after 2.261 cards, Shadow-Rite Priest is likely underdrafted because it appears to be a build-around (the card only seems good if it’s played with other cleric-type cards). However, this card is likely worth pivoting for, and should be prioritized highly. Even with only a few other clerics, its second ability is very powerful, letting players find the best cards in their deck and put them directly into play.
Drafted after - 4.721 cards | Winrate - 0.633 | uncommon
Wingmantle Chaplain is known as one of the best cards in the format. Similar to Shadow-Rite Priest, it is a build-around (defender type), but with so many defenders in the set and a win rate over 63%, there are basically no rares that should be picked over this uncommon early in the draft.
Drafted after - 7.519 cards | Winrate - 0.601 | uncommon
Drafted after - 9.941 cards | Winrate - 0.588 | common
A trend is beginning to arise. In this format, defenders are great, and players are rewarded if they can solidify themselves into a focused deck as opposed to general “good stuff”. Similar win rate cards, even at similar rarities, are being drafted a full two picks before Walking Bulwark and Shield-Wall Sentinel. As with Wingmantle Chaplain, these have the same caveat. These cards are only good in a specific deck type (defenders), meaning late in the draft players pass these if they’ve decided to build another type of deck.
Drafted after - 9.783 cards | Winrate - 0.582 | common
Lastly, we have Battle-Rage Blessing. On the surface this card screams mediocre (it’s an unexciting two mana combat trick that doesn’t do anything on its own). The pick rate reflects this: on average, 10 cards are picked before it in every pack. However, a 58.2% win rate indicates this could be a good addition to decks that can support it later in the draft.
Overdrafted and Underperforming
Drafted after - 5.061 cards | Winrate - 0.348 | rare
Drafted after - 4.63 cards | Winrate - 0.4 | rare
Drafted after - 2.683 cards | Winrate - 0.447 | rare
Here are three cards that are similar to Shadow-Rite Priest on the surface: they all buff cards of similar types (goblins, merfolk, and elves, respectively) and have an additional ability. While they’re on average picked later than the Priest, they perform significantly worse. None manage to pass a 45% win rate, despite being drafted very early. The data makes it pretty clear that even if these cards are drafted for use in other formats, they should be avoided in draft play.
Where do these cards differ from Shadow-Rite Priest?
Stats. Shadow-Rite Priest is only a 2/2, but these other three are 1/1s, making them significantly weaker in combat.
Shadow-Rite Priest’s activated ability is a lot more powerful and lets players find and play their best cards (non-clerics as well).
Here are two of the most disappointing cards in the set that tick almost all the boxes for cards that are overdrafted: they’re playable in other formats, they have high ceilings, low floors, and seem incredibly exciting when they work. They’re both higher rarities.
Jodah, the Unifier
Drafted after - 1.613 cards | Winrate - 0.391 | mythic
Jodah is a powerful card when certain conditions are met (5 colors, other legendary creatures) but those conditions are much harder to replicate in the draft format where players can’t anticipate what other cards they’ll have to work with. Forcing Jodah into a deck can make players play other, weaker cards to support him, all while running the risk of never drawing Jodah in any of their games.
Drafted after - 2.097 cards | Winrate - 0.4 | rare
Temporary lockdown is also incredibly situational. Playing four of them in a control deck might work in other formats, but drafting one in limited doesn’t really solve a problem as players won’t reliably draw the card before losing. It’s also a white card that hits many of the small aggressive creatures that white decks usually play in the draft format.
Magic the Gathering is a very complicated game, but with access to data, we can demystify some of these complications. Today we’ve shown one strategy to assess draft picks, and we’ve highlighted both key pitfalls and opportunities. But we’ve only scratched the surface - if you’re interested in reading more about card win rates (and some of the dangers of relying only on win rates) check out this article by 17Lands.
With Datawisp, we can do it without ever writing a line of code, opening the door for more players (and game developers) to gain insights into that data. And when data for the next set, The Brothers’ War (BRO), is released, we’ll be able to recreate this analysis easily by simply switching the data sets out.
We plan to dive even deeper into BRO by looking into how win rates change based on a particular card being drawn, and how combinations of cards affect win rates. These insights will in turn lead to new questions we can attempt to answer, and so on. Using this framework will allow us to discover deeper insights into each dataset as more data becomes available.
A huge thank you to 17Lands for their open source draft data and to Scryfall for Gatherer text and card images.