Is net worth still the best win indicator?
New buildings, neutral items, free wards: The DotA2 Outlanders Update was released late in November 2019 and introduced significant changes to the game. Game updates and patches are a common feature in esports. They keep the game fresh, accessible to new users, and balanced - for casual players. For the pro scene, each update forces the players to adapt to the new rules. The meta game shifts. The game dynamics change.
And if the dynamics change, then so do the predictions. At Bayes Esports Solutions one of our core businesses is to predict the outcome of DotA2 matches. After every large update we will sit down and evaluate our current models. Are they still performing well? Are there new potential features that we should consider? Only if we react to meta changes quickly can we continue providing the best betting odds within the industry.
This is especially true for the Outlanders update, which we consider to have the most impact since the Dota 7.00 patch. Here is why: It's hardly a secret that net worth and XP gaps are the two most significant indicators for who will win the game. While that is still true, we have seen a significant number of "net worth upsets" - teams winning despite being at a significant disadvantage - in the past month of professional game play. So before looking further into the game mechanics, let us consider how the importance of net worth and XP has changed with patch 7.23.
So of course our very first line of analysis is to see how exactly the impact of net worth and xp has changed. As usually, we start by looking at correlations. For this analysis we have taken roughly 2000 matches from patches 7.21-7.22 and roughly 600 matches from patch 7.23. Only professional esports matches were considered. Here we are looking at the entire match, from the start (where all values are equal for the teams) until one teams successfully wins.
Let's first look at the old meta. We see that XP and net worth gap are strongly* correlated. Throughout the markets, we have a very decent correlation between net worth gap and the outcome, and a slightly lower correlation beteen the XP gap and the outcome - which makes sense within the game.
Now let's look at what is going on in patch 7.23. We see that net worth and XP gap are still strongly correlated, no changes there. But looking at the correlation coefficient between net worth gap and map winner, we go from 0.48 to 0.36 - that's a very significant drop! The change is even greater when it comes to Roshan. What used to be a solid 0.35 has dropped to 0.2! For our Data Science team this means that they will need to have a closer look at the market and find better features that determine when a team goes for Roshan.
Interestingly, almost no change can be seen when it comes to who will get the first tower. This makes sense within the game, as you will usually get the tower early on, before you have a chance to get a high tier item drop.
* note that the notions strongly/moderately/weakly correlated are not uniformly defined. For DotA2 features, we consider any correlation above 0.6 to be strong, anything within 0.3 - 0.6 to be moderately correlated. Anything below 0.3 is a weak correlation.
About the author
Dr. Darina Goldin is the Director Data Science at Bayes Esports. She started playing competitive Team Fortress 2 in grad school. While no longer competing, she is still an avid Esports fan. At Bayes, she has created numerous predictive models for Counter Strike, DotA2, and League of Legends. When not crunching numbers, you can find her at the gym training Brazillian Jiu Jitsu.