These stats are derived from games submitted here. As such, they are subject to clerical error. The form has also been revised many times, so some older data was submitted when newer options were not available to select. Of course, the least "popular" heroes are the ones that were added to the game the most recently.
This is derived from data generously provided by Handelabra. To eliminate extra counts from multiplayer games and Weekly One-Shots, any games submitted on the same day that have exactly the same setup and conclusion are counted as just one game. Win statistics are based only on completed games, so there's a bias in that games where people quit after deciding they don't like the way things are going aren't counted. It also doesn't count games on Android because of inconsistent reporting.
Because there's a lot more data from the video game than from the user-submitted form, the logistic regression model is calculated for each hero, villain, and environment. This is visible on the Details tab and the Multiverse page. This is helpful in, for instance, determining which hero is most successful against a particular villain, and which heroes really shine against a particular villain compared to their average across all villains. However, data is still limited in some cases (mostly content that was added to the game the most recently), so consider all data in that light.
As of June 2019, only data from 2019 onward is shown. The dataset is just too large to manage with the computing power I have available. However, the results still appear to be a representative sample of the whole.
The "Search" field in each table uses regular expressions. Mostly this works like a normal text search, but you can use some extra tricks with it. For instance, try: