Posts from Statistics: figures and conclusions.

Hello to the hero community!
Today we have a moment of numbers and interesting facts.

I've always been curious about how certain aspects of the game affect the final result, and even more interesting was to look at the specific numbers.
I hope you'll enjoy it too and join me in a discussion. Don't take my conclusions too strictly; these are just guesses, and I'm sure the veterans will correct me. After all, everyone loves to discuss topics like "which town is the strongest"?

All data was taken from the resource http://heroes3stat.tk/rating/, for which many thanks to the creators.
From this, all implications follow - only games with a confirmed report were considered.
It should be noted that during the analysis, I tried to only take games of top players (top-50) against players of roughly their own level (rating difference does not exceed 250 points).

Using this information, I'll try to answer the following questions:
1) how big of an advantage does color give?
2) which towns are stronger than others on certain templates?
3) does the choice of starting / main hero affect the result?
4) and also look at the statistics of top players on individual templates.

For the first three questions, the sample will be for two canonical templates - JC and 6lm10a. Don't ask why those were chosen, I just needed to settle on something and I picked these (as some of the most popular).

So, let's go. I hope the info will be useful.
Statistic 1, color.

Here I'll immediately make one remark, which applies to all the data presented. People have been playing Heroes for a long time and realized quite a while ago that the game has some imbalance. This is partially compensated by the trading stage before the start of the game. The deviations in the table indicate that the color red is not just stronger on average—it is stronger even considering the trading stage, which is typical for almost all seasons. In other words, on Jebus Cross the color is valued almost correctly (giving about a 1% advantage), while on 6lm10a the color red is significantly undervalued (giving more than a 4.2% advantage). Perhaps this can be partly explained by the story of the biased road generation for the color blue.

Statistic 2, towns on JC



The commonly recognized bottom-2 according to all top players are Inferno and Tower. Castle is certainly strong on Jebus Cross, but before conducting the analysis, my opinion on town strength was different. One of the main conclusions I managed to draw: Dungeon is a heavily overrated town. Conflux has risen greatly in my eyes, while Rampart has fallen. As for the rest, judge for yourselves.

Statistic 2, towns on 6lm10a



Castle is turning into... no, not trousers. Into a bottom-2 town. This is a general trend—on resource-poor XL templates with complex structure, Castle simply lacks resources. But the striking power of Dungeon does make itself felt.

Statistic 3, table for starting heroes on JC



Here WinD is the deviation of the win percentage from the overall win percentage for the town. For example, picking Solmyr is definitely good, because it adds 0.72% to the chances of winning for Tower. At the same time, picking Mephala reduces the chances of winning for Rampart by 12.81%, making it a bad choice. Heroes with a small number of games should be treated with caution: with few games, the outcome could be anything. Here it makes sense to discuss popular hero picks:
Gretchin gives a significantly better start than Crag Hack (the 20% difference is still substantial); for Inferno, the best pick is Octavia (64.71%) vs. Marius (45%); heroes with the specialty of mysticism are heavily undervalued (I don't know why, but it really is the case :)). Tazar is also not in favor in the statistics, because he is indeed weak at the start—Alkin and Dragon give a better start. If the situation hasn't become clear, we move on to the next point.

Statistic 4, table for main heroes on JC



There is one point worth noting. Often the main hero is one freed from a prison in the center. This likely applies to heroes with a small number of games played. As for the rest, you can notice a number of very popular veterans with fairly poor winrates—Mullich, Gundula, Crag Hack, Shakti, Tyraxor, and others. The table also once again confirms Solmyr's uselessness as a main hero due to his... skill-pump(?). Alkin, Dragon, and Gretchin are on top again, joined by Broghild (due to fast and giant stacks of fairies on Jebus Cross) and a couple of heroes with the ballista specialty (which I personally find a bit strange). If it ever occurred to you to make Gundula, Marius, or Damacon your main—think again..

Moving on to 6lm10a.
Statistic 3, table for starting heroes on 6lm10a



I'm certainly no expert on 6lm, but it seems there's no fluff there. Seeing Vok at position 1 surprised me. It could also be the HotA effect (I think there's no Vok there, but the report site still draws him). Anyway, 7 games is still too few for serious conclusions.
Interesting is the presence of veteran Sir Mullich in the top (apparently zombies become significantly stronger at the start and turn into a full-fledged stack), many Conflux heroes (though it was already known that Conflux has top-tier mages), also the specialty in dendroids (I understand they can be built quickly without resource cost and have serious striking power), and of course Jeremy is imbalanced (by the way, it's possible that Illor is better). Ein slips into the top for Tower, which by the way matches FireToSky's picks on XL templates.

Statistic 4, table for main heroes on 6lm10a



With a large number of games - popular veterans are logisticians, Necropolis heroes, many heroes with offense - Damacon, Nimrus, Shakti, Gretchin, Shiva (it seems she's here for a reason, also in the top on Jebus Cross)... the rest are often situational mages (especially starting +gold, possibly some from prisons).

And for dessert: statistics of outstanding individuals by template. Just for fun, without analysis. A bit of hype for the topic won't hurt :D

== Helllight ==



== Amiloo ==



== FireToSky ==

Thank you, this is very interesting statistics. I also wanted to create something similar, but there's one nuance. It's mixed statistics from SoD and HotA. It's like mixing the statistics of black and white pieces in chess and checkers. Heroes3stat doesn't have a flag for SoD or HotA, which is a pity.

Thank you for your hard work!

Thank you, this is very interesting statistics.
I also wanted to create something similar, but there is one nuance. It contains mixed statistics from SoD and HotA. It's like mixing the statistics of black and white pieces in chess and checkers.
Heroes3stat doesn't have a flag for SoD or HotA, which is a pity.

Unfortunately, there is nothing that can be done about this (or is there?). We can only hope that the influence of HotA is not so global, which is quite possible, because the number of games played on HotA is not that large: the number of games played is approximately 57/500 of the expected number, i.e., about 11-12% (it's difficult to estimate precisely because of various castle regenerations and other things, but I think the error is at most a factor of 2).

Unfortunately, there's nothing that can be done about it (or is there?). We can only hope that the impact isn't so global, which is quite possible because the number of games on the site isn't that large: the number of games is approximately 57/500 of the expected number, i.e., about 11-12% (it's difficult to estimate, because of things like regenerating castles and so on, but I think the error is at most a factor of 2).

I actually wrote to nitro right before heroes3stat went down, asking if it was possible to add a separator, because I started writing statistics (I wrote a website parser) and realized that I would end up with a mess if I analyzed everything together.
By the way, how did you do the statistics?

Right before the fall of heroes3stat, I wrote to nitro asking if a delimiter could be added, because I started writing statistics (wrote a site parser) and realized I'd produce nonsense analyzing everything together.
By the way, how did you make the statistics?

You don't need to be a genius to understand that the ratings cannot be separated. The ХВ system does not recognize differences between them. To implement any kind of separation, the ХВ report system would need to be changed, which is problematic at the moment.

You don't have to be a genius to understand that ratings can't be split. The HeroesWorld system doesn't recognize the differences between them. To implement any kind of split, the HeroesWorld report system would have to be changed, which is problematic at the moment.

For gathering statistics, you could simply add an extra column with the game version (without changing the rating calculation), since the version is already indicated in the report.

You don't need to be a genius to understand that you can't separate the rankings. The HotS system doesn't understand the differences between them. To make some kind of separation, you need to change the HotS reporting system, which is problematic at the moment.

I'm not talking about separating the ranking; I don't need it in principle (plus, it's clear that if a player plays well in one version of the game, they can probably play just as well in the other).
When we submit a report, we enter the game version. It's true that it's not displayed afterward, but I really hope it's written somewhere in the database. And to make a distinction, as boovli pointed out, you can simply add a column to the table. If, of course, it's not written to the database and the field is just decorative, then yes, that's a shame.
I said that statistics compiled for both versions of the game at the same time reflect the "average temperature in the ward," showing some kind of overall overpowered element or overall weakness, but if someone dominates in one version and sucks in the other, and vice versa, then it will look average, even though it would be important to understand where they dominate and where they suck.
Something like that.

For collecting statistics, you could simply add an extra column with the game version (without changing the rating calculation), since the version is indicated in the report.

And to make the distinction, as boovli noted, you could just add a column to the table. If, of course, it isn't written to the database and the field is just decorative — then yes, that's unfortunate.

Let me try to clarify. The HeroesWorld report system and the rating calculation website are two completely unrelated resources. The site that calculates the rating simply scans the forum and reads reports written according to a template. The fact that we select a version when submitting a report — that is not displayed in the report itself. I am almost certain that this information does go somewhere into the database, but nothing more. In other words, to somehow implement a separation of ratings, the report system itself would need to be reworked so that the game version is displayed in the report (as a template, for example). But for that to happen, someone needs to do it, and the person who could do it hasn't visited this site for months, so even implementing such a simple feature is highly questionable.

Okay, then here's a question. Is it possible to get a dump of the tables containing forum reports (which likely indicate the game version being used)? If so, who should I contact?

I actually wrote to nitro right before heroes3stat went down, asking if it was possible to add a separator, because I started writing statistics (I wrote a website parser) and realized that I would mess things up by analyzing everything together.
By the way, how did you create the statistics?

Well, I parsed some website using a library (CsQuery for C#) and then used LINQ to process the data.