Deaths Statistics are attempting to show how often players are dying on fights, to what mechanics they are dying, and how much they are being overkilled by. Unlike other statistics on the site, there are a number of caveats, since involving wipes in statistics does create some interesting problems to solve.
For Deaths Statistics, you can view Wipes and Kills together (the default), only Wipes, or only Kills. When using wipes, a death cutoff is used after which no deaths will be counted. This cutoff is computed by starting at 30 seconds before the end of the pull. Any combat resurrections that occur in the last 30 seconds that result in the player getting up and continuing to fight will shift the cutoff to that resurrection point.
The first death cluster that occurs after the cutoff is included, and then all other deaths are omitted on the assumption that the wipe has been called.
All deaths are included for kills; the cutoff mentioned above is only used for wipes. This means that if you filter to just kills, every death involved in a kill will be tracked.
The metric displayed in the bars and graphs is a Death %. This represents the total number of deaths of that spec divided by the total number of players of that spec seen across all the fights. For example, if a pull contains three Subtlety Rogues, and one of them dies, but the other two do not, the Death % is 33%.
Multiple deaths will penalize the spec further, meaning that it is entirely possible for a Death % to exceed 100%. If a fight has two rogues and they die four times total, then the Death % is 200%.
Other statistics on the site - because they only involve kills - undergo a process called de-duplication. For kills, the site does a bunch of work to figure out the real identities of the characters involved as well as the guild that should be credited for the kill. This means that for statistics that only involve kills, it is possible to figure out that two uploaded logs contain the same kill and to only include that kill once in statistics.
As of right now, it is prohibitively expensive to do this for wipes, and in order not to weight wipes more strongly than kills in the death statistics, neither wipes nor kills are de-duplicated. This does introduce some inaccuracy into these statistics, although with enough samples, this duplication stops being particularly relevant. Still, it is something to keep in mind when viewing these statistics; even for kills there is a level of imprecision present that does not exist for the other metrics on the site.
Note this is a technical problem I hope to solve, even though it is pretty minor, i.e., I don't expect solving the problem will have much of an impact on the relative ordering of specs.
If a player death saves within the wipe cutoff, but then dies shortly after the cutoff, the player's death is included in the death statistics, since this is considered a "failed" death save.
When viewing a specific spec and boss, you will see an ability breakdown, i.e., what the spec died to and what the average overkill was for that ability. Keep in mind that some abilities are simply small ticking DoTs, and so even though that was the final killing blow, it may have been something else that really did the player in. I may eventually retool this section to try to infer what ability truly did the player in, but for now, keep that in mind when viewing these statistics.
Continuing to use Fallen Avatar as an example, you may see a spec dying to Dark Mark first and Molten Fel second. The Molten Fel deaths, however, are (unless you blew up all the platforms) probably the result of a Dark Mark knock into the lava anyway. Before blaming a specific killing ability, think through the mechanics of the fight to work out why that killing blow appears.
Deaths that involve Purgatory and Spirit of Redemption will show the ability that triggered them, so what actually killed you is properly recognized.
If a shaman reincarnates within a reasonable timeframe following a death, that death is treated as a death save and is not included in these statistics. This way shaman Reincarnation properly reduces their deaths in these statistics, since it is part of their survivability toolkit.
I would caution against using this data to make grand pronouncements regarding spec balance. Keep in mind the difference between farm (when people take silly risks to get high parses) and progression. For any given boss, some of the wipes will be progression and some will be screwups from guilds that have the fight on farm. All skill levels are encompassed by these statistics, so in that sense it's best to think of these results as being how much the "average" player dies.
In other words, there are no skill percentiles to rely on here. The best players, worst players, and average players are all just in the mix together, so these results should be considered to be what is happening to the "average" players of the spec. These average players may be dying to things that a high skill player considers trivial.
You can filter by item level if you so choose. This is not the item level of the individual player (again it's too tricky to store every player item level separately for wipes), so instead what is being used is the average item level of the raid.
Feel free to send us feedback on Discord or in the forums. Hopefully this helps you as players to see what the most dangerous parts of various fights are going to be for your respective specs so you can go into fights more prepared to survive!