Log in

Previous Entry | Next Entry

Stat Testing - Part II (MEVA)


In this second post on stat testing, I want to focus on XIV's current concept of magic evasion (MEVA) and a little bit on its counterpart magic accuracy (MACC).  Before starting, I want to state clearly that my intention in this post is not to offer a MEVA (or MACC) formula with regards to resistance rates.  As you'll see later on, the trial sizes in these tests were pretty large, but not near the amount required to reach that level of precision.  To contrast this with the 5,400 trial series of tests I'm presenting in the post, Lodeguy's original MACC testing in XI took about 23,000 trials.  What we can hopefully get out of this preliminary testing, though, is a rough idea of how MEVA and MACC are related to resistance rates, what other factors play a role in resistance rates, and finally, a ball park idea of how useful these stats are.

Special thanks to Miko Neversleeps for helping me with parts of the testing in this post.

As with my other math-heavy posts, I have sectioned off the methodology and discussion sections so that you can simply skip down to the "conclusions" section if the math does not interest you.



The overlying assumptions that I began with when testing MEVA were as follows:

    (1) When a "resist" occurs, there is some % cut in the final damage taken.
    (2) There are likely multiple tiers or strengths of resists.
    (3) MEVA applies not only to spells, but also elemental damage attacks and WSs.

The first 2 of these assumptions are probably pretty non-controversial; however, the third may take some quick convincing.  The issue with the game right now is that there are clearly "partial resists" on elemental attacks like certain enemy ranged attacks, but the log will only show the damage and never the "partial resists" message that you'll see with spells.  This can be misleading in that a resist occurs, but is not shown in the log unless one knows the damage cut.

With these assumptions in mind, I set out to find the idea test target, and eventually decided on the rank 35 Cursed Eyes in Nanawa Mines.  My reasons for choosing this target are:

    (1) Seismic Scream, the Ahriman default ranged attack, is Earth elemental
    (2) Seismic Scream is a normal attack that is spammed, making it high frequency
    (3) There are many Cursed Eyes in 1 small room, making pulling easy
    (4) All the eyes in the room are the same level, making it possible to control mob stats and level.

I would have liked to use a parser to streamline data collection; however, there was none that could log the exact amount of damage taken for each attack.  Because of this, I followed a similar procedure to the Cure testing in part I where I manually logged the information into Excel.  Goal trial sizes were 400+, though there are admittedly some that fall well short of this.


First Impressions and Raw Data Collection

My initial trial involved my rank 50 Conjurer with a MEVA of 420.  The Earth resist was "baseline" 341; however, the eyes' Scourge attack gives -8 Umbral resists and never misses, so the actual Earth resist for these trials is 333 usually.  Protect was not used.  A screenshot showing the summary data collection and first analysis is shown below:

image host

To clarify some points here...  The first thing I did was take the 420 trials and count the number of times each number that appeared at least once was seen.  I then separated them (by shades of gray) into distinct ranges (11-12, 22-24, 33-36, 44-48).  Looking at the distinct ranges produced, I reached the conclusion that this represented 4 "tiers" of resist:

    (1) No Resistance (or "None" in the screenshot above)
    (2) Single Resistance (or "Single") which appeared to be a -25% damage cut
    (2) Double Resistance (or "Double") which appeared to be a -50% damage cut
    (2) Triple Resistance (or "Triple") which appeared to be a -75% damage cut

I then sum up each "kind" of resistance seen and give a percentage.  Finally, I calculate a percentage labeled as "Effective" in the screenshot.  This is the percent of damage you will sustain at these resistance rates compared to a situation where you did not resist anything (or where "None" is 100%).


Does Elemental Resist Affect Resist Rate?

This is a simple test to see if resist rate is affected by the elemental resistance.  Because Seismic Scream is Earth based, I set out to create an Earth elemental resist set with some jewelry and some single socket Earth Veil materia.  I think just controlled for the MEVA and changed the earth resistance.  Results are below:

image host

We find here that the resist rates don't really change significantly, which leads me to believe that resist rate is not affected by elemental resists.  I will be the first to admit the trial size for the +Earth resist data set is fairly small, but feel comfortable with the result.  We can verify that Seismic Scream is indeed Earth based though, as the raw damage taken has gone down significantly (in fact, over halved).  How exact elemental resists function to reduce damage is an interesting question, but entirely separate from the discussion of this post, so I'll defer further discussion on that subject for later.


Plotting Resist Rates as a Function of Magic Evasion (MEVA)

Using the same procedure as outlined above, I completed the following data summary table using R35 Cursed Eyes in Nanawa Mines in addition to R52 Bloated Eyes found in the Gridania Treespeak Grand Company Leves (R52 at 5 stars).  Results below:

image host

Most of the data should be easy to interpret at this point with the exception of the "dMEVA floor" column, which I will explain later in the discussion.


R35 Cursed Eye Data Set

If we take the R35 Cursed Eye trials and plot them with resist rates as a function of the magic evasion (MEVA), we end up with a plot that looks like this:

image host

The most important aspect of this plot in my opinion is that the resist rate does not seem to have a premature cap, meaning you can reach a 100% "partial resist" rate (as seen in this plot occurring at 425 Magic Evasion).  This is different than XI, where the resist rate was always capped at 95%.  There was also a 5% floor in XI as well.  Simply from eyeballing MATK testing, I can prematurely confirm that there is no such floor in this game and you can reach 100% land rate (0% resist rate). 

Another thing I found interesting is that the distribution of resist types is not as dominated by "single resists" as I would have first guessed.  In XI, there were also 3 tiers of resist (though they were -50%, -75%, and -88% instead), but this distribution was calculated by a geometric distribution.  This meant that resists were calculated as a series of chances; say your resist rate was 50%.  Then the chance of you getting a "single resist" was 50%.  "Double resists" were calculated by 50% x 50% = 25%.  Triples 50% x 50% x 50% = 12.5%.  If we look at the distribution of resist types above, we can see that the rate of double and triples resists is much more favorable than in XI.

Finally, although not listed in the data collection summary, I was able to verify that 425 MEVA is the exact magic evasion required to hit the 100% partial resist cap.  I did this by "eyeballing" from 420 MEVA and stopping the instant I saw a non-resist (40+ damage).  There were non-resists up to 424 magic evasion (though very rare), but none were seen at 425 for 300+ trials.  This makes me pretty confident that for specifically R35 Cursed Eyes, the minimum MEVA to hit 100% partial resists is exactly 425 MEVA.


R52 Bloated Eye Data Set

I chose another higher level Ahriman target to try to get a better feel for how the magic evasion and resist rate relationship worked for lower lower resist rates.  I ended up going with the R52 Bloated Eyes since it was fairly easy to control level through choosing 5 star on the Grand Company Leve.  Other than switching mobs, the data collection process was the same.  Conjurer is used for self-healing.  A summary of the data was presented earlier and a plot of the Bloated Eye data is shown below:

image host

There is nothing really critical here to note looking at the above plot.  The only thing I'll mention here is that the rate of double resists was again, a bit higher than I would have expected.  The rate is triples was extremely low though.


Combinding Data Sets and Plots

Although we do not know if dLVL plays a role in the resist rates for a particular magic evasion, I felt that it would be nice to see what would happen if I was able to combine the R35 and R52 eye data sets.  However, because these 2 mobs have clearly different plots with respect to the amount of absolute MEVA required to reach a certain resist rate, I had to create a relative MEVA stat to combine them.  This is the purpose of the "dMEVA floor" stat mentioned earlier.

As discussed earlier, it takes exactly 425 MEVA to reach the 100% partial resist rate cap.  Using this information, I converted each of the R35 Cursed Eye data sets from absolute MEVA to MEVA relative to the amount needed to hit this cap (425).  This makes the old value 425 now 0, 420 is now -5, etc. 

Unfortunately, I was unable to reach MEVA values high enough to cap partial resists on the R52 data set.  I did, however, find a data set at an absolute MEVA of 544 which had an extremely similar resist rate profile to the R35 Cursed Eye data set at an absolutely MEVA of 385.  I took these sets to be near equivalent in terms of their relative MEVA, or "dMEVA floor" stat (-40).  This is a pretty heavy assumption to make, but bear with me as I put the graph together (shown below).

image host

Thumbnail link for larger copy:
image host

This plot here now gives us a pretty idea of how increases (or decreases) in magic evasion alter the resist rates.  From eyeballing, it looks as though it takes 200 MEVA to go from 0% partial resists to 100% partial resist.  It is difficult to tell the exact relationship between MEVA and resist rate when going from -200 to 0 here.  It appears to be -somewhat- linear; however, there may be an inflection point at -100 relative resist, which would make the curve non-linear and more complex.  We also don't know if -200 relative MEVA is the exact amount of MEVA required to first see absolutely no resists at all.  This is just guess-work on my part since there's no actual data here.

There is an added variable to this plot in "average % damage taken", which is essentially the "Effective Damage" stat mentioned near the start of the post.  If we look at this variable, we find that the effective damage reduction begins to drop off after reaching the partial resist cap.

Regarding the "usefulness" of this plot as applied to in game scenarios, its major drawback is that it uses "relative magic evasion" instead of absolute.  This means that for any mob, we will need a single point of data to orient ourselves onto the plot before we can get any information from it.  For instance, if we wanted to know the resist rate at 500 MEVA on the R58 Ifrit instance, we would first need a single point of data before we could predict how much lowering or raising MEVA would affect things.

This concludes the actual data collection and manipulation with regards to specifically magic evasion.  I'll be referring back to this plot later when discussing other areas such as magic accuracy (MACC).


Sample of MEVAs Needed to Cap "Partial Resists"

As just previously discussed, the data plot using "relative MEVA" is only applicable if we can find a single data point for a particular mob we are interested in.  I quickly went out to test a couple of mobs to find a data point and made this quick chart of the estimated minimum MEVA needed to "cap."

image host

The most pertinent one here is probably the R30 Ifrit (easy fight) at 610.  For an R30 mob, I was not even able to cap "partial resists" in my highest set at 544.  Though I did not collect sufficient data to post on the R58 Ifrit (hard fight), I went something like 1/20 on any kind of resist in 544 and just called it at that point.  It seems like MEVA can be very useful for normal sets of mobs, but for NMs with enhanced stats and MACC, it may be impractical to impossible to reach sufficient MEVA for reasonable resist rates


Effect of Enfeebling Resist Equipment

Now that we have a ball park idea of how MEVA works and the gains from stacking a certain amount, we can move to testing the "status resistance" stats and materia.  The easiest target for me to test this were the R50 Kobold Prelates just outside of the stronghold, which are THM that love to spam Bio, Poison, and Scourge.  I created a baseline using 544 MEVA without pants, then switched into single and double socket Poison Resistance (Aurelia's Kiss) pants to contrast the difference.  Trials took substantially longer to obtain since these are spell casts, so the trial sizes are low, which limits the interpretation of data.  Raw Data summary below:

image host

There are a couple interesting notes here:

    (1) Although you cannot see from this summary, the DoT damage was 14/tick regardless of "resist" value
    (2) The duration was always 45 seconds if Poison landed, always 60 seconds if Bio landed
    (3) As can been see in the graph, the resist rate did decrease with an increase in "poison resistance"

The specific amount of resistance to Poison gained with +21 poison resistance appears to roughly follow a 1:1 ratio with MEVA, meaning +21 poison resistance is the same as if you had +21 magic evasion for specifically poison.  There is no apparent decrease in DoT damage or duration.  This is a pretty underwhelming result if accurate as there are many of these "status resistance" pieces are on the head and feet, which can be socketed with Manaflight.  I'm not exactly sure why you would socket a status resist piece over Manaflight given these results since MEVA materia would give you global resistance and a higher value (IVs 25 max versus 15 max for resistance materia).

On another note, it's interesting to see that the spells have different "land rates".  For instance, we can clearly see that Bio has a higher land rate in the control test than Poison.  If we want to verify this with a Chi Squared test, Chi Square = 47.8 with a p value < 0.0001 This implies that spells may have a specific land rate modifier, making the land rate for each spell potentially different.  This makes sense since even pre-1.18 it was clear that certain spells seemed to universally land better than others.


Conversion from Magic Evasion to Magic Accuracy

This starts to get tricky in that when converting from 1 stat (MEVA) to another (MACC), we have to make a number of assumptions.  The key one here is the concept that +1 MACC equals -1 enemy/target MEVA.  I think this is a fairly safe assumption to make, but it is an assumption nonetheless.  For the following discussion and plot to be meaningful, we must first accept this assumption of a 1:1 MEVA to MACC ratio.  This being said, the conversion from MEVA to MACC is shown below:

image host

Thumbnail link for larger copy:
image host

I chose to define "0" as the amount of MACC needed to reach 0% resist rates since it seemed like the most important cap for magic accuracy.  There is an extra data point added here for the "0" value which is based solely on the assumption that it takes 200 magic evasion to go from 0% partial resists to 100% partial resists to add clarity to the plot.

Similar to the relative magic evasion plot in which this one is derived from, remember that we must first have a single point of data to orient ourselves onto the plot before we can draw any conclusions in game regarding the usefulness of MACC on a particular enemy.  This limits its use; however, it provides a nice estimation of how much "gain" one can achieve by globally increasing their MACC since the vast majority of situations will involve players between -200 and 0 on the relative MACC graph.


Stone "Magic Evasion Down" Effect

Moving on to more random game properties, I want to briefly talk about the "magic evasion down" effect attributed to the CNJ spell stone.  I tested this by using the R40 Lemurs near Treespeak and their "Soil Smear" Contagion effect.  I found that the effect produced -29 MEVA down that was the same for variation of +/-80 PIE (or +/-20 enfeebling skill).  This -29 MEVA amount could be a % of the mob's total MEVA, so it could vary on the mob, but it is not varied by the player's stats.



Moving straight to the summary of conclusions for this post...

    (1) All magic spells and elemental based normal/TP attacks are subject to a MEVA check.

The in game log currently will only openly notify you if a partial resist for magic spells.  For elemental attacks and TP moves, you will only notice the damage reduction of the resist, but will not be notified of the resist in the log.

    (2) There are 3 "tiers" of resists, labeled in this post as "Single" (-25%), "Double" (-50%), and Triple (-75%)

The game will only give you notice on magic spells that a resist occurs, but you will not be notified of exactly of the strength of the resist.  There is no apparent -100% resistance tier with the exception of using "Decoy" and direct non-damage enfeebling spells like "Bio".

    (3) Elemental Resistances only affect direct damage taken and not MEVA or resist rate.

There is a clear damage decrease when the correct elemental resistance gear is used on particular attacks, but not a change in the resist type rate.  This is a completely seperate kind of testing that I did not really go into in this post beyond the fact that it does not have anything to do with MEVA.

    (4) The resist rates for elemental damage attacks (specifically Seismic Scream) is plotted and shown here.
         This plot can be used to estimate how potential increases and decreases in MEVA will affect resists.

         The plots requires that you have at least 1 data point to be useful on particular mobs.  A brief table
         outlining a couple of mobs tested in this post is shown here as well.

image host

image host

    (5) There is no pre-mature resist rate floor or cap.  You can potentially resist 100% or 0%.

    (6) "Status Resist" equipment gear like Aurelia's Kiss does not change duration or potency of enfeebles.
         There is a 1:1 ratio where 1 "status resist" gives 1 MEVA for that status ailment.

Just to add clarify and emphasis, yes this does appear make MEVA the better stat since it adds global resistance as opposed to just resistance to a specific element.  The amount of testing done on this is pretty small so I could be easily missing something but based on my tests, I don't see how this is useful, especially on head and feet pieces which can receive Manaflight materia.

    (7) Enfeebling spells (maybe all magic spells) have their own unique land rates for a given MEVA.

This just means that if you resist say 50% of 1 spell at a particular MEVA, you may not resist 50% for all spells.

    (8) Enfeebling spells (e.g. Poison) likely only need to have any resist to fully resist.  There is currently no
         such thing as a "partial resist" for enfeebles right now.

    (9) The "Magic Evasion Down" effect on R50 CNJ's Stone gave -29 MEVA on R40 Lemurs.

This gives a rough estimate of the MEVA down effect for an R50.  This value did not change with a +20 increase in enfeebling skill (+80 PIE).  It could still change based on the mob's statistics or the CNJ's level though.

That just about does it! 

Taking briefly about potential application of these results to in-game scenarios, I feel that stacking Manaflights and MEVA probably won't be too useful for high-end mobs like Ifrit.  We can see that the MEVA to 100% "partial resists" even for the R30 Ifrit is extremely high and near unreachable currently without amazing triple socket gear at least.  The R50 Ifrit would be pretty impossible to reach a decent resist rate on.  However, for normal mobs in general instances like Darkhold, I can see a MEVA set being potentially useful, specifically for stopping enfeebles.

We also talked about in the post turning around the MEVA plot and plotting for the opposite stat in magic accuracy.  If we make the assumption that there is a direct 1 for 1 check between MACC and MEVA, we can get a plot like this:

image host

Ultimately, this may be the biggest "prize" piece of information coming out of this post due to the importance of avoiding resists for magic combos on mage jobs.  Seeing how most endgame situations have mages fall between the -200 and 0 range on the plot, it shows how potential gear increases in MACC roughly correlate to gains in the land rate of spells.

I'll probably end up coming back to some of the concepts tested in this post later on as I have more time.  For the most part, these are ball-park, qualitative ideas and statements and it would be interesting to try to add more quantitative data to back up the ideas here.  But moving on to what's ahead, there are other big, global concepts on stat increases I'd like to get to first.  Til part III.