Please upgrade your browser for the best possible experience.

Chrome Firefox Internet Explorer
×

Has anyone tested the Alacrity break points? Are they no longer needed?

STAR WARS: The Old Republic > English > Public Test Server
Has anyone tested the Alacrity break points? Are they no longer needed?

phalczen's Avatar


phalczen
10.15.2019 , 06:13 AM | #11
So dipstik if I am reading your tables correctly, every time you compare an alacrity rating that is below and above our known thresholds for the 1.4s GCD (1212) or 1.3s GCD (3207) the p value is 0 indicating that the observation of different APM is unlikely to be the result of random chance. Every time you compare two alacrity ratings that are on the same side of a threshold I see the p-value much higher, suggesting the observed APM differences are much more likely to be the result of random chance.

This suggests to me that rounding is still occurring to two significant digits, at least as far as GCD is concerned.
If you think I've made a good contribution with this post, I kindly ask that you use my Refer a Friend link! Here is more information about the program.

Lhancelot's Avatar


Lhancelot
10.15.2019 , 09:01 AM | #12
Quote: Originally Posted by dipstik View Post
here are results for gcd casts. would be nice to have this data for channels (diagnostic scan/ lightning strike)

using data from https://docs.google.com/spreadsheets...gid=1526455495

summary:
https://pasteboard.co/IzZn0PG.png


Two-Sample T-Test and CI: 0, 862

Two-sample T for 0 vs 862

N Mean StDev SE Mean
0 20 1.5465 0.0500 0.011
862 20 1.4988 0.0170 0.0038


Difference = mu (0) - mu (862)
Estimate for difference: 0.0478
95% CI for difference: (0.0233, 0.0722)
T-Test of difference = 0 (vs not =): T-Value = 4.04 P-Value = 0.001 DF = 23


Boxplot of 0, 862


Two-Sample T-Test and CI: 0, 755

Two-sample T for 0 vs 755

N Mean StDev SE Mean
0 20 1.5465 0.0500 0.011
755 21 1.50029 0.00901 0.0020


Difference = mu (0) - mu (755)
Estimate for difference: 0.0463
95% CI for difference: (0.0226, 0.0699)
T-Test of difference = 0 (vs not =): T-Value = 4.08 P-Value = 0.001 DF = 20


Boxplot of 0, 755


Two-Sample T-Test and CI: 0, 1293

Two-sample T for 0 vs 1293

N Mean StDev SE Mean
0 20 1.5465 0.0500 0.011
1293 22 1.4120 0.0366 0.0078


Difference = mu (0) - mu (1293)
Estimate for difference: 0.1346
95% CI for difference: (0.1069, 0.1623)
T-Test of difference = 0 (vs not =): T-Value = 9.88 P-Value = 0.000 DF = 34

Two-Sample T-Test and CI: 0, 1294

Two-sample T for 0 vs 1294

N Mean StDev SE Mean
0 20 1.5465 0.0500 0.011
1294 22 1.4133 0.0341 0.0073


Difference = mu (0) - mu (1294)
Estimate for difference: 0.1333
95% CI for difference: (0.1062, 0.1604)
T-Test of difference = 0 (vs not =): T-Value = 10.00 P-Value = 0.000 DF = 33

Two-sample T for 0 vs 1724

N Mean StDev SE Mean
0 20 1.5465 0.0500 0.011
1724 22 1.4030 0.0213 0.0045


Difference = mu (0) - mu (1724)
Estimate for difference: 0.1435
95% CI for difference: (0.1187, 0.1683)
T-Test of difference = 0 (vs not =): T-Value = 11.90 P-Value = 0.000 DF = 25

Two-sample T for 0 vs 2155

N Mean StDev SE Mean
0 20 1.5465 0.0500 0.011
2155 22 1.4012 0.0222 0.0047


Difference = mu (0) - mu (2155)
Estimate for difference: 0.1453
95% CI for difference: (0.1203, 0.1703)
T-Test of difference = 0 (vs not =): T-Value = 11.97 P-Value = 0.000 DF = 25

Two-sample T for 862 vs 755

N Mean StDev SE Mean
862 20 1.4988 0.0170 0.0038
755 21 1.50029 0.00901 0.0020


Difference = mu (862) - mu (755)
Estimate for difference: -0.00149
95% CI for difference: (-0.01027, 0.00730)
T-Test of difference = 0 (vs not =): T-Value = -0.35 P-Value = 0.732 DF = 28

Two-sample T for 862 vs 1293

N Mean StDev SE Mean
862 20 1.4988 0.0170 0.0038
1293 22 1.4120 0.0366 0.0078


Difference = mu (862) - mu (1293)
Estimate for difference: 0.08685
95% CI for difference: (0.06912, 0.10457)
T-Test of difference = 0 (vs not =): T-Value = 10.01 P-Value = 0.000 DF = 30

Two-sample T for 862 vs 1724

N Mean StDev SE Mean
862 20 1.4988 0.0170 0.0038
1724 22 1.4030 0.0213 0.0045


Difference = mu (862) - mu (1724)
Estimate for difference: 0.09575
95% CI for difference: (0.08378, 0.10773)
T-Test of difference = 0 (vs not =): T-Value = 16.17 P-Value = 0.000 DF = 39

Two-sample T for 862 vs 2155

N Mean StDev SE Mean
862 20 1.4988 0.0170 0.0038
2155 22 1.4012 0.0222 0.0047


Difference = mu (862) - mu (2155)
Estimate for difference: 0.09757
95% CI for difference: (0.08527, 0.10988)
T-Test of difference = 0 (vs not =): T-Value = 16.05 P-Value = 0.000 DF = 38

Two-sample T for 862 vs 3017

N Mean StDev SE Mean
862 20 1.4988 0.0170 0.0038
3017 22 1.3983 0.0139 0.0030


Difference = mu (862) - mu (3017)
Estimate for difference: 0.10048
95% CI for difference: (0.09070, 0.11027)
T-Test of difference = 0 (vs not =): T-Value = 20.82 P-Value = 0.000 DF = 36


Two-sample T for 1293 vs 3017

N Mean StDev SE Mean
1293 22 1.4120 0.0366 0.0078
3017 22 1.3983 0.0139 0.0030


Difference = mu (1293) - mu (3017)
Estimate for difference: 0.01364
95% CI for difference: (-0.00350, 0.03078)
T-Test of difference = 0 (vs not =): T-Value = 1.64 P-Value = 0.114 DF = 26

Two-sample T for 1724 vs 3017

N Mean StDev SE Mean
1724 22 1.4030 0.0213 0.0045
3017 22 1.3983 0.0139 0.0030


Difference = mu (1724) - mu (3017)
Estimate for difference: 0.00473
95% CI for difference: (-0.00625, 0.01571)
T-Test of difference = 0 (vs not =): T-Value = 0.87 P-Value = 0.388 DF = 36

Two-sample T for 2155 vs 3017

N Mean StDev SE Mean
2155 22 1.4012 0.0222 0.0047
3017 22 1.3983 0.0139 0.0030


Difference = mu (2155) - mu (3017)
Estimate for difference: 0.00291
95% CI for difference: (-0.00843, 0.01425)
T-Test of difference = 0 (vs not =): T-Value = 0.52 P-Value = 0.606 DF = 35

Two-sample T for 2155 vs 3448

N Mean StDev SE Mean
2155 22 1.4012 0.0222 0.0047
3448 23 1.3080 0.0270 0.0056


Difference = mu (2155) - mu (3448)
Estimate for difference: 0.09327
95% CI for difference: (0.07843, 0.10811)
T-Test of difference = 0 (vs not =): T-Value = 12.68 P-Value = 0.000 DF = 42
I am sure this is a lot of important information but sadly it looks like an alien life form wrote it out to me. That or maybe Einstein, lol. I wish mathematical analytics was my strong point.
TRUE
Quote: Originally Posted by DarthSpuds View Post
RNG is counterproductive because it massively increases player dissatisfaction.
FALSE
Quote: Originally Posted by olagatonjedi View Post
As I detailed in another thread, RNG give the players more control over their gearing.

phalczen's Avatar


phalczen
10.15.2019 , 09:28 AM | #13
Quote: Originally Posted by Lhancelot View Post
I am sure this is a lot of important information but sadly it looks like an alien life form wrote it out to me. That or maybe Einstein, lol. I wish mathematical analytics was my strong point.
P-values are one way to compare two sets of data for differences. You want to use different types of statistical tests depending on the nature of the data and the variable being compared. p-values indicate the likelihood that the results or outcome you are observing are due to random chance, with lower p-values being less likely the observed results are due to random chance. A p-value near zero means it’s highly unlikely the result is due to random chance.
In this case, dipstik tested several sets of Abilities per minute just spamming a basic attack, at different alacrity ratings. The goal being to prove whether or not GCD rounding to tenths or hundredths of a second was occurring. Based on dipstik’s results, the differences between the test runs of basic attack spamming at different alacrity ratings is only significant when the tested alacrity ratings are on different sides of an AR breakpoint. In other words the results pretty much confirm that the GCD is still rounded to tenths of a second.

EDIT: this is proven by the low-breakpoints dipstik chose. An Alacrity Rating of 755, under the new formulae, translates into an Alacrity Percentage of 4.67%. An Alacrity Rating of 862 translates into an Alacrity Percentage of 5.27%. These percentages, if the GCD was rounding to hundredths of a second, would mean a GCD of 1.44 and 1.43 seconds respectively. Dipstik's results show no statistically significant difference between APM at either of those Alacrity Rating levels. Its also followed up in the high breakpoints dipstik chose. The comparisons between an Alacrity Rating of 1293, 1724, and 3017, all between the 1.4 and 1.3s GCD, show no statistically significant difference.
If you think I've made a good contribution with this post, I kindly ask that you use my Refer a Friend link! Here is more information about the program.

dipstik's Avatar


dipstik
10.15.2019 , 04:49 PM | #14
Quote: Originally Posted by phalczen View Post
P-values are one way to compare two sets of data for differences. You want to use different types of statistical tests depending on the nature of the data and the variable being compared. p-values indicate the likelihood that the results or outcome you are observing are due to random chance, with lower p-values being less likely the observed results are due to random chance. A p-value near zero means it’s highly unlikely the result is due to random chance.
In this case, dipstik tested several sets of Abilities per minute just spamming a basic attack, at different alacrity ratings. The goal being to prove whether or not GCD rounding to tenths or hundredths of a second was occurring. Based on dipstik’s results, the differences between the test runs of basic attack spamming at different alacrity ratings is only significant when the tested alacrity ratings are on different sides of an AR breakpoint. In other words the results pretty much confirm that the GCD is still rounded to tenths of a second.

EDIT: this is proven by the low-breakpoints dipstik chose. An Alacrity Rating of 755, under the new formulae, translates into an Alacrity Percentage of 4.67%. An Alacrity Rating of 862 translates into an Alacrity Percentage of 5.27%. These percentages, if the GCD was rounding to hundredths of a second, would mean a GCD of 1.44 and 1.43 seconds respectively. Dipstik's results show no statistically significant difference between APM at either of those Alacrity Rating levels. Its also followed up in the high breakpoints dipstik chose. The comparisons between an Alacrity Rating of 1293, 1724, and 3017, all between the 1.4 and 1.3s GCD, show no statistically significant difference.
good show ol chap

to add: the summary table png is a table of the 95% confidence intervals of difference between means between the row and column headers. for example, for row (left side) 0 column (top) 755 we have an interval of 0.0226 to 0.0699. That means there is a 95% chance the the difference between the gcd times if you have 0 alacrity and 755 alacrity will be between 0.0226 and 0.0699 seconds. For row 1294 and column 3017 we have an interval from -0.006 and 0.016 seconds. This means that there is a 95% chance that the difference is less than 0.02 seconds (essentially zero).

TrixxieTriss's Avatar


TrixxieTriss
10.15.2019 , 09:37 PM | #15
Quote: Originally Posted by dipstik View Post
good show ol chap

to add: the summary table png is a table of the 95% confidence intervals of difference between means between the row and column headers. for example, for row (left side) 0 column (top) 755 we have an interval of 0.0226 to 0.0699. That means there is a 95% chance the the difference between the gcd times if you have 0 alacrity and 755 alacrity will be between 0.0226 and 0.0699 seconds. For row 1294 and column 3017 we have an interval from -0.006 and 0.016 seconds. This means that there is a 95% chance that the difference is less than 0.02 seconds (essentially zero).
If I understand correctly between what you and Phal have said, 1.3 lvls will make little to no difference anymore and we will be making 1.5 to 1.4 builds if alacrity is rounding down?

ottffsse's Avatar


ottffsse
10.15.2019 , 10:42 PM | #16
There are pretty powerful set bonuses in the game now tied to certain longish cooldown abilities like entrench and mental alacrity which you do not want to delay or elongate the cd on those - you want those to be as short as possible actually without making your crit chance plumet bellow like 35-38%. just a 5 sec difference on such a cooldown makes a huge difference in dps. Check the set boni damage boosts based on your class if your wearing such a set, because most of your dps increase comes in that window when something like entrench or alacrity etc is active. I believe many of the classes not just those two have such a set now. Activating such a cd (so the ability refreshed faster) while under something like an alacrity proc relic in scaled down content will be a petty standard trick I think.

On pure burst no dot damage specs yeah the baseline 1.4gcd / 7.5% alac is fine, rest crit usually. On dot classes I hate going under 10% alac because dots start to "tick" slower and I do think there is at that point a noticeable dps decrease. Maybe in Pvp where you prioritize bigger hits over sustain you can drop it... but there again some neat cooldowns on longer (1 min+) cd abilities will be noticeably slower at 7.5% alac vs 10%.
ref http://www.swtor.com/r/gFbhFZ Melisen / Sage Zrella / Sorc

phalczen's Avatar


phalczen
10.16.2019 , 06:53 AM | #17
Quote: Originally Posted by TrixxieTriss View Post
If I understand correctly between what you and Phal have said, 1.3 lvls will make little to no difference anymore and we will be making 1.5 to 1.4 builds if alacrity is rounding down?
No. The breakpoints still exist, if you have enough alacrity rating you could run with a 1.3s GCD. It will be significantly different APM than at 1.4s GCD, i.e. it works just fine. The results were only to prove or disprove if GCD was being rounded to tenths or hundredths of a second. It’s rounding just like LIVE, up to tenths of a second.
If you think I've made a good contribution with this post, I kindly ask that you use my Refer a Friend link! Here is more information about the program.

TrixxieTriss's Avatar


TrixxieTriss
10.16.2019 , 01:05 PM | #18
Quote: Originally Posted by phalczen View Post
No. The breakpoints still exist, if you have enough alacrity rating you could run with a 1.3s GCD. It will be significantly different APM than at 1.4s GCD, i.e. it works just fine. The results were only to prove or disprove if GCD was being rounded to tenths or hundredths of a second. It’s rounding just like LIVE, up to tenths of a second.
Ok, cool. Just checking.

Lhancelot's Avatar


Lhancelot
10.16.2019 , 06:02 PM | #19
Quote: Originally Posted by phalczen View Post
P-values are one way to compare two sets of data for differences. You want to use different types of statistical tests depending on the nature of the data and the variable being compared. p-values indicate the likelihood that the results or outcome you are observing are due to random chance, with lower p-values being less likely the observed results are due to random chance. A p-value near zero means it’s highly unlikely the result is due to random chance.
In this case, dipstik tested several sets of Abilities per minute just spamming a basic attack, at different alacrity ratings. The goal being to prove whether or not GCD rounding to tenths or hundredths of a second was occurring. Based on dipstik’s results, the differences between the test runs of basic attack spamming at different alacrity ratings is only significant when the tested alacrity ratings are on different sides of an AR breakpoint. In other words the results pretty much confirm that the GCD is still rounded to tenths of a second.

EDIT: this is proven by the low-breakpoints dipstik chose. An Alacrity Rating of 755, under the new formulae, translates into an Alacrity Percentage of 4.67%. An Alacrity Rating of 862 translates into an Alacrity Percentage of 5.27%. These percentages, if the GCD was rounding to hundredths of a second, would mean a GCD of 1.44 and 1.43 seconds respectively. Dipstik's results show no statistically significant difference between APM at either of those Alacrity Rating levels. Its also followed up in the high breakpoints dipstik chose. The comparisons between an Alacrity Rating of 1293, 1724, and 3017, all between the 1.4 and 1.3s GCD, show no statistically significant difference.
Very nice translation, I appreciate your work here lol.
TRUE
Quote: Originally Posted by DarthSpuds View Post
RNG is counterproductive because it massively increases player dissatisfaction.
FALSE
Quote: Originally Posted by olagatonjedi View Post
As I detailed in another thread, RNG give the players more control over their gearing.