Its not random at all. Just like the dmg you deal as a hero in DotA. Lets say u are Chaos Knight with 30-70 dmg. The computer decides your dmg in a way that you have to do the last hit to get the creep kill, whiich shows that although the dmg was meant to be random, its not

Heh looks like sanjikun is the only one that graps the concept

Anyway here is how it goes guys, if the dmg dealt by your hero to an enemy unit is really random, do you think you still need to go for the last hit? Imagine, as a Centaur with a dmg of 53-63, you have to fight last hit with a range unit that has a dmg of 22-29. Go figure. Other examples of random stuffs that are not really random at all: Greater Bash as sanjikun mentioned, chances of critical etc. You can go do a statistical analysis if you want.

As for the stunning stuff, it really depends on the mechanics of DotA. Lets say that the ulti of Tidehunter actually creates a unit that causes the stun, then as long as he has produced the unit, even if u have stunned him, he can still stun you. Although I do not fully understand the mechanics of DotA. According to the scenario descriped this would be a more probably explaination rather than the "lag" issue

This post has been edited by Yukito: Mar 29 2006, 12:35 PM

Basically in statistical analysis, extrapolating is always an estimate and not accurate at best.


Yes it is true you do not know which 16 hits will bash. But if a large enough sample size is taken, approximately 16% of your hits consists of Greater Bash, if it is truly random. Plus minus some percentage. The larger your sample size, the smaller you deviation from the ideal figure would be.


Yes I agree. Sorry for not doing so but here I go

Background
Actually back when me and my friends were crazy about DotA (well now we really lost all our enthusiasm), we did many many tests on it. One of it was the random-ness of the dmg dealth by a certain hero.

Hero of choice
Chaos Knight, at lvl 1 he has a dmg range of 33-73 (as according to 5.67), which was the largest range around among the DotA heroes, and seemed the most ideal candidate for a random-ness test by measuring his damage over time.

Experiment Design
The sample involves 50 level 1 melee creeps. We had a team of 10 people and each of us would note down the HP of the creep before and after the damage was dealt, which after mathematical processing would show the damage dealt. Regen of the creep was considered negliegeble. If the Chaos Knight should level up, the game would be restarted and the experiment continue. Any waves of creeps that are suspected to be relatively close to the tower would not be calculated in fear that should the creeps come close to the tower for any reason it would make the calculation difficult. Another 4 heroes was selected and placed in the same lane as chaos knight to delay its leveling up time.

Results were mainly separated into 2 sections. Mainly, the dmg dealt before the creeps HP is above 200 (Table 1) and below (Table 2). Note that the creep had approximately 800 HP as its maximum HP thus our hypothesis was that the dmg dealt would be different as the creep's HP hits below 25% and therefore be shown not random as all. The figure 25% was selected simply because this number being relatively significant as according to our opinions. The same judgement was made for the sample size chosen.

Results
A graph of score against damage dealth was plotted. The score represents the number of times a certain dmg has been dealt. According to the graoh, table 1 results yielded an almost perfect bell-shape curve (Figure 1) whereas for table 2 a graph that was unequally skewed to both lefft and right was yielded (Figure 2). A shape that is in sharp contrast to the former.

Discussion
According to figure 1, the yielding of an almost perfect bell-shape curve, in contrast to figure 2, an unequally skewed graph, shows a sharp contrast in the scores between the tables. The results obtained was in accordance to the hypothesis. If the dmg dealt was random, there should not be a significant difference between the 2 graphs. The sample size chosen was believed to be large enough to accomodate such a conclusion.

Conclusion
In conclusion, the damage dealt by a DotA hero, is assumed to not be random, with Chaos Knight as a reference.

Although not much explanation is done right here, I believe the data would be enough.

This post has been edited by Yukito: Mar 30 2006, 09:55 AM

1000 samples is usually a number that is, IHMO, relatively significant. This is shown that a normal distribution can be achieved simply by the data from the table 1. Remember that it is the difference in both graphs that is the key point and demonstrates and non-randomness of damage dealing.


When checking random-ness, normal distribution is usually what comes into my mind.


So there are 5 heroes right? 2 buy flask of sapphire water, 3 buy four gaunlets of str (the 12 additional dmg is welcomed as there are more variables to etst with). Basically what we do is we go for the next waves of creep if possible, those where creeps are not involved and would not interfere with the results. Flask to heal.

walabies, remember the time when I was promoting the greatness of DotA and you were just lying on the bed promoting normal warcraft and saying how lame was it? It was at that time I did the test

P/S now doing watch movie. I have something else to say but I would tell u guys later.

This post has been edited by Yukito: Mar 31 2006, 03:58 PM

Dude as I said the results were separated into two tables based on the creeps HP. Refer back to my post.


I wish to remind you guys that the notion possessed by AI to "attack lowest health creep" is only implemented after v5.67, the version I tested on. Before that, the AI was really simple, frag all enemies they see, its a no-brainer for them.


Don't worry, I am not THAT crazy yet

After being through DotA since the old times, the balancing issue is always there, no such thing is "now is more balance than last time". Its how you see it.


No such thing as few heroes dominating the game, except for clan matches. And its still the case in ver 6

Version 5 has quite a number of comboes available, if you check out the replays of tournaments To say they are dull, is inaccurate at best.

In v6 clan matches do you guys think hyperstone > mask of madness for Sven?