STRATZ
Published in

STRATZ

Dota 2 Item Timings

Defining the most important item timings

When we went to Captains Draft 4.0 in Washington DC, we spoke to Nahaz of datdota about the lack of good information out there on how much impact item timing has on winrate. Whilst many people who play DotA 2 have considered that there are items that could be more effective if purchased by a specific time, we have yet to see anyone attempt to quantify item timing in a meaningful enough way that we can make direct comparisons and rank times of purchase against one another.

Since we have detailed item purchase data on all heroes at STRATZ, we are in a good position to delve into this data and publish our findings.

When we asked the question, “By how much does winrate change for each minute the item timing is missed?”, we postulated that the faster the winrate falls off, the more important that item must be for attaining victory.

In mathematics, the slope or gradient of a line is a number that describes both the direction and the steepness of the line. Thus, the bigger the winrate-gradient for a hero-item combo, the more important it is.

In essence, we are equating item-timing importance with the rate of change of winrate, and expressing the rate of change as the mathematical gradient.

Let’s check out a sample winrate-over-time for Phantom Assassin Black King Bar purchases.

Phantom Assassin BKB. Blue points show win percentages of item purchases in that minute [error bars indicate 95% confidence interval, calculated by assuming win-loss can be represented by Bernoulli trials]. Orange line represents the distribution of when item was purchased. The dashed lines indicate the times between which approximately 69% of the results lie. Solid lines indicate the mean purchase time of the item on the hero (vertical), as well as hero average winrate (horizontal).

This winrate shape is similar to what we see for many hero-item combos. Unpredictable behaviour often occurs very early and very late.

This unpredictability can be attributed to a combination of fewer games where purchases were made at odd times, and a higher proportion of ‘weird’ item build orders. As an example, some very early timings on items occur when players skip boots and other important early game items to rush their first big item out. This tactic is effective in far fewer situations than it is attempted.

Note that in the central 69% of timings (purchases between 1 standard deviation above or below the average purchase timing), we tend to get a sensible well defined slope. In other words, the winrate decreases every minute the item gets delayed. Therefore, we define our winrate gradient as being the [linear] gradient in the area between these 69% boundary times.

For some of these graphs you will notice that any purchase time for an item has a higher winrate than average hero winrate. How is this possible? Simply completing an important item can be crucial to winning. If the game ends before a hero can finish building their item, there is a very high probability that they lost the game.

How do different build orders affect results?

Why could this be a problem? We will use two examples:

a) Armlet on Huskar

b) Orchid on Storm Spirit

a) Huskar Armlet (Averages: 49.2% winrate, 13min 25s timing)
b) Storm Spirit Orchid (Averages: 49.2% winrate, 27min 31s timing)

In our first example, armlet on Huskar is an essential first major item for two reasons:

  • Huskar’s passive gives magic resistance, so he only requires armour and HP to survive, both of which are provided by armlet.
  • Huskar’s passive increases his attack speed as his health decreases, so the ideal situation is to maintain low enough HP to deal tonnes of damage while avoiding death. Since armlet can be toggled to control his HP, it’s arguably the perfect item.

This is why armlet is purchased in over 98% of Huskar games at Divine level.

In our second example, Orchid is a great item on Storm because the hero requires mana regen and attack speed, both of which are provided by Orchid.

Storm Spirit players will sometimes go Orchid straight after boots, especially against enemy cores with important escape spells like Puck or Ember Spirit. However, seeing Storm attempting to build Orchid will trigger either Euls Scepter, BKB or Manta Style purchases to dispel the Orchid silence. This is a situation where poor timing can greatly decrease Orchid’s usefulness against those cores. Since kills on supports aren’t as valuable in the early stages of the game, the Orchid rush becomes very detrimental.

Due to this risk, many Storm Spirit players will opt for items that don’t offer as much kill potential but increase their farm speed significantly, like Bloodstone or Kaya, and purchase Orchid later on. We have already explained that it is a great item on Storm in general, and can assist in late game pickoffs on defenseless supports; a more useful objective when the game is 5v4.

In a case like this, an item build that delays purchase may have a winrate as high or higher than builds that rush the item.

This presents an issue where we are not 100% sure if a 25 minute Orchid purchase is a really poor timing on a rushed Orchid or a really good timing on a delayed Orchid.

These conflicting factors should naturally lead to shallower slopes or more weirdly shaped graphs in general. The timings will be more spread out as possible build variation increases.

Therefore a) will naturally have a bigger winrate gradient than b) due to greater item build variation.

Does this invalidate our approach for defining ‘most important item timings’? We don’t believe so. Why?

If an item can be purchased at multiple points in response to different circumstances, then it is inherently less important for a hero than an item that MUST always be purchased at a specific point.

Even though we are not controlling for precisely where in a build the items are purchased, it does not invalidate the question "What are the most important item timings in Dota 2?"

Highest win-rate gradient items

What items were considered?

We analysed the winrates of every hero, looking at their top 3 most commonly purchased expensive (more than 2000 gold) items throughout the game.

Out of the 345 hero/item combinations, the following ranked top 20 across last week’s matches:

For an item with -2%/min gradient, with an average timing of ~15 minutes (I.e. Armlet on DK), this means if you got your armlet 5 minutes late at 20 minutes your chance of victory drops by 10%.

The graphs for every one of these items can be found in the appendix at the end of the article.

Graphs for Top 3

Arc Warden Hand of Midas (Averages: 48.0% winrate, 11min 19s timing)

Hand of Midas: The ultimate greedy item in DotA 2. It takes 15 minutes with 100% efficiency to achieve return on your investment in a Midas, when other items like Force Staff or Drums of Endurance can be purchased for roughly the same price. Because buying this item effectively puts your team behind for the next 15 minutes in many scenarios, it’s often only purchased when heroes hugely benefit from the extra XP the item provides.

Arc Warden can use Midas on both his main hero and his tempest double, greatly increasing the item’s efficiency.

Invoker Hand of Midas (Averages: 49.8% winrate, 10min 36s timing)

All of Invoker’s spells become stronger with each level. Therefore he is more XP dependent than any other hero. Whilst the extra gold is a nice bonus, he primarily purchases midas for the bonus XP it provides.

Anti-mage Battlefury (Averages: 47.2% winrate, 16min 19s timing)

Commonly Occurring Items

In the table above we see a lot of:

  • Hand of Midas
  • Blink Dagger
  • Armlet of Mordiggian

Hand of Midas: We have already discussed the item in the context of Arc Warden and Invoker, yet we also see Lone Druid on the list.

Midas is typically important for Lone Druid since it comes as a core part of
the “Radiance rush” build. Because Radiance is such an expensive item, purchasing Midas beforehand only slows the Radiance down by a couple
of minutes.

Since the Spirit Bear also benefits greatly from the attack speed and better late game scaling granted by the Midas, it has become a staple of the “bear build” to go Midas into Radiance.

Blink Dagger: This doesn’t require much explanation. For heroes with large AoE [Area of Effect] melee range disables like Axe or Magnus, this item is absolutely vital to effectively use those spells.

Armlet: This item is commonly bought on strength carry heroes because it provides better stats for the cost than any other item. The heroes we see it being important on are carries that cant quickly flash farm because they lack AoE creep clearing capabilities, and have to make up for it by killing heroes instead. This makes them very ‘snowballey’ heroes, meaning victory for them is heavily dependent on maintaining momentum and keeping ahead of the opponent carry’s net worth. This idea of ‘staying ahead’ is what leads these early armlet timings to be more impactful on game outcome than one would initially expect.

Restricting results to Ancient/Divine ranks

Things to note:

  • At higher ranks many of the hero-item combos are still in the list, with Anti-Mage Battle Fury and Invoker Midas remaining top three.
  • Tinker Blink Dagger is most important on this list, after not being featured at all in other ranks. This is not too surprising. In the hands of an experienced player, Tinker is a hero that can feel impossible to kill. At lower levels, any mistake in positioning can mean his quick demise. This leads to less experienced players squandering the advantages they get from good Tinker item timings, which is far less common at higher levels.
  • Item timings for Divine/Ancient ranks have larger winrate gradients in general. This is proof that hitting the timings is more impactful at higher levels, due to a greater understanding of how to convert item strengths into victory.

Some interesting anti-timings

We also investigated some item timings that appeared to lead to decreased winrates the earlier they were picked up.

Lion Blink Dagger

Lion Blink Dagger could be an example of seeing two overlapping builds in the data. Around 19 minutes there is a peak in winrate from players who rush an early Blink after boots and achieve that goal. Winrate drops off as players who may have their timing delayed due to deaths complete it. The winrate increases again, however, when players who went Drums, Aether Lens, Force Staff, or any utility support item first complete their second item Blink at a decent time.

Note almost all data points are above the average winrate, because Blink is a pretty useful item on Lion. Games where you never manage to save up 2250 gold to complete your Blink are probably games you have lost.

Pudge Aether Lens

The dip at 20 minutes before the rise can possibly be attributed to Blink dagger having a greater game impact, meaning players who rush Aether Lens before Blink win less than those who go Blink first.

Should I be skipping smaller items to build important items at the right time?

If these timings are so important, will building Magic Wand or Ring of Aquila beforehand delay these timings and decrease your chance of victory?

If we take Terrorblade Manta Style as an example [one of the top 20 item timings], this is shown to not be true.

Looking at the detailed purchase patterns for Terrorblade, we see both Magic Wand and Aquila purchases having a positive impact on winrate.

In fact, purchasing Magic Wand increases winrate on 101 out of 115 heroes!

How can I use this info in my games?

Are you planning on playing one of these heroes with a top core item? First ask yourself if you are likely to hit your timing this match. If you see an enemy mid Viper or Queen of Pain with a roaming Pudge, hitting that X minute Midas timing on Invoker seems fairly farfetched. In this situation, it might be advisable to opt for a safer pick that has no such timing dependency.

If you’ve already picked your hero and need to hit your timing, be selfish. Ask your supports to stack jungle if you are Tinker, or switch lanes to get a good matchup if you are Huskar or Lifestealer looking for a powerful Armlet timing.

If you are supporting and have these cores on your team, you can use it to prioritise which lanes should be helped most. For example, the data shows Euls Scepter timings on Death Prophet being far more important than people might believe. Therefore, helping Death Prophet out mid by either babysitting or rotating to gank their midlaner, is often far more impactful than staying in the safelane to secure your carry farm.

We’ve discussed the items that have the most impact based on time of purchase, but you can research the impact of buying any item during any time period by following these steps:

  1. Navigate to STRATZ.com and select a hero page.

2. Scroll down to the Purchase Pattern section, and select “Detailed View”.

3. Once in Detailed View, select either the min or max time slider under the Purchase Pattern banner to move it.

4. Choose any time period of interest, and sort by cost, pick percentage, win percentage, or differential.

Many thanks for taking the time to read!

STRATZ_ThePianoDentist

Appendix (Bit for people who like graphs a little too much)

Here are the remaining graphs for the top hero-item combos.

Legion Commander Blademail
Axe Blademail
Invoker Aghanim’s Scepter
Death Prophet Euls Scepter
Dragon Knight Armlet of Mordiggian
Chaos Knight Armlet of Mordiggian
Chaos Knight Echo Sabre
Huskar Armlet of Mordiggian
Shadow Fiend Blink Dagger
Sven Blink Dagger
Dragon Knight Shadow Blade
Axe Blink Dagger
Bristleback Vanguard
Earthshaker Blink Dagger
Magnus Blink Dagger
Lone Druid Hand of Midas
Terrorblade Manta Style

--

--

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store