Cheers mate - some interesting stuff again.
1 Like
That’s quite interesting.
The insight that instigation and threat projection are important is intriguing. If that’s the case, why not brew of haste and mead of madness?
I wonder why top players avoid the blade of slashing though.
That’s the one that throws me. Could be noise? I may look into how I could incorporate a larger dataset from multiple tournaments. Would require some data cleaning but seems doable.
Think it may just be that the 5pt stuff is used at the end to fill a gap, rather than part of possibly better planning?
Looking at what was taken by the top percentile (outside of boots and GGS which seem to help!) could be interesting.
I’d guess that the artifacts were really focused on enhancing good units, rather than buffing mid rank ones or using leftover points at the end.
Also bit odd is that since Blade and Mace are (excluding a few situations) statistically the same, the former was taken a lot more.
It could be that blade of slashing is a go-to for everyone with 5 points to spare, but higher rated players have been favouring fire oil instead?
1 Like
For low att heroes the reroll is effectively an extra attack, while the mace doesn’t matter if you don’t hit (or have the CS to damage with all hits often).
That the ‘feeling’ the items give - similarly that I’d give elite to units with better Me and vicious to those with better CS - even if ultimately, from a stats perspective they both increase your chance of ultimately causing a wound.
1 Like
I went ahead and added in aussie masters and french masters, where I had data ready to go. Beefed up the data set.
Here’s the full table (sorry for the formatting)
| item_name | Top_% | Bottom_% | Gap |
|---|---|---|---|
| Sir Jesse’s Boots of Striding | 50 | 28.6 | 21.4 |
| Gnome-Glass Shield | 16.7 | 0 | 16.7 |
| Blade of Slashing | 10.4 | 26.5 | 16.1 |
| Fire-Oil | 14.6 | 2 | 12.6 |
| Healing Brew | 2.1 | 14.3 | 12.2 |
| Tome of Darkness | 12.5 | 2 | 10.5 |
| Hammer of Measured Force | 4.2 | 14.3 | 10.1 |
| Pipes of Terror | 2.1 | 12.2 | 10.1 |
| Sacred Horn | 4.2 | 14.3 | 10.1 |
| Conjurer’s Staff | 8.3 | 0 | 8.3 |
| Inspiring Talisman | 4.2 | 12.2 | 8 |
| Dwarven Ale | 8.3 | 2 | 6.3 |
| Diadem of Dragonkind | 0 | 6.1 | 6.1 |
| Staying Stone | 22.9 | 18.4 | 4.5 |
| Lute of Insatiable Darkness | 16.7 | 12.2 | 4.5 |
| Trickster’s Wand | 12.5 | 8.2 | 4.3 |
| Axe of the Giant Slayer | 4.2 | 0 | 4.2 |
| Shroud of the Saint | 2.1 | 6.1 | 4 |
| Aegis of the Elohi | 2.1 | 6.1 | 4 |
| Ej Periscope | 6.2 | 10.2 | 4 |
| Brew of Haste | 10.4 | 8.2 | 2.2 |
| Blade of the Beast Slayer | 4.2 | 2 | 2.2 |
| Talisman of Silence | 4.2 | 2 | 2.2 |
| Boots of Levitation | 4.2 | 2 | 2.2 |
| Hann’s Sanguinary Scripture | 6.2 | 4.1 | 2.1 |
| Liliana’s Tear | 2.1 | 0 | 2.1 |
| Mace of Crushing | 2.1 | 4.1 | 2 |
| Piercing Arrow | 2.1 | 4.1 | 2 |
| Crystal Pendent of Retribution | 2.1 | 4.1 | 2 |
| Banner of Eternal Darkness | 0 | 2 | 2 |
| Chalice of Wrath | 10.4 | 10.2 | 0.2 |
| The Boomstick | 2.1 | 2 | 0.1 |
| Wings of Honeymaze | 2.1 | 2 | 0.1 |
| Wine of Elvenkind | 2.1 | 2 | 0.1 |
| War-Bow of Kaba | 4.2 | 4.1 | 0.1 |
| The Scrying Gem | 2.1 | 2 | 0.1 |
| Zephyr Crown | 2.1 | 2 | 0.1 |
| Amulet of the Fireheart | 6.2 | 6.1 | 0.1 |
| Mead of Madness | 2.1 | 2 | 0.1 |
| Skirmisher’s Boots | 4.2 | 4.1 | 0.1 |
| Scythe of the Harvester | 4.2 | 4.1 | 0.1 |
4 Likes
I think this is right. It reminds me of the furor around the monty hall problem–people feel something intuitively, and it’s hard to shift. It’s an easy math problem to test and see, but on fanatics you have people who feel passionately one way or the other.
Also the blade is the first 5-pt item listed, so that makes it easier to select.
1 Like
A quick data analytics dive into 2,600 tournament games to see how lists perform.
2 Likes
Interesting!
I wonder if the number of scoring units and volume of attacks migh interfere with each other?
More units come more attacks and units with US usually have more attacks than heroes.
Trash units/ofjective sitters often have lots of bad attacks, which might drive up the average?
Is the number of attacks only the att stat (i.e. melee) or does it include shooting too?
I track both total attacks and total ranged attacks separately. I’m sure theyre correlated with US. But frankly all of the effects I see are pretty minor–even with 2,600 games in the dataset, the best model I can generate is only marginally better than a coin flip. So list building matters, but it looks to me like player skill matters much more.