TvP late game phenomenon: an empirical perspective

TvP is arguably the most frequently discussed match up in the current metagame. At a point, it was said that the match up is imbalance and favors Protoss. The current understanding is that Protoss has an advantage in the late game, while Terran has to leverage their strength in the early and mid game.

Recently, Blizzard made an statement regarding the state of this match up. This topic was intensively discussed on TeamLiquid forum. In summary, Blizzard acknowledged that Protoss has an advantage over Terran in the late game. Terran has to take advantage before it goes into the late game.

We do agree that if both sides take few to no losses going into the late game, protoss can have an advantage. That said, we also know that terran players have a lot of offensive capability and harassment options at their fingertips in the mid-game. If terran players press that mid-game advantage, then protoss can’t necessarily get into the late game at their full potential, which can nullify any advantage they might have had.

Startale_PartinG is probably the best PvT Protoss player in the world now, and he has a win rate of 87.5% in GSL against Terran. He has extremely high win rate in late game against Terran. This reinforces Blizzard’s statement above.

The purpose of this study is to put this generally agreed on understanding of the current TvP into test. Before moving into the empirical analysis, there are some key terms required proper definition for the sake of this study.

First of all, what is late game?

In general, players group different stages of the game into three categories, “early game”, “mid game” and “late game”. However, there is no common consensus of how early is early? Or how late is late? Most players will answer along the line of “beyond X min, then it’s late game”. X varies across individuals.

The argument that I want to bring across here is not what X should be, but rather should the state of the game be judged by the duration of the game? Instead of looking at time, another plausible approach is looking at it from the supply count. Again, it goes back to what’s the supply threshold then. Further argument about using supply as a benchmark is that it is too unstable since it is not consistent as it may fluctuate as the players battle.

Another plausible approach is by looking at the tech. Let say, when Protoss has Colossus, it is late game. However, this highly depends on the build and other extraneous factors. Therefore, time is arguably the best measurement. Although time also has the same problem as other approaches mentioned above,  it is still the most accountable for quantitative analysis.

So, how should the three different states of the game be differentiated based on time?

Let’s start with the threshold between early game and mid game. In Day[9] Daily #458, the focus is on the mid game, which makes it a good reference point. Day[9] repeatedly mentioned that Terran moves out with two medivacs, which is approximately 10-11th min. This is based on the assumption of using the most common TvP build now. Indeed, it is the most commonly used build in TvP in GSL this season. Before the first pair of medivacs are out, Terran generally sit back macro up. The objective of Terran’s play changes when the medivacs are out. Therefore, I propose that early game is ranged from 1st min to 10th min and mid game starts from 11th min.

Next is the money question, what is the threshold between mid game to late game? For the purpose of this study, I will define it as when both sides have the key infrastructures constructed, basic upgrades completed (e.g. charge and stim etc.), and have three or more mining bases. Based on this definition, there is a wide range of time as it differs more from game to game compared to early to mid game. In general, it falls around 15th to 19th min. Therefore, the threshold between mid to late game used in this study is 18th min.

To sum up, the state of the game is operationalised based on duration,
– Early game: 1st to 10th
– Mid game: 11th to 17th
– Late game : 18th and beyond

H1: Protoss has a higher win rate in late game than early and mid game.

H2: Terran has a higher win rate in mid game than late game. (Based on Blizzard’s statement)

Since it is an empirical research, data collection is required. The sample used in this study is the TvP games in Code S of GSL Season 2 of  2012. It is hard to be comprehensive by covering as many tournaments as possible due to various restrictions. Code S represents the top players in the world and it’s a tournament that lasted for months. There are 67 TvP games played in season 2. This should give a reasonable representation of the top pro level scene.

The time here is referring to the in-game time in Starcraft. The stated time is the time when “GG” was typed by the defeated players.

As mentioned earlier, there may be limitation to this method using duration of the game for justification. MVP did a 2x proxy barracks play, and it resulted in a low econ game in game 7 against Squirtle in the final. However, it  dragged on to the 11th min, which is considered a mid game not early game based on my operational definition. It is hard to justify if one should consider such unorthodox game as early or mid game.

One major assumption of this study is that a player wins at the time frame when he has advantage. In other words, let say Terran has an advantage before 18th min and opponent may not concede at that point. The advantage will accumulate and eventually wins the Terran player the game after 20th min, which may affect the result of this study. However, based on the hypothesized statements by majority of the community that Protoss is stronger as the game moves into the later game, then probably there is a linear relationship to that. Both actual metric measurement and categorical measurement (early/mid/late) are used.

Therefore, this study should still shed light about this TvP late game phenomenon.

Map is included as a plausible variable since the duration of the game may be affected by the map. Map may also directly have an impact on the win rate as some maps may be favorable to a race in the match up.

H3: Map affects the duration of the game.

H4: Map affects the races’ win rate.

It shows that duration of the game does not have an statistical significant effect on the match up for both metric and categorical measurements. This means that both H1 and H2 are not supported

Map does not have statistical significant effect on the winner of the match up and duration of the game. This means that H3 and H4 are not supported.

In conclusion, it may not be true that Protoss has an advantage in the late game and Terran has an advantage before that. This phenomenon may be intuitive but it is not statistically supported. With that being said, Starcraft is a dynamic game with depth, it is hard to come to a conclusion based on what we think we have observed. There are other unobserved variables which are hard to generalise across different games.

In the end, it may just be “The better player wins in this match up.”, as suggested by quite a number of pro gamers in the interviews during this GSL season.

3 thoughts on “TvP late game phenomenon: an empirical perspective

  1. Hmm… But your sample size from just the GSL is a bit small. Each category is only represented around 30 times on duration, and 10 times on map. Also, these series of Bo3 / Bo5 / Bo7 play a lot different to Bo1 scenarios… There is a lot of metagaming going on in these that you wouldn’t see, for example, in ladder games, or in larger, faster tournaments like MLG where they don’t have time to prepare for a certain opponent over a week’s time.

    Still, interesting to see that there was no significance here.

    You should look into getting sc2gears, if you’ve heard of it – you can download mass numbers of replays an analyze them pretty quickly :D

What do you think?

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s