Chatter Plots: Game Length in GSL Season 2

Published on 09/24/2019 05:24 PDT by Firnafth

The average game in the 2019 GSL Season 2 lasted 10.7 minutes, but since there were a few moderately long games, the most common game duration was 8-10 minutes.

We tested the statistical significance of differences in game duration between all matchups with a Kruskal-Wallis test and a Dunn's Test for multiple comparisons, taking the average duration of all games between the same players for the analysis (n=58). Game duration differed significantly between matchups (p=0.0002), but the only difference between individual pairs of matchups was PvT's (11.54±1.76 minutes, n=17) significantly greater length than PvZ (8.55±1.42 minutes, n=18). This is quite different from our findings for WCS Winter Americas, where TvZ was longer than all other matchups.

In WCS Winter Americas, there were no very short (<5 minute long) TvZs, and furthermore several TvZs ran over 20 minutes; in GSL Season 2, three TvZs were close to only 4 minutes long and only a few ran over 20 minutes, making this matchup much more similar in its extremes to all of the others. A lower average and range of duration for PvZ in GSL helped differentiate it from PvT; and while ZvZ games were relatively short in GSL, the small number of examples made it difficult to make any firm declarations about the state of ZvZ overall. The same can be said for TvT.

There was no significant difference between the lengths of non-mirror matchups (PvZ, TvZ, PvT) when one race won vs. the other (t-tests, all p > 0.25).

If we compare the overall shape of the histograms of game duration in this tournament vs. those of WCS Winter Americas and the first Super Tournament, while there are some differences, overall the distributions do appear fairly similar.

We took the average duration of games between the same players on the same maps and found that there was no statistically significant difference in the duration of games between maps (n=169, Kruskall-Wallis test on square-root transformed values, p=0.22). This is similar to our findings for WCS Winter Americas, although that tournament did have a partially different map pool. It seems that even though some maps are known as 'rush-favoring' or 'macro' maps, this is not reflected in the durations of games. To explore this more, we will have to look at strategies used on the maps and their success rates, which would be an analysis for another time.

In our next article, we'll take a look at buildings proxied, of all kinds, in GSL Season 2!

Data Collection and Analysis

We collected data by watching VODs of all games played for 2019 GSL Season 2 (n=182 games). For each game, we notated the players, matchup, winner, and game duration; during the games we also notated the first four nonsupply buildings built by each player and any proxies attempted; and we tallied all attacks, their initiator, whether the attack was harassment or not, and its general unit composition as well as the unit composition used to defend against it.

All statistical analyses were done in R 3.5 with the stats and PMCMRplus packages.

For fully detailed methods, see this document.

For raw data, see this spreadsheet.