Analysis of consumer behaviour among DotA 2 players and its effect on individual performance

Abstract

This thesis examine the possible e ect of consumer behaviour, specifically gold sources distribution and e ectiveness in spending, on rank between DotA 2 players. Relevant literature from the past years was summarised and dataset containing more than 91 thousand data points from OpenDota API used to determine the relationship was constructed. Ordered Logistic Regression and K-Means Clustering algorithms were applied on four datasets of 12 variables divided based on the player's position and the outcome of the game, conclud- ing that only a fraction of the variation (McFadden's R2 of 6.2% and Cox & Snell's R2 11.7% on average) has been explained and cluster analysis resulted in a mean accuracy of 38.25%. Those results were supported by the correlation analysis indicating low correlation coe cient values of dependent variable rank. Despite such output, patterns were discovered across individual rank clusters from metrics acclaiming higher e ectiveness of more experienced players re- garding consumer behaviour with strong statistical significance. It has been concluded that the e ect is present, yet its magnitude by itself is not su cient to predict the regressand. The outcome implies possible presence of other vari- ables that have an e ect on the explained variable which are not included in the...

Tato bakalá ská práce zkoumá moûn˝ vliv spot ebního chování, konkrétn dis- tribuce zdroj zlata a efektivity b hem nakupování, hrá hry DotA 2 na jejich v˝konnosti. Byla shrnuta relevantní literatura z posledních let a byl sestaven datov˝ soubor obsahující více neû 91 tisíc datov˝ch bod z rozhraní Open- Dota API, jeû byl pouûit ke zkoumání tohoto vztahu. Na ty i datové sady obsahující 12 prom nn˝ch rozd len˝ch na základ pozice hrá e a v˝sledku hry byly pouûity algoritmy uspo ádané logistické regrese a K-Means algoritmus se záv rem, ûe byl vysv tlen pouze zlomek variability (pr m rné hodnoty McFad- denova R2 6,2% a Cox & Snellova R2 11,7%) a shluková anal˝za s p esností 38,25%. Tyto v˝sledky byly podpo eny korela ní anal˝zou, která ukázala nízké hodnoty korela ního koeficientu závislé prom nné v˝konnost. Navzdory t mto v˝sledk m byly v jednotliv˝ch v˝konnostních skupinách objeveny zákonitosti potvrzující vyööí efektivitu zkuöen jöích hrá v oblasti spot ebního chování se silnou statistickou v˝znamností. Záv rem byla p ítomnost tohoto efektu, avöak jeho velikost sama o sob není dostate ná k p edpov di závislé prom nné. V˝sledek nazna uje moûnou p ítomnost dalöích prom nn˝ch, které mají vliv na vysv tlovanou prom nnou, jeû nejsou v anal˝ze zahrnuty.