3 thoughts on “Understanding Creep Score in LoL Stats”

  1. Hi Tim,

    Great creep score anaytical approach explanation and justified criticism to what Riot announced they want to change in neutral monster camps calculation to cs total. I personally strongly dislike the idea as well but first things first I wanted to add my own thoughts on creep score analytics in League of Legends.

    It is right that creep score is mostly a representation of gold it gives and most of people treat it exactly like that but there is one more thing: while using creep score to evaluate gold is not entirely precise there is another analytically reasonable way to look at creep score. It fact, and you mentioned that in your video, cs is a count of successful last hitting (landing the killing blow) on a minion. Let alone from it’s gold afiliation, cs can represent how good player’s micromanagement is by telling how many successful last hits player earned or rather how little they missed (i will expand this thought later). When comparing players’ creep scores (eg showing CSD@10) what I see is not the difference in gold count from a major source, but a difference in lane management, micro skill, push power, relfection to roamer’s laning, difference in pressure/matchup – all these combined in one value. In this approach losing canon minion should equal losing caster minion because they both show a single situation of for example being pressured or single effort to properetly time you auto attack. Of course in terms of economics it can be different, but when unable to have cs gold value it is still reasonable to use cs number when it counts only minion last hits. Furthermore, even when cs gold would be available data I would consider using both cs count and cs gold to see the whole story.

    How about monsters then? Well, in fact they fit neither to cs count nor cs gold. When killing jungler monsters in most cases player is the only damage source to them so he does not have to worry about finishing blow. Therefore monster kills dont reflect anything what cs count does. This considered, accurate measures should always separate minions and jungle monsters count/gold. Moreover, junglers and laners might require complete different set of measures. Proper jungler assessment requires knowing number of buffs taken and neutral monster gold in own and enemy jungle separately because taking enemy jungle is theoreticly worth more as it often denies that camp’s gold from enemy jungler. To see the whole story proximity measures (how much time jungler spent top/mid/bot) are needed as well.

    There is one measure I came up for that can be used for all roles showing their “standard” non event income, although it requires separating gold values into sources as you’ve mentioned in the video – the data that is not available right now (Riot please, its so crutial). I’ve already written to you about this idea, though – the measure is gold earned without kills, assists and turrets (gold w/o KAT). Do you consider this a good idea?

    One more thing I’d like to add here that expands the thought cs should rather be presented as how little is missed than how much was scored (in the early game at least). You said you ‘kinda like’ CSD@15 or @10. The question is what ‘kinda’ means in this sentence. Well, I think those scores have a big issue that make them not fully accurate. The reason of that is that they capture gamestate at one moment and because of that they might lead to incorrect conclusions. For example, pushing lanes like Caitlyn bot at excact 10 or 15 minutes might be pushing hard so that while she killed all enemy minions, her own accumulated in ~2 full waves. Enemy adc didn’t have chance to score those cs, however in the following minute or two he should be able to get most of them under his tower. As a result CSD@X measure sometimes might favor pushing lane but in fact the temporary difference will equalize in short time. I’m curious whether the massive CSD@10 Caitlyn leads at worlds group stage wasn’t partly caused by this factor.

    In the ideal scenario where we have all the data available the most accurate measures in my opinion that evaluate early game cs/gold would be calculated as: number of last hits player scored on minions in his lane to total number of minions died in player’s lane and similarly to that gold from minons last hit by player in his lane to total value of gold from minions that died in player’s lane. This could obviously be presented @10,15 and as a difference between player and his lane opponent just like CSD.

    I would love to hear what you think about my findings.
    Regards,
    Patryk

    1. Hey Patryk. Thanks for the thoughts. I do agree that it would be very useful to be able to measure gold from different sources, for example measuring gold from farm and gold from kills/assists and gold from objectives (towers, Baron, etc.) separately.

      As for the potential for creeps to be killed shortly before or after the 10:00 or 15:00 marks or whenever you are measuring, this is of course a possible issue with any time-stamped statistic. The hope is that these small advantages/disadvantages would balance out as you increase the sample size, but ultimately it’s an imperfection of timestamp stats that we simply accept and live with.

      1. Thanks for the reply,

        The problem with LoL esports is that sample sizes are entirely to small to count on Bernoulli’s law (large numbers) to balance the odds… This leaves us with really sad reality: we use inaccurate measures all around and try to build analytics on them. For analytical purposes, LoL data needs massive changes and extensions to be any worthy.

        If someone were to strongly apply to Riot to consider the analytical needs of their game’s data, they have my vote. I’ll be first to sign under this request.

Join the Discussion