Power creep is a familiar phenomenon in video games, where new introductions in the game begin to outpace the old by virtue of better stats, abilities, and/or relevance to the current meta. Pokemon has been around for 9 generations, and near 1,000 Pokemon have been added through multiple games. The wide diversity of Pokemon to choose from and the multiple generations of Pokemon to pick from lead to a likely conclusion: the more recent Pokemon will likely be stronger due to better stats, abilities, or meta relevance. I am curious to see how disparate the power of Generation I Pokemon are from Generation VII, since the competitive relevance of past Pokemon is important in determining whether they would need adjustments to make them fit better in later installments.
I would like to look across the multiple generations of Pokemon and determine the extent to which power creep has taken effect. The dataset I will be using for this analysis is a set of all Pokemon and their unique forms and their attributes from Generation I to Generation VII. This covers all available Pokemon and any unique forms from those generations, as it was only Generation VIII and beyond that removed Pokemon, moves, abilities, and other elements from past games in their respective games.
Data Dictionary:
id: the unique identifier of a Pokemon or a unique form of a Pokemon
ndex: the number assigned to the Pokemon in the National Pokedex
species: the name of the Pokemon
forme: the unique form of the Pokemon (form is in parenthesis, no alternate forms otherwise)
type1: the Pokemon’s primary type
type2: the Pokemon’s secondary type, may be absent
ability1: the Pokemon’s primary ability, equally likely to appear as the secondary ability if present
ability2: the Pokemon’s secondary ability, may be absent
abilityH: the Pokemon’s hidden ability, a rarer possible ability, may be absent
hp: the Pokemon’s HP stat, how much damage it can take in a battle before fainting
attack: the Pokemon’s attack stat, related to how much damage the Pokemon can do with physical attacks
defense: the Pokemon’s defense stat, related to how much damage from physical attacks are reduced by
spattack: the Pokemon’s special attack stat, related to how much damage the Pokemon can do with special attacks
spdefense: the Pokemon’s special defense stat, related to how much damage from special attacks are reduced by
speed: the Pokemon’s speed stat, determines the order in which Pokemon use their moves
total: the Pokemon’s stat total, the sum of all above stats
weight: the weight of the Pokemon as listed by the Pokedex
height: the height of the Pokemon as listed by the Pokedex
dex1: the Pokedex description for one of the games in its generation of origin
dex2: the other Pokedex description for the other game in its generation of origin
class: the descriptive aspect of the Pokemon in the Pokedex
percent-male: the percentage likelihood of encountering a male of that Pokemon, blank if the Pokemon is genderless or if the Pokemon is 100% female
percent-female: the percentage likelihood of encountering a female of that Pokemon, blank if the Pokemon is genderless or if the Pokemon is 100% male
pre-evolution: the previous Pokemon that the current Pokemon evolved from, may be absent
egg-group1: the primary egg group of the Pokemon, related to breeding, may be absent in Pokemon that cannot breed
egg-group2: the secondary egg group of the Pokemon, related to breeding, may be absent
[1] "HP"[1] 1.00000 70.04147 255.00000[1] "Attack"[1] 5.00000 79.60226 190.00000[1] "Defense"[1] 5.00000 73.73044 230.00000[1] "Special Attack"[1] 10.00000 74.55042 194.00000[1] "Special Defense"[1] 20.0000 72.9114 230.0000[1] "Speed"[1] 5.00000 70.32139 180.00000[1] "Total"[1] 175.0000 441.1574 780.0000In the above summary statistics, the minimum value is on the left, the mean value is in the middle, and the maximum is on the right. Some interesting minimums are HP at 1, Attack, Defense, and Speed at 5, Special Attack at 10, Special Defense at 20, and a stat total of 175. The minimum HP value is due to a second stage Pokemon named Shedinja, but is still viable due to its unique ability. The other unique stat minimum is Speed, which is a three-way tie between Shuckle, Munchlax, and Pyukumuku. All other stats are represented by first stage Pokemon that you would expect to have the lowest of that stat. All stats minus total which is the sum of all stats are around 70. The highest average stat is Attack at around 79.60, suggesting that more Pokemon would be classified as physical attackers than special attackers. The average stat total is arround 441, suggesting that Pokemon with stat totals above 441 would likely be better. Both kinds of attack stat max out around 190, whereas both defense stats max out at 230, suggesting that how damage is calculated favors the attack stats and that greater defense is needed to counteract that. A max HP of 256 gives further credence to the idea that attack stats are favored in damage calculation.
For the above graph, the average stat totals for each generation generally lean towards power creep, but defy the trend twice. First, in Generation II, the average stat total actually lowered, and is the lowest generation stat total overall. The second point where the stat total lowered was in Generation V, though this is likely due to it following Generation IV, the highest overall average stat total at near 500. Regardless, despite the two points where the average stat total lowered, the graph generally follows an increasing trend of greater stat totals every generation. However, the significance of the increases may not be as great, since several generations, namely Generations I, V, and VI, have very similar average stat totals and thus could be argued as a point against power creep since the change in their averages were not significant enough.
`stat_bin()` using `bins = 30`. Pick better value with `binwidth`.
$x
[1] "Stat Total"
$y
[1] "Count of Stat Totals"
$title
[1] "Histograms of Stat Totals by Generation"
attr(,"class")
[1] "labels"The distributions of stat totals, specifically the peak(s) of their distributions, are interesting in defining a generation’s average stat total. For example, Generation I is double-peaked at around 300 and 500, and does not fall off immediately in the other surrounding values. In Generation II, however, its peak is far and away the highest single group and is slightly over 300, which would explain why Generation II has the lowest overall stat total. Generation IV’s high average stat total can be explained by the sheer amount of Pokemon with stat totals over 700 (around 18 instead of between 5 - 10) in addition to having a high amount of Pokemon with stat totals around 500. As a rule, there is generally a peak at around 500 for all generations minus Generation II. Also, as the generations advance in time, the peaks and valleys become more exaggerated, with Generation II minus its lone peak and Generation III being fairly evenly distributed while Generation IV onward has seen increasingly wider gaps in stat total spread and more Pokemon concentrated in a couple peaks. Ultimately, the distribution supports and goes against power creep, as newer generations push more Pokemon into the 500 range, but the Pokemon around 600 decrease with newer generations.
$x
[1] "Generation"
$y
[1] "Ratio of Attack and Defense ((Attack + Defense)/Total)"
$title
[1] "Ratio of Attack and Defense by Generation"
attr(,"class")
[1] "labels"In general, despite a couple exceptions to the rule, Pokemon generally range between 20% to 50% of their stat total being devoted to their “physical” stats, that being attack and defense. Most points hover around 30% regardless of the generation, suggesting that across generations average attack and defense is a common/desirable design choice. The generation with greatest divergence between high and low physical Pokemon is Generation III, with a little under 10 points above 50% and more than that below 20%, suggesting this is the generation with the highest specialization. The next highest would be Generation II, while all other generations seem to fall within the limits with great consistency or fall outside the limits in a specific area, such as Generation IV having physical, low-stat total Pokemon and Generation VI’s divergence mainly occurring in high-stat total Pokemon. This result does work against the power creep hypothesis, as newer generations see less specialization than Generation III.
$x
[1] "Generation"
$y
[1] "Difference in Attack to Special Attack (Attack - Special Attack)"
$title
[1] "Difference of Attack to Special Attack by Generation"
attr(,"class")
[1] "labels"Specialization of offensive stats can also be used to determine power creep. A Pokemon specialized in a specific offensive stat is generally more effective than a Pokemon that is not specialized. This is due to the opportunity cost of more of the Pokemon’s stat total being invested into a stat it will not use. As a general trend, as the stat total increases, the split between special attack and attack widens, suggesting that low-tier Pokemon are generally unspecialized or lack large enough stats to have meaningful differences between them and that high-tier Pokemon can see greater specialization in their mode of offense.
For most generations, most Special Attack specialized Pokemon rarely go below - 50, or a Special Attack 50 points above its Attack stat. The generations that do most frequently are Generation I and Generation V. Conversely, Attack much more easily passes 50, suggesting that Pokemon with high Attack stats more often have a corresponding low Special Attack than vice versa. The two generations with the highest degree of Attack specialization are Generation I, Generation IV, and Generation V, with both having a decent number of points greater than 75 scattered consistently across its stat total spread. This would suggest that Attack - Special Attack divergence was greatest at Generation I and Generation V, which both contradicts and supports power creep. If Generation I’s divergence was larger than newer generations, this would argue against power creep, suggesting the original Pokemon were more optimized than newer additions, while Generation V suggests that power creep cropped up there with greater optimization of offensive stats in a newer generation.
$x
[1] "Generation"
$y
[1] "Ratio of Bulkiness ((Defense + Special Defense + HP)/Total)"
$title
[1] "Ratio of Bulkiness by Generation"
attr(,"class")
[1] "labels"“Bulk”, or the ratio of Defense, Special Defense, and HP to a Pokemon’s stat total, is meant to represent the general defensiveness and survivability of a Pokemon. Divergence in “bulk” also represents specialization that makes a Pokemon more viable, as the extra “bulk” represents an opportunity cost that prevents the Pokemon from having greater offensive stats like Attack, Special Attack, and Speed. Like the physical stats mentioned above, most Pokemon fall within 37.5% to 62.5% of their stats being HP, Defense, and Special Defense. The generations with the most “bulk” divergence are Generation III and Generation V. For Generations I, II, and IV the divergence in “bulk” was mostly unidirectional. In Generation I, most of the divergence in “bulk” is below 37.5%, and hardly past the line. In Generation II, there was more consistent high “bulk” Pokemon but only low-stat total low “bulk” Pokemon. In Generation IV, most of the divergence in bulk is above 62.5%, suggesting more “bulky” Pokemon but less diversity in low “bulk” Pokemon.
In all, this set of scatter plots is the most promising argument for power creep in Pokemon. All generations follow the same 37.5% to 62.5% pattern, yet newer generations go against this rule more often, which can be an indication of defensive power creep. Newer generations see increased divergence in “bulk”, offering greater opportunity for specialization to Pokemon in newer generations.
$x
[1] "Generation"
$y
[1] "Ratio of Speed (Speed/Total)"
$title
[1] "Ratio of Speed by Generation"
attr(,"class")
[1] "labels"Differences in Speed also help pave the way for power creep, as stellar speedsters can attack first and theoretically finish their opponent before they finish them, whereas slower Pokemon generally have less of their stat total devoted to Speed and more towards “bulk” or offense meant to outlast and overpower the competition. The scatter plot of Speed appears to be the most variable of the percentage stat total analyses, though this is likely down to Speed being the only variable represented in the numerator. The Speed ratios on the scatter plot generally fall between 5% and 25%. Generation I has a lot of low-level high Speed Pokemon, but hardly any low Speed Pokemon. Generation III, Generation V, and Generation VII all share greater numbers of high Speed Pokemon, but only Generation III and Generation V have a significant enough number of low Speed Pokemon.
This set of scatter plots also supports the power creep hypothesis through increased slow Pokemon in newer generations. While it sounds disadvantageous to go last, some moves and items use being slow to their benefit by changing turn order or by having greater priority in the turn. While fast Pokemon are fairly common, especially at lower stat totals, having a higher number of slow Pokemon creates a greater likelihood of strategies where the goal is not to simply “be faster”. Having a greater divergence of Speed stats allows for more competitive optimization in newer generations, between “glass cannons” which are fast and strong and “bruisers” that are slow but strong and bulky.
The above graph represents box plots of the expected value (EV) of each move from the generation that they were introduced in. The median EV rises with each generation, supporting the conclusion that moves introduced in newer generations are better than past generations by virtue of being more powerful and/or more accurate. Generation I has two outliers, but those two points represent Self-Destruct and Explosion, two moves designed to KO the user and obviously of niche if any strategic use. The spreads of the generations also tend to shrink, with the exception of Generation VI, suggesting that the lower EV moves specifically are becoming more accurate, stronger, or simply less prevalent in later generations. Generation VII is the generation with the most outliers, as most of the EVs appear to lay within 75 to 100.
Ultimately, while there is compelling arguments for both sides of the power creep argument, I conclude that Pokemon has experienced a degree of power creep. With the average stat totals, bulk ratios, speed ratios, and the move EV ratios most in support and conflicting results in most other graphs, most graphs point to a general power creep in later games. While the results may not be alarmingly different across generations, multiple graphs detailed a gradual increase in their values as the generations get newer. This result makes sense, because were Game Freak to create new Pokemon that completely blow the others out of the water, it would ruin the competitive balance of the games. A gradual increase in stats gives time for each generation’s meta to adjust, develop, and adapt, while reducing the viability of poorer-performing Pokemon slowly instead of making them completely irrelevant in the span of a game or two. Ultimately, I believe that how Game Freak has handled power creep in the series was right, making sure that not all of the original cast of Pokemon are rendered unused, but incentivizing gradually using newer Pokemon through small increases in their overall effectiveness.