Steamゲームの好評率に関する分析
蘇逸文
2026-06-30
1 Data
本分析では、Steamゲームに関するデータを用いて、価格、recommendations、発売年、無料ゲームであるかどうか、多人数ゲームであるかどうかなどが、好評率とどのような関係にあるかを確認する。
games <- read_csv("steam_games.csv.csv", show_col_types = FALSE)データに含まれる変数名を確認する。
names(games)## [1] "appid" "name"
## [3] "release_date" "required_age"
## [5] "price" "dlc_count"
## [7] "detailed_description" "about_the_game"
## [9] "short_description" "reviews"
## [11] "header_image" "website"
## [13] "support_url" "support_email"
## [15] "windows" "mac"
## [17] "linux" "metacritic_score"
## [19] "metacritic_url" "achievements"
## [21] "recommendations" "notes"
## [23] "supported_languages" "full_audio_languages"
## [25] "packages" "developers"
## [27] "publishers" "categories"
## [29] "genres" "screenshots"
## [31] "movies" "user_score"
## [33] "score_rank" "positive"
## [35] "negative" "estimated_owners"
## [37] "average_playtime_forever" "average_playtime_2weeks"
## [39] "median_playtime_forever" "median_playtime_2weeks"
## [41] "discount" "peak_ccu"
## [43] "tags" "pct_pos_total"
## [45] "num_reviews_total" "pct_pos_recent"
## [47] "num_reviews_recent"分析に使用する変数のみを抽出する。
games2 <- games[, c(
"name",
"price",
"recommendations",
"categories",
"genres",
"pct_pos_total",
"positive",
"negative",
"num_reviews_total",
"release_date"
)]2 Organizing Data
好評率が0または欠損しているゲームを除外する。
games2 <- subset(games2, !is.na(pct_pos_total) & pct_pos_total > 0)Demo、Playtest、DLC、Soundtrackなど、通常のゲーム本体とは性質が異なるものを除外する。
games2 <- games2[
!grepl("Demo|Playtest|DLC|Soundtrack", games2$name, ignore.case = TRUE),
]データの概要を確認する。
summary(games2)## name price recommendations categories
## Length:52954 Min. : 0.000 Min. : 0 Length:52954
## Class :character 1st Qu.: 0.990 1st Qu.: 0 Class :character
## Mode :character Median : 4.990 Median : 0 Mode :character
## Mean : 7.869 Mean : 1686
## 3rd Qu.: 9.990 3rd Qu.: 159
## Max. :199.990 Max. :4401572
## genres pct_pos_total positive negative
## Length:52954 Min. : 1.00 Min. : 0 Min. : 0.0
## Class :character 1st Qu.: 67.00 1st Qu.: 12 1st Qu.: 1.0
## Mode :character Median : 81.00 Median : 36 Median : 10.0
## Mean : 77.11 Mean : 2120 Mean : 351.9
## 3rd Qu.: 91.00 3rd Qu.: 182 3rd Qu.: 47.0
## Max. :100.00 Max. :7480813 Max. :1135108.0
## num_reviews_total release_date
## Min. : 10 Min. :1997-06-30
## 1st Qu.: 21 1st Qu.:2018-03-15
## Median : 55 Median :2021-02-22
## Mean : 2225 Mean :2020-09-04
## 3rd Qu.: 238 3rd Qu.:2023-07-27
## Max. :8632939 Max. :2025-03-10recommendationsは分布の偏りが大きいため、対数変換を行う。recommendationsが0の場合にも対応できるように、log(1 + recommendations)を用いる。
games2$log_rec <- log1p(games2$recommendations)発売日から発売年を表すrelease_yearを作成する。
games2$release_date <- as.Date(games2$release_date)
games2$release_year <- as.numeric(format(games2$release_date, "%Y"))価格が0ドルのゲームを無料ゲームとみなし、free_dummyを作成する。
games2$free_dummy <- ifelse(games2$price == 0, 1, 0)categoriesにMulti-playerまたはMultiplayerが含まれる場合、多人数ゲームとみなし、multiplayerダミーを作成する。
games2$multiplayer <- ifelse(
grepl("Multi-player|Multiplayer", games2$categories, ignore.case = TRUE),
1,
0
)Steamのgenreは1つのゲームに複数付与される場合がある。本分析では、主要ジャンルとしてAction、Adventure、RPG、Simulation、Strategy、Indieについて、それぞれダミー変数を作成する。
games2$action <- ifelse(grepl("Action", games2$genres, ignore.case = TRUE), 1, 0)
games2$adventure <- ifelse(grepl("Adventure", games2$genres, ignore.case = TRUE), 1, 0)
games2$rpg <- ifelse(grepl("RPG", games2$genres, ignore.case = TRUE), 1, 0)
games2$simulation <- ifelse(grepl("Simulation", games2$genres, ignore.case = TRUE), 1, 0)
games2$strategy <- ifelse(grepl("Strategy", games2$genres, ignore.case = TRUE), 1, 0)
games2$indie <- ifelse(grepl("Indie", games2$genres, ignore.case = TRUE), 1, 0)3 Descriptive Statistics
分析に用いる主な変数の記述統計量を確認する。
summary(games2[, c(
"price",
"recommendations",
"log_rec",
"pct_pos_total",
"num_reviews_total",
"release_year",
"free_dummy",
"multiplayer"
)])## price recommendations log_rec pct_pos_total
## Min. : 0.000 Min. : 0 Min. : 0.000 Min. : 1.00
## 1st Qu.: 0.990 1st Qu.: 0 1st Qu.: 0.000 1st Qu.: 67.00
## Median : 4.990 Median : 0 Median : 0.000 Median : 81.00
## Mean : 7.869 Mean : 1686 Mean : 2.008 Mean : 77.11
## 3rd Qu.: 9.990 3rd Qu.: 159 3rd Qu.: 5.075 3rd Qu.: 91.00
## Max. :199.990 Max. :4401572 Max. :15.297 Max. :100.00
## num_reviews_total release_year free_dummy multiplayer
## Min. : 10 Min. :1997 Min. :0.0000 Min. :0.0000
## 1st Qu.: 21 1st Qu.:2018 1st Qu.:0.0000 1st Qu.:0.0000
## Median : 55 Median :2021 Median :0.0000 Median :0.0000
## Mean : 2225 Mean :2020 Mean :0.2147 Mean :0.2104
## 3rd Qu.: 238 3rd Qu.:2023 3rd Qu.:0.0000 3rd Qu.:0.0000
## Max. :8632939 Max. :2025 Max. :1.0000 Max. :1.00004 Distribution of Recommendations
recommendationsは一部のゲームで非常に大きな値を取るため、まず10,000以下の範囲に限定して分布を確認する。
hist(
games2$recommendations[games2$recommendations <= 10000],
breaks = 50,
main = "Distribution of Recommendations (<= 10,000)",
xlab = "Recommendations",
ylab = "Number of Games"
)
次に、対数変換後のrecommendationsの分布を確認する。
hist(
games2$log_rec,
breaks = 50,
main = "Distribution of log(Recommendations + 1)",
xlab = "log(Recommendations + 1)",
ylab = "Number of Games"
)
5 Regression Analysis
まず、価格、recommendations、発売年、無料ゲームダミー、多人数ゲームダミーを説明変数として回帰分析を行う。
model1_all <- lm(
pct_pos_total ~ price + log_rec + release_year + free_dummy + multiplayer,
data = games2
)
summary(model1_all)##
## Call:
## lm(formula = pct_pos_total ~ price + log_rec + release_year +
## free_dummy + multiplayer, data = games2)
##
## Residuals:
## Min 1Q Median 3Q Max
## -79.906 -9.300 3.287 12.455 38.442
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -2.333e+03 4.671e+01 -49.949 <2e-16 ***
## price -7.157e-02 7.298e-03 -9.807 <2e-16 ***
## log_rec 1.239e+00 2.630e-02 47.119 <2e-16 ***
## release_year 1.192e+00 2.312e-02 51.576 <2e-16 ***
## free_dummy 4.536e-01 1.987e-01 2.282 0.0225 *
## multiplayer -3.833e+00 1.844e-01 -20.781 <2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 16.88 on 52948 degrees of freedom
## Multiple R-squared: 0.07832, Adjusted R-squared: 0.07823
## F-statistic: 899.8 on 5 and 52948 DF, p-value: < 2.2e-16次に、主要ジャンルのダミー変数を加えたモデルを推定する。
model2_all <- lm(
pct_pos_total ~ price + log_rec + release_year + free_dummy + multiplayer +
action + adventure + rpg + simulation + strategy + indie,
data = games2
)
summary(model2_all)##
## Call:
## lm(formula = pct_pos_total ~ price + log_rec + release_year +
## free_dummy + multiplayer + action + adventure + rpg + simulation +
## strategy + indie, data = games2)
##
## Residuals:
## Min 1Q Median 3Q Max
## -75.590 -9.157 3.138 11.982 43.881
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -2.496e+03 4.605e+01 -54.205 < 2e-16 ***
## price -2.033e-02 7.289e-03 -2.789 0.005294 **
## log_rec 1.341e+00 2.593e-02 51.705 < 2e-16 ***
## release_year 1.274e+00 2.280e-02 55.870 < 2e-16 ***
## free_dummy 7.090e-01 1.961e-01 3.615 0.000301 ***
## multiplayer -3.514e+00 1.885e-01 -18.639 < 2e-16 ***
## action -1.907e+00 1.539e-01 -12.386 < 2e-16 ***
## adventure -1.322e+00 1.525e-01 -8.673 < 2e-16 ***
## rpg -2.840e+00 1.832e-01 -15.498 < 2e-16 ***
## simulation -6.458e+00 1.747e-01 -36.971 < 2e-16 ***
## strategy -1.925e+00 1.866e-01 -10.317 < 2e-16 ***
## indie 2.771e+00 1.634e-01 16.961 < 2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 16.54 on 52942 degrees of freedom
## Multiple R-squared: 0.1147, Adjusted R-squared: 0.1146
## F-statistic: 623.8 on 11 and 52942 DF, p-value: < 2.2e-166 Interpretation
回帰分析の結果、priceの係数が負で有意である場合、価格が高いゲームでは好評率が低い傾向が見られると解釈できる。
また、log_recの係数が正で有意である場合、recommendationsが多いゲームでは、好評率も高い傾向が見られる。
free_dummyの係数が正で有意である場合、無料ゲームでは有料ゲームに比べて好評率が高い傾向があると考えられる。
multiplayerの係数が負で有意である場合、多人数ゲームでは好評率が低い傾向が見られる。
7 Additional Analysis
レビュー数が少ないゲームでは好評率が不安定になりやすいため、追加分析として、総レビュー数の水準ごとにサンプルを分けた回帰分析を行う。
games2$total_reviews <- games2$num_reviews_total
games2$review_group <- cut(
games2$total_reviews,
breaks = quantile(games2$total_reviews, probs = c(0, 0.25, 0.5, 0.75, 1), na.rm = TRUE),
include.lowest = TRUE,
labels = c("low", "medium_low", "medium_high", "high")
)
table(games2$review_group)##
## low medium_low medium_high high
## 13766 12890 13060 13238レビュー数グループごとに、価格やrecommendationsなどの変数と好評率の関係を確認する。
model_low <- lm(
pct_pos_total ~ price + log_rec + release_year + free_dummy + multiplayer,
data = subset(games2, review_group == "low")
)
model_medium_low <- lm(
pct_pos_total ~ price + log_rec + release_year + free_dummy + multiplayer,
data = subset(games2, review_group == "medium_low")
)
model_medium_high <- lm(
pct_pos_total ~ price + log_rec + release_year + free_dummy + multiplayer,
data = subset(games2, review_group == "medium_high")
)
model_high <- lm(
pct_pos_total ~ price + log_rec + release_year + free_dummy + multiplayer,
data = subset(games2, review_group == "high")
)
summary(model_low)##
## Call:
## lm(formula = pct_pos_total ~ price + log_rec + release_year +
## free_dummy + multiplayer, data = subset(games2, review_group ==
## "low"))
##
## Residuals:
## Min 1Q Median 3Q Max
## -77.286 -11.878 4.571 16.106 41.020
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -4.403e+03 1.290e+02 -34.123 < 2e-16 ***
## price -1.323e-02 1.342e-02 -0.986 0.324
## log_rec -2.194e-02 3.945e+00 -0.006 0.996
## release_year 2.216e+00 6.385e-02 34.705 < 2e-16 ***
## free_dummy -2.665e-01 4.562e-01 -0.584 0.559
## multiplayer -2.252e+00 4.844e-01 -4.649 3.37e-06 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 20.44 on 13760 degrees of freedom
## Multiple R-squared: 0.08367, Adjusted R-squared: 0.08334
## F-statistic: 251.3 on 5 and 13760 DF, p-value: < 2.2e-16summary(model_medium_low)##
## Call:
## lm(formula = pct_pos_total ~ price + log_rec + release_year +
## free_dummy + multiplayer, data = subset(games2, review_group ==
## "medium_low"))
##
## Residuals:
## Min 1Q Median 3Q Max
## -78.557 -10.516 3.232 13.521 44.518
##
## Coefficients: (1 not defined because of singularities)
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -3.120e+03 1.037e+02 -30.082 < 2e-16 ***
## price -1.359e-01 1.870e-02 -7.266 3.92e-13 ***
## log_rec NA NA NA NA
## release_year 1.582e+00 5.134e-02 30.819 < 2e-16 ***
## free_dummy -3.123e+00 3.961e-01 -7.884 3.43e-15 ***
## multiplayer -3.619e+00 4.113e-01 -8.801 < 2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 17.34 on 12885 degrees of freedom
## Multiple R-squared: 0.078, Adjusted R-squared: 0.07771
## F-statistic: 272.5 on 4 and 12885 DF, p-value: < 2.2e-16summary(model_medium_high)##
## Call:
## lm(formula = pct_pos_total ~ price + log_rec + release_year +
## free_dummy + multiplayer, data = subset(games2, review_group ==
## "medium_high"))
##
## Residuals:
## Min 1Q Median 3Q Max
## -68.640 -8.749 2.756 11.594 35.637
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -2.373e+03 8.106e+01 -29.278 < 2e-16 ***
## price -1.119e-01 1.888e-02 -5.928 3.15e-09 ***
## log_rec 2.671e-01 5.904e-02 4.524 6.13e-06 ***
## release_year 1.214e+00 4.015e-02 30.232 < 2e-16 ***
## free_dummy -1.614e+00 3.956e-01 -4.079 4.55e-05 ***
## multiplayer -3.974e+00 3.333e-01 -11.922 < 2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 15.02 on 13054 degrees of freedom
## Multiple R-squared: 0.07837, Adjusted R-squared: 0.07802
## F-statistic: 222 on 5 and 13054 DF, p-value: < 2.2e-16summary(model_high)##
## Call:
## lm(formula = pct_pos_total ~ price + log_rec + release_year +
## free_dummy + multiplayer, data = subset(games2, review_group ==
## "high"))
##
## Residuals:
## Min 1Q Median 3Q Max
## -69.504 -6.294 2.734 9.183 23.005
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -831.02866 63.28836 -13.131 < 2e-16 ***
## price -0.10845 0.01051 -10.323 < 2e-16 ***
## log_rec 1.20231 0.06481 18.552 < 2e-16 ***
## release_year 0.44967 0.03132 14.358 < 2e-16 ***
## free_dummy 2.59852 0.47534 5.467 4.67e-08 ***
## multiplayer -5.76976 0.24654 -23.403 < 2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 12.69 on 13232 degrees of freedom
## Multiple R-squared: 0.0835, Adjusted R-squared: 0.08315
## F-statistic: 241.1 on 5 and 13232 DF, p-value: < 2.2e-168 Interpretation
レビュー数が最も少ないグループlowでは価格の係数は負で有意であった。一方、medium_highでは価格の係数は正で有意であった。しかし，medium_lowおよびhighでは有意な関係は見られなかった。
このことから，価格と好評率の関係はレビュー数の規模によって異なる可能性がある。ただし，いずれのモデルでも決定係数は非常に小さく，価格のみで好評率を説明する力は限定的である。