適切な統計検定を選ぶ
Pythonで学ぶ実験計画法
James Chapman
Curriculum Manager, DataCamp
適切な検定の選択
- データセットの特徴
- データ型
- 分布 → 多くの検定で正規性を仮定
- 変数数
- 仮説
結果: 正確で信頼できる結論！
- t検定、ANOVA、カイ二乗
データセット：運動成績
- トレーニング法と食事が運動成績に与える影響
athletic_perf.sample(n=5)
Athlete_ID Training_Program Diet_Type Initial_Fitness Performance_Inc
167 Endurance Plant-Based High 9.113040
289 Endurance Keto Low 11.039744
164 Endurance Plant-Based Medium 11.614835
30 Strength Keto Medium 7.384686
186 HIIT High-Protein Low 6.776078
独立2標本t検定
- _2群_ の平均を比較
- 仮定: 正規分布、等分散
from scipy.stats import ttest_ind group1 = athletic_perf[athletic_perf['Training_Program'] == 'HIIT']['Performance_Inc'] group2 = athletic_perf[athletic_perf['Training_Program'] == 'Endurance']['Performance_Inc']t_stat, p_val = ttest_ind(group1, group2) print(f"T-statistic: {t_stat}, P-value: {p_val}")
T-statistic: 0.20671020082911742, P-value: 0.8364563849070663
p_val > $\alpha$ → 平均差の証拠は不十分
一元配置分散分析（One-way ANOVA）
- 複数（>2）群 の平均を比較
- 仮定: 群間の等分散
from scipy.stats import f_oneway program_types = ['HIIT', 'Endurance', 'Strength'] groups = [athletic_perf_data[athletic_perf_data['Training_Program'] == program] ['Performance_Increase'] for program in program_types]f_stat, p_val = f_oneway(*groups) print(f"F-statistic: {f_stat}, P-value: {p_val}")
F-statistic: 1.5270022393256704, P-value: 0.2188859009050602
p_val > $\alpha$ → 平均差の証拠は不十分
カイ二乗検定（連関）
- カテゴリ変数 間の関係を検定
- 分布の仮定なし
from scipy.stats import chi2_contingency
import pandas as pd
contingency_table = pd.crosstab(athletic_perf['Training_Program'],
athletic_perf['Diet_Type'])
Diet_Type High-Protein Keto Plant-Based
Training_Program
Endurance 33 28 33
HIIT 27 32 40
Strength 38 29 40
カイ二乗検定（連関）
chi2_stat, p_val, dof, expected = chi2_contingency(contingency_table)
print(f"Chi2-statistic: {chi2_stat}, P-value: {p_val}")
Chi2-statistic: 2.154450885821988, P-value: 0.7073764021451127
p_val > $\alpha$ → 連関の証拠は不十分
Vamos praticar!
Pythonで学ぶ実験計画法
