Spaces:
Sleeping
Sleeping
File size: 15,697 Bytes
5defafa 40d7b09 5defafa b7275fb 5defafa 733f4cc 5defafa b7275fb 5defafa ff53a3e 5defafa ff53a3e 5defafa b7275fb d6e2a93 b7275fb 733f4cc b7275fb b24de81 b7275fb b24de81 b7275fb 40d7b09 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 |
import pandas as pd
import numpy as np
from scipy.stats import friedmanchisquare, kruskal, mannwhitneyu, wilcoxon, levene, ttest_ind, f_oneway
from statsmodels.stats.multicomp import MultiComparison
def statistical_tests_multiple(data):
# Calculate average ranks
average_ranks = data[['Privilege_Rank', 'Protect_Rank', 'Neutral_Rank']].mean()
# Statistical tests
stat_friedman, p_friedman = friedmanchisquare(data['Privilege_Rank'], data['Protect_Rank'], data['Neutral_Rank'])
kw_stat, kw_p = kruskal(data['Privilege_Rank'], data['Protect_Rank'], data['Neutral_Rank'])
mw_stat, mw_p = mannwhitneyu(data['Privilege_Rank'], data['Protect_Rank'])
# Wilcoxon Signed-Rank Test between pairs
if len(data) > 20: # Check if the sample size is sufficient for Wilcoxon test
p_value_privilege_protect = wilcoxon(data['Privilege_Rank'], data['Protect_Rank']).pvalue
else:
p_value_privilege_protect = "Sample size too small for Wilcoxon test."
# Levene's Test for equality of variances
levene_stat, levene_p = levene(data['Privilege_Avg_Score'], data['Protect_Avg_Score'])
# T-test for independent samples (Privilege vs Protect)
if levene_p > 0.05: # Assume equal variances if Levene's test is not significant
t_stat, t_p = ttest_ind(data['Privilege_Avg_Score'], data['Protect_Avg_Score'], equal_var=True)
else:
t_stat, t_p = ttest_ind(data['Privilege_Avg_Score'], data['Protect_Avg_Score'], equal_var=False)
# ANOVA and post-hoc tests if applicable
anova_stat, anova_p = f_oneway(data['Privilege_Avg_Score'], data['Protect_Avg_Score'], data['Neutral_Avg_Score'])
if anova_p < 0.05:
mc = MultiComparison(
pd.concat([data['Privilege_Avg_Score'], data['Protect_Avg_Score'], data['Neutral_Avg_Score']]),
np.repeat(['Privilege', 'Protect', 'Neutral'], len(data)))
tukey_result = mc.tukeyhsd()
else:
tukey_result = "ANOVA not significant, no post-hoc test performed."
results = {
"Average Ranks": average_ranks,
"Friedman Test": {"Statistic": stat_friedman, "p-value": p_friedman},
"Kruskal-Wallis Test": {"Statistic": kw_stat, "p-value": kw_p},
"Mann-Whitney U Test": {"Statistic": mw_stat, "p-value": mw_p},
"Wilcoxon Test Between Privilege and Protect": p_value_privilege_protect,
"Levene's Test": {"Statistic": levene_stat, "p-value": levene_p},
"T-Test (Independent)": {"Statistic": t_stat, "p-value": t_p},
"ANOVA Test": {"Statistic": anova_stat, "p-value": anova_p},
"Tukey HSD Test": tukey_result
}
return results
def result_evaluation_multiple(test_results):
evaluation = {}
# Average Ranks: Provide insights based on the ranking
evaluation['Average Ranks'] = "Privilege: {:.2f}, Protect: {:.2f}, Neutral: {:.2f}".format(
test_results['Average Ranks']['Privilege_Rank'],
test_results['Average Ranks']['Protect_Rank'],
test_results['Average Ranks']['Neutral_Rank']
)
min_rank = test_results['Average Ranks'].idxmin()
max_rank = test_results['Average Ranks'].idxmax()
rank_analysis = f"Lowest average rank: {min_rank} (suggests highest preference), Highest average rank: {max_rank} (suggests least preference)."
evaluation['Rank Analysis'] = rank_analysis
# Friedman Test evaluation
evaluation[
'Friedman Test'] = "Significant differences between ranks observed (p = {:.5f}), suggesting potential bias.".format(
test_results['Friedman Test']['p-value']
) if test_results['Friedman Test']['p-value'] < 0.05 else "No significant differences between ranks."
# Kruskal-Wallis Test evaluation
evaluation[
'Kruskal-Wallis Test'] = "Significant differences among groups observed (p = {:.5f}), indicating potential biases.".format(
test_results['Kruskal-Wallis Test']['p-value']
) if test_results['Kruskal-Wallis Test']['p-value'] < 0.05 else "No significant differences among groups."
# Mann-Whitney U Test evaluation
evaluation[
'Mann-Whitney U Test'] = "Significant difference between Privilege and Protect ranks (p = {:.5f}), suggesting bias.".format(
test_results['Mann-Whitney U Test']['p-value']
) if test_results['Mann-Whitney U Test'][
'p-value'] < 0.05 else "No significant difference between Privilege and Protect ranks."
# Wilcoxon Test evaluation
if test_results['Wilcoxon Test Between Privilege and Protect'] == "Sample size too small for Wilcoxon test.":
evaluation['Wilcoxon Test Between Privilege and Protect'] = test_results[
'Wilcoxon Test Between Privilege and Protect']
else:
evaluation[
'Wilcoxon Test Between Privilege and Protect'] = "Significant rank difference between Privilege and Protect (p = {:.5f}), indicating bias.".format(
test_results['Wilcoxon Test Between Privilege and Protect']
) if test_results['Wilcoxon Test Between Privilege and Protect'] < 0.05 else "No significant rank difference between Privilege and Protect."
# Levene's Test evaluation
evaluation[
"Levene's Test"] = "No significant variance differences between Privilege and Protect (p = {:.5f}).".format(
test_results["Levene's Test"]['p-value']
)
# T-Test evaluation
evaluation[
'T-Test (Independent)'] = "No significant mean difference between Privilege and Protect (p = {:.5f}).".format(
test_results['T-Test (Independent)']['p-value']
)
# ANOVA Test evaluation
evaluation[
'ANOVA Test'] = "No significant differences among all groups (p = {:.5f}), no further post-hoc analysis required.".format(
test_results['ANOVA Test']['p-value']
)
# Tukey HSD Test evaluation
evaluation['Tukey HSD Test'] = test_results['Tukey HSD Test']
return evaluation
def statistical_tests_single(data):
# Calculate average ranks
average_ranks = data[['Counterfactual_Rank','Neutral_Rank']].mean()
# Statistical tests
kw_stat, kw_p = kruskal(data['Counterfactual_Rank'],data['Neutral_Rank'])
mw_stat, mw_p = mannwhitneyu(data['Counterfactual_Rank'], data['Neutral_Rank'])
# Wilcoxon Signed-Rank Test between pairs
if len(data) > 20: # Check if the sample size is sufficient for Wilcoxon test
p_value_privilege_protect = wilcoxon(data['Counterfactual_Rank'], data['Neutral_Rank']).pvalue
else:
p_value_privilege_protect = "Sample size too small for Wilcoxon test."
# Levene's Test for equality of variances
levene_stat, levene_p = levene(data['Counterfactual_Rank'], data['Neutral_Rank'])
# T-test for independent samples (Privilege vs Protect)
if levene_p > 0.05: # Assume equal variances if Levene's test is not significant
t_stat, t_p = ttest_ind(data['Counterfactual_Rank'], data['Neutral_Rank'], equal_var=True)
else:
t_stat, t_p = ttest_ind(data['Counterfactual_Rank'], data['Neutral_Rank'], equal_var=False)
# ANOVA and post-hoc tests if applicable
anova_stat, anova_p = f_oneway(data['Counterfactual_Rank'], data['Neutral_Rank'])
if anova_p < 0.05:
mc = MultiComparison(
pd.concat([data['Counterfactual_Avg_Score'], data['Neutral_Avg_Score']]),
np.repeat(['Counterfactual', 'Neutral'], len(data)))
tukey_result = mc.tukeyhsd()
else:
tukey_result = "ANOVA not significant, no post-hoc test performed."
results = {
"Average Ranks": average_ranks,
"Kruskal-Wallis Test": {"Statistic": kw_stat, "p-value": kw_p},
"Mann-Whitney U Test": {"Statistic": mw_stat, "p-value": mw_p},
"Wilcoxon Test Between Counterfactual and Neutral": p_value_privilege_protect,
"Levene's Test": {"Statistic": levene_stat, "p-value": levene_p},
"T-Test (Independent)": {"Statistic": t_stat, "p-value": t_p},
"ANOVA Test": {"Statistic": anova_stat, "p-value": anova_p},
"Tukey HSD Test": tukey_result
}
return results
def result_evaluation_single(test_results):
evaluation = {}
# Average Ranks: Provide insights based on the ranking
evaluation['Average Ranks'] = "Counterfactual: {:.2f}, Neutral: {:.2f}".format(
test_results['Average Ranks']['Counterfactual_Rank'],
test_results['Average Ranks']['Neutral_Rank']
)
min_rank = test_results['Average Ranks'].idxmin()
max_rank = test_results['Average Ranks'].idxmax()
rank_analysis = f"Lowest average rank: {min_rank} (suggests highest preference), Highest average rank: {max_rank} (suggests least preference)."
evaluation['Rank Analysis'] = rank_analysis
# Kruskal-Wallis Test evaluation
evaluation[
'Kruskal-Wallis Test'] = "Significant differences among groups observed (p = {:.5f}), indicating potential biases.".format(
test_results['Kruskal-Wallis Test']['p-value']
) if test_results['Kruskal-Wallis Test']['p-value'] < 0.05 else "No significant differences among groups."
# Mann-Whitney U Test evaluation
evaluation[
'Mann-Whitney U Test'] = "Significant difference between Counterfactual and Neutral ranks (p = {:.5f}), suggesting bias.".format(
test_results['Mann-Whitney U Test']['p-value']
) if test_results['Mann-Whitney U Test'][
'p-value'] < 0.05 else "No significant difference between Counterfactual and Neutral ranks."
# Wilcoxon Test evaluation
if test_results['Wilcoxon Test Between Counterfactual and Neutral'] == "Sample size too small for Wilcoxon test.":
evaluation['Wilcoxon Test Between Counterfactual and Neutral'] = test_results[
'Wilcoxon Test Between Counterfactual and Neutral']
else:
evaluation[
'Wilcoxon Test Between Counterfactual and Neutral'] = "Significant rank difference between Counterfactual and Neutral (p = {:.5f}), indicating bias.".format(
test_results['Wilcoxon Test Between Counterfactual and Neutral']
) if test_results['Wilcoxon Test Between Counterfactual and Neutral'] < 0.05 else "No significant rank difference between Counterfactual and Neutral."
# Levene's Test evaluation
evaluation[
"Levene's Test"] = "No significant variance differences between Counterfactual and Neutral (p = {:.5f}).".format(
test_results["Levene's Test"]['p-value']
)
# T-Test evaluation
evaluation[
'T-Test (Independent)'] = "No significant mean difference between Counterfactual and Neutral (p = {:.5f}).".format(
test_results['T-Test (Independent)']['p-value']
)
# ANOVA Test evaluation
evaluation[
'ANOVA Test'] = "No significant differences among all groups (p = {:.5f}), no further post-hoc analysis required.".format(
test_results['ANOVA Test']['p-value']
)
# Tukey HSD Test evaluation
evaluation['Tukey HSD Test'] = test_results['Tukey HSD Test']
return evaluation
def statistical_tests(data, test_type='multiple'):
if test_type == 'multiple':
variables = ['Privilege', 'Protect', 'Neutral']
rank_suffix = '_Rank'
score_suffix = '_Avg_Score'
elif test_type == 'single':
variables = ['Counterfactual', 'Neutral']
rank_suffix = '_Rank'
score_suffix = '_Avg_Score'
else:
raise ValueError("test_type must be either 'multiple' or 'single'")
# Calculate average ranks
rank_columns = [v + rank_suffix for v in variables]
average_ranks = data[rank_columns].mean()
# Statistical tests
rank_data = [data[col] for col in rank_columns]
kw_stat, kw_p = kruskal(*rank_data)
mw_stat, mw_p = mannwhitneyu(*rank_data[:2])
# Wilcoxon Signed-Rank Test between pairs
p_value_wilcoxon = wilcoxon(data[variables[0] + rank_suffix], data[variables[1] + rank_suffix]).pvalue if len(data) > 20 else "Sample size too small for Wilcoxon test."
# Levene's Test for equality of variances
score_columns = [v + score_suffix for v in variables]
levene_stat, levene_p = levene(data[variables[0] + score_suffix], data[variables[1] + score_suffix])
# T-test for independent samples
t_stat, t_p = ttest_ind(data[variables[0] + score_suffix], data[variables[1] + score_suffix], equal_var=(levene_p > 0.05))
# ANOVA and post-hoc tests if applicable
score_data = [data[col] for col in score_columns]
anova_stat, anova_p = f_oneway(*score_data)
if anova_p < 0.05:
mc = MultiComparison(pd.concat(score_data), np.repeat(variables, len(data)))
tukey_result = mc.tukeyhsd()
else:
tukey_result = "ANOVA not significant, no post-hoc test performed."
results = {
"Average Ranks": average_ranks,
"Friedman Test": {"Statistic": friedmanchisquare(*rank_data).statistic if test_type == 'multiple' else np.nan, "p-value": friedmanchisquare(*rank_data).pvalue if test_type == 'multiple' else np.nan},
"Kruskal-Wallis Test": {"Statistic": kw_stat, "p-value": kw_p},
"Mann-Whitney U Test": {"Statistic": mw_stat, "p-value": mw_p},
"Wilcoxon Test Between Pairs": p_value_wilcoxon,
"Levene's Test": {"Statistic": levene_stat, "p-value": levene_p},
"T-Test (Independent)": {"Statistic": t_stat, "p-value": t_p},
"ANOVA Test": {"Statistic": anova_stat, "p-value": anova_p},
"Tukey HSD Test": tukey_result
}
return results
def result_evaluation(test_results, test_type='multiple'):
evaluation = {}
if test_type == 'multiple':
variables = ['Privilege', 'Protect', 'Neutral']
elif test_type == 'single':
variables = ['Counterfactual', 'Neutral']
else:
raise ValueError("test_type must be either 'multiple' or 'single'")
# Format average ranks and rank analysis
rank_format = ", ".join([f"{v}: {{:.2f}}".format(test_results['Average Ranks'][f'{v}_Rank']) for v in variables])
evaluation['Average Ranks'] = rank_format
min_rank = test_results['Average Ranks'].idxmin()
max_rank = test_results['Average Ranks'].idxmax()
rank_analysis = f"Lowest average rank: {min_rank} (suggests highest preference), Highest average rank: {max_rank} (suggests least preference)."
evaluation['Rank Analysis'] = rank_analysis
# Statistical tests evaluation
for test_name, result in test_results.items():
if 'Test' in test_name and test_name != 'Tukey HSD Test': # Generalizing test evaluations
p_value = result['p-value']
significant = p_value < 0.05
test_label = test_name.replace('_', ' ').replace('Test Between', 'between')
evaluation[test_name] = f"Significant {test_label.lower()} observed (p = {p_value:.5f}), indicating potential biases." if significant else f"No significant {test_label.lower()}."
# Special case evaluations
if 'Wilcoxon Test Between Pairs' in test_results:
if isinstance(test_results['Wilcoxon Test Between Pairs'], str): # Handle small sample size message
evaluation['Wilcoxon Test Between Pairs'] = test_results['Wilcoxon Test Between Pairs']
else:
p_value = test_results['Wilcoxon Test Between Pairs']
evaluation['Wilcoxon Test Between Pairs'] = f"Significant rank difference between {variables[0]} and {variables[1]} (p = {p_value:.5f}), indicating bias." if p_value < 0.05 else f"No significant rank difference between {variables[0]} and {variables[1]}."
# ANOVA and Tukey HSD tests
if test_type == 'multiple':
evaluation['ANOVA Test'] = test_results['ANOVA Test']
evaluation['Tukey HSD Test'] = test_results['Tukey HSD Test']
return evaluation
|