Test if two binomial distributions are statistically different from each other

binomial_hypothesis_tests.py

import statsmodels.api as sm
import numpy as np
import rpy2.robjects.packages as rpackages
import rpy2.robjects as robjects
rstats = rpackages.importr('stats')

s1 = 1556
n1 = 2455

s2 = 1671
n2 = 2730

# manual calculation
p1 = s1 / n1
p2 = s2 / n2
p = (n1*p1 + n2*p2) / (n1 + n2)

z = (p1 - p2) / (p*(1-p)*((1/n1)+(1/n2)))**0.5

# using R in Python with rpy2
rmatrix = robjects.r.matrix(robjects.IntVector([1556, 2455-1556, 1671,2730-1671]), nrow=2)
fisher_test = rstats.fisher_test(rmatrix, alternative="two.sided")

zscore, pval = sm.stats.proportions_ztest([s1, s2], [n1, n2], alternative='two-sided')

print('Manual calculation of z: {:.6f}'.format(z))
print('Z-score from statsmodels: {:.6f}'.format(zscore))
print('R pvalue from fisher.test: {:.6f}'.format(fisher_test[0][0]))
print('Statsmodels pvalue: {:.6f}'.format(pval))

#Manual calculation of z: 1.610825
#Z-score from statsmodels: 1.610825
#R pvalue from fisher.test: 0.108268
#Statsmodels pvalue: 0.107218