Translates Gelman's retrodesign function from R to Python

retroDesign.py

from scipy.stats import t


def retroDesign(A, s, alpha=0.05, df=1000000, n_sims=10000):
 """

 :param A: the hypothesized true effect size
 :param s: standard error
 :param alpha: confidence level
 :param df: degree of freedom
 :param n_sims: number of simulations
 :return: power, typeS, exaggeration
 """
 z = t.ppf(1-alpha/2, df)
 p_high = 1-t.cdf(z-A/s, df)
 p_low = t.cdf(-z-A/s, df)
 power = p_high + p_low
 typeS = p_low/power
 estimate = A + s*t.rvs(df, size=n_sims)
 significant = [e for e in estimate if abs(e)>s*z]
 exaggeration = sum([abs(x)/A for x in significant])/len(significant)
 return power, typeS, exaggeration


if __name__ == '__main__':
 # Example: true effect size of 0.1, standard error 3.28, alpha=0.05
 print(retroDesign(.1, 3.28))

 # Example: true effect size of 2, standard error 8.1, alpha=0.05
 print(retroDesign(2, 8.1))