# Holm-Bonferroni (Holm) correction chosen for multiple hypothesis testing vs. the default Holm-sidak (hs) in statsmodels
# as hs assumes the individual tests performed are independent, which may not apply in this situation. Holm does not make
# this assumption but offers less statistical power.

def df_p_value_calc(df):
    """Using p-values provided, determines the alpha level required for a multi-hypothesis test and determines whether
    each p-value meets the necessary threshold and performs the A/B test. ALL test results must by True for statistical
    significance to be established.

    This function uses the Holm-Bonferroni method to address the multiplicity problem.

    Parameters: df is the dataframe containing the publications, article counts, and rates for each category.

    Output: a list of tuples, where the first element of the tuple is the category name and the second is a tuple
    containing: an array of Booleans (True if the alt. hypothesis is accepted, False if not), an array of floats
    (p-values corrected for multiple tests), the corrected alpha under Sidak, and the corrected alpha under Bonferroni.
    """

    result_list = []

    for c in df['category'].unique():
        mask = df['category'] == c
        cat_df = df[mask]
        p_values = cat_p_value_calc(cat_df)
        result_list.append((c, multipletests(p_values, method = 'holm')))

    return result_list