def _powered_effect_from_summary_df()

in spotify_confidence/analysis/frequentist/confidence_computers/sample_size_computer.py [0:0]

• 32 lines of code
• 5 McCabe index (conditional complexity)

def _powered_effect_from_summary_df(df: DataFrame, **kwargs: Dict) -> DataFrame:
    if (df[ADJUSTED_POWER].isna()).any():
        df[REQUIRED_SAMPLE_SIZE_METRIC] = None
    else:
        all_weights = kwargs[TREATMENT_WEIGHTS]
        control_weight, treatment_weights = all_weights[0], all_weights[1:]

        current_number_of_units = kwargs[FINAL_EXPECTED_SAMPLE_SIZE]

        binary = df[kwargs[IS_BINARY]].values[0]
        z_alpha = st.norm.ppf(
            1
            - df[ADJUSTED_ALPHA_POWER_SAMPLE_SIZE].values[0] / (2 if df[PREFERENCE_TEST].values[0] == TWO_SIDED else 1)
        )
        z_power = st.norm.ppf(df[ADJUSTED_POWER].values[0])
        non_inferiority = is_non_inferiority(df[NIM].values[0])

        max_powered_effect = 0
        for treatment_weight in treatment_weights:
            kappa = control_weight / treatment_weight
            proportion_of_total = (control_weight + treatment_weight) / sum(all_weights)

            this_powered_effect = df[POWERED_EFFECT] = confidence_computers[ZTEST].powered_effect(
                df=df.assign(kappa=kappa)
                .assign(current_number_of_units=current_number_of_units)
                .assign(proportion_of_total=proportion_of_total),
                z_alpha=z_alpha,
                z_power=z_power,
                binary=binary,
                non_inferiority=non_inferiority,
                avg_column=POINT_ESTIMATE,
                var_column=VARIANCE,
            )

            max_powered_effect = max(this_powered_effect.max(), max_powered_effect)

        df[POWERED_EFFECT] = None if max_powered_effect == 0 else max_powered_effect

    return df