Skip to content

Power Analysis

Statistical power for z- and t-tests computed by hand with SciPy, cross-checked against statsmodels, plus Cohen's d helpers.

compute_power_z(effect_size, alpha=0.05, n=None, two_tailed=True)

Compute statistical power for a one-sample Z-test.

Parameters:

Name Type Description Default
effect_size float

Standardized effect size (Cohen's d).

required
alpha float

Significance level (Type I error rate).

0.05
n int

Sample size.

None
two_tailed bool

If True, uses two-tailed critical value; otherwise one-tailed.

True

Returns:

Type Description
float

Statistical power in [0, 1]. Power > 0.8 is typically desirable.

Notes

Uses the normal distribution (appropriate when population variance is known or n is large enough for CLT to apply).

Source code in utils/power_utils.py
def compute_power_z(
    effect_size: float,
    alpha: float = 0.05,
    n: int | None = None,
    two_tailed: bool = True,
) -> float:
    """
    Compute statistical power for a one-sample Z-test.

    Parameters
    ----------
    effect_size : float
        Standardized effect size (Cohen's d).
    alpha : float, default=0.05
        Significance level (Type I error rate).
    n : int
        Sample size.
    two_tailed : bool, default=True
        If True, uses two-tailed critical value; otherwise one-tailed.

    Returns
    -------
    float
        Statistical power in [0, 1]. Power > 0.8 is typically desirable.

    Notes
    -----
    Uses the normal distribution (appropriate when population variance is known
    or n is large enough for CLT to apply).
    """
    if n is None:
        raise ValueError("n (sample size) is required")
    z_alpha = norm.ppf(1 - alpha / 2 if two_tailed else 1 - alpha)
    z_power = z_alpha - (effect_size * np.sqrt(n))
    power = 1 - norm.cdf(z_power)
    return power

compute_power_t(effect_size, alpha=0.05, n=None, two_tailed=True)

Compute statistical power for a one-sample t-test.

Parameters:

Name Type Description Default
effect_size float

Standardized effect size (Cohen's d).

required
alpha float

Significance level.

0.05
n int

Sample size (must be >= 2).

None
two_tailed bool

If True, uses two-tailed critical value.

True

Returns:

Type Description
float

Statistical power in [0, 1].

Notes

Uses the t-distribution with df = n - 1. Preferred over Z-test power when population variance is unknown.

Source code in utils/power_utils.py
def compute_power_t(
    effect_size: float,
    alpha: float = 0.05,
    n: int | None = None,
    two_tailed: bool = True,
) -> float:
    """
    Compute statistical power for a one-sample t-test.

    Parameters
    ----------
    effect_size : float
        Standardized effect size (Cohen's d).
    alpha : float, default=0.05
        Significance level.
    n : int
        Sample size (must be >= 2).
    two_tailed : bool, default=True
        If True, uses two-tailed critical value.

    Returns
    -------
    float
        Statistical power in [0, 1].

    Notes
    -----
    Uses the t-distribution with df = n - 1. Preferred over Z-test power
    when population variance is unknown.
    """
    if n is None:
        raise ValueError("n (sample size) is required")
    df = n - 1
    t_alpha = t.ppf(1 - alpha / 2 if two_tailed else 1 - alpha, df)
    t_power = t_alpha - (effect_size * np.sqrt(n))
    power = 1 - t.cdf(t_power, df)
    return power

statsmodels_power(effect_size, alpha=0.05, n=None, alternative='two-sided')

Compute t-test power using the statsmodels TTestPower API.

Parameters:

Name Type Description Default
effect_size float

Standardized effect size (Cohen's d).

required
alpha float

Significance level.

0.05
n int

Number of observations.

None
alternative str
'two-sided'

Returns:

Type Description
float

Statistical power in [0, 1].

Source code in utils/power_utils.py
def statsmodels_power(
    effect_size: float,
    alpha: float = 0.05,
    n: int | None = None,
    alternative: str = "two-sided",
) -> float:
    """
    Compute t-test power using the statsmodels TTestPower API.

    Parameters
    ----------
    effect_size : float
        Standardized effect size (Cohen's d).
    alpha : float, default=0.05
        Significance level.
    n : int
        Number of observations.
    alternative : str, default='two-sided'
        Direction: 'two-sided', 'larger', or 'smaller'.

    Returns
    -------
    float
        Statistical power in [0, 1].
    """
    if n is None:
        raise ValueError("n (sample size) is required")
    power_obj = TTestPower()
    power = power_obj.power(effect_size=effect_size, nobs=n, alpha=alpha, alternative=alternative)
    return power

compute_cohens_d(mean1, mean2, std_dev)

Compute Cohen's d effect size from two group means.

Parameters:

Name Type Description Default
mean1 float

Mean of the first group.

required
mean2 float

Mean of the second group.

required
std_dev float

Pooled or common standard deviation.

required

Returns:

Type Description
float

Cohen's d = (mean1 - mean2) / std_dev. Interpretation: 0.2 small, 0.5 medium, 0.8 large.

Source code in utils/power_utils.py
def compute_cohens_d(mean1: float, mean2: float, std_dev: float) -> float:
    """
    Compute Cohen's d effect size from two group means.

    Parameters
    ----------
    mean1 : float
        Mean of the first group.
    mean2 : float
        Mean of the second group.
    std_dev : float
        Pooled or common standard deviation.

    Returns
    -------
    float
        Cohen's d = (mean1 - mean2) / std_dev.
        Interpretation: 0.2 small, 0.5 medium, 0.8 large.
    """
    return (mean1 - mean2) / std_dev