osl_dynamics.analysis.statistics#

Functions to perform statistical significance testing.

Functions#

`evoked_response_max_stat_perm`(data, n_perm[, ...])	Statistical significance testing for evoked responses.
`group_diff_max_stat_perm`(data, assignments, n_perm[, ...])	Statistical significance testing for difference between two groups.
`paired_diff_max_stat_perm`(data, n_perm[, metric, n_jobs])	Statistical significance testing for paired difference.

Module Contents#

osl_dynamics.analysis.statistics.evoked_response_max_stat_perm(data, n_perm, covariates=None, metric='copes', n_jobs=1)[source]#

Statistical significance testing for evoked responses.

This function fits a General Linear Model (GLM) with ordinary least squares and performs a sign-flip permutation to build a null distribution using maximum statistic across the target dimensions.

Parameters:

data (np.ndarray) – Data array for baseline corrected evoked responses. Shape is (n_samples, *target_dims).
n_perm (int) – Number of permutations.
covariates (dict, optional) –
Dictionary of continuous covariates to include as confound regressors. Each key is a covariate name and each value is a 1D array of length n_samples. These are z-scored internally. Example:
```
covariates = {
    "age": np.array([25, 30, 28, ...]),
    "head_size": np.array([55.1, 57.3, 54.8, ...]),
}
```
metric (str, optional) – Statistic to compute p-values. Options are 'copes' or 'tstats'.
n_jobs (int, optional) – Number of jobs to run in parallel.

Returns:

pvalues – P-values. Shape is (*target_dims,).

Return type:

np.ndarray

osl_dynamics.analysis.statistics.group_diff_max_stat_perm(data, assignments, n_perm, covariates=None, metric='tstats', n_jobs=1)[source]#

Statistical significance testing for difference between two groups.

This function fits a General Linear Model (GLM) with ordinary least squares and performs a row-shuffle permutation to build a null distribution using maximum statistic across the target dimensions (target_dims) for the difference between two groups.

Parameters:

data (np.ndarray) – Data array for baseline corrected evoked responses. Shape is (n_samples, *target_dims).
assignments (np.ndarray) – Group assignments. Shape is (n_samples,). Must have exactly two unique values.
n_perm (int) – Number of permutations.
covariates (dict, optional) –
Dictionary of continuous covariates to include as confound regressors. Each key is a covariate name and each value is a 1D array of length n_samples. These are z-scored internally. Example:
```
covariates = {
    "age": np.array([25, 30, 28, ...]),
    "head_size": np.array([55.1, 57.3, 54.8, ...]),
}
```
metric (str, optional) – Statistic to compute p-values with. Options are 'copes' or 'tstats'.
n_jobs (int, optional) – Number of jobs to run in parallel.

Returns:

group_diff (np.ndarray) – Difference between two groups. Shape is (*target_dims,).
pvalues (np.ndarray) – P-values. Shape is (*target_dims,).

Return type:

Tuple[numpy.ndarray, numpy.ndarray]

osl_dynamics.analysis.statistics.paired_diff_max_stat_perm(data, n_perm, metric='copes', n_jobs=1)[source]#

Statistical significance testing for paired difference.

This function fits a General Linear Model (GLM) with ordinary least squares and performs sign-flip permutations to build a null distribution with the maximum statistic across the target dimensions for the paired differences.

Parameters:

data (np.ndarray) – Data array for baseline corrected evoked responses. Shape is (n_samples, *target_dims).
n_perm (int) – Number of permutations.
metric (str, optional) – Statistic to compute p-values with. Options are 'copes' or 'tstats'.
n_jobs (int, optional) – Number of jobs to run in parallel.

Returns:

paired_diff (np.ndarray) – Paired differences. Shape is (*target_dims,).
pvalues (np.ndarray) – P-values. Shape is (*target_dims,).

Return type:

Tuple[numpy.ndarray, numpy.ndarray]