Configure t-test sigma value through the CONFIG

docs/source/refs.bib

@@ -128,4 +128,15 @@ @article{ding2006granger
   author={Ding, Mingzhou and Chen, Yonghong and Bressler, SL},
   journal={Handbook of Time Series Analysis [Internet]. Wiley, Wienheim},
   year={2006}
+}
+
+@article{ridgway2012problem,
+  title={The problem of low variance voxels in statistical parametric mapping; a new hat avoids a ‘haircut’},
+  author={Ridgway, Gerard R and Litvak, Vladimir and Flandin, Guillaume and Friston, Karl J and Penny, Will D},
+  journal={Neuroimage},
+  volume={59},
+  number={3},
+  pages={2131--2141},
+  year={2012},
+  publisher={Elsevier}
 }

frites/config.py

@@ -57,3 +57,10 @@
     'R_PFrd', 'R_PFrm', 'R_OFCvl', 'R_OFCv', 'R_OFCvm', 'R_PFCvm', 'R_ACC',
     'R_Insula', 'L_Thal', 'L_Cd', 'L_Put', 'L_GP', 'L_Hipp', 'L_Amyg', 'L_NAc',
     'R_Thal', 'R_Cd', 'R_Put', 'R_GP', 'R_Hipp', 'R_Amyg', 'R_NAc']
+
+
+"""
+Default sigma for the hat correction when performing the t-test using
+MNE-Python
+"""
+CONFIG['TTEST_MNE_SIGMA'] = 0.001

frites/stats/stats_param.py

@@ -12,7 +12,7 @@
                 trimmed=lambda x, axis=0: trim_mean(x, .2, axis=axis))
 
 
-def ttest_1samp(x, pop_mean, axis=0, method='mne'):
+def ttest_1samp(x, pop_mean, axis=0, method='mne', sigma=0.001):
     """One-sample t-test.
 
     Parameters
@@ -25,11 +25,17 @@ def ttest_1samp(x, pop_mean, axis=0, method='mne'):
         Axis along which to perform the t-test
     method : {'mne', 'scipy'}
         Use either the scipy or the mne t-test
+    sigma : float | 0.001
+        Hat adjustment method, a value of 1e-3 may be a reasonable choice
 
     Returns
     -------
     tvalues : array_like
         Array of t-values
+
+    References
+    ----------
+    Ridgway et al., 2012 :cite:`ridgway2012problem`
     """
     if method == 'scipy':
         from scipy.stats import ttest_1samp as sp_ttest
@@ -38,7 +44,7 @@ def fcn(x, pop_mean, axis):  # noqa
     elif method == 'mne':
         from mne.stats import ttest_1samp_no_p as mne_ttest
         def fcn(x, pop_mean, axis):  # noqa
-            return mne_ttest(np.moveaxis(x, axis, 0) - pop_mean, sigma=1e-3)
+            return mne_ttest(np.moveaxis(x, axis, 0) - pop_mean, sigma=sigma)
 
     return fcn(x, pop_mean, axis)
 
@@ -97,13 +103,16 @@ def rfx_ttest(mi, mi_p, center=False, zscore=False, ttested=False):
     # get the mean of surrogates
     _merge_perm = np.r_[tuple([mi_p[k].ravel() for k in range(n_roi)])]
     pop_mean_surr = np.mean(_merge_perm)
-    logger.info(f"    T-test across subjects (pop_mean={pop_mean_surr}; "
-                f"center={center}; zscore={zscore})")
     # perform the one sample t-test against the mean both on the true and
     # permuted mi
-    t_obs = np.stack([ttest_1samp(mi[k], pop_mean_surr) for k in range(n_roi)])
-    t_obs_surr = np.stack([ttest_1samp(mi_p[k], pop_mean_surr,
-                                       axis=1) for k in range(n_roi)])
+    from frites.config import CONFIG
+    kw = dict(method='mne', sigma=CONFIG['TTEST_MNE_SIGMA'])
+    logger.info(f"    T-test across subjects (pop_mean={pop_mean_surr}; "
+                f"center={center}; zscore={zscore}; sigma={kw['sigma']})")
+    t_obs = np.stack([ttest_1samp(
+        mi[k], pop_mean_surr, axis=0, **kw) for k in range(n_roi)])
+    t_obs_surr = np.stack([ttest_1samp(
+        mi_p[k], pop_mean_surr, axis=1, **kw) for k in range(n_roi)])
     t_obs_surr = np.swapaxes(t_obs_surr, 0, 1)
 
     return t_obs, t_obs_surr, pop_mean_surr

frites/workflow/wf_stats_ephy.py

@@ -128,7 +128,9 @@ def fit(self, effect, perms, inference='rfx', mcp='cluster', tail=1,
                 assert rfx_suj, "For RFX, `n_subjects` should be > 1"
                 # modelise how subjects are distributed
                 es, es_p, pop_mean = rfx_ttest(effect, perms)
-                self.update_cfg(ttest_pop_mean=pop_mean)
+                from frites.config import CONFIG
+                sigma = CONFIG['TTEST_MNE_SIGMA']
+                self.update_cfg(ttest_pop_mean=pop_mean, ttest_sigma=sigma)
             tvalues = es
 
         # ---------------------------------------------------------------------