fix : perform PCA only on the original data and apply same transform to

original and defaced data instead of running a PCA separately for
defaced and original

statistical_analysis/IQMs/LoadIQMs.py

@@ -1,4 +1,4 @@
-"""Load IQMs from MRIQC output into a dataframe"""
+# # Load IQMs from MRIQC output into a dataframe
 
 import os
 import json
@@ -94,6 +94,6 @@ def get_key(my_dict,val):
 
 iqms_df = pd.DataFrame(np.hstack((i_merge,site)), columns = ['subject'] + iqms_keys + ['defaced'] + ['site'])
 
-# Repeated measure MANOVA is only implemented in R
+# Repeated-measures MANOVA is only implemented in R
 # Thus we save the dataframe so we can load it in R
 iqms_df.to_csv('iqms_df.csv')

statistical_analysis/ManualRatings/ANOVA.ipynb

@@ -610,7 +610,7 @@
    "id": "354ce05a",
    "metadata": {},
    "source": [
-    "For the assumption of sphericity to be verified, the test should not reject the null hypothesis, which is that the variance between group is equal. The assumption is violated by the interaction defaced:rater, thus anova_test can apply automatically sphericity correction"
+    "For the assumption of sphericity to be verified, the test should not reject the null hypothesis, which is that the variance between group is equal. If any assumption is violated, we will switch to the alternative analysis using linear mixed-effects models."
    ]
   },
   {

statistical_analysis/ManualRatings/README.md

@@ -2,15 +2,17 @@
 ## Statistical analysis associated to the defacing pre-registration.
 
 Manual ratings
-- `simulate_data.R` is a script to simulate raters manually assigning quality grades to subjects in original and defaced conditions. The simulated data have a positive bias towards defaced ratings.
-
-
-- `VisualizeRatings.ipnyb` is a notebook to visualize the distributions of ratings in the defaced vs original conditions. 
 
 
 - `ANOVA.ipynb` is the notebook to run repeated-measures ANOVA and check its assumptions.
 - In `VaryDatasetParameters_ANOVA.ipynb` I performed a sensitivity analysis to verify how many ratings do we need for the model to detect a significant effect of defacing.
 
 
 - `LinearMixedEffectModels.ipynb` is the alternative notebook we will use in case one of the assumption of the repeated-measures ANOVA is violated.
+
+
+- `VisualizeRatings.ipnyb` is a notebook to visualize the distributions of ratings in the defaced vs original conditions.
+
+
+- `simulate_data.R` is a script to simulate raters manually assigning quality grades to subjects in original and defaced conditions. The simulated data have a positive bias towards defaced ratings.
 <!-- #endregion -->

statistical_analysis/ManualRatings/VisualizeRatings.ipynb

@@ -32,8 +32,8 @@
    "outputs": [],
    "source": [
     "## Simulate data with bias\n",
-    "n_rated <- 580\n",
-    "n_sub <- 580 #nbr of subjects available in the dataset\n",
+    "n_rated <- 185\n",
+    "n_sub <- 185 #nbr of subjects available in the dataset\n",
     "n_rater <- 4 #nbr of raters\n",
     "perc_biased <- c(2,10,60,90) #4 raters + bias\n",
     "file_name=\"SimulatedData/SimulatedDefacedRatings_4raters_bias.Rda\"\n",
@@ -298,8 +298,8 @@
    "outputs": [],
    "source": [
     "## Simulate continuous data with bias\n",
-    "n_rated <- 580\n",
-    "n_sub <- 580 #nbr of subjects available in the dataset\n",
+    "n_rated <- 185\n",
+    "n_sub <- 185 #nbr of subjects available in the dataset\n",
     "n_rater <- 4 #nbr of raters\n",
     "perc_biased <- c(2,10,60,90) #4 raters + bias\n",
     "ratings_range <- seq(0,1,length.out=11)\n",