I've been performing a LME longitudinal analysis on a dataset with two groups (controls, clinicals), where each participant completed two MRI scans (pre, post).

I was examining a particular surface cluster that arose from a time*group interaction effect. I used the --olab option from the mri_surfcluster function to determine the vertex IDs from the particular cluster, and used the following code to determine the cluster means of (control_pre, clinical_pre, control_post, clinical_post) from the cluster:

[Y,mri] = fs_read_Y(lh.thickness_sm15.mgh'); % get surface data

Qdec = fReadQdec(sprintf('%s/long.qdec.table.dat', qdec_table_path));
Qdec = rmQdecCol(Qdec,1); % (removes first column)
sID = Qdec(2:end,1); % (grabs the subjects' IDs)
Qdec = rmQdecCol(Qdec,1); % (removes the subjects' Is ID column)
M = Qdec2num(Qdec); % (converts to a numeric matrix)
[M,Y,ni] = sortData(M,1,Y,sID); % (sorts the data)

time = M(:,1);
group = M(:,4);

label_file = 'lh_cluster_vertices-0001.label' % from mri_surfcluster output

dataStruct = importdata(sprintf('%s/%s', label_file.folder, label_file.name));
vertices = dataStruct.data(2:5:end); % select the rows containing the surface vertex IDs

cluster_data = Y(:, vertices);

control_pre = cluster_data(group == 0 & time == 0, :);
clinical_pre = cluster_data(group == 1 & time == 0, :);
control_post = cluster_data(group == 0 & time > 0, :);
clinical_post = cluster_data(group == 1 & time > 0, :);

From this code I get the following averages for each designation: 2.4397, 2.4340, 2.4069, and 2.4423.

When, however, I plot the data with lme_lowessPlot(time ,cluster_data, 0.70, group), the values are dissimilar from what I calculate. The LME documentation warns that "lowess plots can also be misleading, as it plots averages that could be affected by dropout etc", but I'm unsure what that means here, as the data has already been processed.

I'm curious what the reason is for this inconsistency.

Thank you.

Dan