See the "trick" to locate the midpoint of stacked bars from the hidden property at How-do-i-label-each-bar-in-bar-group-with-a-string-on-top
I don't know if it's yet been made visible or not; add to the complaints to TMW about how poorly-designed the BAR function user interface is in general; I've railed for 20+ years.
ADDENDUM
Stab at adapting to your variables above...
H=bar(LTSIR);
hold on
X=[]; % placeholder for X position from bar
for i=1:length(H) % iterate over number of bar objects
X=[X H(i).XData+hB(i).XOffset]; % and get the center bar group positions
end
hE=errorbar(X,[LTSIR LTSIRe],'.')
You'll have to check on orientation/size of X vector vis a vis the error data arrays but the idea is to retrieve the X locations of the midpoints of the bar and then plot the error bars versus it instead of the nominal bar position.
Again, this kind of thing should be available as properties/methods of the bar object already.
ADDENDUM 2
OK, I had a few minutes...
AB=randi([110 160],4,2); % make some sample data
hB=bar(AB); % make default bar plot
ylim([0 200]) % scale same for rough appearance similar
e=std(AB); % get a sample error value
X=[]; % collect the X values with offset for each
for i=1:length(hB)
X=[X;hB(i).XData+hB(i).XOffset];
end
>> X=X.' % rearrange X by column and show what this is...
X =
0.8571 1.1429
1.8571 2.1429
2.8571 3.1429
3.8571 4.1429
>> hold on % so can add to bar plot
>> hEB=errorbar(X,AB,repmat(e,4,1),'.'); % add error bars
>>
The above gives the attached figure...
You'll have to clean up to match your colors, etc., etc., etc., but shows how to get the correct X positions to plot on each bar in the multiple-bar style.
ADDENDUM 2
Reflection made me realize don't need the loop to retrieve the X coordinates--
X=cell2mat(get(hB,'XData')).' + [hB.XOffset];
will do the same thing using the fact that get will retrieve the same property across multiple handles and return as cell array and knowing that the .XOffset property is a constant for each bar series then recent releases will do the addition using implicit singleton expansion. At the time I wrote the Answer linked to, the latter feature was not yet implemented in the release available to me at the time.
