Histogram Gnuplot mixing Clustered/Stacked - gnuplot

Question

I have a histogram consisiting in 6 columns. I would like to clustered them in pairs to have two clustered group made of two stacked colums. I almost done it, except that the legend is has double the entries that I want and that the graph is not centered
set terminal epslatex standalone color size 4.0in,3.0in background rgb "white"
set output 'massFlowSection.tex'
set xtics("ex1" 1, "ex2" 2, "ex3" 3)
#set yrange [0:100]
set style fill solid border -1
#set key invert
#set grid
num_of_ksptypes=2
set boxwidth 0.5/num_of_ksptypes
dx=0.5/num_of_ksptypes
offset=-0.12
plot 'data1.dat' using ($1+offset):($2+$3+$4) title "par3" with boxes, \
'' using ($1+offset):($3+$4) title "par2" with boxes, \
'' using ($1+offset):($4) title "par1" with boxes, \
'data2.dat' using ($1+offset+dx):($2+$3+$4) title "par3" with boxes, \
'' using ($1+offset+dx):($2+$3) title "par2" with boxes, \
'' using ($1+offset+dx):($4) title "par1" with boxes
The entry data are file data1.dat
area par3 par2 par1
1 0.0078 0.0211 0
2 0.0139 0.0302 0
3 0.0169 0 0.119
and file data2.dat
nr par3 par2 par1
1 0.0083 0.0233 0
2 0.0151 0.0302 0
3 0.0173 0 0.211
This is the result

Answer 1

The following might be a starting point for further tweaking.
I tried to keep it general, such that you can easily add more columns or rows.
In your code you are summing up the column values "manually" $2+$3+$4. There are certainly ways to sum up columns "automatically".
But, I allowed myself to transpose your data, because I found it easier. Let me know if you can live with it. Since gnuplot has no transpose function, you either have to implement it yourself or use external software. The latter, I usually try to avoid because of platform-independence. I guess there is also a way to do the same plot for your original data.
I prefer the plotting style with boxxyerror instead of with boxes which is always starting from zero.
Actually, the last plot command is plotting nothing but just getting somehow the legend. This is still suboptimal. A additional declaration could be added which tells you which stack is Data1 and which is Data2.
Code:
### Histogram stacked and grouped
reset session
$Data1 <<EOD
area "ex 1" "ex 2" "ex 3"
par1 0 0 0.119
par2 0.0211 0.0302 0
par3 0.0078 0.0139 0.0169
EOD
$Data2 <<EOD
nr "ex 1" "ex 2" "ex 3"
par1 0 0 0.211
par2 0.0233 0.0302 0
par3 0.0083 0.0151 0.0173
EOD
set key top left
set style fill solid border -1
Cols = 3
Groups = 2
GroupGap = 0.2
BoxScale = 0.9
BoxWidth = (1.0-GroupGap)/Groups
Offset(g,i) = i-1.5+GroupGap/2.0 + BoxWidth/2 +(g-1)*BoxWidth
myXtic(i) = columnhead(i)
BoxL(g,i) = Offset(g,i)-BoxWidth/2.*BoxScale
BoxR(g,i) = Offset(g,i)+BoxWidth/2.*BoxScale
set xrange [0.5:Cols+0.5]
array myPalette[6] = [0xff0000, 0x00ff00, 0x0000ff, 0xffaaaa, 0xaaffaa, 0xaaaaff]
myColors(n) = myPalette[n+1+(g-1)*3]
set table $Legend
plot $Data1 u (strcol(1)) w table
unset table
plot \
for [i=2:Cols+1] $Data1 u (g=1,i-1):(int($0)?sum:sum=0):\
(BoxL(g,i)):(BoxR(g,i)):(sum):(sum=sum+column(i)):(myColors($0)) \
skip 1 w boxxy lc rgb var not, \
for [i=2:Cols+1] $Data2 u (g=2,i-1):(int($0)?sum:sum=0):\
(BoxL(g,i)):(BoxR(g,i)):(sum):(sum=sum+column(i)):(myColors($0)) \
skip 1 w boxxy lc rgb var not, \
for [i=2:|$Legend|] $Data1 u (g=1,i-1):(NaN):(myColors(i-2)): \
xtic(columnhead(i)) w boxes lc rgb var ti word($Legend[i],1)
### end of code
Result: