One way to simulate a two-dimensional array is to keep the rows as strings in a one-dimensional array and unpack them at each iteration. You will have to choose a suitable delimiter that doesn't appear in the data. Since you mention CSV, I'll use a comma, but this won't be smart enough to handle data like this with embedded commas:
name, start date, visits, games, balance
"Williamson, Dennis", "January 11, 2007", 12, 42, 17000
Here's a simple example of iterating over the values in a simulated two-dimensional array:
# avg, stddev, max, min
data_array=(
            "17,18,19,1"
            "12,14,16,2"
            "6,8,10,3"
            )
saveIFS=$IFS
for row in ${data_array[@]}
do
    IFS=","
    cols=($row)
    IFS=$saveIFS
    for col in ${cols[@]}
    do
        newval=$(do_something $col)
    done
done
Making changes to the contents of the array is possible:
rowidx=2
colidx=2
IFS=","
cols=(${data_array[rowidx]})
cols[colidx]=$some_value
data_array[rowidx]="${cols[*]}"
IFS=$saveIFS
As you can see, it gets complicated fast and there are lots of gotchas which I haven't mentioned. Use Python.