I found two solutions myself:
1) Normalize the database. I get two tables:
table first uid | information1 | metainformation2
table second uid | col | result_of_col
Then I can easily use the standard STDDEV function.
2) Use R. Data is a de-standardized format, as it should be used in statistical analysis. Thus, it is easy to get into R and use the following code.
sd (t (data set [1: 4,3: 8]))
Note that I just take the numerical part of this data.frame, leaving the selection of columns 3-8. And don't get too much data (so this time I use only the first couple of rows). t () transfers the data that is needed because sd () only works with columns.
There is a rowSds function in the vsn package that should work similarly to rowMean and rowSum, but for some reason this might be deprecated. At least these packages were not available in the Swiss CRAN mirror;).
Hth someone else.
source share