[EpiData-list] RE: Sorting of 2x2 tables (Oswego example discussion)

epidata-list at lists.umanitoba.ca epidata-list at lists.umanitoba.ca
Fri Nov 9 02:54:28 CST 2007

The problem of the /sd option in relation to
    tab ill cakes vanilla chocolate /oa /sd /ex
   tab ill cakes vanilla chocolate /oa /ex

is related to how boolean variables are sorted. Currently in the version
2.0 released there is a bug in sorting of boolean 2x2 variables.
The error will be fixed as soon as possible.

For numerical variables coded 1/0 or 2/1 there is no problem. The reason I
did not catch the boolean Y/N error is, that the tab command gives correct
estimates for boolean as well as numerical variables:
   tab ill cakes vanilla chocolate /oa /ex

But if the user adds the sorting (as was needed earlier) :
   tab ill cakes vanilla chocolate /oa /ex /sd
reveals the bug.

The way development is done is that before release a number of tests are
made, currently around 1250, e.g. after the mentioned tabulation 24 tests
were applied for the tabulation of oswego rr values with different options
(but not the /sd) option. The oswego rr test has now been extended to
cover this issue as well.  (18 more tests).  A given test can be like this
using result variables:

tab ill vanilla rolls /o

assert samenum(round($OR*100)/100,22.15)
assert samenum(round($ORLL*100)/100,5.22)
assert samenum(round($ORUL*100)/100,138.39)

assert samenum(round($OR1*100)/100,12.26)
assert samenum(round($ORLL1*100)/100,1.76)
assert samenum(round($ORUL1*100)/100,150.86)

assert samenum(round($OR2*100)/100,38.19)
assert samenum(round($ORLL2*100)/100,4.06)
assert samenum(round($ORUL2*100)/100,1983.85)

assert samenum(round($SumOR*100)/100,20.61)
assert samenum(round($SumORLL*100)/100,4.93)
assert samenum(round($SumORUL*100)/100,126.46)

The "hidden" bug arising by the sort option is a good example of why many
tests are needed, but I think also that it shows that even with a number
of test releases the policy of "silent release" is appropriate.


Jens Lauritsen
Coordinator and initiator of EpiData Project

More information about the EpiData-list mailing list