Re: McNemar Test for genetic data

Hi Corinna,

I'm still not completely sure that what you want is the TDT test. Your
original post said:

I like to test if there is a favourite combination of the aa, ab or bb
from the parents in children with a disease.

So it sounds like your question is possibly more basic than what the
TDT tests.

If one could tabulate your data like this:

child parental freq-
disease alleles uency
--------------------------
yes aa aa n1
yes aa ab n2
yes ab ab n3
yes aa bb n4
yes ab bb n5
yes bb bb n6
no aa aa n7
no aa ab n8
no ab ab n9
no aa bb n10
no ab bb n11
no bb bb n12

then you could analyze this as a 2 x 6 frequency table (that is, two
levels of child disease status by six allele patterns) to see if there
is a statistical association. You could use an exact test (e.g.,
StatXact) to assess row/column association. (Maybe SPSS also has a
suitable exact test).

I think an alternative way ist to count the allel frequency?!

Yes, you could arrange data like this:

Child
disease No of "a" parental alleles
status 0 1 2 3 4
+-----+-----+-----+-----+-----+
yes | n1 | n2 | n3 | n4 | n5 |
+-----+-----+-----+-----+-----+
no | n6 | n7 | n8 | n9 | n10 |
+-----+-----+-----+-----+-----+

Again, you could analyze this as a 2 x 5 frequency table using an exact
test--but here you could place an ordinal restriction on the column
variable. If there is a statistical association, it would then be
simple to observe the direction of the trend -- whether the probability
of child disease is positively or negatively associated with number of
"a" parental alleles (and therefore has the reverse association with
number of "b" parental alleles.

Notice that I'm not considering child allele status here, which doesn't
seem to relate to your original question (though obviously the possible
association of child alleles to child disease status is a relevant
question at some level).

I hope I'm not making this more complicated than the problem actually
is. I'm just trying to give you some other ideas. It really seems like
the first step is to clarify, in statistical terms, what your
hypothesis is. For example, can one state the null hypothesis here in
words?

John
--
John Uebersax PhD

.

Relevant Pages