I have recently done an analysis using the chi squared test for trend and
got such a high chi squared value I suspect something is very wrong. I
obtained a chi sq value of over 600 and a resulting p value of 2 E -147
which is ridiculous.
Does anyone know of any limitations on the size of frequencies when using
this test ? My data contains some large frequencies ( >4000) and I wondered
if this could be a reason for such a bizarre answer. Incidentally the trend
when graphed does slope, but not so much that I would expect an answer like
this.
I have carried out this test numerous times before and never had any cause
for concern (albeit with smaller frequencies). I am concerned at this value.
The data I used is shown below.
-2 -1 0 1 2 3 4 5
6 7 8 class
16 1754 4025 3365 2339 1765 1395 881 319
132 9 yes
4 1824 7629 8828 7422 4683 4455 3056 975
363 37 no
20 3578 11654 12193 9761 6448 5850 3937 1294 495
46 total
I thought that the chi sq test for trend looked at the proportion of
successes in different groups over time and so I can't understand why I get
such a bizarre answer. I have checked my working numerous times.
If anyone thinks that this data is not suitable for analysis in this way I
would be happy to hear from them, alternatively if anyone has a text they
would recommend outlining the theory behind the test I would also be
interested.
Thanks in advance
Regards
Jon Cooke
0203 3834284
[log in to unmask]
Care of the Elderly
Imperial College School of Medicine
Hammersmith Campus
Du Cane Road
London
W12 0NN
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|