# Count data and exposure

#### jimme

##### New Member
Hi, I have seven categories, 'monday','tuesday','wednesday'...
Each category contains count data, however I have a different number of observations on each day - some days have 10 observations, some 12, some 21.
I want to test to see if the day of the week has an influence on the count and decided a Chi-Square test is most appropriate (...good choice?).
I am unsure what to use as observed and expected in the test. Should I average the count on each day and use this as the observed values? Or should I sum the observed values on each day and then adjust the expected values to reflect the difference in exposure?
Any help would be really appreciated,
Thanks

#### katxt

##### Well-Known Member
It looks to me like - get obs and count for each day - get total obs and total count -
Then for each day, expected = day count/total count*total obs. Observed is always the actual count.

#### jimme

##### New Member
Thanks a lot for your answer, that is great to know that I should use the actual count and not take the average. I am a bit confused how you are calculating the expected, is that actually a mistake? Would it not be:
expected = Number of observations this day / total number of observations * total count
This way we get the proportion of the total count that we would expect to occur on this day if we assume equality?

#### katxt

##### Well-Known Member
You're perfectly right. Sorry.
I think I probably got mixed up with the word "observed" which usually means the number seen which in this case is "count". In any event, with chi square, it only works with the raw count data, never averages or percentages.

#### jimme

##### New Member
No problem, thanks again for your help! When I ran the incorrect test using the averages I got p=0.2, now with the correct expected values I get p<0.0001, which makes sense because the categories appear to be quite different. So I'm glad this is now working as expected.