If I have the number of people presenting with a particular disorder in two particular years, lets say 12 in 2005 and 28 in 2006. Can I us the chi squared goodness of fit with 20 as my expected value (i.e. if the average of the two - no difference between them) to tell if there is a significant difference between the two? Is that legitimate thing to do? Thanks!
Hi Kevin!
I think in general a chi-square test are not used in time series data. If the particular disorder happened during more than one year, testing the hypothesis that there are no difference among them is meaningless, since the people with particular disorder is still the same.
Please try to find another suitable test. Thanks.
Hi, Kevin
No, it is not correct, sorry. Chi-square is for categorical variables, and number of people is an ordered variable. And, as Ezra noted, time is an ordered variable. It is impossible to compute an average of two qualitative classes. Say you have black patients and white patients. What is the mean ethnicity?
- https://odajournal.com/2013/09/19/unioda-vs-chi-square-ordinal-data-sometimes-feign-categorical/
Thanks folks for taking the time to answer, it's much appreciated! I thought maybe that you could have presented/didn't present and in that particular group (year)/not in that group as dichotomous variables and do a 2x2 chi squared either but that makes sense re: time being an ordered variable etc. I might be able to do without stats altogether and just say the number increased and do graphs etc but I'll have a think about it. Thanks again!
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics