Hi,
I want to check the test I can use for determining association between a count/discrete variable (Eg: No of health visits for particular disease) and categorical variable (types of providers).. I dont have any background info as variables so little unsure if can use poisson type model.
I have got suggestions to use chi square but im unsure
pls help.
Dear Marian Abraham ,
This sounds like a one way ANOVA or, if its assumptions are not met, a Kruskal Wallis.
If you know the number of non-visits (per disease) as well as the number of (yes-)visits, then you can run a chi-square test. Otherwise (if you do not know the number of non-visits), indeed, an ANOVA (if normaly distributed counts) or a Kruskal-Wallis. A Poisson regression (or negative binomial) would be indicated if the there are very much low (or zero) counts (i.e. the distribution is Poisson or Negative Binomial). See e.g.
https://stats.idre.ucla.edu/r/dae/negative-binomial-regression/
https://stats.stackexchange.com/questions/60643/difference-between-binomial-negative-binomial-and-poisson-regression
If your data came from a normally distributed population, you can perform the One-way ANOVA test. But if your population cannot pass the test of the normality then you have to perform the Kruskal-Wallis test.
However, if you categorize the continuous variable of yours i.e. no. of health visits for a particular disease, then you can perform the chi-square test between these 2 variables.
The test of the normality of your data can be checked in several different ways. But I prefer the 1-sample K-S test and Shapiro Wilk test. If you got a p-value less than 0.05, which indicates your data came from a normally distributed population. If it is more than or equal to 0.05 than the data would not pass the test of normality.
Thank you so much Dr. Samuels, Dr. Ahmed and Mr. Feys for these responses. They have been very helpful.
If my independent variable is a binary variable, instead of Kruskal-Wallis H test ( equivalent of anova), should I use Mann Whitney (equivalent of t test)?
Hello dr Marian , yes your answer is right regarding application of Man Whitney test which is equivalent to independent T test. I upload 2 pictures to explain difference between parametric and non parametric tests.
Few advise that the Kruskal Wallis test should be done when shape of distributions of groups are the same. I have also tried the levene to check equality of variance. Is there any other test I could use when these assumptions are violated?
- Badges
- Science topic