Can I use ANOVA for counts (integers)?

S

stweb

Guest
#1
Hello,



I have a crossed experimental design with two fixed effects: Neighbour/Stranger and Female/Male.



The response variable is counts. Actually, the counts are background subtracted and some are negative.

Sample sizes are fairly large and balanced = 30 x 4 = 120.

Boxplots of counts are symmetrical for the different fixed effects N/S and F/M.

Counts range from -2 to +10.

The magnitude of median differences is 1 for both effects (n.b. integer data)

Boxplots are a bit wider for N and for M.



I want to test: is the difference between average counts for N/S or F/M significant?



Is ANOVA suitable, followed by t-tests for the significant effects?

Strictly, ANOVA is for continuous, normal variables.



Is there a non-parametric equivalent for two-way ANOVA?

Or perhaps a generalised linear model?
The data are not Poisson because there are negative results.
I am also interested in testing for interactions between N/S and F/M., which is why I chose two-way ANOVA.


What more should I do?

The could be some unequal variance between groups.


I use R for statistical computing
and can try more sophisticated analyses.

Thanks.
 
#2
(I) Well we know that ANOVA is used to compare means of more than two independent (x) variables with equal variances and normal distributed sample .
Please note that the dependent variable (y) is continuous

The question about integer appears to be a kind of small confusion with discrete and continuous

(II) testing for interactions between N/S and F/M Generalized Linear Model of ANOVA can be used.
GLM or repeat measure ANOVA can be used when we want to observe the interactions in within the subject along with in between the subjects

(III) if there could be some unequal variance between groups then KRUSKAL WALLIS test