Hi,
I am working in R and know of the T-test and Tukey-test to compare levels of a variable, but these require equal sample sizes. Is there a test I can use where equal sample sizes are not a requirement?
in the attached file there's a table in which I'd like to make a few comparisons, namely does the amount of viruses present (#viruses) affect the value of the different breeding values? The amount of viruses represents an undependant variable (factorial) and the other ones are all dependent numerics.
As an example; is there a significant difference in varroa resistance between groups that are infected with 0,1,2 or 4 viruses?
I conducted a one-way ANOVA for every independant variable and concluded that there are significant differences for the following variables: 'defensive behaviour', 'chalkbrood' and 'total breeding value'.
PROBLEM: if I conduct a Tukey-test to see WHICH levels vary I don't see any significant difference at all. What test should I use at this point that does not require equal sample sizes?
If you are using anova, use car::Anova for the anova table, and the lsmeans package with a Tukey adjustment for the comparison of treatments.
You still need to be sure that residuals from your model meet normality and homoscedasticity assumptions.
See the following for examples:
http://rcompanion.org/handbook/I_05.html
http://rcompanion.org/rcompanion/d_05.html
BTW, a two sample t-test can be used for unequal sample sizes. Tukey-Kramer comparisons as well, though the implementation in R is limited.
Best to use the multcomp package or lsmeans package for multiple comparisons with unequal sample sizes.
... and you should also improve your data, if it is possible, because there is only 1 case with the level "4 viruses" and only 3 cases with level "2 viruses"! I think, that these numbers of observations are insufficient for ANOVA and other above mentioned tests.
I think, that with this dataset it is only relevant to test the significant difference between the levels of 0 and 1 viruses - if the data meet normality, you can use t-test, if not, then you can use Wilcoxon nonparametric test.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics