#### mili

##### New Member
Hello :wave: ,

I think of ANOVA analysis as being useful when we want to see if there is a significant difference between the means of levels. Particularly with the one-way-ANOVA, there is one categorical vaiable ... with a few levels .... and one continuous, with a mean number for each level being used in calculation.

But what about when there is one categorical variable, say three levels, but for each level I don't have a mean but a total. So say 12 selected level A, 13 selected level B, and 40 selected level C of variable ABC. So its not like each participant selected a number for each option and then the means were calculated but instead each participant just selects an option over the others. Is some kind of non-parametric design needed to statistically find out if there is a significant difference between the numbers of people that selected a particular option? Can it even be analysed ?

Many many thanks,

Coco

p.s. I was looking to analyse this in excel, any support on how I might be able to do that would also be extremely useful.

#### Karabiner

##### TS Contributor
Looks like you have the H0 that the proportion of each option is the same
(i.e. 0.333..)? You could evaluate that H0 by using a Chi square goodness
of fit test. If Excel can't do this then you might want to look for web based
application here: http://statpages.org/#CrossTabs

Kind regards

K.

#### mili

##### New Member
thanks so much for the reply Yes, chi test for goodness of fit sounds right. I am unsure how to find where the difference is though. So for example, if I have three levels, and the chi goodness of fit is less than the alpha, where are the significantly different values... between a and b, b and c, a and c ... a combination? Is there a follow up test?

Thanks again,

Coco

#### Karabiner

##### TS Contributor
You could do 3 pairwise comparisons with a Bonferroni-corrected alpha.

Kind regards

K.

#### mili

##### New Member
Do you know if this can be done in excel? The pairwise comparisons....

I am having trouble finding any instructions online on how to do this in excel....

Thanks Coco

#### segunstarr

##### New Member
Yes, you can actually use kruskal wallis methods, it's a non-parametric method of dealing with one-way Anova when the data is categorical.