Trying to run an ANOVA or Kruskall Wallis but struggling

Question

[abb525]

Hi there

So I seem to be struggling with running any of the statistical tests, I believe because my data is not normally distributed (according to shapiro wilkes test) I technically need to run a Kruskall Wallis test(?). Originally I realised I needed to restructure my matrix into a vector, which I have now done, so that I have the 1st column containing the Simpsons index values across all the samples, and then the 2nd (group column) containing labels of the taxanomic phyla these values are associated with.

I am trying to run a statistical significance test to see if there is a significant difference between one of the taxanomic phyla groups and the other phyla groups, in terms of the simpsons Index values.

    tmp_vector=as.vector(tmp)

tmp_vector

labels=as.vector(sapply(colnames(tmp), function(i){rep(i, dim(tmp)[1])}))

labels

fungi_vector = cbind(tmp_vector, labels)

fungi_vector

The former is the code I used to generate my vector.

And here is a snippet of the output.

Could do with some help as to figuring out how to run the test I need to as I'm not very experienced with R.

Regards

Answer 1

[Kevin Blighe]

Hi,

Please permit that I first reproduce some data in the same format as yours:

fungi_vector <- data.frame(
  tmp_vector = c(rnorm(10, 0.7, 0.1), rnorm(10, 0.2, 0.1)),
  labels = c(rep('Ascomycota', 10), rep('Basidiomycota', 10)))
fungi_vector$tmp_vector <- as.character(fungi_vector$tmp_vector)
str(fungi_vector)
'data.frame':   20 obs. of  2 variables:
 $ tmp_vector: chr  "0.651365595291301" "0.661684503596836" "0.755288193037626" "0.666716503201382" ...
 $ labels    : chr  "Ascomycota" "Ascomycota" "Ascomycota" "Ascomycota" ...

---

---

Então / So, the first issue is that your column of numbers is encoded as characters; so, we first need to convert that to numerical values:

fungi_vector$tmp_vector <- as.numeric(fungi_vector$tmp_vector)

To be proper, we should also convert the second column to categorical values:

fungi_vector$labels <- factor(fungi_vector$labels,
  levels = c('Ascomycota','Basidiomycota'))

Then, a Kruskal Wallis test can be done like this:

kruskal.test(tmp_vector ~ labels, fungi_vector)

    Kruskal-Wallis rank sum test

data:  tmp_vector by labels
Kruskal-Wallis chi-squared = 14.286, df = 1, p-value = 0.0001571

For my 'manufactured' data, I already knew that the p-value would be statistically significant due to the fact that I had pre-selected different means via rnorm() (see above).

We can also plot this:

boxplot(tmp_vector ~ labels, fungi_vector)

I will leave the remainder to you.

Kind regards,

Kevin

[abb525]

So would this then compare if there's a statistically significant difference across all of the groups? There are several more phyla groups in the dataset.

The objective was to basically determine if there was a statistically significant difference between Ascomycota and the other phyla in the data.

Although using the code you showed me has definitely worked in giving me a P value when I run the Kruskall Wallis on the whole dataset.

[Kevin Blighe]

So would this then compare if there's a statistically significant difference across all of the groups? There are several more phyla groups in the dataset.

Yes.

The objective was to basically determine if there was a statistically significant difference between Ascomycota and the other phyla in the data.

In this case, please try Dunn's post-hoc test via FSA::dunnTest() ( see https://www.rdocumentation.org/packages/FSA/versions/0.8.32/topics/dunnTest ). This can be run in the same way as kruskal.test()

[abb525]

Would the Pairwise Wilcoxon test also be relevant? Since I am trying to find out specifically the significant differences between groups? That looks like it shouldn't be too hard to code for either.

[Kevin Blighe]

If they are logically paired, then no problem using the Wilcoxon, of course.