If you are looking for impact assessment then it may be useful to follow Generalized linear model with gamma distribution with log as a link function. Soil erosion, Fire killed and resprouting can be factorial as 1,2,3 where Low (1) can be taken as a default. You can check whether or not it is due to fire or soil erosion with model selection using AICc after you run GLM. You can also see which species is affected more.

Thank you very much for your input on using 'gamma distribution with log as a link function. I had a concern about converting the variables in factorial. I am running the analysis and will get back to you

I am focusing only on one species and intend to look soil erosion and fire effect on that particular species (Cymbopogon coloratus).

Hi Saneesh,

Congrats for the work.

I would suggest you may try to fit a Multinomial logistic regression model by modelling the response variable y with three classes (Fire killed =High, Medium, Low), and the other categorical data as categorical explanatory variables x. For instance, If you have a data point y_1 = (1, 0, 0), that is Fire killed High class has a success (1) and others failure (0's), with the categorical explanatory variables x_j as indicators.

You can use some R package like this one to do it:

https://stats.idre.ucla.edu/r/dae/multinomial-logistic-regression/

or in python:

https://scikit-learn.org/stable/modules/generated/sklearn.linear_model.LogisticRegression.html

And test your hypothesis using some recommended test in the literature. If you search, you may find several on publications about hypothesis tests with multinomial logistic regression models. For sure, many of them are available either in R or Python.

If the reply is useful, please, click to recommend it for others,

Success on the research,

Thank you very much Jhonathan Pedroso Rigal dos Santos for your inputs. Would it be okay to run a chi square test to measure the association of variables?

I am attempting to run multinomial regression by representing the ordered categories ("high", "low", "medium", "no") as 4 levels (1,2,3,4).

Saneesh Cherapurath Soman I would recommend you to perform a likelihood ratio test, you can test the full model (all effects) against a reduced model (all effects, less the one you want to test), and see if it differs from zero.

If the reply was useful, please, recommend for others,