My experiment have categorical variables such as moisture (wet +, wet -, normal, dry- and dry +) and lightness (low-, low+, normal, hight- and hight +). But I have continuous variables such as pH and temperature. These variables are being measured in soil under four plant's canopy and bare soil. In these conditions were planted seedlings.My aim is to test facilitation and to use these variables to this.
Dear Jhonny,
It looks like your moisture variable is not simply 'categorical' (or nominal scale) but in fact on an ordinal scale, since there is a clear ranking in soil moisture content. See
http://stattrek.com/statistics/measurement-scales.aspx?Tutorial=AP
for more details. Hence, a simple approach would then be to associate a number to each class, for instance: Replace
wet + > wet - > normal > dry- > dry +
by
2 > 1 > 0 > -1 > -2
and then run your multivariate analysis on that basis. The same would apply to 'lightness'. You may need to think about the numerical values to use, as well as their spacing, and to decide whether to use the same (or different) scales for both of these variables.
Lastly, these two variables (moisture and lightness) could, in principle, be measured on a ratio scale, so why are they provided as 'categorical' values in the first place?
I hope this helps. Cheers, Michel.
Ordinal variables are customary in human depending responses like taste, attention, reaction time awareness, and do not imply proportionality in scaling. You may see wine example on https://cran.r-project.org/web/packages/ordinal/vignettes/clm_intro.pdf to get an idea, andif it is quite like your case apply r-project ordinal package
Dear Jhohny,
another way how to deal with categorical or ordinal explanatory variables in your model is to use so-called dummy variables. It means that you code the presence or occurrence of respective category as 1 and the opposite as 0. As a result, for instance, instead of one variable such as soil moisture of five categories you will get exactly five variables composed only of unities and zeroes. You have to leave out one of these five variables (maybe the first), otherwise your model will suffer from collinearity. This way, such a type of variables is treated, for instance, in the 'vegan' R package.
More details on dealing with these variables can be explored through the examples related to RDA given in the following book for ecologists:
Borcard, D., Gillet, F., Legendre, P., 2011. Numerical Ecology with R, 1. ed. ed, Use R! Springer, New York, NY.
Hope this helps.
Best regards from Czechia,
Ondrej
Thank you all for help. I will study the possibilities offered.
- Badges
- Science method
- Similar topics
- Biological Science
- Ecology