I have examined the vegetation cover for 9 sites, using a quadrant and estimating % cover in the quadrant. Each site has 6 measures. There are 3 treatments: ungrazed, extensive grazing and intensive grazing. 3 sites for each grazing regime.
Which statistical tests can I use to examine if treatment impacts the % cover?
Recap:
9 sites in total
3 sites ungrazed
3 sites extensively grazed
3 sites intensively grazed
6 data points per site
cover is estimated as % cover
Thank you!
You could use a simple linear regression model and analyse it using an ANOVA, i.e. 'Cover ~ Treatment (Grazing) + Site'. You would probably expect there to be some type of effect of the grazing treatment but also some type of effect between the sites. Furthermore, you can use a post-hoc test (e.g. Tukey HSD) to see differences between the individual grazing treatments or sizes in a pair-wise context.
Thank you Carl Alexander Frisk . My hesitation lies in the fact that I have percentages. Can I treat them like any other count data? I read somewhere that it is appropriate to transform percentage data before using an ANOVA. For instance with: y = ln(p/(100-p)) where p is the percentage and y is the final data to use. However, when I tried this I had a problem with values that are 100%.
All my percentages lie between 80%-100%.
There are no special rules for % data or fractions. If your residuals look normally distributed (when plotting the linear model) you don't need to transform your data. I have recently analysed %-cover data for a paper and all model assumptions were met just fine without transformations.
- Badges
- Science topic
- Similar topics
- Geoscience
- Asia
- India