I surveyed 4 sites (A, B, C and D) during six times along a year. On each site I measured abundance in six sampling transects. I have a total of 36 sampling units on each site.
Sites A and B are very impacted and sites C and D are reference with almost no human disturbance but topography is different A=C ≠ B=D (i.e. A and C present steep slope; B and D present gentle slope). I have 3 factors: Time (6 levels); Impact (2 levels) and Topography (2 levels). Data found to be normal and homocedastic. I´m interested in determining the relative contribution of Human disturbance and Topography to abundance variation of this species.
Is it possible lumping data together in order to perform two ANOVAS to determine the effect size attributed to each factor and then compare them?
I mean to perform a bifactorial ANOVA grouping data to compare Impact versus Reference sites and to calculate the effect size with its confidence intervals. To perform another bifactorial ANOVA grouping data to compare Gentle slope versus Steep slope and to calculate the effect size with its confidence intervals. Then comparing the effect size for each analysis in order to determine the relative contribution to variation of each factor
Note that: 1) variation due to time should be the same for both ANOVA; 2) in the first ANOVA the two sites on each level present different topography; 3) in the second ANOVA the two sites on each level present different disturbance.
- If it’s not possible to do that, there is any other alternative to determine relative contribution of Impact and Topography to variations in abundance with this sampling design?
why not doing an ANOVA, and then a stepwise selection of variables? it is based on parcimony, or the relative importance of each factor? it will give you which of your factor explained the variance, and its level of explanation
I have a question. When you are mentioned that you made measures in six times along year, is these measures made at the same sites? If your reply is yes then you need to study the possibility to perform "repeated measures ANOVA". Additionally when you talk about slope, apparently the model is unbalanced since you can not obtain all the possible combinations between your factors. As marie mentioned in her answer you need to look the most important factors, this you can found using a variance contribution analyses.
The data can be analyzed using a two-factor 2X2 ANOVA. One factor is slope (high, low) and the other factor is disturbance (reference, exposure). Unfortunately, your experimental unit is the site, which means that you have no replication in your experiment. You can analyze the data using each transect as a replicate (n=6), but be aware that the experiment is pseudoreplicated. In any case, do not use n=36... the samples along each transect are subsamples and should be averaged to produce a transect value.
The time series data are not required to address your main hypothesis, although you could undertake a separate analysis of the time series data using a non-parametric Mann-Kendall trends test.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Modeling
- Analyses of Variance
- ANOVA