I feel like maybe this question is so easy that it's hard and I keep doubting every choice I make, so I thought I would just finally ask online. I have one dependent variable with three levels: percentage of caches a bird make in one of three different types of substrates (sand, gravel, or other). My independent variable has two levels (the three conditions the birds are in whilst caching).
I pretty much know my pattern of results and I have figures drawn up and everything (I can post a small subset of the data if that would be helpful to people in answering the question), but I cannot, for the life of me, figure out if I have chosen the right test to analyse this data and I keep having doubts every time I make any progress.
I'd ran a similar study not too long ago that was almost identical, except the independent variable only had two levels (gravel and sand). Therefore, I could just run a Friedman ANOVA (since the data violated normality assumptions) for percentage of caches in one or the other substrate.
Right now the only solution I have been able to come up with for this is running multiple Friedman ANOVAs, but, as mentioned, I am really starting to doubt this. This is primarily because--unlike when I only had 2 levels in my DV--, I cannot just run a test for one of the levels (e.g. in my first experiment, if 25% of the caches were in sand, you automatically knew the other 75% were in gravel; however, for this experiment, if 25% are in sand, that only means that 75% are in gravel or "other" and you do not know how it is divided up between the these two).
I know there are more advanced forms of analysis as well, like mixed models, but I've been specifically advised against doing those in this particular instance.
It's just that I keep writing and writing, then I get freaked out about my stats, so I do a bunch of research, then I try to re-calculate and re-write….and it's getting to the point where I can't do that anymore. So, I kind of need to know now if I need to change my method of analysis.
Could anyone give me advice?
As with most regression decisions, everything depends on your DV. If you leave it as it is this is a three group classification problem IMO. The question then becomes is the DV ordinal or simply categorical.i can't tell from your description. If ordinal then you need ordinal logistic regression. If categorical (that is nominal) you need multinomial logistic regression. There are good programs available in most stat packages for both of these methods. Google search will give you full details. Now here's my guess. From your description I would guess that your D V is ordinal. But you may not want to leave it in three groups. This would change methods completely. My suggestion is try the appropriate logistic regression and look at the results. At that point you may wish to try a different regression model based on the properties of the new DV. Further questions about the STATISTICAL methods can be asked at any point. Best wishes, David Booth
I also think that your questions are best adressed by a multinominal regression. If "sand," "gravel", "other" has some natural order (e.g. the ganularity) and this is actually the variable of interest, then it should be an ordinal logistic regression (in this case it would have been more informative to measure the actual, say, granularity, e.g. as average particle size or something like that, rendering the categorization unnnessesary and allowing for the analysis of a functional relationship between granularity and your IV).
https://www.google.com/url?sa=t&source=web&rct=j&url=https://m.youtube.com/watch%3Fv%3DDcS6_I63PHs&ved=2ahUKEwjUjfOlpMvxAhXCgv0HHSO5AasQFjAFegQIIBAF&usg=AOvVaw2C7CmokIk5BSS7-KcJNrjR
Just to clarify a few things mentioned in several people's answers:
1) David Eugene Booth -- I believe my DV is ordinal, as the main variable I am looking at is number of caches. What may have been confusing is that I am measuring the number of caches in the sand, number in the gravel, and number in the other locations. I've attached a photo that shows an example of how my data is set up. It's just a small subset as it is only showing one measurement for each condition per substrate. Perhaps I might have two DVs...?
2) Jochen Wilhelm -- The decision to use sand and gravel was because I am looking am looking at the noise the substrates make (sand is quite silent, whereas gravel is very noisy, when birds interact with it). The "other" location has been added because, in previous studies, they'd been making some caches in neither substrate. I have heard many people argue that caches made in "other" locations are even more quiet than sand, so perhaps there actually is a natural order: noise associated with substrate? This is a very interesting idea and I will look into this form of analysis as it is not something I had thought of.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Modeling
- Analyses of Variance
- ANOVA