I have been trying to get answers for a while, and have looked up every help section and tried other stats sites. I have a thesis that I need to be completing shortly, but one set of data has been completely out of my league due to a small sample size.

Basically, I have four broods (birds with chicks - precocial so they can peck at hatch e.g. chickens). I measured the proportion of time each brood fed every other day (approximately) until they flew. I had three broods in the ocean front and one brood in the mudflat habitats.

Each data point within a brood is non-independent, so the only way to compare the mf brood against the OF broods in terms of time spent feeding (and other behaviours) would be to use the mean and use a Kruskal-Wallis or a one-sample Wilcoxon (mu=mean for MF brood). Of course the sample is too small to get anywhere near significance even though there is a clear pattern of MF feeding more and being less disturbed when I look at all the data points.

It was suggested that I could use a mixed effects model so that each data point for a brood could be used and each brood would be considered a block. The person typed out some ideas for models but no one could tell me how to do it step-by-step and I don't really understand them. I found a source which would let me do it in Minitab (https://onlinecourses.science.psu.edu/stat502/node/72) but when I tried it it told me that it is unbalanced and I think it is because the first block is in one 'treatment' and the other three in the other 'treatment' but there is no overlapping of either.

Another issue is that now I may have figured what the models do, I think they just try to explain the reasons why things are as they are, but I would still like to be able to say that MF feed significantly more than OF broods and that MF brood is disturbed less often. The model would only try to explain the reasons why rather than if they are significantly different (which is all I am really trying to figure out).

If there is an easier way to compare two groups that have independent and non-independent data then I would greatly appreciate it. I actually thought of using a point biserial correlation using the means with MF coded as 0 and OF coded as 1. It works but is it valid? (http://faculty.vassar.edu/lowry/pbcorr.html)

Is there anyone out there who could possibly help me with this? I am reaching my deadline and know that if I could just do this part then the rest should be okay. I am also a foreign student and my professor has said he can't help me long distance.

I also have flush rates vs number of days to hatch. I wanted to see if birds were less likely to flush off their nest closer to hatch so used correlation. I have data for 7 nests and plotted each data point (n=77). The issue again is that not all points are independent but I need to show the flush rate for nest age (and not all nests were measured at the same age). Should I do a correlation for each nest and then is there someway of combining them to give me a p-value, or should I leave it as it is?

  • $\begingroup$ It would be better to split this into two questions. The flush rates one seems a separate problem to the one about feeding of mud flat birds. $\endgroup$ – Peter Ellis Mar 20 '12 at 4:00
  • $\begingroup$ On the mudflat question, how many birds per brood? And does the eating behavior change over time, or is the intention to merge together the measurements which happen to be taken over time into a single measure of a bird's overall eating behaviour? If you could post your data somewhere, or an example of it, it might help. $\endgroup$ – Peter Ellis Mar 20 '12 at 4:07
  • $\begingroup$ and looking at the flush rate question, again it's not clear what your data is. Do you have 11 observations on 7 individual birds, each observation taken a bit closer to hatch day? $\endgroup$ – Peter Ellis Mar 20 '12 at 4:17
  • $\begingroup$ and what is flush rate measured in, what sort of distribution does it have, etc. $\endgroup$ – Peter Ellis Mar 20 '12 at 4:20
  • $\begingroup$ Back to the mud flats - it seems to me that there is a problem that with only one brood on the mud flat you will not be able to separate a mudflat effect from the effect of the individual brood. You might be able to say that brood are big eaters, but that might be due to either the brood (eg genetics) or location. I would imagine a mixed effects model with brood as a random effect and mudflat as a fixed effect would return an error because of this. If you had at least two mudflat broods however, the mixed effects model would be a possible means to answer your question about significance. $\endgroup$ – Peter Ellis Mar 20 '12 at 4:25

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.