1
$\begingroup$

I have two waterbodies that were sampled for a single water quality variable over time. Both waterbodies were sampled in the fall of each year from 2009-2012. In each year of sampling, 5 replicate samples were taken (so I would have 20 samples in total for each waterbody). I want to compare the two waterbodies over time to see if the water quality variable in each waterbody follows a similar trend. I feel like this is similar to an ANCOVA problem where a person could test for equal slopes. But this problem is a bit different because instead of having two continuous time series, I have 5 replicates of samples for each waterbody in each year. Anyway I'm wondering if ANCOVA is not reasonable, what is the best method to proceed with? A colleague suggested planned linear contrasts might help but I don't understand how that would be applicable to comparing 2 different waterbodies.

$\endgroup$
  • $\begingroup$ What is the correlation between the replicate samples? For parsimony you might benefit from using only one sample if they are nearly identical. Alternatively, you could potentially create a latent variable using the measurements to more accurately reflect water quality. $\endgroup$ – Behacad Apr 2 '13 at 22:04
  • $\begingroup$ correlation is low , I guess. I'm more looking for a general plan of attack I can use for these types of problems in the future. It needs to be easily communicated to biologists, so I think making latent variables is out. $\endgroup$ – Jimj Apr 2 '13 at 22:10
  • $\begingroup$ What is the purpose of the replicate samples? Are they supposed to measure the same thing? $\endgroup$ – Behacad Apr 3 '13 at 0:51
  • $\begingroup$ Yes, they are supposed to measure the same thing. We take replicates because of natural variability in the samples. $\endgroup$ – Jimj Apr 3 '13 at 1:04
  • $\begingroup$ And the 5 samples differ substantially (the measure is reliable?) If the measures are decently correlated then maybe you could just use one. Alternatively, you could perhaps use the mean of the 5. Is that a possibility? What is the correlation between the 5 measurements? $\endgroup$ – Behacad Apr 3 '13 at 2:11

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.