Stream temperature over time: is a repeated measures ANOVA my best bet?

[dsilver]

I am trying to analyze temperature data across 5 sites taken over approximately 4 months and am having trouble working out the best test to use (I'm out-of-practise with stats!).

The data:
Five temperature data loggers were set out along a river. Three were spaced along tributary A, one on tributary B, and one downstream of where the two tributaries meet (don't ask about the study design - I'm analyzing someone else's data). The data loggers recorded temperature hourly over 4 months, resulting in 2640 temperature readings for each of the 5 sites. We are trying to find out if there is a significant difference in temperature among sites

Thoughts on analysis:
1) I know (or assume) I should do a repeated measures analysis with sampling event (hourly measurements) as the repeated measure. Is this correct?

2) Do I need to nest the data logger sites within the tributary somehow? Can I, with 3 loggers in one tributary and only 1 in the other? And, if so, how do I deal with the 1 logger downstream of the confluence of the two tributaries?

3) Finally, I'm doing the analysis in Systat, so if you have any advice on how to set up the data, I would appreciate it! (I also have Statistica available, if that's a better option for any reason).

Thank you so much for any help you can provide - I really appreciate it!

 

[bugman]

If this were my data, I would pool the three loggers on tributary A (I doubt they are different unless they are placed up or downstream of an STP or another point source) - I would then base my analysis on weekly means - daily data is going to introduce a lot of noise and redundant data - I would then do a one-way ANOVA (temp. data is almost always normally distrubted in the residuals). Thats me. Other may argue otherwise. If in doubt do it this way and comapreit to a repeat measures analysis and let us know how you go.

I use R not systat so I can't help with your last question.

P.

 

[dsilver]

Thanks, bugman!

After thinking about it further, I had thought I might compare the three sites on Tributary A to see if I could combine them so I'm glad to see that's your recommendation also. I was also thinking about using daily means rather than the hourly data - I'll see what weekly does for me. I hadn't considered just using a one-way ANOVA rather than the repeated measures, but I'll try it out (I might do both and see what kind of differences I get - if I can figure out how to make the repeated measures analysis happen in the programs I have access to).

Thanks again.

 

[PeterFlom]

If it was me, I would run a multilevel model. They make fewer assumptions than RM ANOVA.