Advice on time series analysis/ ancova on temporal data
Nov 22, 2017 2:18 AM(2706 views)
I would like some advise on how I should analyse my time series data.
I have hourly measurements of water temperature data for 5 months across 5 sites. I have summarized my data in following way
2.Day (0-150) 2. Month (June-October, 5 levels) 3. Daily Mean Temperature(Response variable 1) 4. Daily Max Temperature (Response variable 2) 5. daily standard deviation,or some other measure of variation (Response variable 3) So I have 5 x 5 x 150 = 750 observations
I want to test the following : - 1. Is my response variable significantly different between sites.
2.Is there a monthly interaction? For example in month 1 there all sites have similar response and in month 2 the one site will have significantly higher response than other sites.
I am planning to do a time - series / or a mixed model analysis by specifying an temporal autocorrelation structure.
For fixed effect I want the explanatory variables to be
Option 1- month (as factor) , site(as factor) and interaction of these two. Option 2 - day (as a continous variable) and site (as factor) and intereaction of these two.
Since this is a time series data I am expecting significant temporal autocorrelation. So when I do this analysis for mean and max temperature I should add a correlation structure. But I am not sure which variable should I use for correlation - month or Day ?
Also I want to do an anova/ancova on standard deviation of daily temperature, so for this do I still need to add a correlation structure ? Or since it is a deviation, the values become independent ?
I'm pretty sure this is not what you'd like to hear, but here's my advice: Consult with a professional statistician familiar with time series analyses with whom you can work through all the context and details of your application.
There are so many questions that come to my mind about your application. In my opinion, working through those isn't one of the strengths of a discussion forum, and your question presents a statistics problem, not necessarily a JMP problem.