I artifically elevated the air temperature in the heated plot by 3-celsius degree compared to that in the control for 7 days.
I want to figure out the effect of the temperature manipulation on soil temperature.
The datalogger collected the soil temperature every 30 minutes.
In that situation, can I calculate daily mean soil temperature using 30 minute-unit data to go ANOVA?
Additionally, can I treated the respective days as repeated factors for repeated measures ANOVA?
You must mention that you presume that the other factors (other than heat) are similar. You can do any thing if you can interpret it.
The design is not very clear to me. Did you observe several soil plots? If so, you might perhaps compute the means of 120 collected temperatures per hour for each of the 7 days and then run a repeated measures ANOVA with 2 Within subjects/plots factors :24 Hours and 7 Days). If you do'nt have many plots, maybe you can further reduce the Hours, e.g. from 0 to 6 A.M., from 6 to 12 A.M., from 1 to 6 P.M. and from 6 to 24 P.M.) You could also run 7 one-way repeated measures anovas, but then correct the p-values for multiple tests.
If the distributions of the means is normal (check this with a histogram and Shapiro tes) you can run the parametric repeated measures ANOVA; if not normal, uou can run the non-parametric lf.f2 from the nparLD R package (see attachment and https://rdrr.io/cran/nparLD/man/ld.f2.html ) (The one-way repeated measures nonparametric equivalent is the Friedman test.)
If you have a time-course, you should use a regression model. ANOVA is a bad choice.
Jos Feys Thanks for your advice. I arrayed overall 54 plots (three temperature levels, three precipitation levels, and six replicates). Temp levels consisted of control, 3-celsius-degree heat, and 6-celsius-degree heat. Prec levels consisted of control, drought, heavy rainfall.
If I understand you have 48 measurements per day and a wondering whether aggregating to a day amount is a) doable and b) good practice. The doable is sure, just choose your statistics. b) is trickier as I assume there are time of day effects. Lots of ways that the economists (and others) deal with cyclical patterns like this in time series. Also, your question suggests that you are worried about having lots of time points. The phrase intensive longitudinal methods is often used and a good book is: https://www.amazon.com/Intensive-Longitudinal-Methods-Introduction-Methodology/dp/146250678X/ref=sr_1_1?dchild=1&hvadid=77996659922595&hvbmt=bp&hvdev=c&hvqmt=p&keywords=intensive+longitudinal+methods&qid=1611418439&sr=8-1&tag=mh0b-20
Jos Feys Thanks. I have one more question. When I perform the normality test, input data should be separated by the treatments? I wonder if I should use the respective data for one normality test or entire data including control, 3-celsius-degree heat plot, and 6-celsius-degree-plot. I could assume the normality when I used separated data, on the other hand, I could not assume it when I used the entire data.
- Badges
- Science topic
- Similar topics
- Geoscience
- Atmospheric Sciences
- Atmosphere
- Air