I measured number of mass spectra peaks from differently fixated material to test for an effect of fixation on number of peaks. I made weekly measurements to see the change over time for 3 Months, which is a relevant time scale for us. From the raw data I calculated median values as I want to test the resulting slope using MannKendall and TheilSen Estimator. To do so, I have to specify the data as a Time Series object. But as I only have 13 observations, I am not sure what frequency to use in ts(). I already tested if the frequency has an influence on results and it is enormous.
This is an example of my data:
Week Meadian_peaks 0 113 1 111 2 110 3 112 4 110.25 5 108.5 6 105.25 7 102 8 103.2 9 104.4 10 105.60 11 106.8 12 108
I specified the time series with weekly frequency:
ts_X25100all = ts(X25100all$Meadian_peaks, frequency = 52, start = 1, end = 13) MannKendall(ts_X25100all) #tau = -0.00614, 2-sided pvalue =0.82478
This result tells me, that there is now significant difference of the slope to zero.
Specifying the time series with
frequency = 1 (yearly) the result is quite different:
ts_X25100all = ts(X25100all$Meadian_peaks, frequency = 1, start = 1, end = 13) MannKendall(ts_X25100all) #tau = -0.462, 2-sided pvalue =0.032736
This is why I want to know, what frequency I have to choose. I am a bit in struggle because I understood that based on the frequency a seasonality is described - which, to my understanding, I do not have in my data.
The example I provide here is a well fixated sample resulting in only minor differences over time. I have others that I want to compare with this one which is why I want to test for monotonic trend. My raw data does not allow an ANCOVA. I already tested different linear models and data transformations. Residuals were never normal distributed. Just in case you would like to suggest this.
So my question is: Which frequency to use for my data?