I have a time series data of 40 year's monthly mean Sea Surface Temperature(SST). I want to perform Man-Kindall(M-K) test to detect the long term trend. But before M-K test,I have to eliminate autocorrelation from the time series data.So,at first I want to check the autocorrelation by performing a Durbin-Watson test.As the time series data has seasonality, do i need to adjust the seasonality first before performing Durbin-Watson test?
also,I am using R language.
I normally visually inspect Autocorrelation and Partial Autocorrelation Function (ACF and PCF) plots, so my response varies slightly from the specifics of your question.
But yes, I adjust for seasonality (using pairs of sine and cosine terms) and then look for residual autocorrelation (in my case with plots, in your case with D-W). I can't remember where I learned this approach, but I believe the reasoning is that seasonality is the most common form of autocorrelation. So basically, account for the biggest source of autocorrelation in your time series and see if anything is left over. I've published a number of papers with this approach and it's never been raised as an issue. Hope that's helpful.
Of course yes, always take into account all the assumptions and more in this object of study. Greetings
Yes, you have to adjust the seasonality present in your data before going for the D-W test.
Why not giving a try to Seasonal Kendall test instead? https://www.rdocumentation.org/packages/EnvStats/versions/2.3.1/topics/kendallSeasonalTrendTest
or see Package "Kendall"
Cristian Popa Thanks for your feedback. Should I worry about auto-correlation in my timeseries data if I try seasonal kendall test using "Kendall" package?
- Badges
- Science topic
- Similar topics
- Statistics
- Time Series