I have 52 station and each station got three different transect. I have GPS location for those stations and do not have GPS location for each transect, so, those different transects are representing with their stations' GPS location. The gls model shows temporal auto-correlation, so I ran the initial gls model with corExp, corGaus, corLin, corRatio, and corSpher to get the best model, and all can not be checked since zero values are present. I guess, this is because I do not have transect gps location and included the stations' GPS locations for each transect. So, in this case, if I add the interaction of each transect to the model that the errors of temporal auto-correlation can be added to the model and problem will be solved?? Or is there any other method to deal with temporal auto-correlation in gls which has zero distance? Thanks in advance. Tharaka
In my opinion, you need to go in field and collect the missing data...
Dear Tharaka,
Hello there,
if the identified auto-correlations are statistically significant, as you mentioned, you may need to remove the auto-correlation structure from the data first and, then, you should continue your analyses based on statistically independent data. As a basic, right, and effective approach to remove the structure of auto-correlation form your observations, you should estimate/fit the time series model (e.g. ARIMA (p,d,q) Models) first and then extract the auto-correlation structure from the actual data. Now, you may have a set of statistically independent residuals which are ready to be used for your concerned analyses. That is, you should apply all your considerations on the residuals.
Hope that helps!
Ali
- Badges
- Science topic
- Similar topics
- Computer Communications (Networks)
- Simulators