Lower DOF with mean UD

[Amelia J]

Hi there,

I have split my GPS data into months for all 11 individuals, and excluded months where the data was not range resident (based on segmentation). I created unique individual-month-year IDs (e.g Bob_Jan21, Bob_Apr22, Zac_Aug23, etc.) - creating 82 IDs.

For every unique individual-month-year ID I calculated UDs, and then after that I used mean() to calculate the mean UDs per individual, to get an idea of their UD across the entire tracking period. While the UDs of the 82 IDs all had plenty of DOF (>50), the mean UD for each of the 11 individuals typically has DOF < 5 each.

Apologies if this is a silly question, but why have the degrees of freedom become drastically smaller? Does the DOF now represent the effective sample size of UDs available to that individual, rather than the effective sample size of datapoints?

I assume the only 'solution' is bootstrapping to try account for the low DOF values but I just wanted to understand first where those values are coming from.

My code:

# Fit models

fit_list <- list()

for (name in names(telemetry)) {
  data <- telemetry[[name]]
  GUESS <- ctmm.guess(data, interactive = FALSE)
  fit_list[[name]] <- ctmm.select(data, GUESS, method = "pHREML")
}

# Calculate AKDEs in one call

UDs_all <- akde(telemetry, fit_list)

# Average UD by individual ID

individual_ids <- sapply(names(UDs_all), function(x) sub("_[^_]+$", "", x))
UDs_by_individual <- split(UDs_all, individual_ids)
mean_UDs <- lapply(UDs_by_individual, mean)

# Output for one individual

DOF[area] for individual-month-year UDs:

SLX1501_Apr22 =  193

SLX1501_Aug22 = 212

SLX1501_Jul22 = 441

SLX1501_Jun22 = 246

DOF[area] for mean UD:

SLX1501 = 1.4

Any help would be really appreciated :) 

Thanks so much!

[Christen Fleming]

Hi Amelia,

In the default mean() with sample=TRUE, the uncertainty relations assume a larger number of unknown ranges that weren't sampled. If you use sample=FALSE, then the uncertainty relations assume that you sampled all of the ranges of that individual. I think this is the issue - the number of unique ranges per individual with sample=TRUE.

Best,

Chris