www.phidot.org

Hi Murray,

I've been troubleshooting density covariates analysis for some time and can't seem to get the code to run properly.

I'm doing mark-resight of butterflies, and looking at 3 covariates Canopy, Elevation, and Ceanothus (host plant density), and interactions between canopy+ceanothus and elevation+ceanothus. Trap type is "count".

Code: Select all

base.args <- list(capthist = Ald2019aCH, mask = clippedmask, trace = FALSE, verify=FALSE, method='Nelder-Mead')
 args.0 <- c(base.args, model = D ~ 1)
 args.Dcan <- c(base.args, model = D ~ 'Canopy2')
 args.Delv <- c(base.args, model = D ~ 'Elvpcl_buff3')
 args.Dnjt <- c(base.args, model = D ~ 'Ceanothus2_num')
 args.Dnjtelv <- c(base.args, model = D ~ 'Ceanothus2_num' + 'Elvpcl_buff3')
 args.Dnjtcan <- c(base.args, model = D ~ 'Ceanothus2_num' + 'Canopy2')
arglist <- list(null = args.0, Dcan = args.Dcan, Delv = args.Delv, Dnjt = args.Dnjt, Dnjtelv = args.Dnjtelv, Dnjtcan = args.Dnjtcan)
fits <- par.secr.fit(arglist)

First I attempted without method=Nelder-Mead and then with it. Both had the same outcome:

Error in getmaskpar(!CL, D, data$m, sessnum, details$unmash, attr(data$capthist, :
bad densities in 'secrloglikfn' (try different optimisation method, link, or model?

I'm unclear how to troubleshoot ie "try different optimisation method, link, or model" exactly. Pretty new to R!

Thank you for any assistance.
Angela

I have no idea either. Maybe some bad covariates in there? If you still want to pursue this, send me the data offline.
Murray

Hi Murray,

I'm reviving this conversation as I've run into a new, but similar, issue relating to density covariate results.
I'm struggling to replicate results from a previous analysis. I suppose it's normal for density estimates to change slightly between analyses? But there's some major problems:

-Top density model ranked in AIC changes when analysis is repeated.

-The s.e. and CL estimates (at different levels of the covariate) are so large it isn't tenable for publication anymore. eg. at a density estimate = 0.0003, s.e.=6e^23.

I didn't have this issue originally back in March 2022 when I produced these results for my thesis. I might have lost some important piece of the code in my attempt to tidy it between now and then? Or maybe some R/Rstudio/secr/dependency updates since then might have thrown off the analysis?

My complete code is below. Please let me know if you notice anything that seems odd.

Code: Select all

captfile <- read.delim("CapAld21_fix.txt")
trapfile <- read.delim("TrapsAld21.txt")
Ald21CH <- read.capthist("CapAld21_fix.txt", "TrapsAld21.txt", fmt=c("trapID"),
                         detector="count", noccasions=43, markocc=c(1,0,1,0,1,0,1,0,1,0,1,0,1,0,1,0,1,0,1,0,1,0,1,0,1,0,1,0,1,0,1,0,1,1,0,1,0,1,0,1,0,1,0))
Ald21CH <- addSightings(Ald21CH, nonID="TmAld21.txt",
                        unmarked="TuAld21.txt", uncertain="TnAld21.txt")
mask <- make.mask(traps(Ald21CH))
UsedHabitat2 <- rgdal::readOGR(dsn=".", layer="Useable_corrected")
traps<- traps(Ald21CH)
clippedmask <- make.mask(traps(Ald21CH), buffer=347, spacing=15, type="trapbuffer", poly=UsedHabitat2, poly.habitat=TRUE)
DENScov5 <- rgdal::readOGR(dsn=".", layer="NewCov_Burn_2021", stringsAsFactors =FALSE)
DENScov5@data$Ceanothus2 <- as.factor(DENScov5@data$Ceanothus2)
DENScov5@data$Canopy3cat <- as.factor(DENScov5@data$Canopy3cat)
clippedmask <- addCovariates (clippedmask, DENScov5, columns = 'Ceanothus2', strict = FALSE, replace = FALSE)
clippedmask <- addCovariates (clippedmask, DENScov5, columns = 'Canopy3cat', strict = FALSE, replace = FALSE)
base.args <- list(capthist = Ald21CH, mask = clippedmask, trace = FALSE, verify=TRUE, method='Nelder-Mead', details = list(nsim = 10000))
args.0 <- c(base.args, model = D ~ 1, g0 ~ 1)
args.Dcan <- c(base.args, model = D ~ Canopy3cat, g0 ~ 1)
args.Dcean <- c(base.args, model = D ~ Ceanothus2, g0 ~ 1)
args.Dceancan <- c(base.args, model = D ~ Ceanothus2:Canopy3cat, g0 ~ 1)
args.Dadd <- c(base.args, model = D ~ Ceanothus2+Canopy3cat, g0 ~ 1)
arglist <- list(null = args.0, Dcan = args.Dcan, Dcean = args.Dcean, Dceancan = args.Dceancan, Dadd = args.Dadd)
fits <- par.secr.fit(arglist,ncores = 4)
covAIC <- AIC(fits, criterion = "AIC", verify=TRUE)
Canopy3cat <- c(0,0,1,1,2,2)
Ceanothus2 <- c(4,5,4,5,4,5)
Canopy3cat <- as.factor(Canopy3cat)
Ceanothus2 <- as.factor(Ceanothus2)
covs <- data.frame(Canopy3cat, Ceanothus2)
covs
CIa21 <- predict(fits$fit.Dadd, newdata=covs,
                 realnames=NULL, type=c("response"),
                 se.fit=TRUE, alpha=0.05, savenew=FALSE)



This question was answered elsewhere