p values for dynamic occupancy models
Christina
Hello,
I ran parboot on a dynamic occupancy model after using mb.gof.test because it never gave a result for p.global, only a p value which equals 0. My dataframe, model, and parboot code is below, showing strong p values for all 3 fit statistics. Should I have run it on each season separately instead to get p values? Or does anyone have insight on why I didn’t get p. global values which would have shown p values for all 4 seasons? Thanks
>coyoteMS.umf <- unmarkedMultFrame(y= coyoteMS.data[,2:329],
siteCovs = coyoteMS.data[,c("natveg", "dev", "psize", "dist",”hp”)],
obsCovs=NULL,
yearlySiteCovs=NULL,
numPrimary=4)
>coyoteMS_global<-colext(psiformula= ~1, gammaformula = ~ 1, epsilonformula = ~ 1, pformula = ~natveg+dev+psize+dist+hp, coyoteMS.umf, se=TRUE)
>mb.gof.test(coyoteMS_global,nsim=1000)
Goodness-of-fit for dynamic occupancy model
Number of seasons: 4
Chi-square statistic:
Season 1 Season 2 Season 3 Season 4
7.619132e+09 2.132495e+44 1.193411e+48 2.267485e+22
Total chi-square = 1.193624e+48
Number of bootstrap samples = 1000
P-value = 0
Quantiles of bootstrapped statistics:
0% 25% 50% 75% 100%
6.3e+13 4.7e+17 3.9e+18 4.6e+19 1.8e+24
Estimate of c-hat = 1.577709e+26
> fitstats <- function(coyoteMS_global) {
+ observed <- getY(coyoteMS_global@data)
+ expected <- fitted(coyoteMS_global)
+ resids <- residuals(coyoteMS_global)
+ sse <- sum((resids^2),na.rm=TRUE)
+ chisq <- sum(((observed - expected)^2 / expected),na.rm=TRUE)
+ freeTuke <- sum(((sqrt(observed) - sqrt(expected))^2),na.rm=TRUE)
+ out <- c(SSE=sse, Chisq=chisq, freemanTukey=freeTuke)
+ return(out)
+ }
> parboot(coyoteMS_global,fitstats,nsim=1000,report=1)
t0 = 378.6994 8585.066 674.4327
Running in parallel on 7 cores. Bootstrapped statistics not reported.
Call: parboot(object = coyoteMS_global, statistic = fitstats, nsim = 1000, report = 1)
Parametric Bootstrap Statistics:
t0 mean(t0 - t_B) StdDev(t0 - t_B) Pr(t_B > t0)
SSE 379 -49.1 48.2 0.836
Chisq 8585 -44.0 408.8 0.513
freemanTukey 674 -16.5 64.8 0.608
t_B quantiles:
0% 2.5% 25% 50% 75% 97.5% 100%
SSE 288 333 395 428 458 519 586
Chisq 7583 7866 8343 8609 8886 9482 10188
freemanTukey 504 566 651 691 734 815 936
Marc J. Mazerolle
Hi Christina,
When you have a dynamic occupancy model fit with mb.gof.test( ), the function returns the global chi-square statistic and corresponding P-value. The global chi-square is computed by summing the chi-squares obtained for each season. This is exactly what you have in the output you included in your original message. If you look into the object a little further, you can also get the summary of observed vs predicted frequencies for each season separately. For example, if you stored the result of mb.gof.test( , nsim = 5000) in an object called outGOF, then you can access these season-specific chi-square tables with:
> dd$chisq.table$tables
These tables can help identify the seasons that are associated with the largest chi-square values. That being said, the huge values you are getting for each season suggest strong lack-of-fit. Also note that the second part of your code with the different test statistics you tried are not appropriate for a model based on detections/non-detections. For the chi-square to work with detection/non-detection data, you need to summarize information (e.g., for detection histories, or rows, or columns).
Good luck,
Marc
Marc J. Mazerolle
2405 rue de la Terrasse
Université Laval
Québec, Québec G1V OA6, Canada
Tel: (418) 656-2131 ext. 407120
Email: marc.ma...@sbf.ulaval.ca
Envoyé : 15 janvier 2021 14:59
À : unmarked
Objet : [unmarked] p values for dynamic occupancy models
You received this message because you are subscribed to the Google Groups "unmarked" group.
To unsubscribe from this group and stop receiving emails from it, send an email to unmarked+u...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/unmarked/f19be378-be7f-46eb-8cf2-80c95db16858n%40googlegroups.com.
Christina
Chi-square statistic = 2.70343e+152
Number of bootstrap samples = 1000
P-value = 0
Quantiles of bootstrapped statistics:
0% 25% 50% 75% 100%
6.8e+37 1.3e+46 3.3e+48 6.0e+50 1.2e+61
Estimate of c-hat = 1.944875e+94
thank you
Marc J. Mazerolle
Christina,
The chi-square is still huge and so is the c-hat estimate. I would not use models with such lack-of-fit for inferences. I think you should back up a bit and do some reading on the approaches you are using. There are many resources available online and in print.
Best,
Marc
Marc J. Mazerolle
2405 rue de la Terrasse
Université Laval
Québec, Québec G1V OA6, Canada
Tel: (418) 656-2131 ext. 407120
Email: marc.ma...@sbf.ulaval.ca
À : unmarked
Objet : Re: [unmarked] p values for dynamic occupancy models