Strange N Values from model.average

[janiskpa]

Hi All,

I am running CMR models on manta ray population. So far, the values are as expected and made sense. However, I recently did model.average on the results. The averaged mdoels showed good and expected values for Phi, p, and pent. However, the N values didn't make sense. In the base model, where N estimates with the top model is ~800, the model averaging gave the N estimate only 95. In the Sex model, where N estimates are ~340 for F, and ~420 for M, the model averaging gave the N estimates of 41 & 48 for F & M.

This the model.average code I used:

Code: Select all

#averaging all models with weight > 0
results.chat.best.all <- remove.mark (results.chat, c(1, 2, 4, 6:8, 10, 12:14, 16, 18:20, 22, 24)) #Removing the models you dont want to average
results.chat.best.all
real.ave.all.Phi = model.average(results.chat.best.all, parameter="Phi", vcf=F)
real.ave.all.p = model.average(results.chat.best.all, parameter="p", vcf=F)
real.ave.all.pent = model.average(results.chat.best.all, parameter="pent", vcf=F)
real.ave.all.N = model.average(results.chat.best.all, parameter="N", vcf=F)


Does anyone have any insight on why it's giving strange N, but decent values for the other parameters?

Thanks!

[jlaake]

A search of this forum would have likely given you the answer. The parameter that is being model averaged is f0 and not N. This has evolved through time so even if it is called N it is actually the number missed. Add the numbers observed to the estimates and confidence intervals and you will have the values you want.

[janiskpa]

Great, thank you so much. I just saw the post on this issue and will follow the guides from there. Thanks!

[cooch]

A search of this forum would have likely given you the answer. The parameter that is being model averaged is f0 and not N. This has evolved through time so even if it is called N it is actually the number missed. Add the numbers observed to the estimates and confidence intervals and you will have the values you want.

Of course Jeff knows this, but...f0 (and encounter probabilities) are what MARK generates as 'reals'. N is now generated as a 'derived' parameter. So, N = M(t+1) + f0, where M(t+1) is a count, and f0 is estimated.

[janiskpa]

Hi there,
Thank you for the explanation. I have checked other posts regarding this topic too, but for some reason, I'm still not getting my N values... From the output of the model averaging that I did with models with a weight > 0, I printed only the N value, which I got a table with these columns:

N g1 a0 t2012

- estimates.par.index: 32
- estimates.estimate: 93.62566922
- estimates.se: 18.73934682
- estimates.lcl: 63.24486889
- estimates.ucl: 138.6004038
- estimates.fixed:
- estimates.group: 1
- estimates.age: 0
- estimates.time: 2012
- estimates.Age : 0
- estimates.Time: 0
- X32: 351.1631194

You mentioned:

N = M(t+1) + f0, where M(t+1) is a count, and f0 is estimated.

Meaning, f0 is the column "estimates.estimate" or "estimates.se"?
And I didn't understand which column M(t+1) was, or how to get that specific number.

Someone also said:

Add the numbers observed to the estimates and confidence intervals and you will have the values you want.

Meaning add estimate.estimate + estimate.se + lcl + ucl? If I do this, the value is still quite far from the actual value (~800).

I am quite new at CMR, apologies for the 'silly' questions. Thank you so much.

[jlaake]

Mt+1 is the total number of animals that you captured. f0 is the number estimated to have been missed during the capture process. The sum of those is your estimate of N. What is confusing you is that what is labeled N is actually f0. So yes the estimate under N is the estimate of f0 and the lcl and ucl are the confidence intervals for f0. Add Mt+1 to each of those values to get the estimate and confidence intervals for N. I doubt that you read someone like saying to add lcl and ucl which makes no sense.

[cooch]

Also, be sure to read section 14.10.1 in the 'Gentle Introduction'. For model-averaged N, the CI is not constructed simply as 1.98 x SE (because N is not a 'normal' parameter -- log-normal distribution, [0,>0], etc.).

[janiskpa]

Thank you! I've figured it out