Variance estimation problems negbin and varprop

[erex]

This is worrying. Appears a varprop issue was left unresolved at the beginning of April. Hence, models with negbin distributions produce absurd (I hope they are absurd) variance estimates

Results produced on 21June

Summary of uncertainty in a density surface model calculated
 by variance propagation.

Quantiles of differences between fitted model and variance model
    Min.  1st Qu.   Median     Mean  3rd Qu.     Max. 
-1.31031 -0.04789  0.17085  0.01322  0.23876  0.27258 

Approximate asymptotic confidence interval:
        2.5%         Mean        97.5% 
    29.57459   2015.81785 137399.06890 
(Using log-Normal approximation)

Point estimate                 : 2015.818 
Standard error                 : 20412.4 
Coefficient of variation       : 10.1261 

Results from same fitted model lifted from Duke workshop notes

summary(var_nb_xy_ms)

## Summary of uncertainty in a density surface model calculated
##  by variance propagation.
## 
## Quantiles of differences between fitted model and variance model
##       Min.    1st Qu.     Median       Mean    3rd Qu.       Max. 
## -5.596e-06 -1.399e-07 -2.096e-08  1.690e-07  1.982e-07  7.159e-06 
## 
## Approximate asymptotic confidence interval:
##     2.5%     Mean    97.5% 
## 1005.079 1589.217 2512.848 
## (Using log-Normal approximation)
## 
## Point estimate                 : 1589.217 
## Standard error                 : 328.2824 
## Coefficient of variation       : 0.2066

Threat from old 27March issue disable var.prop() for negbin models

• don't know if that action was taken
• whether taken or not, should practical be re-written so that negbin is not fit?

[dill]

I think this will require some maths to fix up (based on discussion with MVB).

I recommend for now we don't allow varprop+nb in Distance UI. I will look into whether we can do this in R once I am able to fix up the maths...

On Wed, 21 Jun 2017, at 16:00, erex wrote:

This is worrying. Appears a varprop issue was left unresolved at the beginning of April. Hence, models with negbin distributions produce absurd (I hope they are absurd) variance estimates

Results produced on 21June

Summary of uncertainty in a density surface model calculated
 by variance propagation.

Quantiles of differences between fitted model and variance model
    Min.  1st Qu.   Median     Mean  3rd Qu.     Max. 
-1.31031 -0.04789  0.17085  0.01322  0.23876  0.27258 

Approximate asymptotic confidence interval:
        2.5%         Mean        97.5% 
    29.57459   2015.81785 137399.06890 
(Using log-Normal approximation)

Point estimate                 : 2015.818 
Standard error                 : 20412.4 
Coefficient of variation       : 10.1261 

Results from same fitted model lifted from Duke workshop notes

summary(var_nb_xy_ms)

## Summary of uncertainty in a density surface model calculated
##  by variance propagation.
## 
## Quantiles of differences between fitted model and variance model
##       Min.    1st Qu.     Median       Mean    3rd Qu.       Max. 
## -5.596e-06 -1.399e-07 -2.096e-08  1.690e-07  1.982e-07  7.159e-06 
## 
## Approximate asymptotic confidence interval:
##     2.5%     Mean    97.5% 
## 1005.079 1589.217 2512.848 
## (Using log-Normal approximation)
## 
## Point estimate                 : 1589.217 
## Standard error                 : 328.2824 
## Coefficient of variation       : 0.2066

Threat from old 27March issue disable var.prop() for negbin models

• don't know if that action was taken
• whether taken or not, should practical be re-written so that negbin is not fit?

-- You are receiving this because you are subscribed to this thread. Reply to this email directly or view it on GitHub: https://github.com/DistanceDevelopment/spatial-workshops/issues/43

[dill]

Solved in dsm repo.