No variance estimates for lambda0 in null model

48 views
Skip to first unread message

jcl...@utk.edu

unread,
Nov 8, 2022, 10:00:11 AM11/8/22
to secr
I am using openCR to estimate lambda for some robust design hair snare data using the PLBl model formulation with lambda0~1, phi~1, lambda~1, and sigma~1 and I get no SE estimates for lambda0...
Beta parameters (coefficients)
               beta    SE.beta         lcl        ucl
lambda0 -2.24694272        NaN         NaN        NaN
phi      1.83172859 0.14303517  1.55138482 2.11207237
lambda   0.01336562 0.01684289 -0.01964585 0.04637708
sigma    7.88270507 0.01393126  7.85540031 7.91000984

Eigenvalues :  1 0.85043 0.00976 0.43143
Numerical rank of Hessian : 4  ( svtol = 1e-05 )

Variance-covariance matrix of beta parameters
              lambda0           phi       lambda         sigma
lambda0 -4.577311e-04  4.152557e-05 1.749149e-06 -6.454664e-05
phi      4.152557e-05  2.045906e-02 9.891112e-04 -1.152844e-05
lambda   1.749149e-06  9.891112e-04 2.836830e-04  1.754022e-06
sigma   -6.454664e-05 -1.152844e-05 1.754022e-06  1.940800e-04

Fitted (real) parameters evaluated at base levels of covariates

 lambda0
 session estimate SE.estimate lcl ucl
    2007 0.105722          NA  NA  NA
    2008 0.105722          NA  NA  NA
    2009 0.105722          NA  NA  NA
    2010 0.105722          NA  NA  NA
    2011 0.105722          NA  NA  NA
    2012 0.105722          NA  NA  NA
    2013 0.105722          NA  NA  NA
    2014 0.105722          NA  NA  NA
    2015 0.105722          NA  NA  NA
    2016 0.105722          NA  NA  NA
    2017 0.105722          NA  NA  NA
    2018 0.105722          NA  NA  NA
    2019 0.105722          NA  NA  NA
    2020 0.105722          NA  NA  NA

The other estimates seem ok and the lambda0 estimate seems okay too, but there is no SE.  Anyone have any idea what is causing this?  Thanks!
Joe

Murray Efford

unread,
Nov 8, 2022, 3:14:12 PM11/8/22
to secr
A total mystery. Note negative variance estimate in vcov matrix. But Hessian appears to be full rank. Maybe you could send me the data offline?
Murray

Murray Efford

unread,
Nov 8, 2022, 3:27:18 PM11/8/22
to secr
You could also try recomputing the variances with openCR.fit(..., method = 'none', start = oldfit)

On Wednesday, November 9, 2022 at 4:00:11 AM UTC+13 jcl...@utk.edu wrote:

jcl...@utk.edu

unread,
Nov 8, 2022, 4:09:46 PM11/8/22
to secr
Will do.  Thanks.

Murray Efford

unread,
Nov 8, 2022, 7:58:08 PM11/8/22
to secr
Joe
When I run your constant-parameter model I get healthy estimates (below). I used a non-spatial fit as a starting point, but I can't see that would have changed much. I predict that you will get OK variances with method = 'none'. I did see some strange behavior during optimisation: it periodically tryied parameters that jumped the likelihood down to about -7171 from -7140. That may have made Newton-Raphson unstable (a case for Nelder-Mead?).
Murray

Beta parameters (coefficients)
              beta     SE.beta         lcl         ucl
lambda0 -2.2772345 0.007637203 -2.29220315 -2.26226587
phi      1.8417506 0.143763130  1.55998001  2.12352113
lambda   0.0145999 0.016824652 -0.01837581  0.04757561
sigma    7.8831279 0.014337079  7.85502769  7.91122801

Eigenvalues :  1 0.22784 0.2074 0.00236

Numerical rank of Hessian : 4  ( svtol = 1e-05 )

Variance-covariance matrix of beta parameters
              lambda0           phi        lambda         sigma
lambda0  5.832687e-05 -1.438050e-06 -5.769353e-07 -3.840832e-05
phi     -1.438050e-06  2.066784e-02  9.857146e-04 -1.544323e-05
lambda  -5.769353e-07  9.857146e-04  2.830689e-04  1.695365e-06
sigma   -3.840832e-05 -1.544323e-05  1.695365e-06  2.055518e-04


Fitted (real) parameters evaluated at base levels of covariates

 lambda0
 session  estimate  SE.estimate       lcl       ucl
    2007 0.1025675 0.0007833286 0.1010436 0.1041143
    2008 0.1025675 0.0007833286 0.1010436 0.1041143
    2009 0.1025675 0.0007833286 0.1010436 0.1041143
    2010 0.1025675 0.0007833286 0.1010436 0.1041143
    2011 0.1025675 0.0007833286 0.1010436 0.1041143
    2012 0.1025675 0.0007833286 0.1010436 0.1041143
    2013 0.1025675 0.0007833286 0.1010436 0.1041143
    2014 0.1025675 0.0007833286 0.1010436 0.1041143
    2015 0.1025675 0.0007833286 0.1010436 0.1041143
    2016 0.1025675 0.0007833286 0.1010436 0.1041143
    2017 0.1025675 0.0007833286 0.1010436 0.1041143
    2018 0.1025675 0.0007833286 0.1010436 0.1041143
    2019 0.1025675 0.0007833286 0.1010436 0.1041143
    2020 0.1025675 0.0007833286 0.1010436 0.1041143
Reply all
Reply to author
Forward
0 new messages