[R] Convergence warnings from zeroinfl (package pscl)

[Daniel E. Bunker]

Dear R-Helpers,

Can anyone please help me to interpret warning messages from zeroinfl 
(package pscl) while fitting a zero inflated negative binomial model?

The console reports convergence and the parameters seam reasonable, but 
these

<<Warning messages:
1: algorithm did not converge in: glm.fit(X, Y, family = poisson())
2: fitted rates numerically 0 occurred in: glm.fit(X, Y, family = 
poisson()) >>

suggest that it did not converge.  (See full output below.)

Could some possibly help me to interpret these results?

Thanks for your time,

Dan


 > zip3=zeroinfl(count=round(modAbun*1000) ~ .,
+ x=~mod*intact,
+ z=~1,
+ dist="negbin",
+ trace=TRUE,
+ data=beetles)
Zero-Inflated Count Model
Using logit to model zero vs non-zero
Using Negative Binomial for counts
dependent variable y:
Y
      0     10     11     14     17     19     21     23     25     28 
    31     33     34     37     40     42     46
   1056      3      9      1      4      1      1      8      2      2 
     5      7      3      5      2     22      1
<<snip>>
385000 407000 416000 470250 519000 537750 560000 754680
      1      1      1      1      1      1      1      1
generating start values...done
MLE begins...
initial  value 10878.408362
iter   2 value 10600.864576
<<snip>>
iter  52 value 9531.456792
final  value 9531.456792
converged
done
Warning messages:
1: algorithm did not converge in: glm.fit(X, Y, family = poisson())
2: fitted rates numerically 0 occurred in: glm.fit(X, Y, family = 
poisson())
 > summary(zip3)
Zero-Inflated Count Model Summary
---------------------------

Call:
zeroinfl(count = round(modAbun * 1000) ~ ., x = ~mod * intact,
     z = ~1, data = beetles, dist = "negbin", trace = TRUE)

Total Log-likelihood: -9531.45679243448

  Zero-Inflated Model was fit with a logit link
Coefficients:
   Estimate Std. Error    z value   Pr(>|z|)
   -0.01853    0.05685   -0.32598    0.74444
---------------------------------------------------------------------------

Count Model (Negative Binomial)
Coefficients:
                  Estimate Std. Error  z value   Pr(>|z|)
(Intercept)      7.316434   0.217397  33.6547 2.659e-248
modes           -0.329699   0.363943  -0.9059  3.650e-01
modls            0.451599   0.267710   1.6869  9.162e-02
modmag           1.713005   0.296333   5.7807  7.440e-09
modpag           1.654261   0.266674   6.2033  5.529e-10
modsl            1.946231   0.308218   6.3145  2.711e-10
modacpas         2.379625   0.275172   8.6478  5.252e-18
intact          -0.002080   0.001198  -1.7368  8.241e-02
modes:intact     0.075579   0.020012   3.7768  1.589e-04
modls:intact     0.036582   0.006215   5.8862  3.952e-09
modmag:intact    0.005915   0.002770   2.1351  3.276e-02
modpag:intact    0.021262   0.004977   4.2721  1.937e-05
modsl:intact     0.008784   0.002573   3.4136  6.411e-04
modacpas:intact -0.009994   0.002735  -3.6546  2.576e-04
log(theta)      -1.405175   0.062412 -22.5145 2.993e-112

Theta = 0.2453



-- 
Daniel E. Bunker
BioMERGE Associate Director
Post-Doctoral Research Scientist
Columbia University
Department of Ecology, Evolution and Environmental Biology
1200 Amsterdam Avenue
New York, NY 10027-5557

deb37ATcolumbiaDOTedu
212-851-1888 phone
212-854-8188 fax