R help - Feb 2009 - using stepAIC with negative binomial regression - error message help

Dear List,
I am having problems running stepAIC with a negative binomial regression model. 
I am working with data on manta ray abundance, using 20 predictor variables. 
Predictors include variables for location (site), time (year, cos and sin of
calendar day, length of day, percent lunar illumination), oceanography (sea
surface temp mean and std, sea surface height mean and std), weather (cos wind
direction, sin wind direction, wind speed, temperature, barometric pressure),
and tides (dummy variables for high, falling, low, and rising).  So predictors
are binary, categorical, and continuous variables.  I ran glm.nb to fit my full
model containing all first order terms.  The model I ran was:
glm.nb.full<glm.nb(mantas~site+year+cosday+sinday+daylength+lunarpercent+sstmean+sststd+
sshmean+sshstd+cosdir+sindir+spd+temp+alt+tideht+high+falling+low+plankton)
 
However, when I use stepAIC on the model I get the message:
 
> stepAIC(glm.nb.full)
Start:  AIC=19240.46
mantas ~ site + year + cosday + sinday + daylength + lunarpercent + 
    sstmean + sststd + sshmean + sshstd + cosdir + sindir + spd + 
    temp + alt + tideht + high + falling + low + plankton
 
Error in dropterm.default(object, ...) : 
  number of rows in use has changed: remove missing values?
 
 
 
I know some of my predictors are missing values for certain dates.  I assume
this is what “number of rows in use has changed” means.  Must I remove all rows
that are missing values?  Or is there an option that does this for me?  Or am I
completely off and there is something else going on?  This is my first time
using R for statistics, so I am not sure if this is a mistake with my data or
with my use of R.  I would appreciate any insight into what I am doing wrong.
 
Thanks,
 
Tim
 
Tim Clark
PhD Candidate
Department of Zoology
University of Hawaii
Honolulu, HI  96816


      
	[[alternative HTML version deleted]]

t c <mudiver1200 <at> yahoo.com> writes:
> 
> Dear List,
> I am having problems running stepAIC with a 
> negative binomial regression model.? I am working with data on
> manta ray abundance, using 20 predictor variables.? 
 [snip]

The model I ran was:
> glm.nb.full<glm.nb(mantas~site+year+
  cosday+sinday+daylength+
  lunarpercent+sstmean+sststd+
> sshmean+sshstd+cosdir+sindir+spd+
  temp+alt+tideht+high+falling+low+plankton)
> ?
> However, when I use stepAIC on the model I get the message:
> ?
> > stepAIC(glm.nb.full)
> Start:? AIC=19240.46
> mantas ~ site + year + cosday + sinday + daylength + lunarpercent + 
> ??? sstmean + sststd + sshmean + sshstd + cosdir + sindir + spd + 
> ??? temp + alt + tideht + high + falling + low + plankton
> ?
> Error in dropterm.default(object, ...) : 
> ? number of rows in use has changed: remove missing values?

   Yes.  As the note in ?stepAIC says,

  The model fitting must apply the models to the same dataset.  This
     may be a problem if there are missing values and an 'na.action'
     other than 'na.fail' is used (as is the default in R). We suggest
     you remove the missing values first.

  The easiest way to remove missing values is with na.omit, e.g.

fullmod <- glm.nb(..., data=na.omit(mydata))
stepAIC(fullmod)

   aren't cosday, sinday, and daylength pretty strongly collinear ... ?

  Ben Bolker