[R-sig-ME] nested effects for only a small subset of situations

Robert Chatfield chatfield at alumni.rice.edu
Fri Apr 21 01:11:13 CEST 2017

I am estimating a model for fire emissions factors (slopes wrt x)

spectype.lmer <- lmer( y ~ x + 1 +  (x - 1 | species.type) + ( 1 | id)  , data=first.long.dat )

where  y actually represents a variety of (chemical) species, alll normalized by mean,

but there can be variations from one observati on  to the next among a few (6) different fire-types
“type.”  So I have oonxtructed the compound factor species.type as commonly described
in Bates’s writings.

id is just the sequential number of the observation of all 10 species. 

Now it turns out that for two of the species, there should be an extra term somthing
like   ( 1 | plume )    

spectype.plume.lmer <-  lmer( y ~ x + 1 + (x - 1 | species.type) + ( 1 | id) +  ( 1 | plume ) , data=first.long.dat ) 

where all the observations with  id  in that   plume (3 to 40 ids per plume)     should have the same value.  
id   is nested withiin   plume  .
A  dditionally, the intercept should depend only on species, not species.type

Only 2/10s of the observations require this value.
I want to “horde” degrees of freedom except for maybe some experimental tests to
see if the other species . Variance explained is pretty good even without including
this term.

This does not seem to fit wtihin the formula choices as I understand them.

(1) Can I create a factor CO.plume.offset which “activates” only when species is  CO  but
varies by plume, a level for each plume ?        (allso   a factor CH4. plume. offset )

(2) How about estimating the offset after spec.type.lmer is calculated using the
residuals.  Then I could re-enter the offsets in a second lmer calculation, by adjusting y .  This does
seem to reduce the meaning of statistical tests, but it’s very simple. The residuals for CO  and CH4 for
spectype.lmer are not high but the residuals corresponding to CH4 especially do seem
to move up and down from plume to plume.

Bob Chatfield

Robert Chatfield
Robert.B.Chatfield at nasa.gov <mailto:Robert.B.Chatfield at nasa.gov>
Earth Science, Atmospherics, NASA Ames
MS 245-5 
Moffett Field CA 94035-1000 USA
660-604-5490   Cell: 650-793-5227

	[[alternative HTML version deleted]]

More information about the R-sig-mixed-models mailing list