Thank you very much!
It helps a lot. You are right about the NA´s in the coefficients, the model
needs some simplification.
Thank you again.
-------------------------------------------------------------------------------------------------------------------------------
Pablo Pita Orduna
Grupo de Recursos Marinos y Pesquerías.
Universidade de A Coruña. Campus da Zapateira s/n. E-15071. A Coruña, Spain.
Tel. +34(981) 167000 ext. 2204 Fax. +34(981) 167065.
www.fismare.net/verdeprofundo
www.recursosmarinos.net
----- Original Message -----
From: joris meys
To: Pablo Pita Orduna
Cc: r-help@r-project.org ; jfre...@udc.es ; ppitaord...@gmail.com
Sent: Saturday, March 07, 2009 12:28 AM
Subject: Re: [R] Interpreting GLM coefficients
One thing I notice immediately is a number of NA values for your
coefficients. If I were you, I would try a model with less parameters, and use
the anova() function to compare models, to see if the extra terms really
improve the model.
e.g.
fit1 <- glm(Y~X1+X2+X3,...)
fit2 <- glm(Y~X1+X2+X3+X1:X2,...)
anova(fit1, fit2, test="F")
If you checked all these, understanding the interaction terms will be most
easy if you normalized your numeric data before the analysis. For the
interpretations, you just fill in some values to get an idea. For example :
given the model : Y= a+b1*X1+b2*X2+b3*X1*X2
Say X1 and X2 are numeric :
interpretation of the main term : "Y increases with b2 for an increase of 1
unit in X2, given X1 is average".
interpretation of the interaction term : "For an X1 value of n units from the
mean, X2 increases with b2+n*b3" (n is negative when value is lower than the
mean).
In a Y ~ X2 plot, you can make this visible by plotting 3 different functions
: one for a low X1 value, one for an average X1 value and one for a high X1
value. This gives you an indication of the effect of X1 on X2.
for an interaction between a categorical terms or a categorical and a
numerical, you follow exact the same reasoning, but keep in mind that the
reference level represents a 0, and the mentioned level represents a 1. Fill in
the values in the equation, and you can understand the meaning of the terms.
Then again, you can plot a seperate function Y~X1 for every level of a certain
factor.
This isn't a straight answer on your question, but I'm afraid there is none.
I hope this helps you with building your model.
Kind regards.
Joris
On Fri, Mar 6, 2009 at 11:04 PM, Pablo Pita Orduna <pp...@udc.es> wrote:
Hi all,
I´m fitting GLM´s and I can´t interprete the coefficients when I run a
model with interaction terms.
When I run the simpliest model there is no problem:
Model1<-glm (Fishes ~ Year + I(Year^2) + Kind.Geographic + Kind.Fishers +
Zone.2 + Hours + Fishers + Month, family = poisson(log)) # Fishes, Year, Hours,
and Fishers are numeric, Kind.Geographic, Kind.Fishers, Zone.2 and Month are
factors with 4, 3, 5 and 12 levels respectively.
Model1$coefficients (whith Helmert contrasts):
(Intercept) Year IYear^2 Kind.Geographic1
Kind.Geographic2 Kind.Geographic3 Kind.Fishers1 Kind.Fishers2
Zone.21 Zone.22 Zone.23 Zone.24
-4.416915e+02 4.758455e-01 -1.270986e-04 -5.436199e-01
-1.068809e-01 -1.498580e-01 2.958462e-01 1.316589e-01
-1.328204e-01 -1.605802e-01 5.281869e-03 7.422885e-02
Hours Fishers Month1 Month2
Month3 Month4 Month5 Month6 Month7
Month8 Month9 Month10
9.772076e-02 -2.709955e-03 -1.586887e-01 -1.887837e-02
-5.183241e-03 5.870942e-02 7.075386e-02 2.061223e-02
7.372268e-03 -1.204835e-02 -5.047994e-03 2.441498e-02
Month11
-5.665261e-03
So I can write, for example:
y = -4.416915e+02 + -1.270986e-04*x^2 + 4.758455e-01*x # And add this
function to a plot(Year,Fishes).
My problem is to understand the coefficients for the model with interaction:
Model2<-glm(Fishes ~ Year + I(Year^2) + Kind.Geographic + Kind.Fishers +
Zone.2 + Hours + Fishers + Month + Year:Kind.Geographic + Year:Kind.Fishers +
Year:Zone.2 + Year:Hours + Year:Fishers + Year:Month + Kind.Geographic:Hours +
Kind.Fishers:Hours + Zone.2:Hours + Hours:Fishers + Hours:Month
+Kind.Geographic:Fishers + Zone.2:Fishers + Fishers:Month , poisson (log))
Model2$coefficients (with Helmert contrast):
(Intercept) Year I(Year^2)
Kind.Geographic1 Kind.Geographic2 Kind.Geographic3
Kind.Fishers1 Kind.Fishers2
1.641473e+03 -1.748703e+00 4.664752e-04
-6.721427e+00 1.856033e+01 -3.762727e-02
2.903564e+01 9.022858e+01
Zone.21 Zone.22 Zone.23
Zone.24 Hours Fishers
Month1 Month2
8.110814e-02 -1.902803e+01 8.335792e+00
-3.661641e+00 -7.824623e+00 7.088065e-01
2.479387e+03 8.346729e+02
Month3 Month4 Month5
Month6 Month7 Month8
Month9 Month10
4.052680e+02 2.384440e+02 1.570644e+02
1.032445e+02 7.930499e+01 6.487925e+01
5.592869e+01 3.888328e+01
Month11 Year:Kind.Geographic1 Year:Kind.Geographic2
Year:Kind.Geographic3 Year:Kind.Fishers1 Year:Kind.Fishers2
Year:Zone.21 Year:Zone.22
4.801656e+01 3.397984e-03 -9.443234e-03
NA -1.449305e-02 -4.470212e-02
-6.269309e-05 9.421045e-03
Year:Zone.23 Year:Zone.24 Year:Hours
Year:Fishers Year:Month1 Year:Month2
Year:Month3 Year:Month4
-4.184866e-03 1.854810e-03 3.257250e-03
-4.103058e-04 -1.264934e+00 -4.255907e-01
-2.069909e-01 -1.216459e-01
Year:Month5 Year:Month6 Year:Month7
Year:Month8 Year:Month9 Year:Month10
Year:Month11 Kind.Geographic1:Hours
-8.015823e-02 -5.278291e-02 -4.054404e-02
-3.313487e-02 -2.846036e-02 -1.973118e-02
-2.410902e-02 1.341231e-01
Kind.Geographic2:Hours Kind.Geographic3:Hours Kind.Fishers1:Hours
Kind.Fishers2:Hours Zone.21:Hours Zone.22:Hours
Zone.23:Hours Zone.24:Hours
5.806418e-02 NA 1.318444e-02
-1.234521e-01 7.961319e-04 1.622411e-02
-5.357266e-04 7.749412e-03
Hours:Fishers Hours:Month1 Hours:Month2
Hours:Month3 Hours:Month4 Hours:Month5
Hours:Month6 Hours:Month7
-6.805803e-03 7.971549e+00 2.627090e+00
1.403430e+00 7.747360e-01 5.163810e-01
3.732549e-01 2.827511e-01
Hours:Month8 Hours:Month9 Hours:Month10
Hours:Month11 Kind.Geographic1:Fishers Kind.Geographic2:Fishers
Kind.Geographic3:Fishers Zone.21:Fishers
2.091230e-01 1.057001e-01 1.159929e-01
NA -3.422994e-02 -3.421851e-03
NA -3.802354e-03
Zone.22:Fishers Zone.23:Fishers Zone.24:Fishers
Fishers:Month1 Fishers:Month2 Fishers:Month3
Fishers:Month4 Fishers:Month5
1.618358e-05 4.197350e-04 -8.773088e-05
7.806815e-01 2.509529e-01 1.280752e-01
7.520139e-02 5.370220e-02
Fishers:Month6 Fishers:Month7 Fishers:Month8
Fishers:Month9 Fishers:Month10 Fishers:Month11
3.786759e-02 2.936135e-02 2.381810e-02
2.422438e-02 NA NA
I would like to know if it is possible to extract the intercepts and the
slopes for Year and Year^2 for this Model2.
Thank you very much for your help.
________________________________________________________________________________
Pablo Pita Orduna
Grupo de Recursos Marinos y Pesquerías.
Universidade de A Coruña. Campus da Zapateira s/n. E-15071. A Coruña, Spain.
Tel. +34(981) 167000 ext. 2204 Fax. +34(981) 167065.
www.fismare.net/verdeprofundo
----------------------------------------------------------------
Correo enviado usando el servicio de Webmail de la UDC.
______________________________________________
R-help@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
[[alternative HTML version deleted]]
______________________________________________
R-help@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
