Message-ID: <7DF2365FF07C0E4E89419D65CCC93C9E01435C5A5C9B@EXCHANGE11.campus.tue.nl>
Date: 2009-01-07T06:43:55Z
From: Serebrenik, A.
Subject: Residual deviance (cross-post from sci.stat.consult)
Dear all,
I'm trying to fit a statistical model to series of measurements.
Unfortunately, my knowledge of statistics is rather limited, so I'm a
bit at loss of what is going on with the model.
First of all, I've prepared a histogram. Then, I've tried to fit a Poisson model to express the relation between the middle points of
classes (mids) and the corresponding frequencies (density). I've got the following Poisson models using R:
> summary(fmDP)
Deviance Residuals:
Min 1Q Median 3Q Max
-1.20831 -0.56363 -0.28010 0.08324 3.19099
Coefficients:
Estimate Std. Error z value Pr(>|z|)
(Intercept) 1.2495 0.1528 8.18 2.84e-16 ***
hD$mids -3.3683 0.4420 -7.62 2.53e-14 ***
---
Signif. codes: 0 ~Q***~R 0.001 ~Q**~R 0.01 ~Q*~R 0.05 ~Q.~R 0.1 ~Q ~R 1
(Dispersion parameter for poisson family taken to be 1)
Null deviance: 129.59 on 99 degrees of freedom
Residual deviance: 55.67 on 98 degrees of freedom
AIC: Inf
Number of Fisher Scoring iterations: 5
> summary(fmDPAll)
Deviance Residuals:
Min 1Q Median 3Q Max
-1.3687 -0.5672 -0.3386 0.0513 4.9680
Coefficients:
Estimate Std. Error z value Pr(>|z|)
(Intercept) 1.478 0.148 9.988 <2e-16 ***
hDAll$mids -4.327 0.501 -8.636 <2e-16 ***
---
Signif. codes: 0 ~Q***~R 0.001 ~Q**~R 0.01 ~Q*~R 0.05 ~Q.~R 0.1 ~Q ~R 1
(Dispersion parameter for poisson family taken to be 1)
Null deviance: 181.646 on 99 degrees of freedom
Residual deviance: 74.457 on 98 degrees of freedom
AIC: Inf
Number of Fisher Scoring iterations: 5
As 55.67 < 74.457 the first model seems to fit better than the second
one, but how good is it? Should I compare these residual deviances
with chi-square? Should I look for some other model with smaller
residual deviance?
Best regards,
Alexander