Jessica L Hite/hitejl/O/VCU
2009-Feb-23 15:24 UTC
[R] Follow-up to Reply: Overdispersion with binomial distribution
THANKS so very much for your help (previous and future!). I have a two
follow-up questions.
1) You say that dispersion = 1 by definition ....dispersion changes from 1
to 13.5 when I go from binomial to quasibinomial....does this suggest that
I should use the binomial? i.e., is the dispersion factor more important
that the
2) Is there a cutoff for too much overdispersion - mine seems to be
huge......Residual deviance: 1580.1 on 123 degrees of freedom
I do have some outliers - but they are legitimate (i.e., not typos).....
I included my data below....if it helps
summary(glm.D93)$dispersion ## 1 (by definition)
Call:
glm(formula = y ~ Pred1, family = "binomial")
Deviance Residuals:
Min 1Q Median 3Q Max
-9.940 -2.778 -0.710 2.130 10.479
Coefficients:
Estimate Std. Error z value Pr(>|z|)
(Intercept) 1.63942 0.07205 22.753 < 2e-16 ***
Pred1F -0.65228 0.11781 -5.537 3.08e-08 ***
Pred1O -3.03239 0.12782 -23.724 < 2e-16 ***
Pred1SN -3.60714 0.11057 -32.623 < 2e-16 ***
Pred1W -1.22131 0.10734 -11.378 < 2e-16 ***
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05
'.' 0.1 ' ' 1
(Dispersion parameter for binomial family taken to be 1)
Null deviance: 3506.7 on 127 degrees of freedom
Residual deviance: 1580.1 on 123 degrees of freedom
(1 observation deleted due to missingness)
AIC: 1863.1
Number of Fisher Scoring iterations: 5
> glm1<-glm(y~Pred1,"quasibinomial")
> summary(glm1)
Call:
glm(formula = y ~ Pred1, family = "quasibinomial")
Deviance Residuals:
Min 1Q Median 3Q Max
-9.940 -2.778 -0.710 2.130 10.479
Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) 1.6394 0.2646 6.197 7.96e-09 ***
Pred1F -0.6523 0.4326 -1.508 0.13415
Pred1O -3.0324 0.4693 -6.461 2.19e-09 ***
Pred1SN -3.6071 0.4060 -8.885 6.51e-15 ***
Pred1W -1.2213 0.3941 -3.099 0.00241 **
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05
'.' 0.1 ' ' 1
(Dispersion parameter for quasibinomial family taken to be 13.48239)
Null deviance: 3506.7 on 127 degrees of freedom
Residual deviance: 1580.1 on 123 degrees of freedom
(1 observation deleted due to missingness)
AIC: NA
Number of Fisher Scoring iterations: 5
DATA
Pred1 PercentSurvival
O 0.181818182
O 0
O 0
O 0
O 0
F 0.766666667
F 0.967741935
F 0.8
F 0.966666667
F 0.833333333
F 0.775
F 0.641025641
F 0.272727273
F 0.606060606
F 0.621621622
F 0.574468085
F 0.918918919
F 0.854166667
F 0.684210526
A 0.438596491
A 0.8
A 0.125
A 0.936507937
A 0.911764706
A 0.75
A 0
A 0.64
A 0.740740741
A 0.703703704
A 0.962962963
A 0.911111111
A 0.97260274
A 0.842105263
A 0.795454545
A 0.970588235
A 0.755555556
A 0.947368421
A 1
A 0.947368421
A 0.785714286
A 0.782178218
A 1
A 0.6
A 0.875
A 0.625
A 0.666666667
A 1
A 1
A 0.611111111
A 0.916666667
A 0.625
A 0.97826087
A 0.975
A 0.933333333
A 1
A 1
A 0.930232558
A 0.810810811
O 0
O 0
O 1
O 0
SN 0
SN 0
SN 0.696969697
SN 0
SN 0.533333333
SN 0
SN 0.027777778
SN 0.6
SN 0.052631579
SN 0
SN 0
SN 0
SN 0.619047619
SN 0
SN 0
SN 0
SN 0.08
SN 0
SN 0.090909091
SN 0
SN 0.5
SN 0
SN 0.78125
SN 0
SN 0
SN 0
SN 0
SN 0.542857143
SN 0
SN 0
SN 0
SN 0
SN 0.4
SN 0
SN 0
SN 0.433333333
O 0.655172414
O 0.238095238
O 0
O 0.409090909
O 0
O 0
O 0.090909091
O 0.310344828
O 0
O 0
O 0
O 0.783783784
W 0.928571429
W 0
W 0.651162791
W 0.3125
W 0.871794872
W 0.511627907
W 0.566666667
W 0.756756757
W 0
W 0.666666667
W 0.55
W 0.826086957
W 0.8
W 0.682926829
W 0.586206897
W 1
W 0.75
W 0.5625
[[alternative HTML version deleted]]
Michael Dewey
2009-Feb-24 18:15 UTC
[R] Follow-up to Reply: Overdispersion with binomial distribution
At 15:24 23/02/2009, Jessica L Hite/hitejl/O/VCU wrote:>THANKS so very much for your help (previous and future!). I have a two >follow-up questions. > >1) You say that dispersion = 1 by definition ....dispersion changes from 1 >to 13.5 when I go from binomial to quasibinomial....does this suggest that >I should use the binomial? i.e., is the dispersion factor more important >that the > >2) Is there a cutoff for too much overdispersion - mine seems to be >huge......Residual deviance: 1580.1 on 123 degrees of freedom >I do have some outliers - but they are legitimate (i.e., not typos)..... > >I included my data below....if it helpsIn your model you have y and Pred1 but in your dataset you have Pred1 and PercentSurvival so that is not the model you fitted. If you fitted PercentSurvival ~ Pred1 did you not get any warning?> summary(glm.D93)$dispersion ## 1 (by definition) > >Call: >glm(formula = y ~ Pred1, family = "binomial") > >Deviance Residuals: > Min 1Q Median 3Q Max >-9.940 -2.778 -0.710 2.130 10.479 > >Coefficients: > Estimate Std. Error z value Pr(>|z|) >(Intercept) 1.63942 0.07205 22.753 < 2e-16 *** >Pred1F -0.65228 0.11781 -5.537 3.08e-08 *** >Pred1O -3.03239 0.12782 -23.724 < 2e-16 *** >Pred1SN -3.60714 0.11057 -32.623 < 2e-16 *** >Pred1W -1.22131 0.10734 -11.378 < 2e-16 *** >--- >Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 > >(Dispersion parameter for binomial family taken to be 1) > > Null deviance: 3506.7 on 127 degrees of freedom >Residual deviance: 1580.1 on 123 degrees of freedom > (1 observation deleted due to missingness) >AIC: 1863.1 > >Number of Fisher Scoring iterations: 5 > > > glm1<-glm(y~Pred1,"quasibinomial") > > summary(glm1) > >Call: >glm(formula = y ~ Pred1, family = "quasibinomial") > >Deviance Residuals: > Min 1Q Median 3Q Max >-9.940 -2.778 -0.710 2.130 10.479 > >Coefficients: > Estimate Std. Error t value Pr(>|t|) >(Intercept) 1.6394 0.2646 6.197 7.96e-09 *** >Pred1F -0.6523 0.4326 -1.508 0.13415 >Pred1O -3.0324 0.4693 -6.461 2.19e-09 *** >Pred1SN -3.6071 0.4060 -8.885 6.51e-15 *** >Pred1W -1.2213 0.3941 -3.099 0.00241 ** >--- >Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 > >(Dispersion parameter for quasibinomial family taken to be 13.48239) > > Null deviance: 3506.7 on 127 degrees of freedom >Residual deviance: 1580.1 on 123 degrees of freedom > (1 observation deleted due to missingness) >AIC: NA > >Number of Fisher Scoring iterations: 5 > > >DATA > > >Pred1 PercentSurvival >O 0.181818182 >O 0 >O 0 >O 0 >O 0 >F 0.766666667 >F 0.967741935 >F 0.8 >F 0.966666667 >F 0.833333333 >F 0.775 >F 0.641025641 >F 0.272727273 >F 0.606060606 >F 0.621621622 >F 0.574468085 >F 0.918918919 >F 0.854166667 >F 0.684210526 >A 0.438596491 >A 0.8 >A 0.125 >A 0.936507937 >A 0.911764706 >A 0.75 >A 0 >A 0.64 >A 0.740740741 >A 0.703703704 >A 0.962962963 >A 0.911111111 >A 0.97260274 >A 0.842105263 >A 0.795454545 >A 0.970588235 >A 0.755555556 >A 0.947368421 >A 1 >A 0.947368421 >A 0.785714286 >A 0.782178218 >A 1 >A 0.6 >A 0.875 >A 0.625 >A 0.666666667 >A 1 >A 1 >A 0.611111111 >A 0.916666667 >A 0.625 >A 0.97826087 >A 0.975 >A 0.933333333 >A 1 >A 1 >A 0.930232558 >A 0.810810811 >O 0 >O 0 >O 1 >O 0 >SN 0 >SN 0 >SN 0.696969697 >SN 0 >SN 0.533333333 >SN 0 >SN 0.027777778 >SN 0.6 >SN 0.052631579 >SN 0 >SN 0 >SN 0 >SN 0.619047619 >SN 0 >SN 0 >SN 0 >SN 0.08 >SN 0 >SN 0.090909091 >SN 0 >SN 0.5 >SN 0 >SN 0.78125 >SN 0 >SN 0 >SN 0 >SN 0 >SN 0.542857143 >SN 0 >SN 0 >SN 0 >SN 0 >SN 0.4 >SN 0 >SN 0 >SN 0.433333333 >O 0.655172414 >O 0.238095238 >O 0 >O 0.409090909 >O 0 >O 0 >O 0.090909091 >O 0.310344828 >O 0 >O 0 >O 0 >O 0.783783784 >W 0.928571429 >W 0 >W 0.651162791 >W 0.3125 > >W 0.871794872 >W 0.511627907 >W 0.566666667 >W 0.756756757 >W 0 >W 0.666666667 >W 0.55 >W 0.826086957 >W 0.8 >W 0.682926829 >W 0.586206897 >W 1 >W 0.75 >W 0.5625 > > [[alternative HTML version deleted]]Michael Dewey http://www.aghmed.fsnet.co.uk