Generalized*Linear*Models* - Columbia Universitymadigan/W2025/notes/GLM.pdf · 2013. 2. 25. ·...

57
Generalized Linear Models based on Zuur et al.

Transcript of Generalized*Linear*Models* - Columbia Universitymadigan/W2025/notes/GLM.pdf · 2013. 2. 25. ·...

Page 1: Generalized*Linear*Models* - Columbia Universitymadigan/W2025/notes/GLM.pdf · 2013. 2. 25. · Observed data Weight cy n e u q re F 10 15 20 25 30 0 50 150 250 Observed data Weight

Generalized  Linear  Models  

based  on  Zuur  et  al.  

Page 2: Generalized*Linear*Models* - Columbia Universitymadigan/W2025/notes/GLM.pdf · 2013. 2. 25. · Observed data Weight cy n e u q re F 10 15 20 25 30 0 50 150 250 Observed data Weight

Kuhnert  and  Venables  

Page 3: Generalized*Linear*Models* - Columbia Universitymadigan/W2025/notes/GLM.pdf · 2013. 2. 25. · Observed data Weight cy n e u q re F 10 15 20 25 30 0 50 150 250 Observed data Weight

Normal  Distribu.on  example:  weights  of  1280  sparrows  

Sparrows<-­‐read.table("/Users/dbm/Documents/W2025/ZuurDataMixedModelling/Sparrows.txt",header=TRUE)  par(mfrow  =  c(2,  2))    hist(Sparrows$wt,  nclass  =  15,  xlab  =  "Weight",  main  =  "Observed  data",xlim=c(10,30))    hist(Sparrows$wt,  nclass  =  15,  xlab  =  "Weight",  main  =  "Observed  data",  freq  =  FALSE,xlim=c(10,30))    Y  <-­‐  rnorm(1281,  mean  =  mean(Sparrows$wt),  sd  =  sd(Sparrows$wt))    hist(Y,  nclass  =  15,  main  =  "Simulated  data",xlab  =  "Weight",xlim=c(10,30))      X  <-­‐seq(from  =  0,to  =  30,length  =  200)  Y  <-­‐  dnorm(X,  mean  =  mean(Sparrows$wt),  sd  =  sd(Sparrows$wt))  plot(X,  Y,  type  =  "l",  xlab  =  "Weight",  ylab  =  "Probabli]es",  ylim  =  c(0,  0.25),  xlim  =  c(10,  30),  main  =  "Normal  density  curve")  par(mfrow  =  c(1,  1))    

Page 4: Generalized*Linear*Models* - Columbia Universitymadigan/W2025/notes/GLM.pdf · 2013. 2. 25. · Observed data Weight cy n e u q re F 10 15 20 25 30 0 50 150 250 Observed data Weight

Observed data

Weight

Frequency

10 15 20 25 30

050

150

250

Observed data

Weight

Density

10 15 20 25 30

0.00

0.10

0.20

Simulated data

Weight

Frequency

10 15 20 25 30

050

100150200250

10 15 20 25 30

0.00

0.10

0.20

Normal density curve

Weight

Probablities

Page 5: Generalized*Linear*Models* - Columbia Universitymadigan/W2025/notes/GLM.pdf · 2013. 2. 25. · Observed data Weight cy n e u q re F 10 15 20 25 30 0 50 150 250 Observed data Weight
Page 6: Generalized*Linear*Models* - Columbia Universitymadigan/W2025/notes/GLM.pdf · 2013. 2. 25. · Observed data Weight cy n e u q re F 10 15 20 25 30 0 50 150 250 Observed data Weight

Poisson  Distribu.on  

In  prac]ce  the  variance  is  o`en  bigger  than  the  mean  -­‐>  overdispersion  

Page 7: Generalized*Linear*Models* - Columbia Universitymadigan/W2025/notes/GLM.pdf · 2013. 2. 25. · Observed data Weight cy n e u q re F 10 15 20 25 30 0 50 150 250 Observed data Weight
Page 8: Generalized*Linear*Models* - Columbia Universitymadigan/W2025/notes/GLM.pdf · 2013. 2. 25. · Observed data Weight cy n e u q re F 10 15 20 25 30 0 50 150 250 Observed data Weight

Gamma  Distribu.on  

Page 9: Generalized*Linear*Models* - Columbia Universitymadigan/W2025/notes/GLM.pdf · 2013. 2. 25. · Observed data Weight cy n e u q re F 10 15 20 25 30 0 50 150 250 Observed data Weight
Page 10: Generalized*Linear*Models* - Columbia Universitymadigan/W2025/notes/GLM.pdf · 2013. 2. 25. · Observed data Weight cy n e u q re F 10 15 20 25 30 0 50 150 250 Observed data Weight

Binomial  Distribu.on  

Page 11: Generalized*Linear*Models* - Columbia Universitymadigan/W2025/notes/GLM.pdf · 2013. 2. 25. · Observed data Weight cy n e u q re F 10 15 20 25 30 0 50 150 250 Observed data Weight
Page 12: Generalized*Linear*Models* - Columbia Universitymadigan/W2025/notes/GLM.pdf · 2013. 2. 25. · Observed data Weight cy n e u q re F 10 15 20 25 30 0 50 150 250 Observed data Weight

Natural  Exponen.al  Family  

to  get,  e.g.,  Poisson:  

Page 13: Generalized*Linear*Models* - Columbia Universitymadigan/W2025/notes/GLM.pdf · 2013. 2. 25. · Observed data Weight cy n e u q re F 10 15 20 25 30 0 50 150 250 Observed data Weight
Page 14: Generalized*Linear*Models* - Columbia Universitymadigan/W2025/notes/GLM.pdf · 2013. 2. 25. · Observed data Weight cy n e u q re F 10 15 20 25 30 0 50 150 250 Observed data Weight

Generalized  Linear  Models  

Page 15: Generalized*Linear*Models* - Columbia Universitymadigan/W2025/notes/GLM.pdf · 2013. 2. 25. · Observed data Weight cy n e u q re F 10 15 20 25 30 0 50 150 250 Observed data Weight
Page 16: Generalized*Linear*Models* - Columbia Universitymadigan/W2025/notes/GLM.pdf · 2013. 2. 25. · Observed data Weight cy n e u q re F 10 15 20 25 30 0 50 150 250 Observed data Weight
Page 17: Generalized*Linear*Models* - Columbia Universitymadigan/W2025/notes/GLM.pdf · 2013. 2. 25. · Observed data Weight cy n e u q re F 10 15 20 25 30 0 50 150 250 Observed data Weight

Generalized  Linear  Models:  Poisson  regression  

Page 18: Generalized*Linear*Models* - Columbia Universitymadigan/W2025/notes/GLM.pdf · 2013. 2. 25. · Observed data Weight cy n e u q re F 10 15 20 25 30 0 50 150 250 Observed data Weight

x  <-­‐  0:100  y  <-­‐  rpois(101,exp(0.01+0.03*x))  MyData  <-­‐  data.frame(x,y)  MyModel  <-­‐  glm(y~x,data=MyData,family=poisson)  summary(MyModel)  coefs  <-­‐  coef(MyModel)  plot(y~x,MyData,ylim=c(0,26))  par(new=TRUE)  plot(x,exp(0.01+0.03*x),ylim=c(0,26),type="l")  par(new=TRUE)  plot(x,exp(coefs[1]+coefs[2]*x),ylim=c(0,26),type="l",col="red")  

Es]mates  parameters  by  maximizing  the  likelihood:  

L = µi × e−µi

yi!i∏ , µi = exp(β0 + β1x1 +…+ βd xd )

Page 19: Generalized*Linear*Models* - Columbia Universitymadigan/W2025/notes/GLM.pdf · 2013. 2. 25. · Observed data Weight cy n e u q re F 10 15 20 25 30 0 50 150 250 Observed data Weight
Page 20: Generalized*Linear*Models* - Columbia Universitymadigan/W2025/notes/GLM.pdf · 2013. 2. 25. · Observed data Weight cy n e u q re F 10 15 20 25 30 0 50 150 250 Observed data Weight
Page 21: Generalized*Linear*Models* - Columbia Universitymadigan/W2025/notes/GLM.pdf · 2013. 2. 25. · Observed data Weight cy n e u q re F 10 15 20 25 30 0 50 150 250 Observed data Weight

   RoadKills  <-­‐  read.table("/Users/dbm/Documents/W2025/ZuurDataMixedModelling/RoadKills.txt",header=TRUE)  RK  <-­‐  RoadKills  #Saves  some  space  in  the  code      plot(RK$D.PARK,  RK$TOT.N,  xlab  =  "Distance  to  park",  ylab  =  "Road  kills")    M1  <-­‐  glm(TOT.N  ~  D.PARK,  family=poisson,  data=RK)  summary(M1)        M2  <-­‐  glm(TOT.N  ~  D.PARK,  family=gaussian,  data=RK)  summary(M2)  

Page 22: Generalized*Linear*Models* - Columbia Universitymadigan/W2025/notes/GLM.pdf · 2013. 2. 25. · Observed data Weight cy n e u q re F 10 15 20 25 30 0 50 150 250 Observed data Weight

Call:  glm(formula  =  TOT.N  ~  D.PARK,  family  =  poisson,  data  =  RK)    Deviance  Residuals:            Min              1Q      Median              3Q            Max      -­‐8.1100    -­‐1.6950    -­‐0.4708      1.4206      7.3337        Coefficients:                              Es]mate  Std.  Error  z  value  Pr(>|z|)          (Intercept)    4.316e+00    4.322e-­‐02      99.87      <2e-­‐16  ***  D.PARK            -­‐1.059e-­‐04    4.387e-­‐06    -­‐24.13      <2e-­‐16  ***  -­‐-­‐-­‐  Signif.  codes:    0  ‘***’  0.001  ‘**’  0.01  ‘*’  0.05  ‘.’  0.1  ‘  ’  1      (Dispersion  parameter  for  poisson  family  taken  to  be  1)            Null  deviance:  1071.4    on  51    degrees  of  freedom  Residual  deviance:    390.9    on  50    degrees  of  freedom  AIC:  634.29    Number  of  Fisher  Scoring  itera]ons:  4  

Page 23: Generalized*Linear*Models* - Columbia Universitymadigan/W2025/notes/GLM.pdf · 2013. 2. 25. · Observed data Weight cy n e u q re F 10 15 20 25 30 0 50 150 250 Observed data Weight

Residual  Deviance:      Difference  in  G2  =  -­‐2  log  L  between  a  maximal  model  where  every  observa]on  has  Poisson  parameter  equal  to  the  observed  value  and  your  model      Null  Deviance:      Difference  in  G2  =  -­‐2  log  L  between  a  maximal  model  where  every  observa]on  has  Poisson  parameter  equal  to  the  observed  value  and  the  model  with  just  an    intercept        Some]mes  useful  to  compare  residual  deviances  for  two  models.  Differences  are  chi-­‐square  distributed  with  degrees  of  freedom  equal  to  the  change  in  the  number  of  parameters    

Page 24: Generalized*Linear*Models* - Columbia Universitymadigan/W2025/notes/GLM.pdf · 2013. 2. 25. · Observed data Weight cy n e u q re F 10 15 20 25 30 0 50 150 250 Observed data Weight
Page 25: Generalized*Linear*Models* - Columbia Universitymadigan/W2025/notes/GLM.pdf · 2013. 2. 25. · Observed data Weight cy n e u q re F 10 15 20 25 30 0 50 150 250 Observed data Weight

MyData  <-­‐  data.frame(D.PARK  =  seq(from  =  0,  to  =  25000,  by  =  1000))  G  <-­‐  predict(M1,  newdata  =  MyData,  type  =  "link",  se  =  TRUE)  F  <-­‐  exp(G$fit)    FSEUP  <-­‐  exp(G$fit  +  1.96  *  G$se.fit)    FSELOW  <-­‐  exp(G$fit  -­‐  1.96  *  G$se.fit)    lines(MyData$D.PARK,  F,  lty  =  1)    lines(MyData$D.PARK,  FSEUP,  lty  =  2)    lines(MyData$D.PARK,  FSELOW,  lty  =  2)  

Page 26: Generalized*Linear*Models* - Columbia Universitymadigan/W2025/notes/GLM.pdf · 2013. 2. 25. · Observed data Weight cy n e u q re F 10 15 20 25 30 0 50 150 250 Observed data Weight

0 5000 10000 15000 20000 25000

020

4060

80100

Distance to park

Roa

d ki

lls

Page 27: Generalized*Linear*Models* - Columbia Universitymadigan/W2025/notes/GLM.pdf · 2013. 2. 25. · Observed data Weight cy n e u q re F 10 15 20 25 30 0 50 150 250 Observed data Weight

Model  Selec.on  in  Generalized  Linear  Models  

Page 28: Generalized*Linear*Models* - Columbia Universitymadigan/W2025/notes/GLM.pdf · 2013. 2. 25. · Observed data Weight cy n e u q re F 10 15 20 25 30 0 50 150 250 Observed data Weight

X

Y

235000

240000

245000

250000

255000

258500

library(AED)  library(lawce)    xyplot(Y  ∼  X,  aspect  =  "iso",  col  =  1,  pch  =  16,  data  =  RoadKills)    

Page 29: Generalized*Linear*Models* - Columbia Universitymadigan/W2025/notes/GLM.pdf · 2013. 2. 25. · Observed data Weight cy n e u q re F 10 15 20 25 30 0 50 150 250 Observed data Weight

Histogram of RK$POLIC

RK$POLIC

Frequency

0 4 8

010

2030

40

Histogram of RK$WAT.RES

RK$WAT.RES

Frequency

0 2 4 6

010

2030

Histogram of RK$URBAN

RK$URBAN

Frequency

0 10 20 30

05

1015

2025

3035

Histogram of RK$OLIVE

RK$OLIVE

Frequency

0 20 40 60

05

1015

2025

30

Histogram of RK$L.P.ROAD

RK$L.P.ROAD

Frequency

0.5 1.5 2.5

05

1015

2025

Histogram of RK$SHRUB

RK$SHRUB

Frequency

0.0 0.5 1.0 1.5

05

1015

Histogram of RK$D.WAT.COUR

RK$D.WAT.COUR

Frequency

0 400 800

02

46

810

Page 30: Generalized*Linear*Models* - Columbia Universitymadigan/W2025/notes/GLM.pdf · 2013. 2. 25. · Observed data Weight cy n e u q re F 10 15 20 25 30 0 50 150 250 Observed data Weight
Page 31: Generalized*Linear*Models* - Columbia Universitymadigan/W2025/notes/GLM.pdf · 2013. 2. 25. · Observed data Weight cy n e u q re F 10 15 20 25 30 0 50 150 250 Observed data Weight
Page 32: Generalized*Linear*Models* - Columbia Universitymadigan/W2025/notes/GLM.pdf · 2013. 2. 25. · Observed data Weight cy n e u q re F 10 15 20 25 30 0 50 150 250 Observed data Weight

How  to  select  variables  for  inclusion  in  the  model  

•  Hypothesis  tes]ng  •  Use  the  z-­‐sta]s]c  produced  by  summary  •  drop1(M2,test=“Chi”)  •  anova(M2,M3,test=“Chi”)  

•  AIC,  BIC,  etc.    -­‐  use  step(M2)  etc.  

Page 33: Generalized*Linear*Models* - Columbia Universitymadigan/W2025/notes/GLM.pdf · 2013. 2. 25. · Observed data Weight cy n e u q re F 10 15 20 25 30 0 50 150 250 Observed data Weight
Page 34: Generalized*Linear*Models* - Columbia Universitymadigan/W2025/notes/GLM.pdf · 2013. 2. 25. · Observed data Weight cy n e u q re F 10 15 20 25 30 0 50 150 250 Observed data Weight
Page 35: Generalized*Linear*Models* - Columbia Universitymadigan/W2025/notes/GLM.pdf · 2013. 2. 25. · Observed data Weight cy n e u q re F 10 15 20 25 30 0 50 150 250 Observed data Weight
Page 36: Generalized*Linear*Models* - Columbia Universitymadigan/W2025/notes/GLM.pdf · 2013. 2. 25. · Observed data Weight cy n e u q re F 10 15 20 25 30 0 50 150 250 Observed data Weight

Overdispersion  

Recall  that  in  the  Poisson  model  we  assume  the  mean  =  variance  

With  no  overdispersion,  residual  mean  deviance  should  be  close  to  1:                  270.23  /  42  =  6.43…not  close  to  1  

 Or,  consider  mean  squared  residuals:    

   sum(resid(M2,  type=“pearson”)^2)  /  42  =  5.93    (suggests  standard  errors  should  be  increased  by  sqrt(5.93)  =  2.44)      Quasi-­‐Poisson  model  accounts  for  overdispersion:        

Page 37: Generalized*Linear*Models* - Columbia Universitymadigan/W2025/notes/GLM.pdf · 2013. 2. 25. · Observed data Weight cy n e u q re F 10 15 20 25 30 0 50 150 250 Observed data Weight
Page 38: Generalized*Linear*Models* - Columbia Universitymadigan/W2025/notes/GLM.pdf · 2013. 2. 25. · Observed data Weight cy n e u q re F 10 15 20 25 30 0 50 150 250 Observed data Weight

Overdispersion  

AIC  not  defined  for  quasi-­‐Poisson  model    Use  drop1(M4,  test=“F”)  is  approximately  correct  or,  be~er,  do  cross-­‐valida]on  

Page 39: Generalized*Linear*Models* - Columbia Universitymadigan/W2025/notes/GLM.pdf · 2013. 2. 25. · Observed data Weight cy n e u q re F 10 15 20 25 30 0 50 150 250 Observed data Weight

0 5000 10000 15000 20000 25000

020

4060

80100

Distance to park

Road

kills

Page 40: Generalized*Linear*Models* - Columbia Universitymadigan/W2025/notes/GLM.pdf · 2013. 2. 25. · Observed data Weight cy n e u q re F 10 15 20 25 30 0 50 150 250 Observed data Weight

Residuals  for  Poisson  Model  

Because  variance  increases  with  the  mean,  standard  residual  doesn’t  make  much  sense.  Can  use  “Pearson  residuals”  

For  quasi-­‐Poisson  should  also  divide  by  the  square  root  of  the    overdispersion  parameter  

Page 41: Generalized*Linear*Models* - Columbia Universitymadigan/W2025/notes/GLM.pdf · 2013. 2. 25. · Observed data Weight cy n e u q re F 10 15 20 25 30 0 50 150 250 Observed data Weight

Nega.ve  Binomial  GLM  

Page 42: Generalized*Linear*Models* - Columbia Universitymadigan/W2025/notes/GLM.pdf · 2013. 2. 25. · Observed data Weight cy n e u q re F 10 15 20 25 30 0 50 150 250 Observed data Weight
Page 43: Generalized*Linear*Models* - Columbia Universitymadigan/W2025/notes/GLM.pdf · 2013. 2. 25. · Observed data Weight cy n e u q re F 10 15 20 25 30 0 50 150 250 Observed data Weight
Page 44: Generalized*Linear*Models* - Columbia Universitymadigan/W2025/notes/GLM.pdf · 2013. 2. 25. · Observed data Weight cy n e u q re F 10 15 20 25 30 0 50 150 250 Observed data Weight
Page 45: Generalized*Linear*Models* - Columbia Universitymadigan/W2025/notes/GLM.pdf · 2013. 2. 25. · Observed data Weight cy n e u q re F 10 15 20 25 30 0 50 150 250 Observed data Weight

M7  <-­‐  glm.nb(formula  =  TOT.N  ~  OPEN.L  +  L.WAT.C  +  SQ.LPROAD  +  D.PARK,    data  =  RK)    plot(M7)  

Page 46: Generalized*Linear*Models* - Columbia Universitymadigan/W2025/notes/GLM.pdf · 2013. 2. 25. · Observed data Weight cy n e u q re F 10 15 20 25 30 0 50 150 250 Observed data Weight

1.5 2.0 2.5 3.0 3.5 4.0 4.5

-3-2

-10

12

Predicted values

Residuals

Residuals vs Fitted

2

1

39

-2 -1 0 1 2

-3-2

-10

12

Theoretical Quantiles

Std

. dev

ianc

e re

sid.

Normal Q-Q

2

1

37

1.5 2.0 2.5 3.0 3.5 4.0 4.5

0.0

0.5

1.0

1.5

Predicted values

Std. deviance resid.

Scale-Location21

37

0.00 0.05 0.10 0.15 0.20 0.25

-2-1

01

2

Leverage

Std

. Pea

rson

resi

d.

Cook's distance

0.5Residuals vs Leverage

19

12

Page 47: Generalized*Linear*Models* - Columbia Universitymadigan/W2025/notes/GLM.pdf · 2013. 2. 25. · Observed data Weight cy n e u q re F 10 15 20 25 30 0 50 150 250 Observed data Weight

1.5 2.0 2.5 3.0 3.5 4.0 4.5

-8-6

-4-2

02

4

Predicted values

Residuals

Residuals vs Fitted

2

137

-2 -1 0 1 2

-3-2

-10

12

3

Theoretical Quantiles

Std

. dev

ianc

e re

sid.

Normal Q-Q

2

1

8

1.5 2.0 2.5 3.0 3.5 4.0 4.5

0.0

0.5

1.0

1.5

Predicted values

Std. deviance resid.

Scale-Location2

18

0.0 0.1 0.2 0.3 0.4 0.5 0.6

-3-2

-10

12

3

Leverage

Std

. Pea

rson

resi

d.

Cook's distance

1

0.5

0.5

1

Residuals vs Leverage

8

12

Contrast  with  quasi-­‐Poisson  model  M4  

Page 48: Generalized*Linear*Models* - Columbia Universitymadigan/W2025/notes/GLM.pdf · 2013. 2. 25. · Observed data Weight cy n e u q re F 10 15 20 25 30 0 50 150 250 Observed data Weight

What  if  zero  is  impossible?  

Page 49: Generalized*Linear*Models* - Columbia Universitymadigan/W2025/notes/GLM.pdf · 2013. 2. 25. · Observed data Weight cy n e u q re F 10 15 20 25 30 0 50 150 250 Observed data Weight
Page 50: Generalized*Linear*Models* - Columbia Universitymadigan/W2025/notes/GLM.pdf · 2013. 2. 25. · Observed data Weight cy n e u q re F 10 15 20 25 30 0 50 150 250 Observed data Weight

Library(VGAM)  data(Snakes)  M3A  <-­‐  vglm(N_days~PDayRain  +  Tot_Rain  +  Road_Loc  +  PDayRain:Tot_Rain,  family  =  posnegbinomial,  data  =  Snakes)      M4A  <-­‐  vglm(N_days~PDayRain  +  Tot_Rain  +  Road_Loc  +  PDayRain:Tot_Rain,  family  =  pospoisson,  data  =  Snakes)    

Page 51: Generalized*Linear*Models* - Columbia Universitymadigan/W2025/notes/GLM.pdf · 2013. 2. 25. · Observed data Weight cy n e u q re F 10 15 20 25 30 0 50 150 250 Observed data Weight

Too  many  zeroes  much  more  common  

try    plot(table(rpois(10000,lambda)))  

Page 52: Generalized*Linear*Models* - Columbia Universitymadigan/W2025/notes/GLM.pdf · 2013. 2. 25. · Observed data Weight cy n e u q re F 10 15 20 25 30 0 50 150 250 Observed data Weight
Page 53: Generalized*Linear*Models* - Columbia Universitymadigan/W2025/notes/GLM.pdf · 2013. 2. 25. · Observed data Weight cy n e u q re F 10 15 20 25 30 0 50 150 250 Observed data Weight

Zero-­‐altered  Models  

Page 54: Generalized*Linear*Models* - Columbia Universitymadigan/W2025/notes/GLM.pdf · 2013. 2. 25. · Observed data Weight cy n e u q re F 10 15 20 25 30 0 50 150 250 Observed data Weight

Zero-­‐altered  Models  

Get  the  es.mates  of  γ  and  β  via  maximum  likelihood  es.ma.on  

   library(pscl)  f1  <-­‐  formula(Intensity~fArea*fYear  +  Length  |  fArea  *  fYear  +  Length)  H1A  <-­‐  hurdle(f1,  dist  =  "poisson",  link  =  "logit",  data  =  ParasiteCod2)  H1B  <-­‐  hurdle(f1,  dist  =  "negbin",  link  =  "logit",  data  =  ParasiteCod2)  AIC(H1A,H1B)  

logis]c  part  count  part  

Can  s]ll  have  overdispersion  in  ZIP/ZAP  models  so  check  ZINB/ZANB  

Page 55: Generalized*Linear*Models* - Columbia Universitymadigan/W2025/notes/GLM.pdf · 2013. 2. 25. · Observed data Weight cy n e u q re F 10 15 20 25 30 0 50 150 250 Observed data Weight

Zero-­‐inflated  Models  

Page 56: Generalized*Linear*Models* - Columbia Universitymadigan/W2025/notes/GLM.pdf · 2013. 2. 25. · Observed data Weight cy n e u q re F 10 15 20 25 30 0 50 150 250 Observed data Weight
Page 57: Generalized*Linear*Models* - Columbia Universitymadigan/W2025/notes/GLM.pdf · 2013. 2. 25. · Observed data Weight cy n e u q re F 10 15 20 25 30 0 50 150 250 Observed data Weight