12/14/24, 1:51 PM Homework 5

Homework 5
STAT 511 – Regression
AUTOR/A FECHA DE PUBLICACIÓN
Maxwell Hahn 14 diciembre 2024, 13:47 EST

Question 1
Problem 1
Normal dist.
–

For a r.v. Y ∼ N (μ, σ )

2
, the pdf is
2
1 (y − μ)
2
f (y ∣ μ, σ ) = exp (− )
2
√ 2πσ 2 2σ

We rewrite as
2 2
y − 2yμ + μ 1
exp (− ) ×
2
2σ √ 2πσ 2

and
1 2 2
yμ − μ y 1
2
exp ( )exp (− ) ×
2 2
σ 2σ √ 2πσ 2

   
c(y,ϕ)

And we see that:

;
θ = μ

ϕ = σ; and
2

A(θ) =
1

2
μ
2
=
1

2
θ
2
.
N.B. dist.
–

For a r.v. Y ∼ NB(r, p) , the pmf is

y + r − 1
r y
f (y ∣ r, p) = ( )p (1 − p)
y

We rewrite as
y + r − 1
exp (r log p + y log(1 − p))( )
y

   
c(y,ϕ)

And we see that:

θ = log(1 − p) ;
ϕ = 1 ; and
A(θ) = −r log(1 − exp(θ)) .
Problem 2
localhost:7358 1/16
 12/14/24, 1:51 PM Homework 5
Normal dist.
–

′
μ = E[Y ] = A (θ) = θ

′′ 2
Var(Y ) = ϕA (θ) = σ

Var(Y ) 2
σ
V(μ) = = 2
= 1
ϕ σ

N.B. dist.
–

′ exp(θ)
E[Y ] = A (θ) = r
1−exp(θ)

′′ exp(θ)
Var(Y ) = ϕA (θ) = −r 2
(1−exp(θ))

Var(Y ) exp(θ)
V(μ) = = −r
ϕ (1−exp(θ)) 2

Problem 3
Normal dist.
–

g(μ) = μ with range . R

N.B. dist.
–

g(μ) = log(μ) with range . R

Problem 4
Normal dist.
–

d(y, μ) = 2{t(y, y) − t(y, μ)} =

2
1 1 (y − μ)
2 2
2 {[ − log(2πσ ) − 0] − [ − log(2πσ ) − ]}
2
2 2 2σ

2
(y − μ)
2
= 2 {+ } = (y − μ)
2
2σ

because is constant for normal distribution.

σ
2

N.B. dist.
–

d(y, μ) = 2{t(y, y) − t(y, μ)} =

y + r − 1 r y y + r − 1 r μ
2 {[ log ( ) + r log ( ) + y log ( )] − [ log ( ) + r log ( ) + y log ( )]}
y r + y r + y y r + μ r + μ

= 2 {[r log(r) − r log(r + y) + y log(y) − y log(r + y)] − [r log(r) − r log(r + μ) + y log(μ) − y log(r + μ)]}

r + μ y r + μ
= 2 {r log ( ) + y log ( ) + y log ( )}
r + y μ r + y

r + μ y(r + μ)
= 2 {r log ( ) + y log ( )}
r + y μ(r + y)

localhost:7358 2/16
 12/14/24, 1:51 PM Homework 5

Question 2
Problem 1
We see that Y i ∼ Bernoulli(π i ) and accordingly has pmf
yi 1−y i
π (1 − π i )
i

which we rewrite as
exp (y i log π i + (1 − y i ) log(1 − π i )) = exp (y i (log π i − log(1 − π i )) + log(1 − π i ))

Accordingly, we have
πi
θi
θ i = log ; ϕ i = 1; A(θ i ) = log(1 + e )
1 − πi

Problem 2
Part (a).
–

We see that
⋆
Pr(Y i = 1 ∣ x i ) = Pr(Y i ≥ τ ∣ xi )

and
Pr(Y i = 1 ∣ x i ) = Pr(α + β 1 x i1 + ⋯ + β p x ip + u i ≥ τ )

where ,
u i ∼ N (0, 1)

which we can rewrite as

Pr(Y i = 1 ∣ x i ) = Pr(−u i ≤ α + β 1 x i1 + ⋯ + β p x ip − τ )

Let β0 = α − τ . Then,
Pr(Y i = 1 ∣ x i ) = Φ(β 0 + β 1 x i1 + ⋯ + β p x ip )

as desired.
Part (b).
–

The link function for probit regression can be found by considering

g

g(π i ) = Φ(β 0 + β 1 x i1 + ⋯ + β p x ip )

as above.
We rewrite as
−1
β 0 + β 1 x i1 + ⋯ + β p x ip = Φ (π i )

and see that the link function is

−1
g(π i ) = Φ (π i )

localhost:7358 3/16
 Problem 3
12/14/24, 1:51 PM Homework 5

df <- read.csv("https://stats.idre.ucla.edu/stat/data/binary.csv")

logistic <- glm(admit ~ ., family = binomial(link = "logit"),

data = df)

summary(logistic)

Call:
glm(formula = admit ~ ., family = binomial(link = "logit"), data = df)

Coefficients:
Estimate Std. Error z value Pr(>|z|)
(Intercept) -3.449548 1.132846 -3.045 0.00233 **
gre 0.002294 0.001092 2.101 0.03564 *
gpa 0.777014 0.327484 2.373 0.01766 *
rank -0.560031 0.127137 -4.405 1.06e-05 ***
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

(Dispersion parameter for binomial family taken to be 1)

Null deviance: 499.98 on 399 degrees of freedom

Residual deviance: 459.44 on 396 degrees of freedom
AIC: 467.44

Number of Fisher Scoring iterations: 4

probit <- glm(admit ~ ., family = binomial(link = "probit"),

data = df)

summary(probit)

Call:
glm(formula = admit ~ ., family = binomial(link = "probit"),
data = df)

Coefficients:
Estimate Std. Error z value Pr(>|z|)
(Intercept) -2.0915037 0.6718360 -3.113 0.00185 **
gre 0.0013982 0.0006487 2.156 0.03112 *
gpa 0.4643598 0.1950263 2.381 0.01727 *
rank -0.3317117 0.0745524 -4.449 8.61e-06 ***
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

(Dispersion parameter for binomial family taken to be 1)

Null deviance: 499.98 on 399 degrees of freedom

Residual deviance: 459.48 on 396 degrees of freedom
AIC: 467.48

Number of Fisher Scoring iterations: 4

localhost:7358 4/16
12/14/24, 1:51 PM Homework 5
Logistic:
exp(−3.450 + 0.002 × gre + 0.777 × gpa − 0.560 × rank)
πi =
1 + exp(−3.450 + 0.002 × gre + 0.777 × gpa − 0.560 × rank)

Probit:
π i = Φ(−2.092 + 0.001 × gre + 0.464 × gpa − 0.332 × rank)

Generally speaking, the coefficients are pretty similar. All predictors (and intercept) are significant in each model, and the p-
values are also, generally speaking, similar.
Problem 4
p1 <- predict(logistic, type = "response")
p2 <- predict(probit, type = "response")
plot(p1, p2,
xlab = "Logistic probs",
ylab = "Probit probs")
abline(0, 1, col = "red")

As most points fall on or close to the red line, we see that most of the probabilities from each model are pretty similar to
each other.
Problem 5
set.seed(511)
sample(c("L","P"), size=1)

[1] "P"

We will use our probit model for this question.

localhost:7358 5/16
12/14/24, 1:51 PM Homework 5
We will use this function to represent the improvement in GRE by 100 in each case
improveGREby100 <- function(dat){
df = transform(dat, gre = gre + 100)
return(df)
}

Case 1:
current_gre650 <- data.frame(
gpa = 3.5, rank = 2, gre = 650
)

improve1 <- improveGREby100(current_gre650)

(currentprob = predict(probit, newdata = current_gre650, type = "response"))

1
0.4126149

(newprob = predict(probit, newdata = improve1, type="response"))

1
0.4677204

newprob - currentprob

1
0.05510552

Case 2:
current_gre450 <- data.frame(
gpa = 3.5, rank = 2, gre = 450
)

improve2 <- improveGREby100(current_gre450)

(currentprob = predict(probit, newdata = current_gre450, type = "response"))

1
0.3083728

(newprob = predict(probit, newdata = improve2, type="response"))

1
0.3591821

newprob - currentprob

1
0.05080932

In both cases, we see that the expected change in probability of admission if the student improves GRE score by 100 is
approximately 0.05.
localhost:7358 6/16
12/14/24, 1:51 PM Homework 5

Question 3
load("homework5.Rdata")
head(lc_danish,4)

Cases Pop Age City

1 11 3059 40-54 Fredericia
2 11 800 55-59 Fredericia
3 11 710 60-64 Fredericia
4 10 581 65-69 Fredericia

Problem 1
We assume ; therefore, the pmf is
Y i ∼ Pois(λ i ⋅ n i )

yi
(λ i ⋅ n i )
exp(−λ i ⋅ n i )
yi !

which we can simplify as

exp(y i log(λ i ⋅ n i ) − λ i ⋅ n i )

yi !

and, accordingly, we have

θ i = log(λ i ⋅ n i ); ϕ i = 1; A(θ i ) = exp(θ i ); c(y i , ϕ i ) = 1/(y i !)

Problem 2
lc_danish$logni <- log(lc_danish$Pop)

model <- glm(

Cases ~ City * Age + offset(logni),
family = "poisson",
data = lc_danish
)

model

Call: glm(formula = Cases ~ City * Age + offset(logni), family = "poisson",

data = lc_danish)

Coefficients:
(Intercept) CityHorsens CityKolding
-4.10264 -1.86606 0.09683
CityVejle Age40-54 Age55-59
-0.37955 -1.52530 -0.18407
Age60-64 Age65-69 Age70-74
-0.06473 0.04048 0.26809
CityHorsens:Age40-54 CityKolding:Age40-54 CityVejle:Age40-54
2.09376 -1.13520 -0.21508
CityHorsens:Age55-59 CityKolding:Age55-59 CityVejle:Age55-59
0.95705 -0.68721 -0.16547
CityHorsens:Age60-64 CityKolding:Age60-64 CityVejle:Age60-64
1.91385 -0.78037 0.11730

localhost:7358 7/16
12/14/24, 1:51 PM Homework 5
CityHorsens:Age65-69 CityKolding:Age65-69 CityVejle:Age65-69
1.50458 -0.19070 0.63347
CityHorsens:Age70-74 CityKolding:Age70-74 CityVejle:Age70-74
1.73347 -0.34732 0.00383

Degrees of Freedom: 23 Total (i.e. Null); 0 Residual

Null Deviance: 129.9
Residual Deviance: -5.773e-15 AIC: 144.4

summary(model)

Call:
glm(formula = Cases ~ City * Age + offset(logni), family = "poisson",
data = lc_danish)

Coefficients:
Estimate Std. Error z value Pr(>|z|)
(Intercept) -4.10264 0.31623 -12.974 < 2e-16 ***
CityHorsens -1.86606 0.77460 -2.409 0.015993 *
CityKolding 0.09683 0.42817 0.226 0.821094
CityVejle -0.37955 0.49280 -0.770 0.441190
Age40-54 -1.52530 0.43693 -3.491 0.000481 ***
Age55-59 -0.18407 0.43693 -0.421 0.673546
Age60-64 -0.06473 0.43693 -0.148 0.882234
Age65-69 0.04048 0.44721 0.091 0.927881
Age70-74 0.26809 0.43693 0.614 0.539496
CityHorsens:Age40-54 2.09376 0.87626 2.389 0.016874 *
CityKolding:Age40-54 -1.13520 0.72405 -1.568 0.116915
CityVejle:Age40-54 -0.21508 0.73059 -0.294 0.768463
CityHorsens:Age55-59 0.95705 0.92605 1.033 0.301383
CityKolding:Age55-59 -0.68721 0.63186 -1.088 0.276767
CityVejle:Age55-59 -0.16547 0.69038 -0.240 0.810580
CityHorsens:Age60-64 1.91386 0.87039 2.199 0.027888 *
CityKolding:Age60-64 -0.78037 0.64583 -1.208 0.226925
CityVejle:Age60-64 0.11730 0.65861 0.178 0.858647
CityHorsens:Age65-69 1.50458 0.89443 1.682 0.092535 .
CityKolding:Age65-69 -0.19070 0.61175 -0.312 0.755249
CityVejle:Age65-69 0.63347 0.64365 0.984 0.325026
CityHorsens:Age70-74 1.73348 0.87991 1.970 0.048832 *
CityKolding:Age70-74 -0.34732 0.62077 -0.559 0.575825
CityVejle:Age70-74 0.00383 0.67732 0.006 0.995488
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

(Dispersion parameter for poisson family taken to be 1)

Null deviance: 1.2991e+02 on 23 degrees of freedom

Residual deviance: -5.7732e-15 on 0 degrees of freedom
AIC: 144.39

Number of Fisher Scoring iterations: 3

We see that many of the candidate predictors and interactions are not significant.
Problem 3
model_wo_int <- glm(
Cases ~ City + Age + offset(logni),
family = "poisson",
localhost:7358 8/16
12/14/24, 1:51 PM Homework 5
data = lc_danish
)
summary(model_wo_int)

Call:
glm(formula = Cases ~ City + Age + offset(logni), family = "poisson",
data = lc_danish)

Coefficients:
Estimate Std. Error z value Pr(>|z|)
(Intercept) -4.21241 0.20819 -20.233 < 2e-16 ***
CityHorsens -0.33006 0.18150 -1.818 0.0690 .
CityKolding -0.37155 0.18781 -1.978 0.0479 *
CityVejle -0.27232 0.18785 -1.450 0.1472
Age40-54 -1.41965 0.25027 -5.672 1.41e-08 ***
Age55-59 -0.31864 0.25205 -1.264 0.2062
Age60-64 0.09896 0.23566 0.420 0.6745
Age65-69 0.34805 0.23343 1.491 0.1359
Age70-74 0.43721 0.23929 1.827 0.0677 .
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

(Dispersion parameter for poisson family taken to be 1)

Null deviance: 129.908 on 23 degrees of freedom

Residual deviance: 23.447 on 15 degrees of freedom
AIC: 137.84

Number of Fisher Scoring iterations: 5

anova(model_wo_int, model, test = "Chisq")

Analysis of Deviance Table

Model 1: Cases ~ City + Age + offset(logni)

Model 2: Cases ~ City * Age + offset(logni)
Resid. Df Resid. Dev Df Deviance Pr(>Chi)
1 15 23.448
2 0 0.000 15 23.448 0.07509 .
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
At α = 0.05 , we reject the null hypothesis that the interaction coefficients are all zero. Accordingly, we have evidence that
our model might benefit overall from including the interactions.
So, yes. The interaction effect between city and age is significant.
Problem 4
library(dplyr)

Attaching package: 'dplyr'

The following objects are masked from 'package:stats':

filter, lag

localhost:7358 9/16
12/14/24, 1:51 PM Homework 5
The following objects are masked from 'package:base':

intersect, setdiff, setequal, union

lc_danish <- lc_danish %>%

mutate(age.q = case_when(
Age == "40-54" ~ 40,
Age == "55-59" ~ 55,
Age == "60-64" ~ 60,
Age == "65-69" ~ 65,
Age == "70-74" ~ 70,
Age == ">74" ~ 75,
TRUE ~ NA_real_
))
lc_danish$age.q2 <- (lc_danish$age.q)**2

model_quantAge <- glm(

Cases ~ City + age.q + offset(logni),
family = "poisson",
data = lc_danish
)
summary(model_quantAge)

Call:
glm(formula = Cases ~ City + age.q + offset(logni), family = "poisson",
data = lc_danish)

Coefficients:
Estimate Std. Error z value Pr(>|z|)
(Intercept) -7.254622 0.352847 -20.560 <2e-16 ***
CityHorsens -0.309326 0.181428 -1.705 0.0882 .
CityKolding -0.354055 0.187705 -1.886 0.0593 .
CityVejle -0.249588 0.187772 -1.329 0.1838
age.q 0.047263 0.005442 8.685 <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: 129.908 on 23 degrees of freedom

Residual deviance: 44.618 on 19 degrees of freedom
AIC: 151.01

Number of Fisher Scoring iterations: 5

model_quantAgeQuad <- glm(

Cases ~ City + age.q + age.q2 + offset(logni),
family = "poisson",
data = lc_danish
)
summary(model_quantAgeQuad)

Call:
glm(formula = Cases ~ City + age.q + age.q2 + offset(logni),
family = "poisson", data = lc_danish)

Coefficients:

localhost:7358 10/16
12/14/24, 1:51 PM Homework 5
Estimate Std. Error z value Pr(>|z|)
(Intercept) -1.399e+01 1.723e+00 -8.123 4.57e-16 ***
CityHorsens -3.328e-01 1.815e-01 -1.834 0.0667 .
CityKolding -3.769e-01 1.878e-01 -2.007 0.0448 *
CityVejle -2.742e-01 1.878e-01 -1.460 0.1443
age.q 2.934e-01 6.098e-02 4.811 1.50e-06 ***
age.q2 -2.142e-03 5.264e-04 -4.069 4.73e-05 ***
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

(Dispersion parameter for poisson family taken to be 1)

Null deviance: 129.908 on 23 degrees of freedom

Residual deviance: 27.529 on 18 degrees of freedom
AIC: 135.92

Number of Fisher Scoring iterations: 5

anova(model_quantAge, model_quantAgeQuad, test = "Chisq")

Analysis of Deviance Table

Model 1: Cases ~ City + age.q + offset(logni)

Model 2: Cases ~ City + age.q + age.q2 + offset(logni)
Resid. Df Resid. Dev Df Deviance Pr(>Chi)
1 19 44.618
2 18 27.529 1 17.089 3.567e-05 ***
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Problem 5
best.model <- model_quantAgeQuad
summary(best.model)

Call:
glm(formula = Cases ~ City + age.q + age.q2 + offset(logni),
family = "poisson", data = lc_danish)

Coefficients:
Estimate Std. Error z value Pr(>|z|)
(Intercept) -1.399e+01 1.723e+00 -8.123 4.57e-16 ***
CityHorsens -3.328e-01 1.815e-01 -1.834 0.0667 .
CityKolding -3.769e-01 1.878e-01 -2.007 0.0448 *
CityVejle -2.742e-01 1.878e-01 -1.460 0.1443
age.q 2.934e-01 6.098e-02 4.811 1.50e-06 ***
age.q2 -2.142e-03 5.264e-04 -4.069 4.73e-05 ***
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

(Dispersion parameter for poisson family taken to be 1)

Null deviance: 129.908 on 23 degrees of freedom

Residual deviance: 27.529 on 18 degrees of freedom
AIC: 135.92

Number of Fisher Scoring iterations: 5

localhost:7358 11/16
12/14/24, 1:51 PM Homework 5
We will choose the last model that we fit (i.e., with a quantitative age predictor and its quadratic term along with city
variable) as our best model. We note that it also has the lowest AIC value, indicating that this might be the best-fitting model.
Additionally, we tested whether the quadratic term results in a significantly different model than without it, and we do have
evidence that the model with the quadratic term performs best.
We will now check the diagnostic plots for this best model to make sure that we are adhering to the assumptions of our
model.
Recall our GLM assumptions are:
Correct link function is used
Linearity - each predictor is included in the linear predictor in the correct form
Correct variance function is used
Dispersion parameter is constant and no overdispersion problem exists
Responses are independent of each other and come from the specified EDM
par(mfrow=c(2,2))
plot(residuals(best.model,type = "working")~predict(best.model,type = "link"),las=1,
main="Working responses residuals",
ylab="Working responses - linear predictors",
xlab="Fitted linear predictors" )
plot(resid(best.model) ~ predict(best.model,type = "link"), las=1,
main="Deviance residuals", ylab="Deviance residuals",
xlab="Fitted linear predictors" )
plot( cooks.distance(best.model), type="h", las=1,
ylab="D", main="Cook's distance")
qqnorm( resid(best.model), las=1,
main="Normal Q-Q plot\ndeviance residuals")
qqline(resid(best.model))

We see no major violations in the plots above.

We also check the dispersion parameter: ::: {.cell}
localhost:7358 12/16
12/14/24, 1:51 PM Homework 5

best.model$deviance / best.model$df.residual

[1] 1.529386

:::
There might be slight overdispersion; however, we do not find this too concerning as the value is not excessively greater
than 1.
We ultimately proceed with our assumptions being reasonably satisfied after viewing the diagnostics and choose this model
as our best model.

localhost:7358 13/16
12/14/24, 1:51 PM Homework 5

Question 4
load("homework5.Rdata")
head(pock,4)

Count Dilution
1 116 1
2 151 1
3 171 1
4 194 1

summary(pock)

Count Dilution
Min. : 5.00 Min. : 1.000
1st Qu.: 18.75 1st Qu.: 2.000
Median : 48.50 Median : 4.000
Mean : 71.04 Mean : 6.417
3rd Qu.:115.25 3rd Qu.: 8.000
Max. :259.00 Max. :16.000

pock$log2Dilution <- log(pock$Dilution, base = 2)

plot(pock$log2Dilution, pock$Count)

Problem 1
Pois.model <- glm(Count ~ log2Dilution, family = "poisson",
data = pock)

localhost:7358 14/16
12/14/24, 1:51 PM Homework 5
Pois.model

Call: glm(formula = Count ~ log2Dilution, family = "poisson", data = pock)

Coefficients:
(Intercept) log2Dilution
5.2679 -0.6809

Degrees of Freedom: 47 Total (i.e. Null); 46 Residual

Null Deviance: 2730
Residual Deviance: 290.4 AIC: 562.4

summary(Pois.model)

Call:
glm(formula = Count ~ log2Dilution, family = "poisson", data = pock)

Coefficients:
Estimate Std. Error z value Pr(>|z|)
(Intercept) 5.26793 0.02255 233.60 <2e-16 ***
log2Dilution -0.68094 0.01544 -44.09 <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: 2729.74 on 47 degrees of freedom

Residual deviance: 290.44 on 46 degrees of freedom
AIC: 562.42

Number of Fisher Scoring iterations: 4

Problem 2
We first check: ::: {.cell}
min(pock$Count)

[1] 5

:::
So we can reasonably use the Goodness of fit test.
(D = deviance(Pois.model))

[1] 290.4387

pchisq(D, df = df.residual(Pois.model), lower.tail = F)

[1] 3.285023e-37

Yes, there is significant evidence to suggest a lack of fit.

Problem 3
localhost:7358 15/16
12/14/24, 1:51 PM Homework 5

qPois.model <- glm(Count ~ log2Dilution, family = "quasipoisson",

data = pock)
qPois.model

Call: glm(formula = Count ~ log2Dilution, family = "quasipoisson",

data = pock)

Coefficients:
(Intercept) log2Dilution
5.2679 -0.6809

Degrees of Freedom: 47 Total (i.e. Null); 46 Residual

Null Deviance: 2730
Residual Deviance: 290.4 AIC: NA

summary(qPois.model)

Call:
glm(formula = Count ~ log2Dilution, family = "quasipoisson",
data = pock)

Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) 5.26793 0.05678 92.78 <2e-16 ***
log2Dilution -0.68094 0.03888 -17.51 <2e-16 ***
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

(Dispersion parameter for quasipoisson family taken to be 6.338981)

Null deviance: 2729.74 on 47 degrees of freedom

Residual deviance: 290.44 on 46 degrees of freedom
AIC: NA

Number of Fisher Scoring iterations: 4

The estimated coefficients are identical. The standard errors in the Poisson model are smaller than those of the quasi-
Poisson model.
Since there is overdispersion (as dispersion ≈ 6.339 > 1), the standard errors and p-values from the Poisson model are
underestimated. The quasi-Poisson model results are more reliable.

localhost:7358 16/16