Scribd
Upload
116 views12 pages

Generalized Linear Models

Generalized linear models (GLM) are made up of three components: a random component that identifies the dependent variable and its probability distribution, a systematic component that identifies explanatory variables, and a link function that connects the mean of the dependent variable to the explanatory variables. Common GLM examples include logistic regression for binary outcomes, Poisson regression for count data, and linear regression for continuous normally distributed data. Parameters in GLM are estimated using statistical software and tested using techniques like Wald tests, likelihood ratio tests, and odds ratios.

Uploaded by

Andika Afriansyah
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPT, PDF, TXT or read online on Scribd
116 views12 pages

Generalized Linear Models

Generalized linear models (GLM) are made up of three components: a random component that identifies the dependent variable and its probability distribution, a systematic component that identifies explanatory variables, and a link function that connects the mean of the dependent variable to the explanatory variables. Common GLM examples include logistic regression for binary outcomes, Poisson regression for count data, and linear regression for continuous normally distributed data. Parameters in GLM are estimated using statistical software and tested using techniques like Wald tests, likelihood ratio tests, and odds ratios.

Uploaded by

Andika Afriansyah
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPT, PDF, TXT or read online on Scribd
You are on page 1/ 12

Generalized Linear Models

Generalized Linear Models (GLM)

General class of linear models that are made up of 3


components: Random, Systematic, and Link unction
!
Random component: "dentifies dependent #aria$le
(Y) and its pro$a$ility distri$ution
!
Systematic %omponent: "dentifies the set of
e&planatory #aria$les (X
'
,(((,X
k
)
!
Link unction: "dentifies a function of the mean
that is a linear function of the e&planatory
#aria$les
k k
X X g + + + =
1 1
) (
Random %omponent

%onditionally Normally distri$uted response )ith constant


standard de#iation * Re+ression models )e ha#e fit so far(

,inary outcomes (Success or ailure)* Random


component has Binomial distri$ution and model is called
Logistic Regression(

%ount data (num$er of e#ents in fi&ed area and-or len+th


of time)* Random component has Poisson distri$ution and
model is called Poisson Regression

.hen %ount data ha#e /(0) 1 2(0), model fit can $e


Negative Binomial Regression

%ontinuous data )ith ske)ed distri$ution and #ariation


that increases )ith the mean can $e modeled )ith a
Gamma distri$ution
%ommon Link unctions

"dentity link (form used in normal and gamma


re+ression models):

Lo+ link (used )hen cannot $e ne+ati#e as


)hen data are Poisson counts):

Lo+it link (used )hen is $ounded $et)een 3 and


' as )hen data are $inary):
= ) ( g
) log( ) ( = g

1
log ) ( g
Lo+istic Re+ression

Lo+istic Re+ression * 4ichotomous Response


#aria$le and numeric and-or cate+orical e&planatory
#aria$le(s)
!
Goal: Model the pro$a$ility of a particular outcome as a
function of the predictor #aria$le(s)
!
5ro$lem: 5ro$a$ilities are $ounded $et)een 3 and '

4istri$ution of Responses: ,inomial

Link unction:

1
log ) ( g
Lo+istic Re+ression )ith ' 5redictor

Response - Presence/Absence of characteristic

Predictor - Numeric variable observed for each case

Model - (x) Probabilit of presence at predictor level x


x
x
e
e
x
1 !
1 !
1
) (

+
+
+
=

1
" ! P(Presence) is the same at each level of x

1
# ! P(Presence) increases as x increases

1
$ ! P(Presence) decreases as x increases
Lo+istic Re+ression )ith ' 5redictor

0
,
1
are unkno)n parameters and must $e estimated
usin+ statistical soft)are such as S5SS, S6S, R or
S7676 (or in a matri& lan+ua+e)
8
5rimary interest in estimatin+ and testin+ hypotheses
re+ardin+
1
8
Lar+e*Sample test (.ald 7est):
8 H
3
:
1
9 3 H
6
:
1
3
) ( %
% & &
% & &
' '
'
1 (
'
'
)
1
)
'
1
)
obs
obs
obs
X P val P
X R R
X S T

:ote: Some soft)are packa+es


perform this as an e;ui#alent <*
test or t*test
=dds Ratio

"nterpretation of Re+ression %oefficient ():


!
"n linear re+ression, the slope coefficient is the chan+e in the
mean response as x increases $y ' unit
! "n lo+istic re+ression, )e can sho) that:

= =
+
) ( 1
) (
) (
) (
) 1 (
x
x
x odds e
x odds
x odds

*hus e

represents the change in the odds of the outcome


(multiplicativel) b increasing x b 1 unit

+f " !( the odds and probabilit are the same at all x levels (e

"1)

+f # ! ( the odds and probabilit increase as x increases (e

#1)

+f $ ! ( the odds and probabilit decrease as x increases (e

$1)
>?@ %onfidence "nter#al for =dds Ratio
Step ': %onstruct a >?@ %" for :

+
) ) )
) ) ) ) ) )
,- & 1 ( ,- & 1 ,- & 1

.tep '% Raise e " '&/10 to the lo1er and upper bounds of the 2+%

+
)
) )
)
) )
,- & 1 ,- & 1
(


e e

+f entire interval is above 1( conclude positive association

+f entire interval is belo1 1( conclude negative association

+f interval contains 1( cannot conclude there is an association


Multiple Lo+istic Re+ression
2&tension to more than one predictor #aria$le (either numeric or
dummy #aria$les)(
.ith k predictors, the model is )ritten:
k k
k k
x x
x x
e
e

+ + +
+ + +
+
=

1 1 !
1 1 !
1
Ad3usted 4dds ratio for raising x
i
b 1 unit( holding
all other predictors constant%
i
e OR
i

=

Man models have nominal/ordinal predictors( and


1idel ma5e use of dumm variables
7estin+ Re+ression %oefficients

7estin+ the o#erall model:


) (
& &
)) log( ' ( )) log( ' ( & &
! all Not %
! %
' '
'
(
'
1 !
'
1 !
obs
k obs
obs
i A
k
X P P
X R R
L L X S T
H
H
=

=
=
= = =

L
!
, L
1
are values of the ma6imi7ed li5elihood function( computed b
statistical soft1are pac5ages& *his logic can also be used to compare
full and reduced models based on subsets of predictors& *esting for
individual terms is done as in model 1ith a single predictor&
5oisson Re+ression

Generally used to model %ount data

4istri$ution: 5oisson (Restriction: 2(0)9/(0))

Link unction: %an $e identity link, $ut typically use the


lo+ link:

( )
k k
X X
k
k k
e X X
X X g


+ + +
=
+ + + = =
&&&
1
1 1 !
1 1 !
(&&&(
&&& ) ln( ) (
7ests are conducted as in Lo+istic re+ression
.hen the mean and #ariance are not e;ual (o#er*dispersion), often replace the
5oisson 4istri$ution replaced )ith :e+ati#e ,inomial 4istri$ution

You might also like