Scribd
Upload
29 views9 pages

4.2 Slides - Generalized Linear Mixed Models Part 1

GLMM1

Uploaded by

Gabi Si Florin Jalencu
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
29 views9 pages

4.2 Slides - Generalized Linear Mixed Models Part 1

GLMM1

Uploaded by

Gabi Si Florin Jalencu
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 9

Generalized Linear

Mixed Models
(GLMMs)
Eventueel ondertitel
Cas Kruitwagen
Combining GLM’s and Mixed Models
In this presentation we will look at:

- A brief recap of the linear model

- The extension to generalized linear models (GLM’s).

2
Linear Regression
• Data
– Continuous outcome variable
Y:
We assume the outcome for
each individual i comes from
N(μi ; σ2).
– Approach: we model μi given
a (set of) predictor variable(s)
X.
• Model
– Yi = β0 + β1 X1i + εi
– εi ~ N(0 ; σ2)
– εi independent for i = 1, …, n

3
Generalized Linear Models
• Data
o Outcome variable Y
o Predictor variable(s) X
• Model
o Left-hand side: Y (continuous, dichotomous, count, ordinal,
categorical, etc., from the exponential family)
o Right-hand side: linear equation β0 + β1 X1i + … + βp Xpi
o Left- and right-hand side are linked together using an
appropriate “link function”

4
Generalized Linear Models
• Example: logistic regression
o Dichotomous outcome variable Y (1/0), e.g.
• pregnant (1 = yes, 0 = no),
• heart disease (1 = yes, 0 = no).
o Assumed distribution of the outcome: binomial.
• Each individual i that is drawn can be seen as the outcome of a
“Bernoulli trial”, with success probability P(Yi=1).
o Principle: we model the success probability P(Yi=1), given a set of
predictor variables.

5
Generalized Linear Models
• Example: logistic regression
o Dichotomous outcome variable Y (1/0).
o Link function: logit
 P( Y  1) 
logit P( Y  1)  ln  
 1  P( Y  1) 
o Model:
 P( Yi  1) 
ln    β0  β1 X 1i  ...  β p X pi
 1  P( Yi  1) 
• For example:
o Y = pregnant (1 = yes, 0 = no), X = age, weight, LHB/CGB genes,
etc.
o Y = heart disease (1 = yes, 0 = no), X = age, weight, exercise,
blood pressure, cholesterol
• Note that 𝑒 𝛽𝑝 is the odds ratio corresponding to the effect of Xp on Y.

6
Generalized Linear Models
• Example: Poisson regression
o Outcome variable Y: count within a given time or space, e.g.
• Y = number of urinary tract infections per year,
• Y = number of telephone calls in NL on a given date,
• Y = number of insects on a plot of land.
o Assumed distribution of the outcome: Poisson.
• Parameter: rate λ (=mean, =variance).
• The Y value for each individual i that is drawn can be seen as a draw
from a Poisson distribution with rate λi
o Principle: we model the rate λi, which is equal to the expected
count E(Yi), given a set of predictor variables.

7
Generalized Linear Models
• Example: Poisson regression
o Count outcome variable Y.
o Link function: natural logarithm.
o Model:
lnE (Yi )    0  1 X 1i  ...   p X pi

• For example:
o Y = number of urinary tract infections per year, X = age, weight,
antibiotics use, cranberry use, etc.
o Y = number of telephone calls in NL on a given date, X =
working day, season, temperature, economy, etc.

8
Generalized Linear Models
• Poisson regression: offset
o Varying exposure window, e.g.
• Insects (not all plots of land which we observe have the same size ->
insects/km2).
• Infections (not all patients were followed for the same length of time
-> infections/year).
o Formula:

 E (Yi ) 
ln    0  1 X 1i  ...   p X pi
 exposure 

lnE (Yi )    0 1 lnexposure   1 X 1i  ...   p X pi

You might also like