You are on page 1of 22

Statistical analysis:

Which to choose?
(Multivariable Analysis)

Professor Dr. Syed Hatim Noor


Dr. Wan Arfah Nadiah Wan Abdul Jamil
Universiti Sultan Zainal Abidin
“Multivariable Analysis”
Why do we need multivariable analysis ?
Purpose of using multivariable analysis
Common multivariable analysis methods in
health related sciences research

Syed Hatim Noor 2


V a ria b le s V a ria b le s

In d e p e n d e n t D ependent
P re d ic to r O u tc o m e
E x p la n a to ry R esponse

C o v a ria te s N o t th e p rim a r y
C o n fo u n d e rs in te re s t.
C o n tro ls M ust be
E ffe c t m o d ifie rs re c o g n iz e d
Syed Hatim Noor 3
Phenomena in statistics
• Confounding (C)
• Multicollinearity (M)
• Interaction (Effect modification) (I)

COMMONESS SEVERITY
C I
M M
I C

Syed Hatim Noor 4


Confounding

?
Risk Factor Disease

Confounder

Syed Hatim Noor 5


Confounding

-distortion of a risk factor-disease


relationship brought about by the
association of other factors with both
risk factor and disease

Syed Hatim Noor 6


Example of confounding

Physical Systolic
Activity
? Blood
level pressure

Age

Syed Hatim Noor 7


Confounding

• Confounding is • Testing the presence of


more common than a confounder does not
interactions need a statistical test
• Confounding is • Comparing parameter
eliminated when estimates and their
multivariable confidence intervals
analysis is carried can answer whether the
out variable is a confounder

Syed Hatim Noor 8


Interaction

Risk Factor ? Disease

Interaction Factor
(=Effect Modifier)

Syed Hatim Noor 9


Example of interaction

Employment in ? Lung cancer


an industry

Cigarette
smoking
Syed Hatim Noor 10
S m o k e rs

R is k o f
lu n g
cancer
N o n -sm o k e rs

Y e a rs o f e m p lo y m e n t in th e in d u s try

Syed Hatim Noor 11


Interactions
• An interaction is rare • Test applied depends
but more serious than on the type of
a confounder multivariable or
multivariate analysis
• Interactions can be • Testing interactions
synergistic or among all variables is
antagonistic not recommended
interactions • Biologically
• Testing the presence (medically) important
of an interaction and possible
needs a particular interactions are
statistical test priority

Syed Hatim Noor 12


Multicollinearity
• Check multicollinearity
• Checked to assess which variable (2 or more) correlate
highly
• Check correlation estimates
• Check standard errors
• May omit the variable if standard error is big
• Decision is subjective (depend on researcher)

Syed Hatim Noor 13


Multivariable Analysis

Purpose
1) To statistically adjust the effect on
variable Y by change in a particular
variable x when others are controlled

X1 Y
X2 X3 X4 …. Statistically adjusted for
E.g diet CHD smoking and
age adjusted for
Syed Hatim Noor 14
2) To discover the variable X which
has most influence on outcome
variable Y

diet
smoking CHD
age
Syed Hatim Noor 15
3) To predict the outcome Y

Clinical
Pathological cancer
Demographic Prognosis
Socio-economic

Syed Hatim Noor 16


Multivariable analysis

• Whole idea of • Common


multivariate multivariate analysis
analysis- “How to methods in health
related sciences
separate research (see
independent tables)
effect of each X • Multivariate models
on Y” • Modeling strategies

Syed Hatim Noor 17


GLM (GENERAL LINEAR MODEL)
Univariate Model
Independent variable
Lecturer academician One way
Master (categorical) ANOVA
Phd.
Young academician + 2 way ANOVA
Middle age groups
Old (categorical)
Male academician + age + Multifactorial
Female sex (categorical) ANOVA

BMI Simple linear BP level


(continuous) regression (continuous
variable)
BMI + Multilpe linear
cholesterol level regression
(continuous)
academician + age + Analysis of
sex + BMI + covariance
cholesterol
(categorical + (ANCOVA)
continuous)
(MANOVA) >1 dependent variable
(continuous)
Multivariate model

Syed Hatim Noor 18


Repeated measures in numerical outcome

• When the dependent variable is a numerical


variable
Independent Dependent Statistical test
variable variable

categorical numerical Repeated measures


ANOVA
(parametric)
categorical numerical Friedman test
(non-parametric)

Syed Hatim Noor 19


Repeated measures in categorical outcome
• When the dependent variable is a categorical
variable
Independent Outcome Statistical test
Variable (s)
Repeated measures
2 measures 2 outcome categories Mc Nemar`s test

2 measures 3+ outcome categories Test of marginal


Homogeneity

3+ measures 2 outcome categories Cochran`s Q test

Syed Hatim Noor 20


Repeated measures Cross-sectional Time-
with independent series (xt)
variables Binary
Logistic
Ordinal Regression
(xt Logit)
Multiple

Count Log-linear regression


(xt Poisson)
Generalized estimating
equations
(GEE) model
(xtgee)

Syed Hatim Noor 21


•Thank you

Syed Hatim Noor 22

You might also like