Session 11 (Comparing)
Session 11 (Comparing)
Leland
Wilkinson
Adjunct
Professor
UIC
Computer
Science
Chief
Scien<st
H2O.ai
[email protected]
Comparing
o Sta<s<cal
methods
exist
for
comparing
2
or
more
groups
o The
classical
approach
is
Analysis
of
Variance
(ANOVA)
o This
method
invented
by
Sir
Ronald
Fisher
o It
revolu<onized
industrial/scien<fic
experiments
o The
researcher
was
able
to
examine
more
than
one
treatment
at
a
<me
o With
only
two
groups,
results
of
Student’s
t-‐test
and
F-‐test
are
equivalent
o Mul<variate
Analysis
of
Variance
(MANOVA)
o This
is
ANOVA
for
more
than
one
dependent
variable
(outcome)
o Hierarchical
modeling
is
for
nested
data
o There
are
several
forms
of
this
mul<level
modeling
2
Copyright
©
2016
Leland
Wilkinson
Comparing
o A
simple
two-‐group
comparison
5
Copyright
©
2016
Leland
Wilkinson
Comparing
o Another
way
at
looking
at
the
independent
groups
test
o The
OLS
regression
model
on
two
groups
y = Xb + e
y = Xb + e
9
Copyright
©
2016
Leland
Wilkinson
Comparing
o Analysis
of
Variance
(ANOVA)
sums
of
squares
g
�
SSB = nj (µ̂j − µ̂)2 between
groups
(regression)
sum
of
squares
j=1
nj
g �
�
SSW = (yij − µ̂j )2 within
groups
(error)
sum
of
squares
j=1 i=1
n
�
SST = (yi − µ̂)2 total
sum
of
squares
i=1
o Ever
see
a
researcher
test
whether
the
within-‐subject
correla<on
is
posi<ve
before
using
the
dependent
t-‐test?
o I
didn’t
think
so
“Variance
analysis
found
a
significant
main
effect
of
gender
on
perceived
dura<on
(F(1,109)=4.29,
p<.05).”
James
J.
Kellaris
and
Susan
Powell
Mantel
(1994)
,"The
Influence
of
Mood
and
Gender
on
Consumers'
Time
PercepHons",
in
NA
-‐
Advances
in
Consumer
Research
Volume
21,
eds.
Chris
T.
Allen
and
Deborah
Roedder
John,
Provo,
UT
:
AssociaHon
for
Consumer
Research,
Pages:
514-‐518.
BULLSHIT
The
story
is
different
for
males
and
females
ANOVA
Poisson ANOVA
Y2
Y1