You are on page 1of 5

11/8/16

• We  use  repeated-­‐measures  ANOVA  to  


evaluate  the  mean  differences  obtained  i n  a  
study  c omparing  two  or  more  treatment  

Anova
conditions  using  the  same  sample  of  
individuals  in  e ach  c ondition.

Repeated-­‐Measures   ANOVA
Psych  110  – Day  21

Sources  of  variability  in  r epeated-­‐ Comparing  the  F-­‐ratio  of  ANOVA  for  
measures   ANOVA independent-­‐measures  &  dependent  measures
• A.  The  numerator   of  the  F-­‐ratio:  between-­‐ F  =  systematic  treatment  effects  +  random,  unsystematic  differences
random,  u nsystematic  differences
treatments  variance
– Systematic  d ifferences  caused  b y  treatments • F  =  Between-­‐treatments  variance  
– Random,  u nsystematic  differences  (error/chance  only;   • Within-­‐treatments  variance
NO  individual  d ifferences)
• *****
F  =  systematic  treatment  effects  +  random,  unsystematic  differences
• B.  The  denominator   of  the  F-­‐ratio:  error   variance random,  u nsystematic  differences

– Random,  u nsystematic  differences  (error/chance;   • F  =  Between-­‐treatments  variance  


individual  d ifferences  exist  b ut  this  is  s ubtracted  from   • error  variance
the  computation  of  w/in  treatments  variance)

1
11/8/16

Notations  
• k =  #   of  treatment   • ΣX =  s um  of  all  the  s cores  
conditions in  the  s tudy  ( Grand  total);  
• n =  n umber  of  s cores  in   ΣX  =  ΣT
each treatment • ΣX2  
• N =  total  #  of  s cores in  the   • M for  each  treatment  ( SS:
entire  s tudy optional)
• T =   sum  of  s cores  for  each   • P =  s um  of  all  the  s cores  
Treatment for  each  Participant in  the  
study  ( Participant/Person  
Totals)
The  partitioning  of  variance  f or  a  
repeated-­‐m easures  experiment.

Consider  the  following  data


Person   Placebo   Drug  A Drug  B Drug  C P  
(Person  
Totals)
A 3 4 6 7 P  = 20 n =  5

B 0 3 3 6 P  =  12 k  =   4
C 2 1 4 5 P  =  12
N  =   20
D 0 1 3 4 P  =  8
ΣX =  60
E 0 1 4 3 P  =  8
ΣX 2=  262
HYPOTHESIS   TESTING   WITH   THE   T1 =   5 T2 =   10 T3 =   20 T4 =   25

REPEATED-­‐MEASURES   ANOVA M1 =  1 M2 =  2 M3 =  4 M4 =  5

The  effect  of   drug  t reatment  on  t he  amount  of  t ime  (in  seconds)
a  painful  stimulus  is  endured.

2
11/8/16

Step2:  Select  alpha  level,  state  type  of  test,  


Step1:  State  the  hypotheses. determine  df’s,  look  up  c ritical  F,  set  DR.
• H0 :  μ1 =  μ2 =  μ3 =  μ4 • α  =  . 05;  repeated-­‐measures  single-­‐factor  ANOVA
• The  type  of  drug  has  no  significant   effect  on  the   • Five  k inds  of  degrees  of  freedom:
amount  of  time  that  the  participants   could  
endure  a  painful   stimulus. – dftotal =  N  – 1  =  2 0  – 1  =  1 9

• H1 :  At  least  one  of  the  treatment   means  is   – dfbet grps =  k  – 1  =  4  – 1  =  3
different.  
• The  type  of  drug  has  a  significant   effect  on  the   – dfwithin =  N  – k  =  2 0  – 4  =  1 6
amount  of  time  that  the  participants   could  
endure  a  painful   stimulus. • dfbet subjects =  n  – 1  =  5  – 1  =  4

• dferror =  dfwithin – dfbet subjects =  16  – 4  =  12

Step2:  Select  alpha  level,  state  type  of  test,  


Step3:  Compute  the  four  terms/values  we  need
determine  df’s,  look  up  c ritical  F,  set  DR.
• For  α  =  .05  & df =  3 ,  1 2  the  c ritical  F  value  i s   3.1. ∑ X 2 = 262
3.49
2 2
3.2. (∑ X ) = 60 = 180
• DR:  Reject  H0 if  Fobs >  3 .49;  otherwise,  retain   N 20
H0. 2 2
3.3. ∑ (T ) = 5 + 10 + 20 + 25 = 230
2 2 2

n 5 5 5 5
2 2 2 2 2 2
3.4.    ∑
  ( P) = 20 + 12 + 12 + 8 + 8 = 204
k 4 4 4 4 4

3
11/8/16

Step4:  Construct   ANOVA   Table  (as  you  


compute   the  values  of  SS,  MS,  and  F )
Step5:  State  your   conclusion
Source   SS df MS F
• Compare  your  observed  F  with  your  c ritical  F  
and  decide  whether  to  reject  or  retain  your  
Between   50 3 16.67 F(3,  12)  =  
treatments 24.88
null  hypothesis.
Within   32 16
treatments • F(3,  1 2)  =  2 4.88,  p  <  .05
between   24 4 • Thus,  we  reject  the  null  hypothesis.
subjects
• The  type  of  drug  has  a  significant  effect  on  the  
error 8 12 0.67 amount  of  time  that  the  participants  c ould  
TOTAL 82 19 endure  a  painful  stimulus.

Measuring  effect  size


SS between
r 2 /η 2 =
SS total − SS betsubjects
50
r 2 /η 2 = = 0.86(or 86%)
82 − 24
r2 =  p ercentage  of  variance  accounted  for.
It  measures  h ow  much  of  the  d ifferences  b etween  
scores  is  accounted  for  b y  the  d ifferences  b etween  treatments.
POST  HOC  TESTS
Thus,  we  s ay  that  8 6%  of  the  variance  in  the  s cores  can  be  
explained  b y  the  d ifferences  between  the  treatment  ( or  b y  the  
treatment).

4
11/8/16

Advantage  of  a  repeated-­‐measures  


ANOVA
• When  the  obtained  F-­‐ratio  i s  significant,  we   • It  eliminates  the  i nfluence  of  i ndividual  
can  determine  exactly  where  the  differences   differences  from  the  analysis.
lie  by  performing  post  hoc  tests  such  as  
Tukey’s  HSD  and  Scheffe’s F  test.
• If  i ndividual  differences  are  extremely  l arge,  a  
treatment  effect  might  be  masked  i n  an  
• Just  remember  to  use  MSerror rather  than   independent-­‐measures  experiment;  thus  a  
MSw/in treatments and  dferror rather  than  dfw/in repeated-­‐measures  design  i s  more  
treatments . appropriate.

Seatwork  
• Study  on  the  effect  of  s leep  d eprivation  ( SD)  on  motor-­‐
Participant 24hrs 36hrs 48hrs P  totals
skills  p erformance.
A 0 0 6 6 n =  5
• Sample   of  5  p articipants  tested  after  2 4  h rs  of  SD,   B 1 3 5 9 N  =  15
tested  again  after   36hrs,  and  tested  once  more  after  48  
hrs. C 0 1 5 6 ΣX = 45
D 4 5 9 18
• DV:  #  of  errors  made  on  the  motor-­‐skills  task ΣX2= 245
E 0 1 5 6
• Do  these  d ata  indicate  that  the  #  of  h rs  of  s leep   T1  =  5 T2=  10 T3  =  30
deprivation  h as  a  s ignificant  effect  on  motor  s kills   M1  =  ? M3  =  ? M3  =  ?
performance?  Test   at    alpha=.05

You might also like