comp formulas a nova
TRANSCRIPT
-
8/12/2019 Comp Formulas a Nova
1/2
PSYCHOLOGY 315 D R. MCFATTER
Computational Formulas for ANOVAOne-Way ANOVA
Let a = # of levels of the independent variable = # of groups N = total # of observations in the experimentn1 = # of observations in group 1, etc.
H0: 1 = 2 = 3 = . . . = a
NOVA analyzes sample variances to draw inferences about population means. Sample variances can always be calculated asSS/df and these sample variances are called mean squares ( MS ):
SS T = 92 + 8 2 + . . . + 1 2 + 5 2 - (80) 2 /15 = 93.333
SS B = (402
+ 252
+ 152
)/5 - 802
/15 = 63.333 SS W = 93.333 - 63.333 = 30.000
An alternative computational approach emphasizing the conceptual basis of ANOVA is given below.
This is the variance of all scores in the experiment = 6.667.
This is the average of the variances within the groups = 2.50.(1.22 2 + 1.87 2 + 1.58 2)/3 = 2.50.
This is n times the variance of the means = 5(6.333) = 31.667.
Multiple Comparisons: t Crit is the critical value from a t -table using the df of the error termfrom the ANOVA table. The error term is always the denominator ofthe F -ratio. Thus, in the above example, the error df would be 12. The
MS Error would be 2.50; n is always the number of observations eachmean youre comparing is based on.
Example . X 1 X 2 X 3 Placebo Drug A Drug B
9 5 28 4 48 5 36 8 19 3 5
Sum 40 25 15 80M 8 5 3 5.333s 1.224745 1.870829 1.581139
ANOVA Summary Table Source SS df MS F pBetween 63.333 2 31.667 12.67 0.0011Within 30.000 12 2.500Total 93.333 14 6.667
SS Total
SS Between SS Within
s SS
df MS 2 = =
( )SS X
X
N Total =
2
2
( ) ( ) ( ) ( )SS
X
n
X
n
X
n
X
N Betweena
a
= + + + 1
2
1
2
2
2
2 2
K
SS SS SS Within Total Between=
df a Between = 1
df N aWithin = F MS
MS Between
Within
=
df N Total = 1
( )$
T Total
X X
N N
MS 22
2
1=
=
$ W a
Within
s s sa
MS 2 12
22 2
=+ + +
=K
$ $
B M Betweenn MS 2 2
= =
LSD t MS
nCrit
Error =
2
-
8/12/2019 Comp Formulas a Nova
2/2
PSYCHOLOGY 315 D R. MCFATTER
Two-Way Factorial ANOVALet:
a = # of levels of the independent variable Ac = # of levels of the independent variable Cac = # of cells in the experiment
N = total # of observations in the experimentn1 = # of observations in cell 1, etc.
SS T = 537 - 932 /18 = 56.50
SS B = 122 /3 + 21 2 /3 + . . . - 93 2 /18 = 26.50
SS W = 56.50 - 26.50 = 30.00 SS A = 45
2 /9 + 48 2 /9 - 93 2 /18 = 0.50 SS C = 30
2 /6 + 33 2 /6 + 30 2 /6 - 93 2 /18 = 1.00
SS AC = 26.50 - 0.50 - 1.00 = 25.00
2 3 factorial design Crime (C)DV = Years Forgery Swindle Burglary
3 6 2Attractive 4 7 4
5 8 6
4 3 4Unattractive 6 4 6
Attractivenessof Offender
(A)
8 5 8
Table of Totals
Forgery Swindle BurglaryMarginalTotals
Attractive 12 21 12 45
Unattractive 18 12 18 48MarginalTotals 30 33 30 93
Table of Means
Forgery Swindle BurglaryMarginalMeans
Attractive 4 7 4 5
Unattractive 6 4 6 5.33MarginalMeans 5 5.5 5 5.17
ANOVA Summary TableSource SS df MS F pA 0.50 1 0.500 0.20 0.6627C 1.00 2 0.500 0.20 0.8214AC 25.00 2 12.500 5.00 0.0263Within 30.00 12 2.500Total 56.50 17 3.324
Null hypotheses:
A main effect H0: A = U C main effect H0: F = S = B AC interaction H0: ( AF - UF) = ( AS - US ) = ( AB - UB)
or equivalentlyH0: parallel lines in the cell mean plot
( )SS X
X
N Total =
2
2
( ) ( ) ( ) (SS
X
n
X
n
X
n
X
N Betweenac
ac
= + + + 1
2
1
2
2
2
2
K
SS SS SS Within Total Between=
( ) ( )SS
for each row
n for each row
X
N A =
2 2
( ) ( )SS
for each column
n for each column
X
N C =
2 2
SS SS SS SS AC Between A C =
2
3
4
5
6
7
8
Forgery Swindle BurglaryCrime (C)
Y e a r s
AttractiveUnattractive
2345678
Unattractive AttractiveAttractiveness (A)
Y e a r s
ForgerySwindleBurglary
df a A = 1
df cC = 1
df a c AC = ( )( )1 1
df N acWithin =
df N Total
= 1