Repeated Masur Te cadoaee
‘ sna ars 3 ghould be moasurocl at ths
ardiruous level Cre they ayy \herval ov cvatio
Nasiabeles) .
2|\nebpendent voatable shotild consist of at least
tata Categovical "crwlated groeips’ oy makcheck pals
Related Groups \nelicates that=ahe Same Subjects
a8 present in beth GAoups “The Meason that ie iss
Possible te have the samo Subjects in each Group
1S because each Subject has been Measurccl an
elo oeabions on the senae dependent variable
3|\ thas should ke no Significant outliers in the
GPorps . Outliers ae Simply Single elala
Ams toi wous data that clo ant Tallow the
Usual pattern.
*) the distaibution af thy dopenclone Variable ie
the uy o mote
melateol props should be
poximately Vrermay elects” cLisial bute of sgChapter 3
Multiple Samples Repeated Measures ANOVA Model
M datredaction
Tn mowt of the animal experiments, more than one treatment group will be under
‘consideration ond observations are recorded from the samme aninnils more than once For
example, while introducing a new drug there will he at feast one standard drug Henee ther
trontment may be new drieg and standard drug. Observation will be taken from the animals before
aderiimistering the drug and after administering the drug. Hence there are two trewtment groups
and two time points. This is simplest case of multiple samples repeated measures ANOVA
motel
32 Medel
In gencral, suppose that repeated measurements at { time points are obeained fom # groups of
subjects. Let ny denote the number of subjects in group hand let m= Thc my Let yyy denowe
the response at time } from the i" subject in group h for h = 1,
1
KL mya =
Table 9 displays the general data layout for this setting. There are at lewst three models for
this situation, all resulting in the same ANOVA table. The stmphest bs
Yop = H+ Yn + 1 + CYTDRG + Many + Ors, any
where,
pis the overall mean,
Yois the fixed effeet of group bh with
1) in the fixed effect uf time j, with Y}
Cy O)yyte the fixed effect for the interaction of the h® group with the /* time
The constraints on the
racthon parameters aie
dives = yr dy 0
: ftane sin the RY group. The thay, =
sheets ~ and variance 057. Finally, y,
The parameters ny ae rae ;
eemmed fo We indeppersientl
am shor ert NAT
with mean rere
th yy = MOD
med 9A =F Ou-he (9 Js
where,
Bt
fy, = Setiarny
5 ‘the mean for group h,
i-n- Py oy 14 Gna - y= +9.)
4s the overall mean ,a
BeBe isthe mean at time j.
ENN yn i 6 fo
Shy = ELMIs the mean for group h at time j, and
Jn) = EPs the mean for the i" subject in group h.
‘The sums of squares are then defined as follows:
Sum of squares due to groups,
SSo = Dh DMLsOn 9)" = Lhe tn - 9.)
Sum of squares of subjects within group,
s Th opt og, ™h (G, a ¥.
Seco) = Dh DM, Lien — Ir)” = Char Laat ~ Ind +
‘Sum of squares due to time,
s J" ae zs}
SS = Eka EM ZaFy-9.)' = 9B }aG1-7)-
‘Sum of squares due to group * time interaction,
SSor = Dhar DM EE aGaj Fn Fy +7.) and
‘Sum of squares of residual ,
SSq = Ther LM Lha(Fni— Ini —Tnn +n)
Note that $Sz, SS7, and SSgr are equal to the sums of squares from a two-factor ANOVA,
model (assuming that all nt observations are independent) with effects for group, time, and the
group * time interaction. The residual sum of squares SSq is due to the subject effect nested
within the cross-classification of group * time.
Table 10. Multiple samples repeated measures ANOVA table
Sun af Squares Degrees of ]
Source : EMS)
(ss) Freedomid f)
iroup 35¢ + OF tye Dy
Subjects(Group B5sea) ms Otay
Time Sh
_—___+—_____
Group» Time
Residual
aTable 10 displays the sum of squares (ss), ot. and expected Ne ae each source ,
in of expected mean the quantilies labeled Da,
differences am
variation, In the colunm squares. : ba
. a
and Dgr represent differences among groups, ong time points group,
time interaction, respectively
iven by
The F statistic for testing for differences among groups is gi
SSep,
with s-1 and nos df.
MSc
MSsic)
This test requires the assumption that the within-group covariance matrices are equal. I
general, this assumption is required for all tests of between-subjects effects.
‘The F statistic for testing differences among time points is piven by
SSr,
us: Ne-1)
=-——
MS, SS
MSa Sf ot 1)]
ly, the F statistic for testing the significance of the
with t— 1 and (n ~ s)(t = 1) df. Similar!
group * time interaction is given by
with (s-1t-1) and (n-sy(t=-1) -£
Both of these tests require the assumption that the within-group covariance matrices are equal
and that the sphericity conditio
4s, An alternative repeated measures ANOVA model for this setting
11 is satisfied. In general, these assumptions are required for all
tests af within-subjects effee
includes an additional random effect for the subject ~ time interaction. This effect is usually
assumed to be uncorrelated with the random subject effect. Although the expected mean squares
for this model are different from those displayed in Table 10, the sums of squares and test
statistics are identical.
‘The illustration of multiple samples repeated measures ANOVA is given in following
section.
22