Professional Documents
Culture Documents
Correct Use of Repeated Measures Analysis of Variance: Eunsik Park, PH.D., Meehye Cho, B.SC., and Chang-Seok Ki, M.D
Correct Use of Repeated Measures Analysis of Variance: Eunsik Park, PH.D., Meehye Cho, B.SC., and Chang-Seok Ki, M.D
DOI 10.3343/kjlm.2009.29.1.1
In biomedical research, researchers frequently use statistical procedures such as the t-test, standard analysis of variance (ANOVA), or the repeated measures ANOVA to compare means between
the groups of interest. There are frequently some misuses in applying these procedures since the
conditions of the experiments or statistical assumptions necessary to apply these procedures are
not fully taken into consideration. In this paper, we demonstrate the correct use of repeated measures ANOVA to prevent or minimize ethical or scientific problems due to its misuse. We also describe
the appropriate use of multiple comparison tests for follow-up analysis in repeated measures ANOVA.
Finally, we demonstrate the use of repeated measures ANOVA by using real data and the statistical software package SPSS (SPSS Inc., USA). (Korean J Lab Med 2008;28:1-9)
Key Words : Ethics, ANOVA, Repeated measures, Multiple comparison
INTRODUCTION
by the user
s lack of fundamental statistical knowledge. These
one population are the same. When we deal with more than
(ANOVA) [4].
PACK tube and Green Cross Green Vac-Tube, with the exis-
each analyte [8]. Since the blood sample from the same subject was used for all three tubes, comparison groups of inter-
rate groups but they have been matched in some way. For
respectively [9].
WilksLambda, Pillai
s Trace, Hotelling-Lawley Trace, and
Roy
s largest root. The only assumption required for valid
ments within the group factor, we say that the data satis-
(on the diagonal of the covariance matrix) are equal and all
statistics give the same F value and P value. All four multi-
Wilk
s have similar power [11]. Wilkslambda is the easiest
Box
s epsilon factor [15]. It has been shown that the estimate
and Geisser method. The output from SPSS (SPSS Inc., Chi-
3. Multiple comparisons
and the inflation of the type I error [17]. The former includes
suspect that your data may not have Type H covariance, all
Bonferroni
s test is a multiple comparison procedure in-
or because one test has much less power than the other does
these statistics can control the type I error rate and ensure
standard ANOVA. We may use well-known multiple comparison procedures or contrast tests, as they are useful to
1. Data structure
compare the means averaged over time or at individual intervals [18]. Little research has been conducted on the impact
power and Type I errors was tested for five a priori tests
type) over day (Day) and tube type (Tube), taken from the
Y_D0
_T1
Y_D0
_T2
Y_D1
_T3
19
...
17
14
...
13
22
...
14
21
...
17
18
...
12
21
...
12
16
...
16
16
...
9
17
...
9
2. Use of SPSS
using SPSS 16.0, after creating the data set with the above
3. Statistical analysis
variables from the left hand list to the right side boxes for
within-subjects variables and between-subject factors if
left hand list to the right side boxes for dependent variables
Tube
. Thus, test results of the significance ofDayand
list to the right side box for post hoc tests and check for the
moves all the data for any subject that has incomplete data,
while MIXED procedure still analyzes data from all the subTable 1. ESR means and SDs
Source
Day
Vacuum tube
Day
Mean
SD
Mean
SD
Mean
SD
Error (day)
BD Vacutainer
8.60
7.96
9.85
9.16
7.10
6.91
Tube
Sekisui INSEPACK
7.80
7.49
8.80
8.51
6.75
6.54
7.16
Green Cross
Green-Vac-Tube
10.70
9.19
10.40
9.16
7.40
Abbreviation: ESR, erythrocyte sedimentation rate; SD, standard deviation; BD, becton dickison.
Unadjusted
Mean
square
Unadjusted
DF
2 109.850
38
5.365
44.517
Error (tube)
38
4.172
Day*Tube
7.842
76
1.014
Error (day*tube)
F
value
Adjust- AdjustUnaded
ed
justed
P value P value
P value
by G-G by H-F
20.477
0.000
0.000
0.000
10.671
0.000
0.000
0.000
7.732
0.000
0.001
0.000
the tests. Since all P values from the four statistics are
This result agrees with that from the univariate test. There-
Since the sphericity condition is not met, the use of follow-up tests should be taken with care. As mentioned above
4. Multiple comparisons
in the section of multiple comparisons, the Bonferroni procedure is the best one to use in terms of type I error control.
Tube
Day*Tube
Statistic
F value P value F value P value F value P value
Pillais trace
Wilks lambda
Hotellings trace
Roys largest root
11.35
11.35
11.35
11.35
0.001
0.001
0.001
0.001
11.66
11.66
11.66
11.66
0.001
0.001
0.001
0.001
7.61
7.61
7.61
7.61
0.001
0.001
0.001
0.001
95% confidence
limits using
Holm-Bonferroni
procedure
0.0608
0.0072
0.0003
(-0.040, 1.640)
(-3.423, -0.777)
(-4.142, -1.658)
(-0.040, 1.640)
(-3.638, -0.562)
(-4.458, -1.342)
0.0252
0.2749
0.0002
0.0504
0.2749
0.0006
(0.146, 1.954)
(-1.574, 0.474)
(-2.319, -0.881)
(-0.001, 2.101)
(-1.574, 0.474)
(-2.501, -0.699)
0.3672
0.5163
0.0115
0.7344
0.5163
0.0345
(-0.443, 1.143)
(-1.249, 0.649)
(-1.137, -0.163)
(-0.572, 1.272)
(-1.249, 0.649)
(-1.260, -0.040)
Vacuum tube
comparison*
Mean
difference
B-S
B-G
S-G
0.80
-2.10
-2.90
0.0608
0.0036
0.0001
B-S
B-G
S-G
1.05
-0.55
-1.60
B-S
B-G
S-G
0.35
-0.30
-0.65
Day
Unadjusted
P value
P value using
Holm-Bonferroni
procedure
pairs of tube types are B-G and S-G at day 1, S-G at day
DISCUSSION
of the Type I error rate and power. It is safe to use the Bon-
trol the type I error rate. However, to ensure the power, the
can be considered.
coronary artery bypass grafting surgery. Clin Chem Lab Med 2006;
44:471-8.
t ( )
12. Huynh H and Feldt LS. Conditions under which mean square ratios
13. Maxwell SE and Delaney HD, eds. Designing experiments and ana-
1990.
14. Atkinson G. Analysis of repeated measurements in physical therapy research. Phys Ther Sport 2001;2:194-208.
REFERENCES
15. Greenhouse SW and Geisser S. On methods in the analysis of profile data. Psychometrika 1959;24:95-112.
16. Huynh H and Feldt LS. Estimation of the box correction for degrees
7. Jung CL, Cho SE, Hong KS. Clinical significance of minor elevation
, . Cardiac Troponin I .
2008;28:339-45.)
. 49 (Sek-
. 2007;27:69-75.)