Example: Data are drawn from test results on file in the records office of the
Laboratory School of the University of Chicago. They consist of scores, obtained
from a cohort of pupils at the eighth through eleventh grade level on alternative
forms of the vocabulary section of the Cooperative Reading Tests. There are 64
students in all, 36 male, 28 female (ordered). Data can be found at [7].
The data consist of repeated measurements, and all 64 students have completed
scores. At first, we need to set up the data for the univariate repeated measures
ANOVA, using data step in SAS software [10] with DO loop, DO TIME = 1 TO 4;
VOCAB = V(TIME); OUTPUT; END. Then, we use PROC GLM with class
variables of “TIME” and “SUBJECT”. The SAS codes are given below.
PROC GLM;
CLASS TIME SUBJECT;
MODEL VOCAB = SUBJECT TIME;
RUN;
The CLASS statement is used to define all variables that are regarded as
categorical factors. MODEL statement specifies the response variable and the
fixed effects. The fixed effects can include both discrete covariates defined in the
CLASS statement and quantitative covariates that are excluded from the CLASS
statement [4, 13].
Now, R-square is equal to 0.87 which indicates that the dependent variables fit
and predict 87 percent of independent variable in the GLM model. The p-values of
“subject” and “grade” (i.e., “time”) variables are both significant (less than 0.01).
Therefore, the result shows that the scores of the vocabulary section of the
Cooperative Reading Tests at the eighth through eleventh grade do differ. If we
explore this analysis in deep, we may find out the average of each score to see
whether their vocabulary skills have increased or decreased. Using PROC MEAN,
we can create the chart below. Obviously, the mean is increasing over time. That is,
the pupils have improved better the vocabulary section of the cooperative reading
tests with grade.
Vocab 1 Vocab 2 Vocab 3 Vocab 4
Mean 1.137 2.542 2.989 3.472
SD 1.889 2.085 2.169 1.926
The main difference between ANOVA and MANOVA is that MANOVA
approach must discard all missing data. The MANOVA needs to be with complete
data because it treats the repeated measures as one vector and the entire data vector
must be complete for the subject to be included in the analysis. Also, the
MANOVA assumes a general form for the correlation of repeated measurements
over time, whereas the ANOVA assumes the much more restrictive
compound-symmetric form. The MANOVA model in one sample case is given by:
yi = μ + εi
where μ defines n×1 mean vector for time points and εi defines n×1 vector of errors
with εi ~ N(0, Σ) where Σ is the variance-covariance matrix. Compared to ANOVA,
the variance-covariance structure for yi is unstructured for MANOVA, which
assumes the covariate to be a general n×n positive definite covariance matrix.
Example: Data are drawn from test results on file in the records office of the
Laboratory School of the University of Chicago. They consist of scores, obtained
from a cohort of pupils at the eighth through eleventh grade level on alternative
forms of the vocabulary section of the Cooperative Reading Tests. There are 64
students in all, 36 male, 28 female (ordered). Data can be found at [7].
The data consist of repeated measurements, and all 64 students have completed
scores. At first, we need to set up the data for the univariate repeated measures
ANOVA, using data step in SAS software [10] with DO loop, DO TIME = 1 TO 4;
VOCAB = V(TIME); OUTPUT; END. Then, we use PROC GLM with class
variables of “TIME” and “SUBJECT”. The SAS codes are given below.
PROC GLM;
CLASS TIME SUBJECT;
MODEL VOCAB = SUBJECT TIME;
RUN;
The CLASS statement is used to define all variables that are regarded as
categorical factors. MODEL statement specifies the response variable and the
fixed effects. The fixed effects can include both discrete covariates defined in the
CLASS statement and quantitative covariates that are excluded from the CLASS
statement [4, 13].
Now, R-square is equal to 0.87 which indicates that the dependent variables fit
and predict 87 percent of independent variable in the GLM model. The p-values of
“subject” and “grade” (i.e., “time”) variables are both significant (less than 0.01).
Therefore, the result shows that the scores of the vocabulary section of the
Cooperative Reading Tests at the eighth through eleventh grade do differ. If we
explore this analysis in deep, we may find out the average of each score to see
whether their vocabulary skills have increased or decreased. Using PROC MEAN,
we can create the chart below. Obviously, the mean is increasing over time. That is,
the pupils have improved better the vocabulary section of the cooperative reading
tests with grade.
Vocab 1 Vocab 2 Vocab 3 Vocab 4
Mean 1.137 2.542 2.989 3.472
SD 1.889 2.085 2.169 1.926
The main difference between ANOVA and MANOVA is that MANOVA
approach must discard all missing data. The MANOVA needs to be with complete
data because it treats the repeated measures as one vector and the entire data vector
must be complete for the subject to be included in the analysis. Also, the
MANOVA assumes a general form for the correlation of repeated measurements
over time, whereas the ANOVA assumes the much more restrictive
compound-symmetric form. The MANOVA model in one sample case is given by:
yi = μ + εi
where μ defines n×1 mean vector for time points and εi defines n×1 vector of errors
with εi ~ N(0, Σ) where Σ is the variance-covariance matrix. Compared to ANOVA,
the variance-covariance structure for yi is unstructured for MANOVA, which
assumes the covariate to be a general n×n positive definite covariance matrix.
การแปล กรุณารอสักครู่..