Multivariate Analysis in SPSS

Analysis of (Co)Variance
• One-way multivariate analysis of variance.
• Two-way multivariate analysis of variance.
• One-way multivariate analysis of covariance.

Variance: average squared deviation from the

mean.
Covariance: a measure of how changes in one
variable are associated with changes in a second
variable
 One-Way MANOVA
• Is used to determine whether there are any
differences between independent groups on
more than one continuous dependent
variable.
• Differs from one-way ANOVA which only
measures one dependent variable.
• Is an omnibus test statistic.
 One-Way MANOVA Example
• Differences in the perceptions of
attractiveness and intelligence of drug users
(non-user, experimenter, regular user) in
movies.
• Differences in students' short-term and long-
term recall of facts based on three different
lengths of lecture
 One-Way MANOVA Assumptions
• Two or more dependent variables should be
measured at the interval or ratio level (i.e.
continuous data).
• Independent variable should consist of two or
more categorical independent groups.
• Independence of observations (there is no
relationship between the observations in each
group or between the group themselves, e.g.
there must be different participants in each group
with no participant being in more than one
group).
• Adequate sample size (need to have more cases
in each group than the number of dependent
variables)
• There are no univariate or multivariate outliers 
boxplots and Mahalanobis distance respectively.
• There is multivariate normality  Shapiro-Wilk.
• There is a linear relationship between each pair
of dependent variables for each group of the
independent variable  scatterplot matrix for
each group of independent variable.
• There is a homogeneity of variance-covariance
matrices  Box’s M test of equality of
covariance.
• There is no multicollinearity [usually when
correlation(s) are too high, greater than 0.9].
 Practice Time
Open your SPSS
• The pupils at a high school come from three different
primary schools. The headteacher wanted to know whether
there were academic differences between the pupils from
the three different primary schools.
• As such, she randomly selected 20 pupils from School A, 20
pupils from School B and 20 pupils from School C, and
measured their academic performance as assessed by the
marks they received for their end-of-year English and
Maths exams.
• Therefore, the two dependent variables were "English
score" and "Maths score", whilst the independent variable
was "School", which consisted of three categories: "School
A", "School B" and "School C".
 Two-Way MANOVA
• Is used if there are two independent variables.
• Example:
Consider an experiment where three teaching
methods were being trialled in schools. The three
teaching methods were called "Regular", "Rote" and
"Reasoning". The researchers primarily wanted to
know whether the effects of the three teaching
methods on students' grades in these two subjects
were different based on students' gender (i.e.,
"male" and "female" students).
• Therefore, in this study the two continuous
dependent variables were "Humanities score" and
"Science score", whilst the two nominal independent
variables were "intervention", which reflected the
teaching methods (i.e., consisting of three groups:
"Regular", "Rote" and "Reasoning") and "gender"
(i.e., consisting of two groups: "males" and
"females").
 One-Way MANCOVA
• Is used if there is one independent variable
and a continuous covariate.
• Can be thought as an extension of one-way
MANOVA.
• Is used to determine whether there are any
statistically significant differences between
the adjusted means of three or more
independent (unrelated) groups, having
controlled for a continuous covariate
 One-Way MANCOVA Example
• To determine whether a number of different exam
performances differed based on test anxiety levels amongst
students, whilst controlling for revision time:
– Dependent variables would be "humanities exam performance",
"science exam performance" and "mathematics exam
performance", all measured from 0-100
– Independent variable would be "test anxiety level", which has
three groups – "low-stressed students", "moderately-stressed
students" and "highly-stressed students“
– Covariate would be "revision time", measured in hours).
• You want to control for revision time because you believe
that the effect of test anxiety levels on overall exam
performance will depend, to some degree, on the amount
of time students spend revising.
 One-Way MANCOVA Assumptions
• Two or more dependent variables should be measured
at the interval or ratio level (i.e. continuous variables).
• One independent variable should consist of two or
more categorical, independent groups (i.e., you have a
nominal variable or an ordinal variable).
• One or more covariates are all continuous variables.
• Independence of observations (there is no relationship
between the observations in each group of the
independent variable or between the groups
themselves)
• There should be a linear relationship between each
pair of dependent variables within each group of the
independent variable  scatterplot matrix with Ioess
Ilnes.
• There should be a linear relationship between the
covariate and each dependent variable within each
group of the independent variable  scatterplot
matrix with Ioess Ilnes.
• There should be homogeneity of regression slopes.
• There should be homogeneity of variances and
covariances  Box’s M test of equality of covariance
matrices.
• There should be no significant univariate outliers
in the groups of your independent variable in
terms of each dependent variable  inspecting
the standardized residuals.
• There should be no significant multivariate
outliers in the groups of your independent
variable in terms of each dependent variable 
Mahalanobis distance.
• There should be multivariate normality 
Shapiro-Wilk.
 Practice Time
• A researcher wanted to determine whether
cardiovascular health was better for normal
weight individuals with higher levels of physical
activity (i.e., as opposed to more overweight
individuals with lower physical activity levels).
• As such, the researcher recruited 120 participants
who were subsequently divided into one of three
groups depending on the amount of physical
activity they performed: low, moderate, high.
There were 40 participants in each group.
• In order to measure cardiovascular health, the
researcher took three measurements from
participants:
– Cholesterol concentration (measured in mmol/L).
– C-Reactive Protein (measured in mg/L).
– Systolic blood pressure (measured in mmHg).
• It was expected that increased levels of physical
activity would have an overall beneficial effect on
cardiovascular health. However, the researcher knows
that body weight also effects cardiovascular health. As
such, the researcher wanted to control for differences
in the body weight of participants.
 Follow-up for Statistically Significant
One-Way MANCOVA
Steps:
1. Test for adjusted mean difference between
groups of independent variable in terms of
each dependent variable.
2. If any of the step 1 is statistically significant,
examine all possible pairwise comparisons
between groups of your independent
variable in terms of adjusted group means on
the dependent variable.
• Example:
to determine whether a number of different exam performances
differed based on test anxiety levels amongst students, whilst
controlling for revision time (i.e., your dependent variables would
be "humanities exam performance", "science exam performance"
and "mathematics exam performance", all measured from 0-100,
your independent variable would be "test anxiety level", which has
three groups – "low-stressed students", "moderately-stressed
students" and "highly-stressed students" – and your covariate
would be "revision time", measured in hours). You want to control
for revision time because you believe that the effect of test anxiety
levels on overall exam performance will depend, to some degree,
on the amount of time students spend revising.
 Linear Regression
(simple and multivariate)
 Linear regression

When do I use it?

• To look at the linear relationship between one
normally distributed interval predictor and one
normally distributed interval outcome variable.
• E.g. to look at the linear relationship between
systolic blood pressure vs. cholesterol level, BMI
and sex
Linear regression

• Click on “regression" under the "Analyze" menu

 Select “linear" from this menu.
• Select the variables you wish to associate and
press "OK."
 Logistic Regression
(simple and multivariate)
 Logistic regression

When do I use it?

• Logistic regression assumes that the outcome
variable is binary
• The first variable listed after the logistic command
is the outcome (or dependent) variable, and all of
the rest of the variables are predictor (or
independent) variables.
• The predictor variables must be either dichotomous
or continuous; they cannot be categorical (i.e. More
than 2 categories).
• Eg. to look at the relationship between the outcome
of MI and hyperlipidemia, cholesterol and systolic
BP.
Logistic regression

• Click on “regression" under the "Analyze" menu

 Select “binary logistic" from this menu.
• Select the variables you wish to associate and
press "OK."