Repeated Measures ANOVA can be viewed as an extension of the paired-samples t-test

because in this case we can compare one group of individuals under more than two conditions
on the levels of a continuous numerical variable. Additionally, the assumption of normality and
independence should be met for applying the specific analysis.

For example, we could apply Repeated Measures ANOVA in order to compare the levels of
depression (interval) of the same group of individuals during the winter, the spring and the
summer period. In this case the “period of the year (winter/spring/summer)” is the within-
subjects independent nominal variable and “depression” is the dependent continuous variable.
Similarly, we could apply Repeated Measures ANOVA in order to compare the levels of anxiety
(interval) of the same group of individuals during the morning, the afternoon and the evening
period of the day. In this example, the “period of the day (morning/afternoon/evening)” is the
within-subjects independent nominal variable and “anxiety” is the dependent continuous
variable.

An additional assumption that we should take under consideration when we apply Repeated
Measures ANOVA is the assumption of sphericity. This assumption is not critical for applying or
not applying the specific analysis. Nonetheless, we should take under consideration the specific
analysis in order to report the correct component of the results.
Sphericity means that the differences between the various pairs of conditions on the levels of
the continuous dependent variable have roughly the same level of variance. For assessing
sphericity we are using the Mauchly’s test. More specifically, if the significance value of the
Mauchly’s test is above the level of 0.05 we should consider that sphericity is met. On the other
hand, if the significance value of the Mauchly’s test is equal or less than 0.05 we should
consider that sphericity is violated.

The Friedman test is the non-parametric equivalent of Repeated Measures ANOVA. It should be
applied in the two following cases: a) we compare one group under more than two conditions
on the levels of an ordinal variable, and b) we compare one group under more than two
conditions on the levels of a continuous variable but the assumptions of normality and/or
independence are violated.

1) We are comparing the letter grades of the same group of students during the 1st, the 2nd
and the last year of their studies. Which statistical test should be applied?
a) One-way ANOVA
b) Repeated Measures ANOVA
 c) Friedman test
d) Kruskal-Wallis test
The Friedman test should be applied because we are comparing one group under more than
two conditions on the levels of an ordinal variable.

2) We are studying the impact of handedness (left-handers/right-handers/mixed-handers)

on anxiety (interval). If normality is met which statistical test should be applied?
a) One-way ANOVA
b) Repeated Measures ANOVA
c) Kruskal-Wallis test
d) Friedman test
One-way ANOVA should be applied because we are comparing more than two groups on the
levels of a continuous variable and normality is met (we have one between-subjects
independent nominal variable and a dependent continuous variable)
3) We are comparing the memory performance (interval) of the same group of individuals
at the age of 60, 65 and 70 years. If normality is met which test should be applied?
a) One-way ANOVA
b) Repeated Measures ANOVA
c) Kruskal-Wallis test
d) Friedman test
Repeated Measures ANOVA should be applied because we are comparing one group under
more than two conditions on the levels of a continuous variable and normality is met.

4) We are studying the impact of place of residence (urban/semi-urban/rural) on the level of

education (ordinal). Which test should be applied?
a) one-way ANOVA
b) Kruskal-Wallis test
c) Repeated Measures ANOVA
d) Friedman test
The Kruskal-Wallis test should be applied because we are comparing three different groups of
individuals on the levels of an ordinal variable. We have one between-subjects independent
nominal variable (place of residence) and a dependent ordinal variable.
5) We study the impact of the level of noise(low/moderate/high) on attention (interval). If we
have a within-subjects design and normality is not met which test should be applied?
a) One-way ANOVA
b) Repeated Measures ANOVA
c) Friedman test
d) Kruskal-Wallis test
The Friedman test should be applied because we have one within-subjects independent
nominal variable (level of noise), a dependent continuous variable (attention) and normality is
not met. If normality was met we should apply Repeated Measures ANOVA.

Two-way ANOVA is applied when we are studying the impact of two between-subjects
independent categorical variables on the levels of a dependent continuous variable.
Additionally, the assumptions of normality and independence should be met for applying the
specific analysis.
For example, we could apply two-way ANOVA in order to study the impact of gender
(males/females) and handedness (right-handers/left-handers) on memory performance
(interval).
When we apply two-way ANOVA we study the main effects of the two independent categorical
variables on the dependent variable as well as the interaction effect of the two independent
categorical variables on the dependent variable.
For example, we can study the main effect of gender on memory performance. This means that
we are comparing males and females on memory performance without taking under
consideration if someone is left-handed or right-handed.

Similarly, we can study the main effect of handedness on memory performance. This means
that we are comparing left-handers and right-handers on memory performance without taking
under consideration if someone is male or female.
A significant interaction effect is present when the impact of the one independent categorical
variable on the dependent variable is affected by the levels of the other independent
categorical variable. For example, we could have a significant interaction effect if the pattern of
difference between right-handers and left-handers on memory performance is affected by
whether someone is male or female.
Interaction graph I indicates a trend for a potentially significant interaction effect because the
impact of handedness on memory appears to depend on the factor of gender. This is because in
the case of males right-handers go better than the left-handers on the memory test whereas in
the case of females we have the opposite pattern with left-handers achieving a better level of
performance than the right-handers.

Interaction graph II also indicates a trend for a potentially significant interaction effect because
the impact of handedness on memory appears to depend on the factor of gender. This is
because in the case of males right-handers and left-handers have a similar level of memory
performance whereas in the case of females we can observe a clear advantage of the right-
handers as compared to the left-handers.
Interaction graph III indicates absence of a trend for a potentially significant interaction effect
because the impact of handedness on memory does not appear to be affected by the factor of
gender. This is because the amount of difference between left-handers and right-handers on
the memory test is the same for males and females. Generally, parallel lines in the interaction
graph imply absence of a trend for a potentially significant interaction effect.

Mixed ANOVA should be applied when we study the impact of at least one between-subjects
independent categorical variable and at least one within-subjects independent categorical
variable on the levels of a dependent variable.
For example, we could apply mixed ANOVA in order to explore if a specific intervention
(pre/post) is more effective on reducing anxiety (interval) either in the case of males or the case
of females.
In this example, gender (males/females) is the between-subjects independent categorical
variable. The intervention (pre/post) is the within-subjects independent categorical variable.
The dependent continuous variable is anxiety (interval).

Similarly, we could apply mixed ANOVA in order to explore if the level of noise (low/high)
affects at a greater extent the memory performance (interval) of older as compared to younger
adults.
In this example, the age group (young-adults/older-adults) is the between-subjects
independent categorical variable. Level of noise (low/high) is the within-subjects independent
categorical variable and memory performance is the dependent continuous variable.
 1) We are studying the impact of gender (males/females) and ethnicity (Greeks/Swedish)
on the levels of extraversion (interval). Which test should be applied?
a) One-way ANOVA
b) Two-way ANOVA
c) Mixed ANOVA
d) Repeated Measures ANOVA
Two-way ANOVA should be selected because we are studying the impact of two between-
subjects independent categorical variables on the levels of a dependent continuous variable.

We want to explore if a specific intervention (pre/post) for reducing school bullying (interval) is
more effective in the case of primary or secondary school students. Which test should be
applied?
a) One-way ANOVA
b) Two-way ANOVA
c) Mixed ANOVA
d) Repeated Measures ANOVA

ANCOVA: we are applying an ANCOVA analysis when apart from the independent categorical
variable or variables we also include at least one continuous variable that should related to the
dependent variable of the study. This continuous variable is called “covariate” and this is why
the corresponding analysis is named ANCOVA. Hence, the covariates in the case of the ANCOVA
analysis are continuous numerical variables that should be related to the dependent variable of
the study.
Example of one-way ANCOVA: we explore the impact of handedness (right-handers/left-
handers/mixed-handers) on memory performance (interval) after controlling for the role of
attention (interval) on memory.
In this example, handedness (right-handers/left-handers/mixed-handers) is the independent
categorical variable, memory is the dependent continuous variable, and attention is the
covariate.
Example of two-way ANCOVA: we explore the impact of ethnicity (Greeks/Swedish) and age-
level (young-adults/older-adults) on the levels of extraversion (interval) after controlling for the
role of anxiety (interval) on extraversion.
In this example, ethnicity (Greeks/Swedish) and age-level (young-adults/older-adults) are the
two independent categorical variables, extraversion is the dependent continuous variable, and
anxiety is the covariate.
By including a covariate in our research design our goal is to explain an additional part of
unsystematic variance that exists in our data. Hence, the unsystematic variance can be reduced
and, therefore, we can increase the F-value for the main independent variables of the study.
This means that it becomes more feasible then main independent variables of the study to
reach the level of statistical significance.

It is also important for the covariate not to be related to the independent categorical variable of
the study. This means that we want the covariate and the independent categorical variable to
be independent from each other.

An additional assumption that should be considered when we apply ANCOVA analysis is the
homogeneity of regression slopes. This means that the level of the correlation between the
covariate and the dependent variable of the study is roughly the same for all categories of the
independent categorical variable. For example, the level of correlation between Stroop1
(covariate) and Stroop2 (Dependent variable) is roughly the same for the groups of left-handers
and right-handers (the two categories of the independent categorical variable).

1) The aim of the present study is to investigate the effect of gender (males/females) and
alcohol consumption (Yes/no) on depression levels. The statistical test that should be applied is:

a) two-way ANOVA
b) one-way ANOVA
c) Repeated-Measures ANOVA
d) independent-samples t-test
Two-way ANOVA should be applied because we are studying the impact of two between-
subjects independent categorical variables on the levels of a continuous variable.
2) If we know that the assumptions for the application of parametric tests are not fulfilled and
we aim at investigating the effect of blood type on cholesterol levels (interval variable) in a
sample of university students, we should apply a:

a) Repeated Measures ANOVA

b) Kruskal-Wallis test
c) Friedman test
d) one-way ANOVA
The Kruskal-Wallis test should be applied because we have one between-subjects independent
nominal variable (blood type) and a dependent continuous variable (cholesterol) but the
assumptions for applying parametric tests are not met. If these assumptions were met we
should apply one-way ANOVA

3) The anxiety scores of a group of older adults is studied in four different conditions. If the
ANOVA analysis that is performed reveals the presence of a significant F-value how many post-
hoc comparisons should be performed if there is no directional hypothesis?

a) none
b) three
c) four
d) six
A simple way to find out the number of the post-hoc comparisons is to sum up all the numbers
that are smaller than the number of groups or conditions that are compared. Hence in this case
we should sum up the numbers 1, 2 and 3. This means that we have six post-hoc comparisons
(1+2+3).
5) For exploring the impact of alcohol consumption (Yes/No) on memory performance (interval)
after controlling for the role of anxiety (interval) on memory performance we should apply:

a) one-way ANOVA
b) repeated-measures ANOVA
c) two-way ANOVA
d) one-way ANCOVA
One-way ANCOVA should be applied because we study the impact of one independent
categorical variable (alcohol consumption) on the dependent variable (memory performance),
after controlling for the role of a covariate, namely anxiety.

6) The mixed ANOVA should be applied when we have:

a) two between-subjects independent variables

b) two within-subjects independent variables
c) at least one between-subjects independent variable and at least one within-subjects
independent variable
d) one ordinal dependent variable

7) What is the new threshold for statistical significance according to the Bonferroni correction
when performing six post-hoc comparisons?

a) 0.05
b) 0.04
c) 0.008
d) 0.006
In the case of the Bonferroni correction we should divide the initial threshold for statistical
significance (typically this value is 0.05) with the number of the post-hoc comparisons that are
conducted: 0.05/6=0.008
8) If the assumptions for the application of parametric tests in our sample are met and we are
comparing the memory performance of one group of individuals under three conditions what
statistical test should be applied?

a) one-way ANOVA
b) repeated-measures ANOVA
c) Friedman’s test
d) Kruskal Wallis test
Repeated measures ANOVA should be applied because we are comparing one group under
three different conditions and the assumptions for applying parametric tests are met. If these
assumptions were not met we should apply the Friedman test.

Simple linear regression & Multiple Linear Regression

In the case of simple linear regression we are using one predictor variable for predicting the
levels of an outcome variable. For example, we may use anxiety in order to predict the levels of
depression. In this example, anxiety is the predictor variable and depression is the outcome
variable.

In the case of multiple linear regression we are using more than one predictor variables for
predicting the levels of an outcome variable. For example, we may use anxiety, physical
exercise and loneliness in order to predict the levels of depression. In this example, anxiety,
physical exercise and loneliness are the three predictor variables. And depression is the
outcome variable of the study.

In both cases, simple linear regression and multiple linear regression, the regression model
corresponds to a straight line. Hence, through the use of this straight line that corresponds to
the regression model we are trying to predict the levels of the outcome variable.

Theoretically we can have numerous straight lines. Nonetheless, from all possible straight lines
we are trying to find the line that makes the best possible prediction of the outcome variable
for the specific set of predictor variables. And this is the regression line.
Any straight line can defined through the use of the following formula:
Y=a*X+b
“a” is the slope or gradient of the line
“b” is the intercept of the line. The point where the line crosses the Y axis.
Hence, if we know the intercept and the slope we can define any straight line.
In the case of simple linear regression, the regression line can be presented in the following
way:
Outcome Variable =a * Predictor Variable + b
Hence, if Depression is the outcome variable and anxiety is the predictor variable, our
regression line would be the following:

Depression= a * Anxiety + b

The residuals correspond to the difference between the prediction and the actual value of the
outcome variable. Among all possible straight lines the regression line corresponds to the line
that on average has the smallest amount of residuals. Hence, the regression line is the line that
makes the best possible prediction of the outcome variable for the specific set of predictors.

The next step in this process is to objectively evaluate the effectiveness of the prediction that is
made by the regression line. This is achieved by comparing the accuracy of the prediction that is
made by the regression line with the effectiveness of the prediction that is made by the mean
value of the outcome variable (that offers a vague and simplistic prediction). What we want to
observe is that regression line improves in a statistically significant way the accuracy of the
prediction that is made by the mean value of the outcome variable. If this is the case, then we
have an effective regression model.
When we apply regression analysis through SPSS we should check the significance value of the
ANOVA table from the corresponding output on SPSS. If the significance value is equal or less
than 0.05 this implies that we have an effective regression model because we are improving in
a statistically significant way the accuracy of our prediction as compared to the prediction that
is made by the mean value of the outcome variable.
The R-square value from the model-Summary Table is an important statistical index because it
corresponds to the part of the variation in the levels of the outcome variable that is explained
by the regression model. In the example that we have in the notes (Stroop1 predictor and
Stroop2 outcome variable) the R-square value is 0.375). This means that the regression model
explains 37.5% of the variation in the levels of the outcome variable, namely of Stroop2
performance.

From the coefficient-Table we can evaluate if a specific predictor has a statistically significant
unique contribution to the regression model by inspecting the corresponding significance value.
Also, another important statistical index that we obtain from the coefficient Table is the
standardized beta value.
The standardized beta value indicates how many that standard deviation units the outcome
variable will change if the corresponding predictor increases by one standard deviation unit.
For example, for Stroop1 the corresponding standardized beta value is equal to 0.613. This
means that if Stroop1 increases by 1 standard deviation unit, then Stroop2 (outcome variable)
will increase by 0.613 standard deviation units.
By using the standardized beta values we can also understand if the association between the
specific predictor variable and the outcome variable is positive or negative. More specifically, if
the standardized beta value is positive this means that we have a positive association between
the specific predictor and the outcome variable. On the other hand, if the standardized beta
value is negative this means that we have a negative association between the specific predictor
and the outcome variable.

Also, if we have multiple predictors by using the standardized beta values we can understand
which of the predictors makes the most important contribution to the regression model. More
specifically, the predictor that makes the most important contribution to the regression model
is the predictor that has the largest standardized beta value in absolute terms (we don’t care if
it is positive or negative)

The Unstandardized B value indicates how many units the outcome variable will change if the
corresponding predictor increases by 1 unit.
Assumptions of regression analysis:
1) The predictor variables can be interval, ratio, and binary
2) The outcome variable should interval or ratio
3) When we apply regression analysis we should avoid the presence of multicollinearity.
Multicollinearity describes a condition where two predictor variables are very strongly
correlated with each other, with a correlation coefficient close to 0.8 or -0.8 or even
closer to 1 or to-1.
4) When we apply regression analysis we should avoid the presence of singularity.
Singularity means that one predictor variable is the combination of other predictors that
are included in the same regression model
5) When we apply regression analysis we should have homoscedasticity. This means that
the variance of the residuals should be roughly the same for all predicted scores of the
outcome variable. Therefore, this implies that the accuracy of our prediction should be
roughly the same for all the scores of the outcome variable.
6) Normality should be also met when we apply regression analysis. In terms of normality
what is important for us are the residuals to be normally distributed with a mean value
close to zero.
7) The regression model is sensitive to the presence of outliers either in the case of the
predictors or in the case of the outcome variable. Hence, the presence of an important
 number of outliers could create problems in the effectiveness of the overall regression
model.

Formulas for estimating the minimum sample size when we apply regression analysis:
Formula for estimating the minimum sample size when we want to evaluate the effectiveness
of the overall regression model: 50 + 8 * k (k corresponds to the number of the predictors)
For example, if we have 4 predictors the minimum sample size for evaluating the effectiveness
of the overall regression model is the following: 50 + 8 * k = 50 + 8 * 4=50+32=82
Formula for estimating the minimum sample size when we want to evaluate the effectiveness
of the individual predictors as well: 104 + k (k corresponds to the number of the predictors)
For example, if we have 6 predictors the minimum sample size for evaluating the contribution
of the individual predictors is the following: 104 + k = 104 + 6 =110
We have two different types of multiple regression analysis:
1) Standard multiple regression: in this case we place all predictors together in the
regression model and we explore their capacity to predict the levels of an outcome
variable. For example, we explore the capacity of age, education, depression and
anxiety to predict quality of life
2) Hierarchical multiple regression: in this case we have different levels of predictors and
our goal is to explore the capacity of a set of predictor variables to predict the outcome
variable after controlling for the role of another set of predictors. For example, we
explore the capacity of anxiety and depression to predict quality of life after controlling
for the role of age and education on predicting quality of life.
For constructing the regression line we should use the unstandardized B values and not the
standardized beta values. More specifically, the unstandardized B value that corresponds to the
“constant” term is the intercept and the unstandardized B value that corresponds to the
predictor is the slope or gradient of the line

Formula for the regression line is the following: Outcome variable= a* Predictor
variable + b
“a” is the intercept of the line and “b” is the slope of the line
In our example, Stroop2 is the outcome variable and Stroop1 is the predictor. Hence, the
regression line is the following:
Stroop2 = a * Stroop1 + b = 0.647 * Stroop1 + 2.302
Hence, for every value on Stroop1 we can make a prediction for Stroop2 according to the
regression model. For example, if Stroop1 is equal to a 100, then the prediction for Stroop2 is
the following: Stroop2= 0.647 * Stroop1 + 2.302=0.647 * 100 + 2.302=64.7+2.302=67.002
Hence, according to the regression model, if Stroop1 is equal to a 100 the prediction for
Stroop2 is a score of 67.002.