Section 1 Data File DescriptionThe fictional data represents a te.docx

Section 1: Data File Description
The fictional data represents a teacher's recording of student
demographics and performance on quizzes and a final exam
across three sections of the course. Each section consists of 35
students which totals to 105 students (sample size, N = 105).
The dataset has 21 variables but in this case only two variables
will be analyzed. These are: gender and gpa variables. The
gender variable is categorized as nominal since the numbers are
arbitrarily assigned to represent group membership. The ‘gpa’
variable belongs to the interval data group since it has a true
zero which is meaningful. Alternatively, gender could be
categorized as a categorical variable while ‘gpa’ ‘as a
continuous variable.
Section 2: Testing Assumptions
1. Articulate the assumptions of the statistical test.
Paste SPSS output that tests those assumptions and interpret
them. Properly integrate SPSS output wher1e appropriate. Do
not string all output together at the beginning of the section.
All statistical tests operate under a set of assumptions. For the t
test, there are three assumptions:
· The first assumption is independence of observations.
· The outcome variable Y is normally distributed.
· The variance of Y scores is approximately equal across groups
(homogeneity of variance assumption)
Figure 1: histogram of GPA
The histogram above shows that the variable is probably not
normally distributed. The bell shape is absent and two peaks are
evident.
Table 1: descriptives
Descriptives
Statistic

Std. Error
GPA
Mean
2,78
,075
95% Confidence Interval for Mean
Lower Bound
2,63
Upper Bound
2,93
5% Trimmed Mean
2,80
Median
2,72
Variance
,583
Std. Deviation
,764
Minimum
1

Maximum
4
Range
3
Interquartile Range
1
Skewness
-,052
,236
Kurtosis
-,811
,467
With reference to the table 1 above, the ‘GPA’ variable is in the
ideal range for skewness due to the fact that its absolute value
for skewness are is less than .50 (approximately symmetric).
The GPA variable is not ideal but acceptable since its kurtosis
value is greater than .50 but less than 1. This new information
gives mixed signals about the data being normal and only a
normality test could iron out the differences.
Table 2: Normality test
Tests of Normality
Kolmogorov-Smirnova
Shapiro-Wilk

Statistic
df
Sig.
Statistic
df
Sig.
GPA
,091
105
,033
,956
105
,001
a. Lilliefors Significance Correction
Looking at the table above, the p-value is less than 0.05.
Therefore, the null hypothesis is rejected and thus it can be
concluded that the variable is not normally distributed.
However, since the sample size is sufficiently large, one does
not need to worry about this violation. On the other hand,
Levene’s test provides a p =0.566 (table 3) meaning that the
null hypothesis should not be rejected. Thus the homogeneity of
variances assumption is not violated. It’s also assumed that
proper research procedures that maintain independence of
observations were followed. Two of the three assumptions are
met.Section 3: Research Question, Hypotheses, and Alpha Level
The research question for this study is whether a relationship
exists between the ‘Gender’ variable and the ‘GPA’ variable.
Does one particular gender dominate the other in terms of
academic performance (gpa)? The null hypothesis states that the
means of the ‘GPA’ based on gender groups do not differ
significantly whereas the alternative hypothesis states that there
is a significant difference in the two ‘GPA’ groups based on the
gender of the participants. The alternative hypothesis is non-
directional meaning that the GPA for males could be higher than

that for females or the GPA for females could be higher than
that for males. The alpha level to be used is the standard .05
level of significance.
Section 4: Interpretation
1. Paste SPSS output for an inferential statistic. Properly
integrate SPSS output where appropriate. Do not string all
output together at the beginning of the section.
2.
Table 3: Independent samples t test results
Independent Samples Test
Levene's Test for Equality of Variances
t-test for Equality of Means
F
Sig.
t
df
Sig. (2-tailed)
Mean Difference
Std. Error Difference
95% Confidence Interval of the Difference
Lower
Upper
GPA
Equal variances assumed

,331
,566
2,004
103
,048
,302
,151
,003
,601
Equal variances not assumed
1,994
83,974
,049
,302
,151
,001
,603
The test was carried out and indicated a p value 0f 0.048. This
means that we shall have to reject the null hypothesis. This
means that the two means are not equal.
The mean GPA for males and females differed significantly,
t(103) =2.004, p = .048 (two-tailed). Mean GPA for the gender
1 group ( M = 2.90, SD = 0.748) was about 0.302 different from
the mean GPA for the gender 2 ( M = 2.59, SD = 0.763). It
could be said that the GPA for gender 2 is lower than that of
GPA for gender 1. The effect size, as indexed by η^2 , was ;
this is a small effect. The 95% CI for the difference between
sample means, M 1 − M 2, had a lower bound of 0.003 and an
upper bound of 0.601. This technically means that the
difference between the two means cannot be zero since zero is
not included in the confidence interval. This is in line with
previous finding where the null hypothesis was rejected and it

had been concluded that a significant statistical difference
between the means was present.
Section 5: Conclusion
In conclusion, it is evident that a student’s previous
performance between gender 1 and gender 2 is different. Gender
1 had higher GPA than Gender 2. A major limitation of this test
is that one cannot compare more than two groups at a go.
Another test has to be used in conjunction or instead of the
independent samples t test.
Chapter 12 - DUMMY PREDICTOR VARIABLES IN
MULTIPLE REGRESSION
12.1 Research Situations Where Dummy Predictor Variables
Can Be Used
Previous examples of regression analysis have used scores on
quantitative X variables to predict scores on a quantitative Y
variable. However, it is possible to include group membership
or categorical predictor variables as predictors in regression
analysis. This can be done by creating dummy (dichotomous)
predictor variables to represent information about group
membership. A dummy or dichotomous predictor variable
provides yes/no information for questions about group
membership. For example, a simple dummy variable to
represent gender corresponds to the following question: Is the
participant female (0) or male (1)? Gender is an example of a
two-group categorical variable that can be represented by a
single dummy variable.
When we have more than two groups, we can use a set of
dummy variables to provide information about group
membership. For example, suppose that a study includes
members of k = 3 political party groups. The categorical
variable political party has the following scores: 1 = Democrat,
2 = Republican, and 3 = Independent. We might want to find out
whether mean scores on a quantitative measure of political
conservatism (Y) differ across these three groups. One way to

answer this question is to perform a one-way analysis of
variance (ANOVA) that compares mean conservatism (Y) across
the three political party groups. In this chapter, we will see that
we can also use regression analysis to evaluate how political
party membership is related to scores on political conservatism.
However, we should not set up a regression to predict scores on
conservatism from the multiple-group categorical variable
political party, with party membership coded 1 = Democrat, 2 =
Republican, and 3 = Independent. Multiple-group categorical
variables usually do not work well as predictors in regression,
because scores on a quantitative outcome variable, such as
“conservatism,” will not necessarily increase linearly with the
score on the categorical variable that provides information
about political party. The score values that represent political
party membership may not be rank ordered in a way that is
monotonically associated with changes in conservatism; as we
move from Group 1 = Democrat to Group 2 = Republican,
scores on conservatism may increase, but as we move from
Group 2 = Republican to Group 3 = Independent, conservatism
may decrease. Even if the scores that represent political party
membership are rank ordered in a way that is monotonically
associated with level of conservatism, the amount of change in
conservatism between Groups 1 and 2 may not be equal to the
amount of change in conservatism between Groups 2 and 3. In
other words, scores on a multiple-group categorical predictor
variable (such as political party coded 1 = Democrat, 2 =
Republican, and 3 = Independent) are not necessarily linearly
related to scores on quantitative variables.
If we want to use the categorical variable political party to
predict scores on a quantitative variable such as conservatism,
we need to represent the information about political party
membership in a different way. Instead of using one categorical
predictor variable with codes 1 = Democrat, 2 = Republican,
and 3 = Independent, we can create two dummy or dichotomous
predictor variables to represent information about political party
membership, and we can then use these two dummy variables as

predictors of conservatism scores in a regression. Political party
membership can be assessed by creating dummy variables
(denoted by D1 and D2) that correspond to two yes/no
questions. In this example, the first dummy variable D1
corresponds to the following question: Is the participant a
member of the Democratic Party? Coded 1 = yes, 0 = no. The
second dummy variable D2 corresponds to the following
question: Is the participant a member of the Republican Party?
Coded 1 = yes, 0 = no. We assume that group memberships for
individuals are mutually exclusive and exhaustive—that is, each
case belongs to only one of the three groups, and every case
belongs to one of the three groups identified by the categorical
variable. When these conditions are met, a third dummy
variable is not needed to identify the members of the third
group because, for Independents, the answers to the first two
questions that correspond to the dummy variables D1 and D2
would be no. In general, when we have k groups or categories, a
set of k − 1 dummy variables is sufficient to provide complete
information about group membership. Once we have represented
political party group membership by creating scores on two
dummy variables, we can set up a regression to predict scores
on conservatism (Y) from the scores on the two dummy
predictor variables D1 and D2:
In this chapter, we will see that the information about the
association between group membership (represented by D1 and
D2) and scores on the quantitative Y variable that can be
obtained from the regression analysis in Equation 12.1 is
equivalent to the information that can be obtained from a one-
way ANOVA that compares means on Y across groups or
categories. It is acceptable to use dichotomous predictor
variables in regression and correlation analysis. This works
because (as discussed in Chapter 8) a dichotomous categorical
variable has only two possible score values, and the only
possible relationship between scores on a dichotomous predictor
variable and a quantitative outcome variable is a linear one.

That is, as you move from a score of 0 = female on gender to a
score of 1 = male on gender, mean height or mean annual salary
may increase, decrease, or stay the same; any change that can be
observed across just two groups can be represented as linear.
Similarly, if we represent political party membership using two
dummy variables, each dummy variable represents a contrast
between the means of two groups; for example, the D1 dummy
variable can represent the difference in mean conservatism
between Democrats and Independents, and the D2 dummy
variable can represent the mean difference in conservatism
between Republicans and Independents.
This chapter uses empirical examples to demonstrate that
regression analyses that use dummy predictor variables (similar
to Equation 12.1) provide information that is equivalent to the
results of more familiar analyses for comparison of group means
(such as ANOVA). There are several reasons why it is useful to
consider dummy predictor variables as predictors in regression
analysis. First, the use of dummy variables as predictors in
regression provides a simple demonstration of the fundamental
equivalence between ANOVA and multiple regression; ANOVA
and regression are both special cases of a more general analysis
called the general linear model (GLM). Second, researchers
often want to include group membership variables (such as
gender) along with other predictors in a multiple regression.
Therefore, it is useful to examine examples of regression that
include dummy variables along with quantitative predictors.
The computational procedures for regression remain the same
when we include one or more dummy predictor variables. The
most striking difference between dummy variables and
quantitative variables is that the scores on dummy variables
usually have small integer values (such as 1, 0, and −1). The use
of small integers as codes simplifies the interpretation of the
regression coefficients associated with dummy variables. When
dummy variables are used as predictors in a multiple regression,
the b raw score slope coefficients provide information about
differences between group means. The specific group means

that are compared differ depending on the method of coding that
is used for dummy variables, as explained in the following
sections. Except for this difference in the interpretation of
regression coefficients, regression analysis remains essentially
the same when dummy predictor variables are included.
12.2 Empirical Example
The hypothetical data for this example are provided by a study
of predictors of annual salary in dollars for a group of N = 50
college faculty members; the complete data appear in Table
12.1. Predictor variables include the following: gender, coded 0
= female and 1 = male; years of job experience; college, coded
1 = Liberal Arts, 2 = Sciences, 3 = Business; and an overall
merit evaluation. Additional columns in the SPSS data
worksheet in Figure 12.1, such as D1, D2, E1, and E2, represent
alternative ways of coding group membership, which are
discussed later in this chapter. All subsequent analyses in this
chapter are based on the data in Table 12.1.
The first research question that can be asked using these data is
whether there is a significant difference in mean salary between
males and females (ignoring all other predictor variables). This
question could be addressed by conducting an independent
samples t test to compare male and female means on salary. In
this chapter, a one-way ANOVA is performed to compare mean
salary for male versus female faculty; then, salary is predicted
from gender by doing a regression analysis to predict salary
scores from a dummy variable that represents gender. The
examples presented in this chapter demonstrate that ANOVA
and regression analysis provide equivalent information about
gender differences in mean salary. Examples or demonstrations
such as the ones presented in this chapter do not constitute
formal mathematical proof. Mathematical statistics textbooks
provide formal mathematical proof of the equivalence of
ANOVA and regression analysis.
Table 12.1 Hypothetical Data for Salary and Predictors of
Salary for N = 50 College Faculty

NOTES: Salary, annual salary in thousands of dollars; years,
years of job experience; gender, dummy-coded gender (0 =
female and 1 = male); college, membership coded 1 = Liberal
Arts, 2 = Sciences, 3 = Business.
Figure 12.1 SPSS Data Worksheet for Hypothetical Faculty
Salary Study
The second question that will be addressed is whether there are
significant differences in salary across the three colleges. This
question will be addressed by doing a one-way ANOVA to
compare mean salary across the three college groups and by
using dummy variables that represent college group membership
as predictors in a regression. This example demonstrates that
membership in k groups can be represented by a set of (k − 1)
dummy variables.
12.3 Screening for Violations of Assumptions
When we use one or more dummy variables as predictors in
regression, the assumptions are essentially the same as for any
other regression analysis (and the assumptions for a one-way
ANOVA). As in other applications of ANOVA and regression,
scores on the outcome variable Y should be quantitative and
approximately normally distributed. If the Y outcome variable
is categorical, logistic regression analysis should be used
instead of linear regression; a brief introduction to binary
logistic regression is presented in Chapter 21. Potential
violations of the assumption of an approximately normal
distribution shape for the Y outcome variable can be assessed
by examining a histogram of scores on Y; the shape of this
distribution should be reasonably close to normal. As described
in Chapter 4, if there are extreme outliers or if the distribution
shape is drastically different from normal, it may be appropriate
to drop a few extreme scores, modify the value of a few extreme
scores, or, by using a data transformation such as the logarithm

of Y, make the distribution of Y more nearly normal.
The variance of Y scores should be fairly homogeneous across
groups—that is, across levels of the dummy variables. The F
tests used in ANOVA are fairly robust to violations of this
assumption, unless the numbers of scores in the groups are
small and/or unequal. When comparisons of means on Y across
multiple groups are made using the SPSS t test procedure, one-
way ANOVA, or GLM, the Levene test can be requested to
assess whether the homogeneity of variance is seriously
violated. SPSS multiple regression does not provide a formal
test of the assumption of homogeneity of variance across
groups. It is helpful to examine a graph of the distribution of
scores within each group (such as a boxplot) to assess visually
whether the scores of the Y outcome variable appear to have
fairly homogeneous variances across levels of each X dummy
predictor variable.
The issue of group size should also be considered in preliminary
data screening (i.e., How many people are there in each of the
groups represented by codes on the dummy variables?). For
optimum statistical power and greater robustness to violations
of assumptions, such as the homogeneity of variance
assumption, it is preferred that there are equal numbers of
scores in each group.1 The minimum number of scores within
each group should be large enough to provide a reasonably
accurate estimate of group means. For any groups that include
fewer than 10 or 20 scores, estimates of the group means may
have confidence intervals that are quite wide. The guidelines
about the minimum number of scores per group from Chapter 5
(on the independent samples t test) and Chapter 6 (between-
subjects [between-S] one-way ANOVA) can be used to judge
whether the numbers in each group that correspond to a dummy
variable, such as gender, are sufficiently large to yield
reasonable statistical power and reasonable robustness against
violations of assumptions.
For the hypothetical faculty salary data in Table 12.1, the
numbers of cases within the groups are (barely) adequate. There

were 20 female and 30 male faculty in the sample, of whom 22
were Liberal Arts faculty, 17 were Sciences faculty, and 11
were Business faculty. Larger group sizes are desirable in real-
world applications of dummy variable analysis; relatively small
numbers of cases were used in this example to make it easy for
students to verify computations, such as group means, by hand.
All the SPSS procedures that are used in this chapter, including
boxplot, scatter plot, Pearson correlation, one-way ANOVA,
and linear regression, have been introduced and discussed in
more detail in earlier chapters. Only the output from these
procedures appears in this chapter. For a review of the menu
selections and the SPSS dialog windows for any of these
procedures, refer to the chapter in which each analysis was first
introduced.
To assess possible violations of assumptions, the following
preliminary data screening was performed. A histogram was set
up to assess whether scores on the quantitative outcome variable
salary were reasonably normally distributed; this histogram
appears in Figure 12.2. Although the distribution of salary
values was multimodal, the salary scores did not show a
distribution shape that was drastically different from a normal
distribution. In real-life research situations, salary distributions
are often positively skewed, such that there is a long tail at the
upper end of the distribution (because there is usually no fixed
upper limit for salary) and a truncated tail at the lower end of
the distribution (because salary values cannot be lower than 0).
Skewed distributions of scores can sometimes be made more
nearly normal in shape by taking the base 10 or natural log of
the salary scores. Logarithmic transformation was judged
unnecessary for the artificial salary data that are used in the
example in this chapter. Also, there were no extreme outliers in
salary, as can be seen in Figure 12.2.
Figure 12.2 Histogram of Salary Scores
A boxplot of salary scores was set up to examine the
distribution of outcomes for the female and male groups (Figure

12.3). A visual examination of this boxplot did not reveal any
extreme outliers within either group; median salary appeared to
be lower for females than for males. The boxplot suggested that
the variance of salary scores might be larger for males than for
females. This is a possible violation of the homogeneity of
variance assumption; in the one-way ANOVA presented in a
later section, the Levene test was requested to evaluate whether
this difference between salary variances for the female and male
groups was statistically significant (and the Levene test was not
significant).
Figure 12.4 shows a scatter plot of salary (on the Y axis) as a
function of years of job experience (on the X axis) with
different case markers for female and male faculty. The relation
between salary and years appears to be linear, and the slope that
predicts salary from years appears to be similar for the female
and male groups, although there appears to be a tendency for
females to have slightly lower salary scores than males at each
level of years of job experience; also, the range for years of
experience was smaller for females (0–15) than for males (0–
22). Furthermore, the variance of salary appears to be
reasonably homogeneous across years in the scatter plot that
appears in Figure 12.4, and there were no extreme bivariate
outliers. Subsequent regression analyses will provide us with
more specific information about gender and years as predictors
of salary. We will use a regression analysis that includes the
following predictors: years, a dummy variable to represent
gender, and a product of gender and years. The results of this
regression will help answer the following questions: Is there a
significant increase in predicted salary associated with years of
experience (when gender is statistically controlled)? Is there a
significant difference in predicted salary between males and
females (when years of experience is statistically controlled)?
Figure 12.3 Boxplot of Salary Scores for Female and Male
Groups
Figure 12.4 Scatter Plot of Salary by Years With Case Markers

for Gender
12.4 Issues in Planning a Study
Essentially, when we use a dummy variable as a predictor in a
regression, we have the same research situation as when we do a
t test or ANOVA (both analyses predict scores on the Y
outcome variable for two groups). When we use several dummy
variables as predictors in a regression, we have the same
research situation as in a one-way ANOVA (both analyses
compare means across several groups). Therefore, the issues
reviewed in planning studies that use t tests (in Chapter 5) and
studies that use one-way ANOVA (in Chapter 6) are also
relevant when we use a regression analysis as the method of
data analysis. To put it briefly, if the groups that are compared
received different “dosage” levels of some treatment variable,
the dosage levels need to be far enough apart to produce
detectable differences in outcomes. If the groups are formed on
the basis of participant characteristics (such as age), the groups
need to be far enough apart on these characteristics to yield
detectable differences in outcome. Other variables that might
create within-group variability in scores may need to be
experimentally or statistically controlled to reduce the
magnitude of error variance, as described in Chapters 5 and 6.
It is important to check that every group has a reasonable
minimum number of cases. If any group has fewer than 10
cases, the researcher may decide to combine that group with one
or more other groups (if it makes sense to do so) or exclude that
group from the analysis. The statistical power tables that
appeared in Chapters 5 and 6 can be used to assess the minimum
sample size needed per group to achieve reasonable levels of
statistical power.
12.5 Parameter Estimates and Significance Tests for
Regressions With Dummy Variables
The use of one or more dummy predictor variables in regression

analysis does not change any of the computations for multiple
regression described in Chapter 11. The estimates of b
coefficients, the t tests for the significance of individual b
coefficients, and the overall F test for the significance of the
entire regression equation are all computed using the methods
described in Chapter 11. Similarly, sr2 (the squared semipartial
correlation), an estimate of the proportion of variance uniquely
predictable from each dummy predictor variable, can be
calculated and interpreted for dummy predictors in a manner
similar to that described in Chapter 11. Confidence intervals for
each b coefficient are obtained using the methods described in
Chapter 11.
However, the interpretation of b coefficients when they are
associated with dummy-coded variables is slightly different
from the interpretation when they are associated with
continuous predictor variables. Depending on the method of
coding that is used, the b0 (intercept) coefficient may
correspond to the mean of one of the groups or to the grand
mean of Y. The bi coefficients for each dummy-coded predictor
variable may correspond to contrasts between group means or to
differences between group means and the grand mean.
12.6 Group Mean Comparisons Using One-Way Between-S
ANOVA12.6.1 Gender Differences in Mean Salary
A one-way between-S ANOVA was performed to assess whether
mean salary differed significantly between female and male
faculty. No other variables were taken into account in this
analysis. A test of the homogeneity of variance assumption was
requested (the Levene test).
The Levene test was not statistically significant: F(1, 48) =
2.81, p = .1. Thus, there was no statistically significant
difference between the variances of salary for the female and
male groups; the homogeneity of variance assumption was not
violated. The mean salary for males (Mmale) was 49.9 thousand
dollars per year; the mean salary for females (Mfemale) was
40.8 thousand dollars per year. The difference between mean

annual salary for males and females was statistically significant
at the conventional α = .05 level: F(1, 48) = 13.02, p = .001.
The difference between the means (Mmale − Mfemale) was +9.1
thousand dollars (i.e., on average, male faculty earned about 9.1
thousand dollars more per year than female faculty). The eta-
squared effect size for this gender difference was .21; in other
words, about 21% of the variance in salaries could be predicted
from gender. This corresponds to a large effect size (refer to
Table 5.2 for suggested verbal labels for values of η2 and R2).
This initial finding of a gender difference in mean salary is not
necessarily evidence of gender bias in salary levels. Within this
sample, there was a tendency for female faculty to have fewer
years of experience and for male faculty to have more years of
experience, as well as salary increases as a function of years of
experience. It is possible that the gender difference we see in
this ANOVA is partly or completely accounted for by
differences between males and females in years of experience.
Subsequent analyses will address this by examining whether
gender still predicts different levels of salary when “years of
experience” is statistically controlled by including it as a
second predictor of salary.
Note that when the numbers of cases in the groups are unequal,
there are two different ways in which a grand mean can be
calculated. An unweighted grand mean of salary can be obtained
by simply averaging male and female mean salary, ignoring
sample size. The unweighted grand mean is (40.8 + 49.9)/2 =
45.35. However, in many statistical analyses, the estimate of the
grand mean is weighted by sample size. The weighted grand
mean in this example is found as follows:
Figure 12.5 One-Way Between-S ANOVA: Mean Salary for
Females and Males
The grand mean for salary reported in the one-way ANOVA in
Figure 12.5 corresponds to the weighted grand mean. (This
weighted grand mean is equivalent to the sum of the 50

individual salary scores divided by the number of scores in the
sample.) In the regression analyses reported later in this
chapter, the version of the grand mean that appears in the
results corresponds to the unweighted grand mean.12.6.2
College Differences in Mean Salary
In a research situation that involves a categorical predictor
variable with more than two levels or groups and a quantitative
outcome variable, the most familiar approach to data analysis is
a one-way ANOVA. To evaluate whether mean salary level
differs for faculty across the three different colleges in the
hypothetical dataset in Table 12.1, we can conduct a between-S
one-way ANOVA using the SPSS ONEWAY procedure. The
variable college is coded 1 = Liberal Arts, 2 = Sciences, and 3 =
Business. The outcome variable, as in previous analyses, is
annual salary in thousands of dollars. Orthogonal contrasts
between colleges were also requested by entering custom
contrast coefficients. The results of this one-way ANOVA
appear in Figure 12.6.
The means on salary were as follows: 44.8 for faculty in Liberal
Arts, 44.7 in Sciences, and 51.6 in Business. The overall F for
this one-way ANOVA was not statistically significant: F(2, 47)
= 2.18, p = .125. The effect size, eta squared, was obtained by
taking the ratio SSbetween/SStotal; for this ANOVA, η2 = .085.
This is a medium effect. In this situation, if we want to use this
sample to make inferences about some larger population of
faculty, we would not have evidence that the proportion of
variance in salary that is predictable from college is
significantly different from 0. However, if we just want to
describe the strength of the association between college and
salary within this sample, we could say that, for this sample,
about 8.5% of the variance in salary was predictable from
college. The orthogonal contrasts that were requested made the
following comparison. For Contrast 1, the custom contrast
coefficients were +1, −1, and 0; this corresponds to a
comparison of the mean salaries between College 1 (Liberal
Arts) and College 2 (Sciences); this contrast was not

statistically significant: t(47) = .037, p = .97. For Contrast 2,
the custom contrast coefficients were +1, +1, and −2; this
corresponds to a comparison of the mean salary for Liberal Arts
and Sciences faculty combined, compared with the Business
faculty. This contrast was statistically significant: t(47) =
−2.082, p = .043. Business faculty had a significantly higher
mean salary than the two other colleges (Liberal Arts and
Sciences) combined.
Figure 12.6 One-Way Between-S ANOVA: Mean Salary Across
Colleges
The next sections show that regression analyses with dummy
predictor variables can be used to obtain the same information
about the differences between group means. Dummy variables
that represent group membership (such as female and male, or
Liberal Arts, Sciences, and Business colleges) can be used to
predict salary in regression analyses. We will see that the
information about differences among group means that can be
obtained by doing regression with dummy predictor variables is
equivalent to the information that we can obtain from one-way
ANOVA. In future chapters, both dummy variables and
quantitative variables are used as predictors in regression.
12.7 Three Methods of Coding for Dummy Variables
The three coding methods that are most often used for dummy
variables are given below (the details regarding these methods
are presented in subsequent sections along with empirical
examples):
1. Dummy coding of dummy variables
2. Effect coding of dummy variables
3. Orthogonal coding of dummy variables
In general, when we have k groups, we need only (k − 1)
dummy variables to represent information about group

membership. Most dummy variable codes can be understood as
answers to yes/no questions about group membership. For
example, to represent college group membership, we might
include a dummy variable D1 that corresponds to the question,
“Is this faculty member in Liberal Arts?” and code the
responses as 1 = yes, 0 = no. If there are k groups, a set of (k −
1) yes/no questions provides complete information about group
membership. For example, if a faculty member reports
responses of “no” to the questions, “Are you in Liberal Arts?”
and “Are you in Science?” and there are only three groups, then
that person must belong to the third group (in this example,
Group 3 is the Business college).
The difference between dummy coding and effect coding is in
the way in which codes are assigned for members of the last
group—that is, the group that does not correspond to an explicit
yes/no question about group membership. In dummy coding of
dummy variables, members of the last group receive a score of 0
on all the dummy variables. (In effect coding of dummy
variables, members of the last group are assigned scores of −1
on all the dummy variables.) This difference in codes results in
slightly different interpretations of the b coefficients in the
regression equation, as described in the subsequent sections of
this chapter.12.7.1 Regression With Dummy-Coded Dummy
Predictor Variables12.7.1.1 Two-Group Example With a
Dummy-Coded Dummy Variable
Suppose we want to predict salary (Y) from gender; gender is a
dummy-coded dummy variable with codes of 0 for female and 1
for male participants in the study. In a previous section, this
difference was evaluated by doing a one-way between-S
ANOVA to compare mean salary across female and male
groups. We can obtain equivalent information about the
magnitude of gender differences in salary from a regression
analysis that uses a dummy-coded variable to predict salary. For
this simple two-group case (prediction of salary from dummy-
coded gender), we can write a regression equation using gender
to predict salary (Y) as follows:

From Equation 12.2, we can work out two separate prediction
equations: one that makes predictions of Y for females and one
that makes predictions of Y for males. To do this, we substitute
the values of 0 (for females) and 1 (for males) into Equation
12.2 and simplify the expression to obtain these two different
equations:
These two equations tell us that the constant value b0 is the best
prediction of salary for females, and the constant value (b0 +
b1) is the best prediction of salary for males. This implies that
b0 = mean salary for females, b0 + b1 = mean salary for males,
and b1 = the difference between mean salary for the male versus
female groups. The slope coefficient b1 corresponds to the
difference in mean salary for males and females. If the b1 slope
is significantly different from 0, it implies that there is a
statistically significant difference in mean salary for males and
females.
The results of the regression in Figure 12.7 provide the
numerical estimates for the raw score regression coefficients
(b0 and b1) for this set of data:
Salary′ = 40.8 + 9.1 × Gender.
For females, with gender = 0, the predicted mean salary given
by this equation is 40.8 + 9.1 × 0 = 40.8. Note that this is the
same as the mean salary for females in the one-way ANOVA
output in Figure 12.5. For males, with gender = 1, the predicted
salary given by this equation is 40.8 + 9.1 = 49.9. Note that this
value is equal to the mean salary for males in the one-way
ANOVA in Figure 12.5.
The b1 coefficient in this regression was statistically
significant: t(48) = 3.61, p = .001. The F test reported in Figure
12.6 is equivalent to the square of the t test value for the null
hypothesis that b1 = 0 in Figure 12.7 (t = +3.608, t2 = 13.02).
Note also that the eta-squared effect size associated with the
ANOVA (η2 = .21) and the R2 effect size associated with the

regression were equal; in both analyses, about 21% of the
variance in salaries was predictable from gender.
When we use a dummy variable with codes of 0 and 1 to
represent membership in two groups, the value of the b0
intercept term in the regression equation is equivalent to the
mean of the group for which the dummy variable has a value of
0. The b1 “slope” coefficient represents the difference (or
contrast) between the means of the two groups. The slope, in
this case, represents the change in the mean level of Y when
you move from a code of 0 (female) to a code of 1 (male) on the
dummy predictor variable.
Figure 12.7 Regression to Predict Salary From Dummy-Coded
Gender
Figure 12.8 shows a scatter plot of salary scores (on the Y axis)
as a function of gender code (on the X axis). In this graph, the
intercept b0 corresponds to the mean on the dependent variable
(salary) for the group that had a dummy variable score of 0
(females). The slope, b1, corresponds to the difference between
the means for the two groups—that is, the change in the
predicted mean when you move from a code of 0 to a code of 1
on the dummy variable. That is, b1 = Mmale − Mfemale, the
change in salary when you move from a score of 0 (female) to a
score of 1 (male). The test of the statistical significance of b1 is
equivalent to the t test of the difference between the mean Y
values for the two groups represented by the dummy variable in
the regression.12.7.1.2 Multiple-Group Example With Dummy-
Coded Dummy Variables
When there are multiple groups (number of groups = k), group
membership can be represented by scores on a set of (k − 1)
dummy variables. Each dummy variable essentially represents a
yes/no question about group membership. In the preceding
example, there are k = 3 college groups in the faculty data. In
this example, we will use two dummy variables, denoted by D1
and D2, to represent information about college membership in a
regression analysis. D1 corresponds to the following question:

Is the faculty member from the Liberal Arts college? 1 = yes, 0
= no. D2 corresponds to the following question: Is the faculty
member from the Sciences college? 1 = yes, 0 = no. For dummy
coding, members of the last group receive a score of 0 on all the
dummy variables. In this example, faculty from the Business
college received scores of 0 on both the D1 and D2 dummy-
coded dummy variable. For the set of three college groups, the
dummy-coded dummy variables that provide information about
college group membership were coded as follows:
Figure 12.8 Graph for Regression to Predict Salary From
Dummy-Coded Gender (0 = Female, 1 = Male)
Now that we have created dummy variables that represent
information about membership in the three college groups as
scores on a set of dummy variables, mean salary can be
predicted from college groups by a regression analysis that uses
the dummy-coded dummy variables shown above as predictors:
The results of the regression (using dummy-coded dummy
variables to represent career group membership) are shown in
Figure 12.9. Note that the overall F reported for the regression
analysis in Figure 12.9 is identical to the overall F reported for
the one-way ANOVA in Figure 12.6, F (2, 47) = 2.17, p = .125,
and that the η2 for the one-way ANOVA is identical to the R2
for the regression (η2 = R2 = .085). Note also that the b0
coefficient in the regression results in Figure 12.9 (b0 =
constant = 51.55) equals the mean salary for the group that was
assigned score values of 0 on all the dummy variables (the mean
salary for the Business faculty was 51.55). Note also that the bi
coefficients for each of the two dummy variables represent the
difference between the mean of the corresponding group and the
mean of the comparison group whose codes were all 0; for
example, b1 = −6.73, which corresponds to the difference
between the mean salary of Group 1, Liberal Arts (M1 = 44.82)
and mean salary of the comparison group, Business (M3 =

51.55); the value of the b2 coefficient (b2 = −6.84) corresponds
to the difference between mean salary for the Science (M =
44.71) and Business (M = 51.55) groups. This regression
analysis with dummy variables to represent college membership
provided information equivalent to a one-way ANOVA to
predict salary from college.
Figure 12.9 Regression to Predict Salary From Dummy-Coded
College Membership
12.7.2 Regression With Effect-Coded Dummy Predictor
Variables12.7.2.1 Two-Group Example With an Effect-Coded
Dummy Variable
We will now code the scores for gender slightly differently,
using a method called “effect coding of dummy variables.” In
effect coding of dummy variables, a score value of 1 is used to
represent a “yes” answer to a question about group membership;
membership in the group that does not correspond to a “yes”
answer on any of the group membership questions is represented
by a score of −1. In the following example, the effect-coded
dummy variable “geneff” (Is the participant male, yes or no?) is
coded +1 for males and −1 for females. The variable geneff is
called an effect-coded dummy variable because we used −1
(rather than 0) as the value that represents membership in the
last group. Our overall model for the prediction of salary (Y)
from gender, represented by the effect-coded dummy variable
geneff, can be written as follows:
Substituting the values of +1 for males and −1 for females, the
predictive equations for males and females become
From earlier discussions on t tests and ANOVA, we know that
the best predicted value of Y for males is equivalent to the
mean on Y for males, Mmale; similarly, the best predicted value
of Y for females is equal to the mean on Y for females,
Mfemale. The two equations above, therefore, tell us that
Mmale = b0 + b1 and Mfemale = b0 − b1. What does this imply

for the values of b0 and b1? The mean for males is b1 units
above b0; the mean for females is b1 units below b0. With a
little thought, you will see that the intercept b0 must equal the
grand mean on salary for both genders combined.
Note that when we calculate a grand mean by combining group
means, there are two different possible ways to calculate the
grand mean. If the groups have the same numbers of scores,
these two methods yield the same result, but when the groups
have unequal numbers of cases, these two methods for
computation of the grand mean yield different results.
Whenever you do analyses with unequal numbers in the groups,
you need to decide whether the unweighted or the weighted
grand mean is a more appropriate value to report. In some
situations, it may not be clear what default decision a computer
program uses (i.e., whether the program reports the weighted or
the unweighted grand mean), but it is possible to calculate both
the weighted and unweighted grand means by hand from the
group means; when you do this, you will be able to determine
which version of the grand mean was reported on the SPSS
printout.
The unweighted grand mean for salary for males and females is
obtained by ignoring the number of cases in the groups and
averaging the group means together for males and females. For
the male and female salary data that appeared in Table 12.1, the
unweighted mean is (Mmale + Mfemale)/2 = (40.80 + 49.90)/2 =
45.35. Note that the b0 constant or intercept term in the
regression in Figure 12.10 that uses effect-coded gender to
predict salary corresponds to this unweighted grand mean of
45.35. When you run a regression to predict scores on a
quantitative outcome variable from effect-coded dummy
predictor variables, and the default methods of computation are
used in SPSS, the b0 coefficient in the regression equation
corresponds to the unweighted grand mean, and effects (or
differences between group means and grand means) are reported
relative to this unweighted grand mean as a reference point.
This differs slightly from the one-way ANOVA output in Figure

12.5, which reported the weighted grand mean for salary
(46.26).
When effect-coded dummy predictor variables are used, the
slope coefficient b1 corresponds to the “effect” of gender; that
is, +b1 is the distance between the male mean and the grand
mean, and −b1 is the distance between the female mean and the
grand mean. In Chapter 6 on one-way ANOVA, the terminology
used for these distances (group mean minus grand mean) was
effect. The effect of membership in Group i in a one-way
ANOVA is represented by αi, where αi = Mi − MY, the mean of
Group i minus the grand mean of Y across all groups.
This method of coding (+1 vs. −1) is called “effect coding,”
because the intercept b0 in Equation 12.5 equals the unweighted
grand mean for salary, MY, and the slope coefficient bi for each
effect-coded dummy variable E represents the effect for the
group that has a code of 1 on that variable—that is, the
difference between that group’s mean on Y and the unweighted
grand mean. Thus, when effect-coded dummy variables are used
to represent group membership, the b0 intercept term equals the
grand mean for Y, the outcome variable, and each bi coefficient
represents a contrast between the mean of one group versus the
unweighted grand mean (or the “effect” for that group). The
significance of b1 for the effect-coded variable geneff is a test
of the significance of the difference between the mean of the
corresponding group (in this example, males) and the
unweighted grand mean. Given that geneff is coded −1 for
females and +1 for males and given that the value of b1 is
significant and positive, the mean salary of males is
significantly higher than the grand mean (and the mean salary
for females is significantly lower than the grand mean).
Note that we do not have to use a code of +1 for the group with
the higher mean and a code of −1 for the group with the lower
mean on Y. The sign of the b coefficient can be either positive
or negative; it is the combination of signs (on the code for the
dummy variable and the b coefficient) that tells us which group
had a mean that was lower than the grand mean.

The overall F result for the regression analysis that predicts
salary from effect-coded gender (geneff) in Figure 12.10 is
identical to the F value in the earlier analyses of gender and
salary reported in Figures 12.5 and 12.7: F(1, 48) = 13.02, p =
.001. The effect size given by η2 and R2 is also identical across
these three analyses (R2 = .21). The only difference between the
regression that uses a dummy-coded dummy variable (Figure
12.7) and the regression that uses an effect-coded dummy
variable to represent gender (Figure 12.10) is in the way in
which the b0 and b1 coefficients are related to the grand mean
and group means.12.7.2.2 Multiple-Group Example With Effect-
Coded Dummy Variables
If we used effect coding instead of dummy coding to represent
membership in the three college groups used as an example
earlier, group membership could be coded as follows:
That is, E1 and E2 still represent yes/no questions about group
membership. E1 corresponds to the following question: Is the
faculty member in Liberal Arts? Coded 1 = yes, 0 = no. E2
corresponds to the following question: Is the faculty member in
Sciences? Coded 1 = yes, 0 = no. The only change when effect
coding (instead of dummy coding) is used is that members of
the Business college (the one group that does not correspond
directly to a yes/no question) now receive codes of −1 on both
the variables E1 and E2.
We can run a regression to predict scores on salary from the two
effect-coded dummy variables E1 and E2:
Figure 12.10 Regression to Predict Salary From Effect-Coded
Gender
The results of this regression analysis are shown in Figure
12.11.
When effect coding is used, the intercept or b0 coefficient is
interpreted as an estimate of the (unweighted) grand mean for
the Y outcome variable, and each bi coefficient represents the

effect for one of the groups—that is, the contrast between a
particular group mean and the grand mean. (Recall that when
dummy coding was used, the intercept b0 was interpreted as the
mean of the “last” group—namely, the group that did not
correspond to a “yes” answer on any of the dummy variables—
and each bi coefficient corresponded to the difference between
one of the group means and the mean of the “last” group, the
group that is used as the reference group for all comparisons.)
In Figure 12.11, the overall F value and the overall R2 are the
same as in the two previous analyses that compared mean salary
across college (in Figures 12.6 and 12.9): F(2, 47) = 2.12, p =
.125; R2 = .085. The b coefficients for Equation 12.9, from
Figure 12.11, are as follows:
Salary′ = 47.03 − 2.205 × E1 − 2.317 × E2.
The interpretation is as follows: The (unweighted) grand mean
of salary is 47.03 thousand dollars per year. Members of the
Liberal Arts faculty have a predicted annual salary that is 2.205
thousand dollars less than this grand mean; members of the
Sciences faculty have a predicted salary that is 2.317 thousand
dollars less than this grand mean. Neither of these differences
between a group mean and the grand mean is statistically
significant at the α = .05 level.
Figure 12.11 Regression to Predict Salary From Effect-Coded
College Membership
Because members of the Business faculty have scores of −1 on
both E1 and E2, the predicted mean salary for Business faculty
is
Salary′= 47.03 + 2.205 + 2.317 = 51.5 thousand dollars per
year.
We do not have a significance test to evaluate whether the mean
salary for Business faculty is significantly higher than the grand
mean. If we wanted to include a significance test for this
contrast, we could do so by rearranging the dummy variable
codes associated with group membership, such that membership
in the Business college group corresponded to an answer of

“yes” on either E1 or E2.12.7.3 Orthogonal Coding of Dummy
Predictor Variables
We can set up contrasts among group means in such a way that
the former are orthogonal (the term orthogonal is equivalent to
uncorrelated or independent). One method of creating
orthogonal contrasts is to set up one contrast that compares
Group 1 versus Group 2 and a second contrast that compares
Groups 1 and 2 combined versus Group 3, as in the example
below:
The codes across each row should sum to 0. For each
orthogonally coded dummy variable, the groups for which the
code has a positive sign are contrasted with the groups for
which the code has a negative sign; groups with a code of 0 are
ignored. Thus, O1 compares the mean of Group 1 (Liberal Arts)
with the mean of Group 2 (Sciences).
To figure out which formal null hypothesis is tested by each
contrast, we form a weighted linear composite that uses these
codes. That is, we multiply the population mean μk for Group k
by the contrast coefficient for Group k and sum these products
across the k groups; we set that weighted linear combination of
population means equal to 0 as our null hypothesis.
In this instance, the null hypotheses that correspond to the
contrast specified by the two Oi orthogonally coded dummy
variables are as follows:
H0 for O1: (+1)μ1 + (−1)μ2 + (0)μ3 = 0.
That is,
H0 for O1: μ1 − μ2 = 0 (or μ1 = μ2).
The O2 effect-coded dummy variable compares the average of
the first two group means (i.e., the mean for Liberal Arts and
Sciences combined) with the mean for the third group
(Business):
or
We can assess whether the contrasts are orthogonal by taking

the cross products and summing the corresponding coefficients.
Recall that products between sets of scores provide information
about covariation or correlation; see Chapter 7 for details.
Because each of the two variables O1 and O2 has a sum (and,
therefore, a mean) of 0, each code represents a deviation from a
mean. When we compute the sum of cross products between
corresponding values of these two variables, we are, in effect,
calculating the numerator of the correlation between O1 and O2.
In this example, we can assess whether O1 and O2 are
orthogonal or uncorrelated by calculating the following sum of
cross products. For O1 and O2, the sum of cross products of the
corresponding coefficients is
(+1)(+1) + (−1)(+1) + (0)(−2) = 0.
Because this sum of the products of corresponding coefficients
is 0, we know that the contrasts specified by O1 and O2 are
orthogonal.
Of course, as an alternate way to see whether the O1 and O2
predictor variables are orthogonal or uncorrelated, SPSS can
also be used to calculate a correlation between O1 and O2; if
the contrasts are orthogonal, Pearson’s r between O1 and O2
will equal 0 (provided that the numbers in the groups are equal).
For each contrast specified by a set of codes (e.g., the O1 set of
codes), any group with a 0 coefficient is ignored, and groups
with opposite signs are contrasted. The direction of the signs
for these codes does not matter; the contrast represented by the
codes (+1, −1, and 0) represents the same comparison as the
contrast represented by (−1, +1, 0). The b coefficients obtained
for these two sets of codes would be opposite in sign, but the
significance of the difference between the means of Group 1 and
Group 2 would be the same whether Group 1 was assigned a
code of +1 or −1.
Figure 12.12 shows the results of a regression in which the
dummy predictor variables O1 and O2 are coded to represent the
same orthogonal contrasts. Note that the t tests for significance
of each contrast are the same in both Figure 12.6, where the
contrasts were requested as an optional output from the

ONEWAY procedure, and Figure 12.12, where the contrasts
were obtained by using orthogonally coded dummy variables as
predictors of salary. Only one of the two contrasts, the contrast
that compares the mean salary for Liberal Arts and Sciences
faculty with the mean salary for Business faculty, was
statistically significant.
Figure 12.12 Regression to Predict Salary From Dummy
Variables That Represent Orthogonal Contrasts
Orthogonal coding of dummy variables can also be used to
perform a trend analysis—for example, when the groups being
compared represent equally spaced dosage levels along a
continuum. The following example shows orthogonal coding to
represent linear versus quadratic trends for a study in which the
groups receive three different dosage levels of caffeine.
Note that the sum of cross products is again 0 [(−1)(+1) +
(0)(−2) + (+1)(+1)], so these contrasts are orthogonal. A simple
way to understand what type of trend is represented by each line
of codes is to visualize the list of codes for each dummy
variable as a template or graph. If you place values of −1, 0,
and +1 from left to right on a graph, it is clear that these codes
represent a linear trend. The set of coefficients +1, −2, and +1
or, equivalently, −1, +2, and −1 represent a quadratic trend. So,
if b1 (the coefficient for O1) is significant with this set of
codes, the linear trend is significant and b1 is the amount of
change in the dependent variable Y from 0 to 150 mg or from
150 to 300 mg. If b2 is significant, there is a quadratic
(curvilinear) trend.
12.8 Regression Models That Include Both Dummy and
Quantitative Predictor Variables
We can do a regression analysis that includes one (or more)
dummy variables and one (or more) continuous predictors, as in
the following example:

How is the b1 coefficient for the dummy variable, D,
interpreted in the context of this multiple regression with
another predictor variable? The b1 coefficient still represents
the estimated difference between the means of the two groups;
if gender was coded 0, 1 as in the first example, b1 still
represents the difference between mean salary Y for males and
females. However, in the context of this regression analysis,
this difference between means on the Y outcome variable for the
two groups is assessed while statistically controlling for any
differences in the quantitative X variable (such as years of job
experience).
Numerical results for this regression analysis appear in Figure
12.13. From these results we can conclude that both gender and
years are significantly predictive of salary. The coefficient to
predict salary from gender (controlling for years of experience)
was b2 = 3.36, with t(47) = 2.29, p = .026. The corresponding
squared semipartial (or part) correlation for gender as a
predictor of salary was sr2 = (.159)2 = .03. The coefficient to
predict salary from years of experience, controlling for gender,
was b = 1.44, t(47) = 10.83, p < .001; the corresponding sr2
effect size for years was .7492 = .56. This analysis suggests that
controlling for years of experience partly accounts for the
observed gender differences in salary, but it does not
completely account for gender differences; even after years of
experience is taken into account, males still have an average
salary that is about 3.36 thousand dollars higher than females at
each level of years of experience. However, this gender
difference is relatively small in terms of the proportion of
explained variance. About 56% of the variance in salaries is
uniquely predictable from years of experience (this is a very
strong effect). About 3% of the variance in salary is predictable
from gender (this is a medium-sized effect). Within this sample,
for each level of years of experience, females are paid about
3.35 thousand dollars less than males who have the same
number of years of experience; this is evidence of possible
gender bias. Of course, it is possible that this remaining gender

difference in salary might be accounted for by other variables.
Perhaps more women are in the college of Liberal Arts, which
has lower salaries, and more men are in the Business and
Science colleges, which tend to receive higher salaries.
Controlling for other variables such as college might help us to
account for part of the gender difference in mean salary levels.
The raw score b coefficient associated with gender in the
regression in Figure 12.13 had a value of b = 3.355; this
corresponds to the difference between the intercepts of the
regression lines for males and females in Figure 12.14. For this
set of data, it appears that gender differences in salary may be
due to a difference in starting salaries (i.e., the salaries paid to
faculty with 0 years of experience) and not due to differences
between the annual raises in salary given to male and female
faculty. The model Results section at the end of the chapter
provides a more detailed discussion of the results that appear in
Figures 12.13 and 12.14.
The best interpretation for the salary data in Table 12.1, based
on the analyses that have been performed so far, appears to be
the following: Salary significantly increases as a function of
years of experience; there is also a gender difference in salary
(such that males are paid significantly higher salaries than
females) even when the effect of years of experience is
statistically controlled. However, keep in mind that statistically
controlling for additional predictor variables (such as college
and merit) in later analyses could substantially change the
apparent magnitude of gender differences in salary.
Figure 12.13 Regression to Predict Salary From Gender and
Years
Figure 12.14 Graph of the Regression Lines to Predict Salary
From Years of Experience Separately for Males and Females
NOTE: This is based on the regression analysis in Figure 12.13:
Salary′ = 34.2 + 3.36 × Gender + 1.44 × Years; gender coded 0

= female and 1 = male.
12.9 Effect Size and Statistical Power
As discussed in Chapter 11, we can represent the effect size for
the regression as a whole (i.e., the proportion of variance in Y
is predictable from a set of variables that may include dummy-
coded and/or continuous variables) by reporting multiple R and
multiple R2 as our overall effect-size measures. We can
represent the strength of the unique predictive contribution of
any particular variable by reporting the estimate of sr2unique
for each predictor variable, as in Chapter 11. The proportion of
variance given by sr2 can be used to describe the proportion of
variance uniquely predictable from the contrast specified by a
dummy variable, in the same manner in which it describes the
proportion of variance uniquely predictable from a continuous
predictor variable.
When only dummy variables are included in regression analysis,
the regression is essentially equivalent to a one-way ANOVA
(or a t test, if only two groups are being compared). Therefore,
the tables presented in Chapter 5 (independent samples t test)
and Chapter 6 (one-way ANOVA) can be used to look up
reasonable minimum sample sizes per group for anticipated
effect sizes that are small, medium, or large. Whether the
method used to make predictions and compare means across
groups is ANOVA or regression, none of the groups should have
a very small n. If n < 20 per group, nonparametric analyses may
be more appropriate.
12.10 Nature of the Relationship and/or Follow-Up Tests
When dummy-coded group membership predictors are included
in a regression analysis, the information that individual
coefficients provide is equivalent to the information obtained
from planned contrasts between group means in an ANOVA.
The choice of the method of coding (dummy, effect, orthogonal)
and the decision as to which group to code as the “last” group
determine which set of contrasts the regression will include.

Whether we compare multiple groups by performing a one-way
ANOVA or by using a regression equation with dummy-coded
group membership variables as predictors, the written results
should include the following: means and standard deviations for
scores in each group, confidence intervals for each group mean,
an overall F test to report whether there were significant
differences in group means, planned contrasts or post hoc tests
to identify which specific pairs of group means differed
significantly, and a discussion of the direction of differences
between group means. Effect-size information about the
proportion of explained variance (in the form of an η2 or sr2)
should also be included.
12.11 Results
The hypothetical data showing salary scores for faculty (in
Table 12.1) were analyzed in several different ways to
demonstrate the equivalence between ANOVA and regression
with dummy variables and to illustrate the interpretation of b
coefficients for dummy variables in regression. The text below
reports the regression analysis for prediction of salary from
years of experience and gender (as shown in Figure 12.13).
Results
To assess whether gender and years of experience significantly
predict faculty salary, a regression analysis was performed to
predict faculty annual salary in thousands of dollars from
gender (dummy-coded 0 = female, 1 = male) and years of
experience. The distribution of salary was roughly normal, the
variances of salary scores were not significantly different for
males and females, and scatter plots did not indicate nonlinear
relations or bivariate outliers. No data transformations were
applied to scores on salary and years, and all 50 cases were
included in the regression analysis.
The results of this regression analysis (SPSS output in Figure
12.13) indicated that the overall regression equation was
significantly predictive of salary; R = .88, R2 = .78, adjusted

R2 = .77, F(2, 47) = 80.93, p < .001. Salary could be predicted
almost perfectly from gender and years of job experience. Each
of the two individual predictor variables was statistically
significant. The raw score coefficients for the predictive
equation were as follows:
Salary′ = 34.19 + 3.36 × Gender + 1.44 × Years.
When controlling for the effect of years of experience on salary,
the magnitude of the gender difference in salary was 3.36
thousand dollars. That is, at each level of years of experience,
male annual salary was about 3.36 thousand dollars higher than
female salary. This difference was statistically significant: t(47)
= 2.29, p = .026.
For each 1-year increase in experience, the salary increase was
approximately 1.44 thousand dollars for both females and
males. This slope for the prediction of salary from years of
experience was statistically significant: t(47) = 10.83, p < .001.
The graph in Figure 12.14 illustrates the regression lines to
predict salary for males and females separately. The intercept
(i.e., predicted salary for 0 years of experience) was
significantly higher for males than for females.
The squared semipartial correlation for years as a predictor of
salary was sr2 = .56; thus, years of experience uniquely
predicted about 56% of the variance in salary (when gender was
statistically controlled). The squared semipartial correlation for
gender as a predictor of salary was sr2 = .03; thus, gender
uniquely predicted about 3% of the variance in salary (when
years of experience was statistically controlled). The results of
this analysis suggest that there was a systematic difference
between salaries for male and female faculty and that this
difference was approximately the same at all levels of years of
experience. Statistically controlling for years of job experience,
by including it as a predictor of salary in a regression that also
used gender to predict salary, yielded results that suggest that
the overall gender difference in mean salary was partly, but not
completely, accounted for by gender differences in years of job
experience.

12.12 Summary
This chapter presented examples that demonstrated the
equivalence of ANOVA and regression analyses that use dummy
variables to represent membership in multiple groups. This
discussion has presented demonstrations and examples rather
than formal proofs; mathematical statistics textbooks provide
formal proofs of equivalence between ANOVA and regression.
ANOVA and regression are different special cases of the GLM.
If duplication of ANOVA using regression were the only
application of dummy variables, it would not be worth spending
so much time on them. However, dummy variables have
important practical applications. Researchers often want to
include group membership variables (such as gender) among the
predictors that they use in multiple regression, and it is
important to understand how the coefficients for dummy
variables are interpreted.
An advantage of choosing ANOVA as the method for comparing
group means is that the SPSS procedures provide a wider range
of options for follow-up analysis—for example, post hoc
protected tests. Also, when ANOVA is used, interaction terms
are generated automatically for all pairs of (categorical)
predictor variables or factors, so it is less likely that a
researcher will fail to notice an interaction when the analysis is
performed as a factorial ANOVA (as discussed in Chapter 13)
than when a comparison of group means is performed using
dummy variables as predictors in a regression. ANOVA does
not assume a linear relationship between scores on categorical
predictor variables and scores on quantitative outcome
variables. A quantitative predictor can be added to an ANOVA
model (this type of analysis, called analysis of covariance or
ANCOVA, is discussed in Chapter 15).
On the other hand, an advantage of choosing regression as the
method for comparing group means is that it is easy to use
quantitative predictor variables along with group membership

predictor variables to predict scores on a quantitative outcome
variable. Regression analysis yields equations that can be used
to generate different predicted scores for cases with different
score values on both categorical and dummy predictor variables.
A possible disadvantage of the regression approach is that
interaction terms are not automatically included in a regression;
the data analyst must specifically create a new variable (the
product of the two variables involved in the interaction) and add
that new variable as a predictor. Thus, unless they specifically
include interaction terms in their models (as discussed in
Chapter 15), data analysts who use regression analysis may fail
to notice interactions between predictors. A data analyst who is
careless may also set up a regression model that is “nonsense”;
for example, it would not make sense to predict political
conservatism (Y) from scores on a categorical X1 predictor
variable that has codes 1 for Democrat, 2 for Republican, and 3
for Independent. Regression assumes a linear relationship
between predictor and outcome variables; political party
represented by just one categorical variable with three possible
score values probably would not be linearly related to an
outcome variable such as conservatism. To compare group
means using regression in situations where there are more than
two groups, the data analyst needs to create dummy variables to
represent information about group membership. In some
situations, it may be less convenient to create new dummy
variables (and run a regression) than to run an ANOVA.
Ultimately, however, ANOVA and regression with dummy
predictor variables yield essentially the same information about
predicted scores for different groups. In many research
situations, ANOVA may be a more convenient method to assess
differences among group means. However, regression with
dummy variables provides a viable alternative, and in some
research situations (where predictor variables include both
categorical and quantitative variables), a regression analysis
may be a more convenient way of setting up the analysis.

Note
1. Unequal numbers in groups make the interpretation of b
coefficients for dummy variables more complex. For additional
information about issues that should be considered when using
dummy or effect codes to represent groups of unequal sizes, see
Hardy (1993).
Comprehension Questions
1.
Suppose that a researcher wants to do a study to assess how
scores on the dependent variable heart rate (HR) differ across
groups that have been exposed to various types of stress. Stress
group membership was coded as follows:
Group 1, no stress/baseline
Group 2, mental arithmetic
Group 3, pain induction
Group 4, stressful social role play
The basic research questions are whether these four types of
stress elicited significantly different HRs overall and which
specific pairs of groups differed significantly.
a.
Set up dummy-coded dummy variables that could be used to
predict HR in a regression.
Note that it might make more sense to use the “no-stress”

group as the one that all other group means are compared with,
rather than the group that happens to be listed last in the list
above (stressful role play). Before working out the contrast
coefficients, it may be helpful to list the groups in a different
order:
Group 1, mental arithmetic
Group 2, pain induction
Group 3, stressful social role play
Group 4, no stress/baseline
Set up dummy-coded dummy variables to predict scores on HR
from group membership for this set of four groups.
Write out in words which contrast between group means
each dummy variable that you have created represents.
b.
Set up effect-coded dummy variables that could be used to
predict HR in a regression.
Describe how the numerical results for these effect-coded

dummy variables (in 1b) differ from the numerical results
obtained using dummy-coded dummy variables (in 1a). Which
parts of the numerical results will be the same for these two
analyses?
c.
Set up the coding for orthogonally coded dummy variables that
would represent these orthogonal contrasts:
Group 1 versus 2
Groups 1 and 2 versus 3
Groups 1, 2, and 3 versus 4
2.
Suppose that a researcher does a study to see how level of
anxiety (A1 = low, A2 = medium, A3 = high) is used to predict
exam performance (Y). Here are hypothetical data for this
research situation. Each column represents scores on Y (exam
scores).
a.Would it be appropriate to do a Pearson correlation (and/or
linear regression) between anxiety, coded 1, 2, 3 for (low,
medium, high), and exam score? Justify your answer. b.Set up
orthogonally coded dummy variables (O1, O2) to represent
linear and quadratic trends, and run a regression analysis to
predict exam scores from O1 and O2. What conclusions can you
draw about the nature of the relationship between anxiety and
exam performance? c.Set up dummy-coded dummy variables to
contrast each of the other groups with Group 2, medium
anxiety; run a regression to predict exam performance (Y) from

these dummy-coded dummy variables. d.Run a one-way
ANOVA on these scores; request contrasts between Group 2,
medium anxiety, and each of the other groups. Do a point-by-
point comparison of the numerical results for your ANOVA
printout with the numerical results for the regression in (2c),
pointing out where the results are equivalent.3.Why is it
acceptable to use a dichotomous predictor variable in a
regression when it is not usually acceptable to use a categorical
variable that has more than two values as a predictor in
regression?4.Why are values such as +1, 0, and −1 generally
used to code dummy variables?5.How does the interpretation of
regression coefficients differ for dummy coding of dummy
variables versus effect coding of dummy variables? (Hint: in
one type of coding, b0 corresponds to the grand mean; in the
other, b0 corresponds to the mean of one of the groups.)6.If you
have k groups, why do you only need k − 1 dummy variables to
represent group membership? Why is it impossible to include k
dummy variables as predictors in a regression when you have k
groups?7.How does orthogonal coding of dummy variables
differ from dummy and effect coding?8.Write out equations to
show how regression can be used to duplicate a t test or a one-
way ANOVA.
(Warner 469-500)
Warner, Rebecca (Becky) (Margaret). Applied Statistics: From
Bivariate Through Multivariate Techniques, 2nd Edition. SAGE
Publications, Inc, 04/2012. VitalBook file.
The citation provided is a guideline. Please check each citation
for accuracy before use.
IBM SPSS Step-by-Step Guide: One-Way ANOVA
Note: This guide is an example of creating ANOVA output in

SPSS with the grades.sav file. The variables shown in this guide
do not correspond with the actual variables assigned in Unit 10
Assignment 1. Carefully follow the instructions in the
assignment for a list of assigned variables. Screen shots were
created with SPSS 21.0.
Creating One-Way ANOVA Output
To complete Section 2 of the DAA for Unit 10 Assignment 1,
you will generate SPSS output for a histogram, descriptive
statistics, and the Shapiro-Wilk test, which are covered in
previous step-by-step guides. The Levene test (homogeneity of
variance) is covered in the steps below.
Refer to the Unit 10 assignment instructions for a list of
assigned variables. The examplevariables year and final are
shown below.
Step 1. Open grades.sav in SPSS.
Step 2. On the Analyze menu, point to Compare Means and
click One-Way ANOVA…
Step 3. In the One-Way ANOVA dialog box:
· Move the assigned dependent variable into the Dependent List
box.
· Move the assigned independent variable into the Factor box.
The examples of final and year are shown below.
· Click the Options button.
Step 4. In the One-Way ANOVA: Options dialog box:
· Select Homogeneity of variance test (for the Levene test for
Section 2 of the DAA).

· Select Descriptive and Means Plot (for Section 4 of the DAA).
· Click Continue.
· Return to the One-Way ANOVA dialog box and select the Post
Hoc button.
Step 5. In the One-Way ANOVA: Post Hoc Multiple
Comparisons dialog box:
· Check the Tukey option for multiple comparisons.
· Click Continue and OK.
Interpreting One-Way ANOVA Output
A string of ANOVA output will appear in SPSS. (The output
below is for the example variable final.)
Step 1. Copy the Levene test output from SPSS and paste it into
Section 2 of the DAA Template. Then interpret it for the
homogeneity of variance assumption.
Test of Homogeneity of Variances
final
Levene Statistic
df1
df2
Sig.
.866
3
101
.462
Step 2. Copy the means plot, paste it into Section 4 of the DAA
Template, and interpret it.
Step 3. Copy the descriptives output. Paste it into Section 4

along with the report of means and standard deviations of the
dependent variable at each level of the independent variable.
Descriptives
final
N
Mean
Std. Deviation
Std. Error
95% Confidence Interval for Mean
Minimum
Maximum
Lower Bound
Upper Bound
Frosh
3
59.33
5.859
3.383
44.78
73.89
55
66
Soph
19
62.42
6.628
1.520

59.23
65.62
48
72
Junior
64
61.47
8.478
1.060
59.35
63.59
40
75
Senior
19
60.89
7.951
1.824
57.06
64.73
43
74
Total
105
61.48
7.943
.775
59.94
63.01
40
75
Step 4. Copy the ANOVA output, paste it into Section 4, and
interpret it.
ANOVA
final

Sum of Squares
df
Mean Square
F
Sig.
Between Groups
37.165
3
12.388
.192
.902
Within Groups
6525.025
101
64.604
Total
6562.190
104
Step 5. Finally, if the overall ANOVA is significant, copy the
post hoc output, paste it into Section 4, and interpret it.
Multiple Comparisons
Dependent Variable: final
Tukey HSD
(I) Year in school
(J) Year in school
Mean Difference (I-J)
Std. Error
Sig.
95% Confidence Interval

Lower Bound
Upper Bound
Frosh
Soph
-3.088
4.993
.926
-16.13
9.96
Junior
-2.135
4.748
.970
-14.54
10.27
Senior
-1.561
4.993
.989
-14.61
11.48
Soph
Frosh
3.088
4.993
.926
-9.96
16.13

Junior
.952
2.100
.969
-4.53
6.44
Senior
1.526
2.608
.936
-5.29
8.34
Junior
Frosh
2.135
4.748
.970
-10.27
14.54
Soph
-.952
2.100
.969
-6.44
4.53
Senior
.574
2.100
.993
-4.91
6.06
Senior
Frosh

1.561
4.993
.989
-11.48
14.61
Soph
-1.526
2.608
.936
-8.34
5.29
Junior
-.574
2.100
.993
-6.06
4.91
5
In Unit 10, we will apply our understanding of the one-way
ANOVA to the SPSS data set.
Proper Reporting of the One-Way ANOVA
Reporting a one-way ANOVA in proper APA style requires an
understanding of several elements. To provide
context for the F test, provide a means plot, as well as the
means and standard deviations for each level of a
given factor. The following elements are included in reporting
the F test:

• The statistical notation for a one-way ANOVA ( F).
• The degrees of freedom.
• The F value.
• The probability value ( p).
• The effect size.
If the omnibus F test is significant, follow with a discussion of
post-hoc tests. Consider the following example
from Warner (2013, p. 254):
The overall F for the one-way ANOVA was statistically
significant, F(3, 24) = 11.94, p < .001. This
corresponded to an effect size of η2 = .60; that is, about 60% of
the variance in anxiety scores was
predictable from the type of stress intervention. This is a large
effect. . . .
In addition, all possible pairwise comparisons were made using
the Tukey HSD test. Based on this
test. . . .
F, Degrees of Freedom, and F Value
The statistical notation for a one-way ANOVA is F, and
following it is the degrees of freedom for this statistical
test, such as (3, 24) reported above. Unlike correlation or a t
test, there are two degrees of freedom reported
for a one-way ANOVA. The first reported df is the between-
groups df, or dfbetween, which is the number of
groups (or levels) minus one ( k − 1). In the example above, the
factor consists of k = 4 levels (4 − 1 = 3). The
second reported df is the within-groups df, or dfwithin, which is
the sample size minus the number of groups or

levels ( N − k). In the example above, N = 28, so 28 − 4 = 24.
The F value is calculated as a ratio of mean
squares, which are both positive. Therefore, any non-zero F
value is always positive.
Probability Value
Appendix C (pp.1058–1061) of the Warner text provides critical
values of F for rejecting the null hypothesis. In
the example above, with (3, 24) degrees of freedom and alpha
level set to .05 (one-tailed versus two-tailed is
not relevant to ANOVA), the table indicates a critical value of ±
3.01 to reject the null hypothesis. The obtained
F value above is 11.94, which exceeds the critical value
required to reject the null hypothesis. SPSS determined
the exact p value to be .000, which is reported as p < .001.
(Remember that SPSS only calculates a p value out
to three decimal places.) This p value is less than .05, which
indicates that the null hypothesis should be rejected
for the alternative hypothesis—that is, at least one of the four
group means is significantly different from the
other group means.
Unit 10 - One-Way ANOVA: Application
INTRODUCTION
Effect Size
The effect size for a one-way ANOVA is eta squared (η2). The
effect size is not provided in SPSS output. It is
calculated by hand by dividing SSbetween by SStotal from the
SPSS ANOVA output. In the example above,

SSbetween = 182.107 and SStotal = 304.107, which means that
182.107 ÷ 304.107 = .60. The effect size is
interpreted using Table 5.2 in the Warner text (p. 208).
Post-Hoc Tests
When the omnibus F is significant, it does not indicate exactly
which pairwise comparisons are significant. A
Tukey's honestly significant difference (HSD) test is one of
many post-hoc tests used. The SPSS output for the
Tukey's HSD indicates which pairwise comparisons are
statistically significant, and this information can be
reported in narrative form (that is, without p values or other
specific statistical notation) as shown in the fourth
paragraph of the "Results" section in the Warner text (p. 254).
The Warner text provides a "Results" example at the end of
each chapter for all statistics studied in this course.
You are encouraged to review these examples and follow their
structure when writing up Section 4,
"Interpretation," of the DAA Template.
Reference
Warner, R. M. (2013). Applied statistics: From bivariate
through multivariate techniques (2nd ed.). Thousand
Oaks, CA: Sage.
OBJECTIVES
To successfully complete this learning unit, you will be
expected to:
1. Interpret the one-way ANOVA output.

2. Apply the appropriate SPSS procedures to check assumptions
and calculate the one-way ANOVA to
generate relevant output.
3. Analyze the assumptions of the one-way ANOVA.
4. Articulate a research question, null hypothesis, alternative
hypothesis, and alpha level.
[u10s1] Unit 10 Study 1- Readings
Use your IBM SPSS Statistics Step by Step text to complete the
following:
• Read Chapter 12, "The One-Way ANOVA Procedure." This
reading addresses the following topics:
◦ Introduction to one-way ANOVA.
◦ SPSS commands.
◦ Post-hoc tests in SPSS.
◦ Planned contrasts in SPSS.
◦ Reporting and interpreting SPSS output.
[u10a1] Unit 10 Assignment 1 - One-Way ANOVA
As with your previous assignments, you will complete this
assignment with the DAA Template. Links to
additional resources are available in the Resources area.
Reminder: The format of this SPSS assignment should be
narrative with supporting statistical output
(table and graphs) integrated into the text in the appropriate
places (not all at the end of the document).
You will analyze the following variables in the grades.sav data
set:
• section
• quiz3

Step 1: Write Section 1 of the DAA.
• Provide the context of the grades.sav data set.
• Include a definition of the specified variables (predictor,
outcome) and corresponding scales of
measurement.
• Specify the sample size of the data set.
• Analyze the assumptions of the one-way ANOVA.
• Paste the SPSS histogram output for quiz3 and discuss your
visual interpretations.
• Paste SPSS descriptives output showing skewness and kurtosis
values for quiz3 and interpret them.
• Paste SPSS output for the Shapiro-Wilk test of quiz3 and
interpret it.
• Report the results of the Levene test and interpret it.
• Summarize whether or not the assumptions of the one-way
ANOVA are met.
• Specify a research question related to the one-way ANOVA.
• Articulate the null hypothesis and alternative hypothesis.
• Specify the alpha level.
• Begin by pasting SPSS output of the means plot and providing
an interpretation.
• Also report the means and standard deviations of quiz3 for

each level of the section variable.
• Next, paste the SPSS ANOVA output and report the results of
the F test, including:
◦ Degrees of freedom.
◦ F value.
◦ p value.
◦ Calculated effect size.
◦ Interpretation of the effect size.
• Finally, if the omnibus F is significant, provide the SPSS post-
hoc (Tukey HSD) output.
◦ Interpret the post-hoc tests.
• Discuss the conclusions of the one-way ANOVA as it relates
to the research question.
• Conclude with an analysis of the strengths and limitations of
one-way ANOVA.
Submit your DAA Template as an attached Word document in
the assignment area.
Resources
One-Way ANOVA Scoring Guide.
DAA Template.
SPSS Data Analysis Report Guidelines.
IBM SPSS Step-by-Step Guide: One-Way ANOVA.
Copy/Export Output Instructions.

APA Style and Format.
Journal Article Summary Scoring Guide
Due Date: End of Unit 9.
Percentage of Course Grade: 12%.
CRITERIA NON-PERFORMANCE BASIC PROFICIENT
DISTINGUISHED
Locate a scholarly
journal article in a
career specialization
that reports a
correlation, a t test,
a one-way ANOVA,
or some
combination.
20%
Does not locate a
scholarly journal
article in a career
specialization that
reports a correlation,
a t test, a one-way
ANOVA, or some
combination.
Locates a scholarly
career specialization,

but the article does not
report a correlation, a t
test, a one-way
ANOVA, or some
combination.
Locates a scholarly
career specialization
that reports a
correlation, a t test, a
one-way ANOVA, or
some combination.
Locates an exemplary
scholarly journal
article in a career
specialization that
reports a correlation, a
t test, a one-way
ANOVA, or some
combination.
Provide context for
the research study,
including a
definition of
variables and scales
of measurement.
15%
Does not provide
context for the
research study.
Provides partial

context for the
research study that
may or may not
include a definition of
variables and scales of
measurement.
Provides context for
the research study,
including a definition
of variables and
scales of
measurement.
Provide exemplary
context for the
research study,
including a definition
of variables and
scales of
measurement.
Identify
assumptions of the
statistic reported in
the journal article.
10%
Does not identify
assumptions of the
Partially identifies
assumptions of the
statistic reported in the

journal article.
Identifies
assumptions of the
Interprets assumptions
of the statistic reported
in the journal article.
Articulate the
research question,
null hypothesis,
alternative
hypothesis, and
alpha level.
15%
Does not articulate
the research
question, null
hypothesis,
alternative
hypothesis, and
alpha level.
Articulates some
combination of, but not
all of, the research
question, null
hypothesis, alternative
hypothesis, and alpha
level.
Articulates the

research question,
null hypothesis,
alternative
level.
Articulates the
research question, null
level in an exemplary
manner.
Report the results of
the journal article,
interpreting the
statistic against the
null hypothesis.
20%
Does not report the
results of the journal
article.
Reports the results of
the journal article, but
does not interpret the
null hypothesis.
the journal article,
interpreting the
null hypothesis.

the journal article in an
exemplary manner,
and interprets the
null hypothesis.
Generate a
conclusion that
includes strengths
and limitations of
10%
Does not generate a
conclusion.
Generates a
conclusion that
includes a partial list of
strengths and
limitations of the
journal article.
Generates a
conclusion that
includes strengths
and limitations of the
journal article.
Generaes an
exemplary conclusion
that includes strengths
journal article.

Communicate in a
manner that is
concise and
professional and
that makes proper
use of APA
formatting.
10%
Does not
communicate in a
manner that is
concise and
professional; does
not make proper use
of APA formatting.
Communicates in a
manner that may or
may not be concise or
professional; often
makes proper use of
APA formatting.
Communicates in a
manner that is
concise and
professional and that
makes proper use of
APA formatting.
Communicates in a
manner that is concise
and professional and
that always makes
proper use of APA

formatting.
For the SPSS data analysis report assignments in Units 6, 8, and
10, you will use the Data Analysis and
Application (DAA) Template with the five sections described
below. As shown in the IBM SPSS step-
by-step guides, label all tables and graphs in a manner
consistent with Capella's APA Style and Format
guidelines. Citations, if needed, should be included in the text
and references included in a reference
section at the end of the report. The organization of the report
should include the following five sections:
Section 1: Data File Description (One Paragraph)
1. Describe the context of the data set. Cite a previous
description if the same data set is used from a
previous assignment. To increase the formal tone of the DAA,
avoid first-person perspective "I."
For example, do not write, "I ran a scatter plot shown in Figure
1." Instead, write, "Figure 1
shows. . . ."
2. Specify the variables used in this DAA and the scale of
measurement of each variable.
3. Specify sample size (N).
Section 2: Testing Assumptions (Multiple Paragraphs)
2. Paste SPSS output that tests those assumptions and interpret
them. Properly embed SPSS output

where appropriate. Do not string all output together at the
beginning of the section. In other
words, interpretations of figures and tables should be near (that
is, immediately above or below)
where the output appears. Format figures and tables per APA
formatting. Refer to the examples in
the IBM SPSS step-by-step guides.
3. Summarize whether or not the assumptions are met. If
assumptions are not met, discuss how to
ameliorate violations of the assumptions.
Section 3: Research Question, Hypotheses, and Alpha Level
(One Paragraph)
1. Articulate a research question relevant to the statistical test.
2. Articulate the null hypothesis and alternative hypothesis for
the research question.
3. Specify the alpha level (.05 unless otherwise specified).
Section 4: Interpretation (Multiple Paragraphs)
1. Paste SPSS output for an inferential statistic and report it.
Properly embed SPSS output where
appropriate. Do not string all output together at the beginning
of the section. In other words,
interpretations of figures and tables should be near (that is,
immediately above or below) where
the output appears. Format figures and tables per APA
formatting.
2. Report the test statistics. For guidance, refer to the "Results"
examples at the end of the
appropriate chapter of your Warner text.
3. Interpret statistical results against the null hypothesis.

Print
SPSS Data Analysis Report Guidelines
Section 5: Conclusion (Two Paragraphs)
1. Provide a brief summary (one paragraph) of the DAA
conclusions.
2. Analyze strengths and limitations of the statistical test.
One-Way ANOVA Scoring Guide
Due Date: End of Unit 10.
Percentage of Course Grade: 12%.
CRITERIA NON-PERFORMANCE BASIC PROFICIENT
DISTINGUISHED
Apply the
appropriate SPSS
procedures to check
assumptions and
calculate the one-
way ANOVA to
generate relevant
output.
20%
No SPSS output is
provided.

Includes some, but
not all, of required the
output. Numerous
errors in SPSS output.
Includes all relevant
output, but also
includes irrelevant
output. One or two
errors in SPSS
output.
Includes all relevant
output; no irrelevant
output is included. No
errors in SPSS output.
Provide context for
the data set,
including a
definition of
required variables
and scales of
measurement.
10%
Does not provide
context for the data
set.
Provides partial
set. May or may not
include a definition of
required variables and

scales of
measurement.
Provides context for
the data set, including
a definition of required
variables and scales
of measurement.
Provides exemplary
set, including a
definition of required
variables and scales of
measurement.
Analyze the
assumptions of the
one-way ANOVA.
20%
Does not analyze the
assumptions of the
one-way ANOVA.
Identifies, but does
not analyze, the
assumptions of the
one-way ANOVA.
Analyzes the
assumptions of the
one-way ANOVA.
Evaluates the
assumptions of the

one-way ANOVA.
Articulate a research
question, null
hypothesis,
alternative
hypothesis, and
alpha level.
10%
Does not articulate a
research question,
null hypothesis,
alternative
level.
Partially articulates a
research question,
null hypothesis,
alternative hypothesis,
and alpha level.
Articulates a research
question, null
hypothesis,
alternative
level.
Articulates a research
question, null
level in an exemplary
manner.

Interpret the one-
way ANOVA output.
20%
Does not interpret the
one-way ANOVA
output.
Partially interprets the
one-way ANOVA
output.
Interprets the one-
way ANOVA output.
Interprets the one-way
ANOVA output in an
exemplary manner.
Generate a
conclusion that
includes strengths
and limitations of
the one-way ANOVA.
10%
Does not generate a
conclusion.
Generates a
conclusion that
includes a partial list
of strengths and
limitations of the one-
way ANOVA.

Generates a
conclusion that
includes strengths
one-way ANOVA.
Generates an
exemplary conclusion
that includes strengths
one-way ANOVA.
Communicate in a
manner that is
scholarly,
professional, and
consistent with the
expectations for
members in the
identified field of
study.
10%
Does not
communicate in a
manner that is
scholarly,
professional, and
consistent with the
expectations for
members in the
identified field of
study.
Inconsistently

communicates in a
manner that is
scholarly,
professional, and
consistent with the
expectations for
members in the
identified field of
study.
Communicates in a
manner that is
scholarly,
professional, and
consistent with the
expectations for
members in the
identified field of
study.
Without exception,
communicates in a
manner that is
scholarly, professional,
and consistent with the
expectations for
members in the
identified field of study.
Running head: DATA ANALYSIS AND APPLICATION
TEMPLATE 1
DATA ANALYSIS AND APPLICATION TEMPLATE 2

Data Analysis and Application (DAA) TemplateLearner
NameCapella University
Data Analysis and Application (DAA) Template
Use this file for all assignments that require the DAA Template.
Although the statistical tests will change from week to week,
the basic organization and structure of the DAA remains the
same. Update the title of the template. Remove this text and
provide a brief introduction.Section 1: Data File Description
Describe the context of the data set. You may cite your previous
description if the same data set is used from a previous
assignment.
Specify the variables used in this DAA and the scale of
measurement of each variable.
Specify sample size (N).Section 2: Testing Assumptions
Paste SPSS output that tests those assumptions and interpret
them. Properly integrate SPSS output where appropriate. Do not
string all output together at the beginning of the section.
Summarize whether or not the assumptions are met. If
assumptions are not met, discuss how to ameliorate violations
of the assumptions.Section 3: Research Question, Hypotheses,
and Alpha Level
1. Articulate a research question relevant to the statistical test.
2. Articulate the null hypothesis and alternative hypothesis.
3. Specify the alpha level.Section 4: Interpretation
1. Paste SPSS output for an inferential statistic. Properly
integrate SPSS output where appropriate. Do not string all
output together at the beginning of the section.
2. Report the test statistics.
3. Interpret statistical results against the null hypothesis.Section
5: Conclusion
1. State your conclusions.
2. Analyze strengths and limitations of the statistical test.

References
Provide references if necessary.

Section 1 Data File DescriptionThe fictional data represents a te.docx

More Related Content

Similar to Section 1 Data File DescriptionThe fictional data represents a te.docx (20)

More from bagotjesusa (20)

Recently uploaded (20)

Section 1 Data File DescriptionThe fictional data represents a te.docx