Chapter 11

Correlational Research

Correlational Studies
Correlational Study: An approach to
research that involves measuring
different variables to see whether there
is a predictable relation among variables
Finding Relationships
The goal of some correlational studies is
simply to find out if variables are related
Is there a relation between whether
students prefer early or late classes and
their grades in those classes?

Are You a Morning Person?

Question
Do student grades relate to whether
they are morning people?
Method
The researchers correlated two kinds
of variables
Grades in early and late classes
Scores on a scale of morningness (13item inventory)

Are You a Morning Person?

Morningness Scale Sample Items
Assuming normal circumstance, how easy
do you find getting up in the morning?
At what time in the evening do you feel
tired and, as a result, in need of sleep?
How long a time does it usually take
before you "recover your senses" in the
morning after rising from a night's sleep?
Source: Smith, C. S., Reilly, C., & Midkiff, K. (1989). Evaluation of three circadian rhythm questionnaires
with suggestions for an improved measure of morningness. Journal of Applied Psychology, 74, 728-738.

Are You a Morning Person?

Result
There was a modest, significant,
correlation between morningness
and grades at 8:00 a.m., r(312) = .
21, p < .01
Students who were not morning
people did less well at 8:00 and
performed better in later classes.

Are You a Morning Person?

Conclusion
The pattern did not hold for classes
starting at 9:00 or later. So even
evening people can do well in early
classes.
The study is correlational, so we do
not know if being a morning person is
the cause of better grades in the
morning.

Source: Guthrie, J. P., Ash, R. A., & Bendapudi, V. (1995). Additional validity evidence for a measure
of morningness. Journal of Applied Psychology, 80, 186-190.

Using the Correlational

Approach

Correlational Studies involve

relationships among variables. It is not
possible to make determinations of
causation with a correlational study.

Using the Correlational

Approach

Problems in Determining Causes with

Correlational Studies
Third Variable Problem--When two
variables are correlated, you cannot say
that one causes the other because there
may be a third variable you do not know
about that affects either or both of the
variables you are investigating.
Directionality Problem--When two
variables are related, you do not know if
the first causes the second or the second
causes the first.

Correlational Studies
Making Predictions
Prediction Study: A correlational study in
which the goal is to predict the value of one
variable (the criterion variable), given the
level of another variable (the predictor
variable) with the predictor variable
occurring first.
Example: Colleges try to predict the GPA of
prospective students (the criterion variable),
given their SAT scores (the predictor variable).

Predictions of behavior are never perfect,

but knowing the relation between the
predictor variable and the criterion variable
will improve the predictions made.

Using the Correlational

Approach

Correlational Analysis refers to the

way data are analyzed and has
nothing to do with the design of a
study.
Experimental data can be subjected to
correlational analysis.
Relations that emerge can be associated
with cause and effect if the study is
experimental, even if analysis is done
with correlational statistics.

Using the Correlational

Approach

Positive and Negative Correlations

With a positive (direct) correlation, when the
value of one variable increases, so does the
value of the second.
Example: Perceived control over life is positively
correlated with life satisfaction: People who feel in
control have greater life satisfaction.

With a negative (indirect) correlation, when the

value of one variable increases, the value of the
second variable decreases.
Example: There is a negative correlation between
how many negative events happens in peoples
lives and their feeling of control: People who go
through many negative events have a feeling of
less control in their lives.
Source: Lang, F. R., & Heckhausen, J. (2001). Perceived control over development and subject
well-being: Differential benefits across adulthood. Journal of Personality and Social
Psychology, 81, 509-523.

Using the Correlational

Approach

Strength of Association

The greater the absolute value of a

correlation, the stronger the relation
between variables (e.g., -.60 represents
a stronger association between
variables than +.50)
By changing the way you score and
label one variable, you can turn a
negative correlation into a positive
correlation, even though the data set
does not change.

Using the Correlational

Approach

Fictitious data showing the relation between (a) time spent

studying for a test and percent correct [r = +.58] and (b) time
spent studying for a test and percent wrong [r = -.58]. The
scatter diagrams are based on the same data and tell the same
story, but the correlation changes from positive in (a) to
negative in (b) because the way the variable is scored (percent
correct versus percent wrong) changes.

Using the Correlational

Approach

Factors Affecting the Size of a Pearson

Product-Moment Correlation
There may be a relation between two
variables, but it may appear less strong
than it really is
When the data show certain
characteristics, the correlation
coefficient may diminish even when two
variables are related

Using the Correlational

Approach

Factors Affecting the Size of a Pearson

Correlation Coefficient
NonlinearityWhen two variables do not have a
linear relation, the Pearson r will be diminish
relative to a situation when there is a linear relation.
Restricted RangeWhen measurements on one
variable do not span a large range of values, the
value of the Pearson r will be smaller.
Heterogeneous subsetsWhen the set of data
consists of subgroups that show different patterns,
the Pearson r might not reflect the pattern within
those subsets.

Using the Correlational

Approach

Nonlinear Relation Between Variables

The scatter diagram here shows a perfect relation

between two variables, but the value of the Pearson r is
0 because the relation between variables is nonlinear.

Using the Correlational

Approach
Restrictions in Range

The scatter diagram for the fictitious data here shows a positive
relation between two variables, but the value of the Pearson r will
be lower if the range of values for Verbal SAT score is restricted. For
the full range of scores, r = .45 but for data in the restricted range
(shown within the circle), the value of the correlation drops, r = .29

Using the Correlational

Approach

Heterogeneous Subsets

The scatter diagram for the fictitious data here shows an

overall positive relation between two variables, but each
of the subsets has a negative correlation. When you
combine subgroups into one overall data set, the
correlation coefficient may not provide useful information.

Traditional Correlational Tests

Varieties of the Correlation Coefficient
Pearson Product-Moment Correlation (Pearson r)The
most widely used correlation coefficient for bivariate
correlations. It is typically cited for data that are on interval or
ratio scales.
Spearman Correlation for Ranks (rs)Correlation
coefficient often used when data are ranked on an ordinal
scale.
Point-Biserial CorrelationCorrelation when one data for
one variable are continuous and for the second variable are
dichotomous (i.e., they can take on two values, like malefemale or yes-no).
Note:
Although some
sources present
different
Phi
Coefficient
()Correlation
used
whenformulas
data forfor
both
these different
types of correlation coefficients, they are all
variables
are dichotomous.
variations on the Pearson r, so you can simply use the regular
formula for the Pearson r for all of them without bothering to
learn different formulas (Howell, 2007) .

Linear Regression: Making

Predictions
Linear Regression
You can make predictions using a linear
regression formula when you have one
predictor variable that you use to estimate
the value of a criterion variable.
The formula is based on the simple formula
for a straight line:

Y = a + bX,
where
Y is the value of the criterion variable,
a is a constant (the Y-intercept), b is the slope
of the line, and
X is the value of the predictor variable

Linear Regression: Making

Predictions

Question
Can you predict how funny a person
will find jokes if you know how good
that persons sense of humor is?
Method
Ask people to rate how funny they
find a set of jokes
Have them complete a sense of
humor scale.

Linear Regression: Making

Predictions
Result
Linear regression analysis showed that
knowing the sense of humor score
helped generate reliable predictions of
joke ratings: people with better sense
of humor provided higher ratings.
Predicted joke rating = Constant + Slope x Sense of
Humor Score
a = constant (Y-intercept) = 1.84
b = slope = 0.76

Linear Regression: Making

Predictions

One person had a sense of humor score of 4.

A second person had a score of 2.
Predicted rating = Intercept + (Slope x score)
Predicted rating = 1.84 + (0.76 x 4) = 1.84 + 3.04 =
4.88
Predicted rating = 1.84 + (0.76 x 2) = 1.84 + 1.52 =
3.36
So we would conclude that on the seven-point scale,
a person with a sense of humor score of 4 would
rate jokes a full point and a half higher than would a
person with a sense of humor score of 2.
Source: Benfante, L., & Beins, B. C. (2007, October). Self-reflection and sense of humor: The big five personality
characteristics and humor. Poster presentation at the annual convention of the New England Psychological Association,
Danbury, CT.

Linear Regression: Making

Predictions
Conclusion
If you know how much of a sense of
humor a person has based on a
paper-and-pencil test, you can make
reliable predictions about how that
person will rate jokes.
People with a better sense of humor
seem to enjoy jokes more.

Correlations with Multiple

Variables

Zero-order correlation: A correlational

analysis involving two variables.
Higher-order correlation: A correlational
analysis involving more than two variables
Multivariate statistics: Statistical approaches
that can accommodate simultaneous analysis
of multiple variables
Multiple regression: A correlational technique
that employs more than one predictor variable
to estimate the value of a criterion variable

Prediction with Multiple

Variables

Question: Can we predict how funny

a person thinks he or she is based on
how they rate jokes and on how
extraverted they are?
Method: Administer sense of humor
and extraversion inventories and ask
participants to rate a set of jokes.
Predictor variables: Extraversion score
and joke ratings
Criterion variable: Self-rating of
funniness (1 = not funny to 10 = very
funny)

Prediction with Multiple

Variables

Results: Multivariate analysis

revealed that knowing how
extraverted people are and how they
rate a set of jokes allows us to
predict how funny they think they
are.

Prediction with Multiple

Variables
Use of Multiple Regression to Predict a Persons SelfRating of Funniness
Predictor Variable 1 = Extraversion score (Score of 1 to
5)
Predictor Variable 2 = Average rating a person gave to
the jokes (Score of 1 to 7)
Criterion Variable = Predicted self-rating of funniness
Constant (Y-intercept) = 3.48
Variable 1 (Extraversion) Slope = 0.521
Variable 2 (Joke rating) Slope = 0.348
Participants Extraversion Score = 3
Participants Average Joke Rating = 2
Regression Equation:
Y = Constant + (Slope1 x Score1) + (Slope2 x Score2)
Y = 3.48 + (0.521 x 3) + (0.348 x 2) = 5.74
A person with an Extraversion score of 3 and an average
joke Benfante,
ratingL.,of& Beins,
2 would
be October).
predicted
to give
a self-rating
offive
Source:
B. C. (2007,
Self-reflection
and sense
of humor: The big

personality characteristics and humor. Poster presentation at the annual convention of the New England

Prediction with Multiple

Variables

Conclusion: We can predict peoples

self-perceptions of how funny they
are by assessing their behavior (i.e.,
how they rated jokes) and their
personality (i.e., extraversion score).
Knowing something about an
individuals personality and their
actual enjoyment of jokes allows us to
make reliable predictions about how
they perceive themselves.

Controversy: The Mental Health

of College Students
College students generally have
good mental health, but reported
stress levels have been increasing.
Can we predict suicidal ideation in
order to minimize actual suicide
attempts?

Controversy: The Mental Health

of College Students
Using multivariate statistics, Gutierrez
et al. (2000) found
a positive relation between negative
themes in a students life and thoughts
of suicide
a negative relation between types of
social relations and thoughts of suicide.

These results could help a counselor

spot potential problems before they
become real problems.

Controversy: The Mental Health

of College Students
Multiple regression has identified several
factors associated with the likelihood of
students seeking help
Gender (women are more likely)
Perceived stigma associated with seeking help
(Less perceived stigma is associated with
greater likelihood)
Level of emotional openness (greater
openness is associated with greater likelihood)
Severity of symptoms (more severe symptoms
are associated with greater likelihood)

Limitations of Multivariate
Correlational Analysis
Limitations of Multivariate Correlational
Analysis
Generalizability
Representative samples are necessary for
confidence in generalizing beyond the sample
It is not always clear that results based on
the measurements made in a given study will
generalize to other measurements
Generalizations might not hold up over time
Confirmation Bias
Creating a model leads to only one possible
model; researchers may fail to test alternate
models