CORRELATIONAL RESEARCH

In general, a correlational study is a quantitative method of research in which you have 2 or

more quantitative variables from the same group of participants, & you are trying to determine if
there is a relationship (or covariation) between the 2 variables.
Types of Correlational Research Designs

1. explanatory design, (it is conducted when researchers want to explore the extents to
which two or more variables co-vary, that is, where changes in one variable are reflected
in changes in the other” (Creswell, 2008, p. 358). When conducting an explanatory
correlational study, researchers typically collect data at one time as their focus is not
based on future or past performance of participants. Thus, when analyzing the findings of
explanatory correlation research, researchers analyze participants as a single group rather
than creating subcategories of participants. Finally, in this type of study researchers
collect two scores from each participant as each score represents each variable being
studied

2. prediction design, is used by researchers when the purpose of the study is to predict
certain outcomes in one variable from another variable that serves as the predictor.
Prediction designs involve two types of variables: a predictor variable and a criterion
variable

a. PREDICTOR variable is utilized to make a forecast or prediction,

b. CRITERION variable is the anticipated outcome that is being predicted.

Prediction studies can usually be identified rather easily by research consumers simply by
taking note of the title of a published study as most published prediction studies include
the word “prediction” in the article’s title.

Characteristics of Correlational Research

Any time a researcher has at least two scores, a graph called a scatterplot can be used to provide
a visual representation of the data that has been collected. Each point on a scatterplot represents
two scores provided by one person. Researchers must select the scores for one variable to be
plotted on the x-axis (the horizontal axis of the graph) while scores for the second variable are
plotted on the y-axis (the vertical axis of the graph). Scatterplots are vitally important to
correlational research as they allow researchers, as well as research consumers, to determine the
following by looking at patterns within the entire group of data points.

1. The form of the relationship

2. The existence of extreme scores

3. The direction of the relationship

4. The degree/strength of the relationship

Conducting a Correlational Study

A correlational research design is useful to researchers who are interested in determining to what
degree two variables are related, however, correlational research “does not “prove” a
relationship; rather, it indicates an association between two or more variables. When a
correlational research design is appropriate for a study, it can be designed by following the steps:

1. Identify two variables that may be related- researchers often select variables to study
with a correlational research design by reading published studies previously conducted by
researchers. Other individuals tend to select variables from real-world situations as they
are interested in findings that are specific to their own situation. When selecting variables
for correlational studies, researchers should avoid the “shotgun approach.” The shotgun
approach occurs when researchers select every and any variable they can think to include.
This approach is discouraged for correlational studies as the greater the amount of
relationships a researcher attempts to examine with correlational research, the more likely
the researcher is to determine that at least one relationship is statistically significant
when, in fact, the significance was only due to chance.

2. Select a sample-Samples should be selected randomly and include at least 30 individuals

willing to partake in the study. Using a greater number of participants allows researchers
to generalize findings to the population of interest. Researchers should also select a
heterogeneous sample that allow for a range of scores. Variability in scores allows
 researchers using correlational methods to determine whether patterns of data exist.
Correlational studies in which a homogenous sample is used and produces a limited range
of scores is describes as having a restriction of range. Wide ranges of scores are
necessary as they allow researchers to determine the true relationship between variables.

3. Select a method of measurement-Often the most complex part of a correlational study is

determining how to effectively measure each variable. As with other research designs,
tools should be determined to be both valid and reliable. (link to validity and reliability
portion of site) Usually in correlational research two measurement tools are required as
each tool measures one of the two variables involved in the study.

4. Collect necessary data-Correlational studies require that researchers obtain data for each
variable from each participant. For instance, if an individual wanted to investigate the
relationship between the length of an intervention and the success of an intervention, for
each student who participates in the study researchers would need to collect data
regarding both the length of the intervention implemented for the student and the success
of the intervention. Collecting data can be especially difficult for researchers conducting
prediction correlational studies as prediction studies require data collection at more than
one point in time. In such cases, researchers often assign numbers to participants to
ensure that data remains confidential while the researcher is waiting to collect data for the
criterion variable.

5. Analyze the data-Data from correlational research is analyzed by using statistical tests
that depend greatly on the type of variables being studied. Variables can be either
continuous, meaning that they change according to small increments (e.g. test scores), or
dichotomous, in which the variable is divided into categories (e.g. gender, grade).

6. Interpret results-When attempting to interpret results, researchers consider both the

strength and the size of the correlation coefficient. Often researchers present findings of
correlational research in a correlational matrix in which each variable is numbered in a
table to show correlations between variables. Asterisks are often used to indicate
correlations that are statistically significant. Statistical significance for correlational
research indicates that the probability that the obtained correlation between two variables
 is due to chance is less than .05 or .01 (depending on which cutoff a researcher has
selected). A lack of statistical significance between two variables means that any
relationship between the variables could be due to chance. However, it is important to
note that statistical significance is likely to occur if a large sample is used by researchers
regardless of whether a true relationship exist between two variables. When making
predictions for individuals, a strong or very strong correlation is typically necessary.
Statistical significance indicates an association between two variables not causation.

Contexts for performing a correlational study:

1) Some independent variables (particularly biographical variables) cannot be manipulated:

gender, age, personality types, ethnic background, height, etc. Note that we have used gender as
an independent variable in a static group design--a type of independent groups design in which
we used the t test for independent means to analyze the data--but in many ways that is essentially
a correlational design because the independent variable is measured, not manipulated. In fact, it
will make no difference whether you perform the t test for independent means (treating the
design as a static group design) or perform the t test for r (treating the design as correlational), in
either case you get the same value for 't'.

2) Some independent variables would be unethical to manipulate: for example: examining the
effects of smoking or pollution levels on cancer rates; examining the effects of exposure to
pornography on sexual deviancy; examining the effects of lowering funding levels for school
systems on student performance; or examining the effects of brain damage in various areas of the
brain on speech production. In all cases you would not be allowed to go in as an experimenter
and manipulate the independent variable. You will have to, instead, examine various levels of the
independent variable as they already exist in the population.

3) Surveys: The nature of surveys is to measure variables, rather than to manipulate them. Unless
the survey has a manipulation build into it (e.g. manipulating the way a question is worded to see
if it influences how it is answered) the survey is essentially a correlational design with many
variables (one for each question). Your analysis will consist of describing the data generated by
each question and of the relationships between those variables.
Tools for measuring the relationship between variables:

Measures of correlation (resulting in a value between -1 and 1)

1) Two cardinal/contineous variables: Pearson r.

2) Two rank variables: Spearman rho. Actually you get the same answer with Pearson r but
Spearman rho is easier to compute because the mean and standard deviation is dependent upon N
(e.g. if you are using rank scores and their are 5 participants than their mean score will be 3).
This is a carryover from the days before computers. If you are using a computer there is no
reason to use Spearman rho rather than Pearson r.

3) One cardinal/contineous and one dichotomous variable: Point biserial. A dichotomous

variable is one that can take on only one of two values (e.g. 'male' or 'female', 'true' or 'false',
'present' or 'absent'). This ends up just being another way of doing the t test on a static group
design (the dichotomous variable is the independent variable).

4) Two ordinal variables: various choices (phi coefficient, gamma, tau, Somer's d). Most work
by counting the number of 'concordant' and 'discordant' pairs of observations. Take any two
participants, if one participant is lower than the other on one variable and also lower on the other
variable that that is 'concordant', or if the participant is higher on one variable and also higher on
the other than that is 'concordant'. But it the participant is lower on one variable and higher on
the other than that is 'discordant'.
 Correlational research.

Advantages Disadvantages

 If correlation is significant then further investigation is justified.  No direct cause and effect
can be inferred - the
 If correlation is not significant then you can probably rule out a association between two
causal relationship. variables could potentially be
explained by a third variable.
 No manipulation of behaviour is required.
 As with experiments, may
 The correlation coefficient can readily quantify observational data. lack internal/external validity.

 Little control on extraneous

variables