UNIT 1 PRODUCT MOMENT COEFFICIENT

OF CORRELATION
Structure
1.0 Introduction
1.1 Objectives
1.2 Correlation: Meaning and Interpretation
1.2.1 Scatter Diagram: Graphical Presentation of Relationship
1.2.2 Correlation: Linear and Non-Linear Relationship
1.2.3 Direction of Correlation: Positive and Negative
1.2.4 Correlation: The Strength of Relationship
1.2.5 Measurements of Correlation
1.2.6 Correlation and Causality

1.3 Pearson’s Product Moment Coefficient of Correlation

1.3.1 Variance and Covariance: Building Blocks of Correlations
1.3.2 Equations for Pearson’s Product Moment Coefficient of Correlation
1.3.3 Numerical Example
1.3.4 Significance Testing of Pearson’s Correlation Coefficient
1.3.5 Adjusted r
1.3.6 Assumptions for Significance Testing
1.3.7 Ramifications in the Interpretation of Pearson’s r
1.3.8 Restricted Range

1.4 Unreliability of Measurement

1.4.1 Outliers
1.4.2 Curvilinearity

1.5 Using Raw Score Method for Calculating r

1.5.1 Formulas for Raw Score
1.5.2 Solved Numerical for Raw Score Formula

1.6 Let Us Sum Up

1.7 Unit End Questions
1.8 Suggested Readings

1.0 INTRODUCTION
We measure psychological attributes of people by using tests and scales in order to
describe individuals. There are times when you realise that increment in one of the
characteristics is associated with increment in other characteristic as well. For example,
individuals who are more optimistic about the future are more likely to be happy. On
the other hand, those who are less optimistic about future (i.e., pessimistic about it)
are less likely to be happy. You would realise that as one variable is increasing, the
other is also increasing and as the one is decreasing the other is also decreasing. In
the statistical language it is referred to as correlation. It is a description of “relationship”
or “association” between two variables (more than two variables can also be correlated,
we will see it in multiple correlation).
5
Correlation and Regression In this unit you will be learning about direction of Correlation that is, Positive and
Negative and zero correlation. You will also learn about the strength of correlation
and how to measure correlation. Specifically you will be learning Pearson’s Product
Moment Coefficient of Correlation and how to interpret this correlation coefficient.
You will also learn about the ramifications of the Pearson’s r. You will also learn the
coefficient of correlation equations with numerical examples.

1.1 OBJECTIVES
After reading and doing exercises in this unit, you will be able to:
z describe and explain concept of correlation;
z plot the scatter diagram;
z explain the concept of direction, and strength of relationship;
z differentiate between various measures of correlations;
z analyse conceptual issues in correlation and causality;
z describe problems suitable for correlation analysis;
z describe and explain concept of Pearson’s Product Moment Correlation;
z compute and interpret Pearson’s correlation by deviation score method and raw
score method; and
z test the significance and apply the correlation to the real data.

1.2 CORRELATION: MEANING AND

INTERPRETATION
Correlation is a measure of association between to variables. Typically, one variable
is denoted as X and the other variable is denoted as Y. The relationship between
these variables is assessed by correlation coefficient. Look at the earlier example of
optimism and happiness. It states the relationship between one variable, optimism
(X) and other variable, happiness (Y). Similarly, following statements are example of
correlations:
As the intelligence (IQ) increases the marks obtained increases.
As the introversion increases number of friends decreases.
More the anxiety a person experiences, weaker the adjustment with the stress.
As the score on openness to experience increases, scores on creativity test also
increase.
More the income, more the expenditure.
On a reasoning task, as the accuracy increases, the speed decreases.
As the cost increases the sales decrease.
Those who are good at mathematics are likely to be good at science.
As the age of the child increases, the problems solving capacity increase.

6 More the practice, better the performance.

All the above statements exemplify the correlation between two variables. The variables Product Moment
Coefficient of Correlation
are shown in italics. In this first section, we shall introduce ourselves to the concept
of correlation.

1.2.1 Scatter Diagram: Graphical Presentation of

Relationship
Scatter diagram (also called as scatterplot, scattergram, or scatter) is one way to
study the relationship between two variables. Scatter diagram is to plot pairs of
values of subjects (observations) on a graph. Let’s look at the following data of five
subject, A to E (Table 1.1). Their scores on intelligence and scores on reasoning task
are provided. The same data is used to plot a scatter diagram shown in Figure 1.1.
Now, I shall quickly explain ‘how to draw the scatter diagram’.
Table 1: Data of five subjects on intelligence and scores of reasoning

Subject Intelligence Scores on reasoning

task
A 104 12
B 127 25
C 109 18
D 135 31
E 116 19

Step 1. Plotting the Axes

Draw the x and y axis on the graph and plot one variable on x-axis and another on
y-axis.
(Although, correlation analyses do not restrict you from plotting any variable on any
axis, plot the causal variable on x-axis in case of implicitly assumed cause-effect
relationship.)
Also note that correlation does not necessarily imply causality.
Step 2. Range of Values
Decide the range of values depending on your data.
Begin from higher or lower value than zero.
Conventionally, the scatterplot is square.
So plot x and y values about the same length.
Step 3. Identify the pairs of values
Identify the pairs of values.
A pair of value is obtained from a data.
A pair of values is created by taking a one value on first variable and corresponding
value on second variable.
Step 4. Plotting the graph
Now, locate these pairs in the graph.
7
Correlation and Regression Find an intersection point of x and y in the graph for each pair.
Mark it by a clear dot (or any symbol you like for example, star).
Then take second pair and so on.
The scatterplot shown below is based on the data given in table 1. (Refer to Figure
1).

40

Scores on Reasoning Task

30

20

10
100 110 120 130 140

Intelligence

Fig. 1: Scatter diagram depicting relationship between intelligence and score on

reasoning task

The graph shown above is scatterplot representing the relationship between intelligence
and the scores on reasoning task. We have plotted intelligence on x-axis because it
is a cause of the performance on the reasoning task. The scores on reasoning have
started from 100 instead of zero simply because the smallest score on intelligence is
104 which is far away from zero. We have also started the range of reasoning scores
from 10 since the lowest score on reasoning is 12. Then we have plotted the pair
of scores. For example, subject A has score of 104 on intelligence and 12 on
reasoning so we get x,y pair of 104,12. We have plotted this pair on the point of
intersection between these two scores in the graph by a dot. This is the lowest dot
at the left side of the graph. You can try to practice the scatter by using the data given
in the practice.

1.2.2 Correlation: Linear and Non-Linear Relationship

The relationship between two variables can be of various types. Broadly, they can
be classified as linear and nonlinear relationships. In this section we shall try to
understand the linear and nonlinear relationships.
Linear Relationship
One of the basic forms of relationship is linear relationship. Linear relationship can
be expressed as a relationship between two variables that can be plotted as a
straight line. The linear relationship can be expressed in the following equation (eq.
1.1):
Y=á+âX (eq. 1.1)
In the equation 1.1,
z Y is a dependent variable (variable on y-axis),
8
z á (alpha) is a constant or Y intercept of straight line, Product Moment
Coefficient of Correlation
z â (beta) is slope of the line and
z X is independent variable (variable on x-axis).
We again plot scatter with the line that best fits for the data shown in table 1. So
you can understand the linearity of the relationship. Figure 2 shows the scatter of the
same data. In addition, it shows the line which is best fit line for the data. This line
is plotted by using the method of least squares. We will learn more about it later
(Unit 4). Figure 2 shows that there is a linear relationship between two variables,
intelligence and Scores on Reasoning Task. The graph also shows the straight line
relationship indicating linear relation.

Fig. 2: Scatter showing linearity of the relationship between Intelligence and Scores on
Reasoning Task

Non-linear Relationship
There are other forms of relationships as well. They are called as curvilinear or non-
linear relationships. The Yorkes-Dodson Law, Steven’s Power Law in Psychophysics,
etc. are good examples of non-linear relationships. The relationship between stress
and performance is popularly known as Yorkes-Dodson Law. It suggests that the
performance is poor when the stress is too little or too much. It improves when the
stress is moderate. Figure 3 shows this relationship. The non-linear relationships,
cannot be plotted as a straight line.
The performance is poor at extremes and improves with moderate stress. This is one
type of curvilinear relationship.

9
Correlation and Regression

Fig. 3: Typical relationship between stress and performance

The curvilinear relationships are of various types (cubic, quadratic, polynomial,

exponential, etc.). The point we need to note is that relationships can be of various
types. This block discussed only linear relationships. Other forms of relationship are
not discussed. The types of correlation presented in this block represent linear
relationships. Pearson’s product-moment correlation, Spearman’s rho, etc. are linear
correlations.
The Stevens’ Power Law states that r = csb where, r is sensation, s is stimulus, c
and b are constants and coefficients, respectively. This is obviously a non-linear
relationship between stimulus and sensation. Although, a reader who can recall some
basic mathematics of 10th grade can easily understand that by taking the log of both
sides, the equation can be converted into linear equation.

1.2.3 Direction of Correlation: Positive and Negative

The direction of the relationship is an important aspect of the description of relationship.
If the two variables are correlated then the relationship is either positive or negative.
The absence of relationship indicates “zero correlation”. Let’s look at the positive,
negative and zero correlation.
Positive Correlation
The positive correlation indicates that as the values of one variable increases the
values of other variable also increase. Consequently, as the values of one variable
decreases, the values of other variable also decrease. This means that both the
variables move in the samer direction. For example,
a) As the intelligence (IQ) increases the marks obtained increases.
b) As income increases, the expenditure increases.
The figure 4 shows scatterplot of the positive relationship. You will realise that the
higher scores on X axis are associated with higher score on Y axis and lower scores
on X axis are generally associated with lower score on Y axis. In the ‘a’ example,
higher scores on intelligence are associated with the higher score on marks obtained.
Similarly, as the scores on intelligence drops down, the marks obtained has also
dropped down.

10
 90
Product Moment
Coefficient of Correlation

80

Marks Obtained
70

60

50

40
80 90 100 110 120 130 140 150

Intelligence

Fig. 4: Positive correlation: Scatter showing the positive correlation between intelligence
and marks obtained.

Negative Correlation
The Negative correlation indicates that as the values of one variable increases, the
values of the other variable decrease. Consequently, as the values of one variable
decreases, the values of the other variable increase. This means that two variables
move in the opposite direction. For example,
a) As the intelligence (IQ) increases the errors on reasoning task decreases.
b) As hope increases, depression decreases.
Figure 5 shows scatterplot of the negative relationship. You will realise that the
higher scores on X axis are associated with lower scores on Y axis and lower scores
on X axis are generally associated with higher score on Y axis.
In the ‘a’ example, higher scores on intelligence are associated with the lower score
on errors on reasoning task. Similarly, as the scores on intelligence drops down,
the errors on reasoning task have gone up.

2
Errors on Reasoning Task

-1

-2

-3
-3 -2 -1 0 1 2 3

Intelligence

Fig. 5: Negative correlation: Scatter showing the negative correlation between intelligence
and errors on reasoning task 11
Correlation and Regression No Relationship
Until now you have learned about the positive and negative correlations. Apart from
positive and negative correlations, it is also possible that there is no relationship
between x and y. That is the two variables do not share any relationship. If they do
not share any relationship (that is, technically the correlation coefficient is zero), then,
obviously, the direction of the correlation is neither positive nor negative. It is often
called as zero correlation or no correlation.
(Please note that ‘zero order correlation’ is a different term than ‘zero correlation’
which we will discuss afterwards).
For example, guess the relationship between shoe size and intelligence?
This sounds an erratic question because there is no reason for any relationship
between them. So there is no relationship between these two variables.
The data of one hundred individuals is plotted in Figure 6. It shows the scatterplot
for no relationship.
10

6
Shoe

0
60 80 100 120 140 160

Intelligence

Fig. 6: Scatter between shoe size and intelligence of the individual

1.2.4 Correlation: The Strength of Relationship

You have so far learnt the direction of relationship between two variables. Any
curious reader will ask a question “how strong is the relationship between the two
variables?” For example, if you correlate intelligence with scores on reasoning, and
creativity, what kind of relationship will you expect?
Obviously, the relationship between intelligence and reasoning as well as the relationship
between intelligence and creativity are positive. At the same time the correlation
coefficient (described in the following section) is higher for intelligence and reasoning
than for intelligence and creativity, and therefore we realise that the relationship
between intelligence and reasoning is stronger than relationship between intelligence
and creativity. The strength of relationship between the two variables is an important
information to interpret the relationship.
Correlation Coefficient
The correlation between any two variables is expressed in terms of a number, usually
called as correlation coefficient. The correlation coefficient is denoted by various
symbols depending on the type of correlation. The most common is ‘r’ (small ‘r’)
indicating the Pearson’s product-moment correlation coefficient.
12
The representation of correlation between X and Y is rxy. Product Moment
Coefficient of Correlation
The range of the correlation coefficient is from –1.00 to + 1.00.
It may take any value between these numbers including, for example, – 0.72, –
0.61, – 0.35, + 0.02, + 0.31, + 0.98, etc.
If the correlation coefficient is 1, then relationship between the two variables is
perfect.
This will happen if the correlation coefficient is – 1 or + 1.
As the correlation coefficient moves nearer to + 1 or – 1, the strength of relationship
between the two variables increases.
If the correlation coefficient moves away from the + 1 or – 1, then the strength of
relationship between two variables decreases (that is, it becomes weak).
So correlation coefficient of + 0.87 (and similarly – 0.82, – 0.87, etc.) shows strong
association between the two variables. Whereas, correlation coefficient of + 0.24 or
– 0.24 will indicate weak relationship. Figure 7 indicates the range of correlation
coefficient.

Fig. 7: The Range of Correlation Coefficient.

You can understand the strength of association as the common variance between
two correlated variables. The correlation coefficient is NOT percentage.

explain this point. See, every variable has variance. We denote it as S x2 (variance
of X). Similarly, Y also has its own variance ( S y2 ). In the previous block you have
learned to compute them. From the complete variance of X, it shares some variance
with Y. It is called covariance.
The Figure 8 shown below explains the concept of shared variance. The circle X
indicates the variance of X. Similarly, the circle Y indicates the variance of Y. The
overlapping part of X and Y, indicated by shaded lines, shows the shared variance
between X and Y. One can compute the shared variance.

X Y

Variance of X Variance of Y

Variance shared by X and Y

Fig. 8: Covariance indicates the degree to which X shares variance with Y

To calculate the percentage of shared variance between X and Y (common variance),

one needs to square the correlation coefficient (r). The formula is given below:

13
Correlation and Regression
Percentage of common variance between X and Y = rxy2 × 100 (eq. 1.2)

For instance, if the correlation between X and Y is 0.50 then the percent of variation
shared by X and Y can be calculated by using equation 1.2 as follows.

Percentage of common variance between X and Y = rxy2 × 100 = 0.502 × 100 =

0.25 × 100 = 25%
It indicates that, if the correlation between X and Y is 0.50 then 25% of the variance
is shared by the two variables, X and Y. You would note that this formula is applicable
to negative correlations as well. For instance, if rxy = – 0.81, then shared variance
is:
Percentage of common variance between X and Y = × 100 = – 0.812 × 100 =
0.6561 × 100 = 65.61%

1.2.5 Measurements of Correlation

Correlation coefficient can be calculated by various ways. The correlation coefficient
is a description of association between two variables in the sample. So it is a
descriptive statistics. Various ways to compute correlation simply indicate the degree
of association between variables in the sample. The distributional assumptions are
not required to compute correlation as a descriptive statistics. So it is not a parametric
or nonparametric statistics.
The calculated sample correlation coefficient can be used to estimate population
correlation coefficient.
The sample correlation coefficient is usually denoted by symbol ‘r’.
The population correlation coefficient is denoted by symbol ‘ñ’.
It is Greek letter rho(ñ), pronounced as row (Spearman’s correlation coefficient is
also symbolised as rho.
This may create some confusion among the readers. Therefore, I shall use symbol
rs for Spearman’s rho as a sample statistics and ñs to indicate the population value
of the Spearman’s rho.
Henceforth, I shall also clearly mention the meaning with which ñ is used in this
block.
z When the population correlation coefficient is estimated from sample correlation
coefficient.
z then the correlation coefficient becomes an inferential statistic.
z Inference about population correlation (ñ) is drawn from sample statistics (r).
z The population correlation (ñ) is always unknown.
z What is known is sample correlation (r).
z The population indices are called as parameters and the sample indices are
called as statistics.
z So ñ is a parameter and r is a statistics.
14
While inferring a parameter from sample, certain distributional assumptions are required. Product Moment
Coefficient of Correlation
From this, you can understand that the descriptive use of the correlation coefficient
does not require any distributional assumptions.
The most popular way to compute correlation is ‘Pearson’s Product Moment
Correlation (r)’. This correlation coefficient can be computed when the data on both
the variables is on at least equal interval scale or ratio scale.
Apart from Pearson’s correlation there are various other ways to compute correlation.
Spearman’s Rank Order Correlation or Spearman’s rho (rs) is useful correlation
coefficient when the data is in rank order.
Similarly, Kendall’s tau (ô) is a useful correlation coefficient for rank-order data.
Biserial, Point Biserial, Tetrachoric, and Phi coefficient, are the correlations that are
useful under special circumstances.
Apart from these, multiple correlations, part correlation and partial correlation are
useful ways to understand the associations (Please note that the last three require
more than two variables).

1.2.6 Correlation and Causality

The correlation does not necessarily imply causality. But, if the correlation between
two variables is high then it might indicate the causality. If X and Y are correlated,
then there are three different ways in which the relationship between two variables
can be understood in terms of causality.
1) X is a cause of Y.
2) Y is cause of X.
3) Both, X and Y are caused by another variable Z.
However, causality can be inferred from the correlations.
Regression analysis, path analysis, structural equation modeling, are some examples
where correlations are employed in order to understand causality.

1.3 PEARSON’S PRODUCT MOMENT

COEFFICIENT OF CORRELATION
The Person’s correlation coefficient was developed by Karl Pearson in 1886. Person
was a editor of “Biometrica” which is a leading journal in statistics. Pearson was a
close associate of psychologist Sir Francis Galton. The Pearson’s correlation coefficient
is usually calculated for two continuous variables. If either or both the variables are
not continuous, then other statistical procedures are to be used. Some of them are
equivalent to Pearson’s correlation and others are not. We shall learn about these
procedure after learning the Pearson’s Correlation coefficient.

1.3.1 Variance and Covariance: Building Blocks of

Correlations
Understanding product moment correlation coefficient requires understanding of mean,
variance and covariance. We shall understand them once again in order to understand
correlation.
15
Correlation and Regression n

Mean : Mean of variable X (symbolised as X ) is sum of scores ( ∑ X i ) divided

i =1

by number of observations (n). The mean is calculated in following way.

∑X i
X= i =1 (eq. 1.3)
n

You have learned this in the first block. We will need to use this as a basic element
to compute correlation.
Variance

The variance of a variable X (symbolised as S X2 ) is the sum of squares of the

deviations of each X score from the mean of X ( ∑ ( X − X ) 2 )divided by number
of observations (n).

(eq. 1.4)

You have already learned that standard deviation of variable X , symbolised as , is

square root of variance of X, symbolised as .
Covariance

The covariance between X and Y ( or S XY ) can be stated as

Cov XY =
∑ ( X − X )(Y − Y ) (eq. 1.5)
n

Covariance is a number that indicates the association between two variables. To

compute covariance, deviation of each score on X from its mean ( ) and deviation
of each score on Y from its mean ( y ) is initially calculated.

Then products of these deviations are obtained.

Then, these products are summated.
This sum gives us the numerator for covariance.
Divide this sum by number of observations (n). The resulting number is covariance.

1.3.2 Equations for Pearson’s Product Moment Coefficient of

Correlation
Having revised the concepts, we shall now learn to compute the Pearson’s Correlation
Coefficient.
Formula
Since we have already learned to compute the covariance, the simplest way to define
Pearson’s correlation is…
16
 Product Moment
CovXY
r= (eq. 1.6)
Coefficient of Correlation
S X SY

Where,

the Cov XY is covariance between X and Y,

S X is standard deviation of X

SY is standard deviation of Y.

Since, it can be shown that CovXY is always smaller than or equal to S X SY , the
maximum value of correlation coefficient is bound to be 1.

The sign of Pearson’s r depends on the sign of CovXY .

If the CovXY is negative, then r will be negative and

if CovXY is positive then r will be a positive value.

The denominator of this formula ( S X SY ) is always positive. This is the reason for a
– 1 to + 1 range of correlation coefficient. By substituting covariance equation (eq.
1.5) for covariance we can rewrite equation 1.6 as

∑ ( X − X )(Y − Y )
r= n
S X SY
(eq. 1.7)

By following a simple rule, a÷ b ÷ c = a ÷ (b × c), we can rewrite equation 1.7

as follows.

r=
∑ ( X − X )(Y − Y )
nS X rSY=
(eq. 1.8)
1.3.3 Numerical Example
Now we shall use this formula to compute Pearson’s correlation coefficient. For this
purpose we will use the following data. The cognitive theory of depression argues
that hopelessness is associated with depression. Aron Beck developed instruments
to measure depression and hopelessness. The BHS (Beck Hopelessness Scale) and
the BDI (Beck Depression Inventory) are measures of hopelessness and depression,
respectively.
Let’s take a hypothetical data of 10 individuals on whom these scales were
administered. (In reality, such a small data is not sufficient to make sense of correlation;
roughly, at least a data of 50 to 100 observations is required). We can hypothesize
that the correlation between hopelessness and depression will be positive. This
hypothetical data is given below in table 2.

17
Correlation and Regression Table 2: Hypothetical data of 10 subjects on BHS and BDI

Subject BHS BDI X − X Y − Y ( X − X )2 (Y − Y ) 2 ( X − X )(Y − Y )

(X) (Y)
1 11 13 0 1 0 1 0
2 13 16 2 4 4 16 8
3 16 14 5 2 25 4 10
4 9 10 -2 -2 4 4 4
5 6 8 -5 -4 25 16 20
6 17 16 6 4 36 16 24
7 7 9 -4 -3 16 9 12
8 12 12 1 0 1 0 0
9 5 7 -6 -5 36 25 30
10 14 15 3 3 9 9 9

n = 10 ∑X ∑Y ∑ ( X − X ) ∑ (Y − Y )
2 2
∑ ( X − X )(Y − Y )
=110 =120 = 156 = 100 = 117
X = 11 Y = 12

SX = ∑ (X − X ) 2
/ n = 4.16

SY = ∑ (Y − Y ) 2
/ n = 3.33

r = ∑ ( X − X )(Y − Y ) / nS X SY = 117 / (10)(4.16)(3.33) = + 0.937

Step 1. You need scores of subjects on two variables. We have scores on ten
subjects on two variables, BHS and BDI.
Step 2. Then list the pairs of scores on two variables in two columns. The order will
not make any difference. Remember, same individuals’ two scores should be kept
together. Label one variable as X and other as Y. We label BHS as X and BDI as
Y.
Step 3. Compute the mean of variable X and variable Y. It was found to be 11 and
12 respectively.

Step 4. Compute the deviation of each X score from its mean ( ) and each Y
score from its own mean ( Y ). This is shown in the column labeled as
X − X and Y − Y . As you have learned earlier, the sum of these columns has to be
zero.

Step 5. Compute the square of x − x and y − y .

This is shown in next two columns labelled as ( x − x )2 and ( y − y )2.

Step 6. Then compute the sum of these squared deviations of X and Y. The sum of
squared deviations for X is 156 and for Y it is 100.
Step 7. Divide them by n to obtain the standard deviations for X and Y. The Sx was
18 found to be 4.16. Similarly, the Sy was found to be 3.33.
Step 8. Compute the cross-product of the deviations of X and Y. These cross- Product Moment
Coefficient of Correlation
products are shown in the last column labeled as ( x − x ) ( y − y ).

Step 9. Then obtain the sum of these cross-products. It was found to be 117. Now,
we have all the elements required for computing r.
Step 10. Use the formula of r to compute correlation. The sum of the cross-product
of deviations is numerator and n, Sx, Sy, are denominators. Compute r. the value of
r is 0.937 in this example.

1.3.4 Significance Testing of Pearson’s Correlation

Coefficient
Statistical significance testing is testing the hypothesis about the population parameter
from sample statistics. When the Pearson’s Correlation coefficient is computed as an
index of description of relationship between two variables in the sample, the significance
testing is not required. The interpretation of correlation from the value and direction
is enough.
However, when correlation is computed as an estimate of population correlation,
obviously, statistical significance testing is required.
“Whether the obtained sample value of Pearson’s correlation coefficient is greater
than the value that can be obtained by chance?” is the question answered by statistical
significance testing about correlation coefficient.
Different values of correlation can be obtained between any two variables, X and Y,
for different samples of different sizes belonging to the same population.
The researcher is not merely interested in knowing the finding in the specific sample
on which the data are obtained. But they are interested in estimating the population
value of the correlation.
Testing the significance of correlation coefficient is a complex issue. It is because of
the distribution of the correlation coefficient. The t-distribution and z-distribution are
used to test statistical significance of r.
The population correlation between X and Y is denoted by ñxy. The sample correlation
is rxy.
As you have learned, we need to write a null hypothesis (HO) and alternative hypothesis
(HA) for this purpose.
The typical null hypothesis states that population correlation coefficient between X
and Y (ñxy) is zero.
HO : ñxy = 0
HA: ñxy ≠ 0
If we reject the HO then we accept the alternative (HA) that the population correlation
coefficient is other than zero. It implies that the finding obtained on the data is not
a sample-specific error.
Sir Ronald Fisher has developed a method of using t-distribution for testing this null
hypothesis.
The degrees of freedom (df) for this purpose are n – 2. Here n refers to number
19
of observations.
Correlation and Regression We can use Appendix C in a statistic book for testing the significance of correlation
coefficient. Appendix C provides critical values of correlation coefficients for various
degrees of freedom. Let’s learn how to use the Appendix C. We shall continue with
the example of BHS and BDI.
The correlation between BHS and BDI is +.937 obtained on 10 individuals. We
decide to do statistical significance testing at 0.05 level of significance, so our á =
.05.
We also decided to apply two-tailed test.
The two-tailed test is used if alternative hypothesis is non-directional, i.e. it does not
indicate the direction of correlation coefficient (meaning, it can be positive or negative)
and one-tail test is used when alternative is directional (it states that correlation is
either positive or negative).
Let us write the null hypothesis and alternative hypothesis:
Null hypothesis

Η O : ρBHS BDI = 0
Η A : ρBHS BDI ≠ 0

Now we will calculate the degree of freedom for this example.

df = n – 2 = 10 – 2 = 8 (eq. 1.9)
So the df for this example are 8. Now look at Appendix C. Look down the leftmost
df column till you reach df = 8. Then look across to find correlation coefficient from
column of two-tailed test at level of significance of 0.05. You will reach the critical
value of r:
rcritical = 0.632
Because the obtained (i.e., calculated) correlation value of + 0.937 is greater than
critical (i.e., tabled) value, we reject the null hypothesis that there is no correlation
between BHS and BDI in the population.
So we accept that there is correlation between BHS and BDI in the population. This
method is used regardless of the sign of the correlation coefficient.
We use the absolute value (ignore the sign) of correlation while doing a two-tailed
test of significance. The sign is considered while testing one-tailed hypothesis.
For example, if the HA: ñ > 0, which is a directional hypothesis, then any correlation
that is negative will be considered as insignificant.

1.3.5 Adjusted r
The Pearson’s correlation coefficient (r) calculated on the sample is not an unbiased
estimate of population coefficient (ñ). When the number of observations (sample
size) are small the sample correlation is a biased estimate of population correlation.
In order to reduce this bias, the calculated correlation coefficient is adjusted. This is
called as adjusted correlation coefficient (radj).

(1 − r 2 )(n − 1)
radj = 1 −
n−2
Where,

20 radj= adjusted r
r2- = the square of Pearson’s Correlation Coefficient obtained on sample, Product Moment
Coefficient of Correlation
n = sample size
In case of our data, presented in table 1.2, the correlation between BHS and BDI
is +.937 obtained on the sample of 10. The adjusted r can be calculated as follows

(1 − .9372 )(10 − 1) (.1220)(9)

radj = 1 − = 1− = 1 − 0.1373 = .929
10 − 2 8
The radj is found to be 0.929. This coefficient is unbiased estimate of population
correlation coefficient.

1.3.6 Assumptions for Significance Testing

One may recall that simple descriptive use of correlation coefficient does not involve
any assumption about the distribution of either of the variables. However, using
correlation as an inferential statistics requires assumptions about X and Y. These
assumptions are as follows. Since we are using t-distribution, the assumptions would
be similar to t.
Assumptions:
Independence among the pairs of score
This assumption implies that the scores of any two observations (subjects in case of
most of psychological data) are not influenced by each other. Each pair of observation
is independent. This is assured when different subjects provides different pairs of
observation.
The population of X and the population of Y follow normal distribution and the
population pair of scores of X and Y has a normal bivariate distribution.
This assumption states that the population distribution of both the variables (X and
Y) is normal. This also means that the pair of scores follows bivariate normal distribution.
This assumption can be tested by using statistical tests for normality.
It should be remembered that the r is a robust statistics. It implies that some violation
of assumption would not influence the distributional properties of t and the probability
judgments associated with the population correlation.

1.3.7 Ramifications in the Interpretation of Pearson’s r

The interpretation of the correlation coefficient depends primarily on two things:
direction and strength of relationship. We have already discussed them in detail and
hence repetition is avoided.
Direction
If the correlation is positive, then the relationship between two variables is positive.
It means that as there is an increase in one there is an increase in another, and as
there is a decrement in one there is a decrease in another. When the direction of
correlation is negative then the interpretation is vice-versa.
Strength
The strength can be calculated in terms of percentage. We have already learned this
formula. So we can convert the correlation coefficient into percentage of common
21
Correlation and Regression variance explained and accordingly interpret. For example, if the correlation between
X and Y is 0.78, then the common variance shared by X and Y is 60.84 percent.
Usually distinct psychological variables do not share much of the common variance.
In fact, the reliability of psychological variables is an issue while interpreting the
correlations. Cohen and Cohen have suggested that considering the unreliability of
psychological variables, the smaller correlations should also be considered significant.
Although, direction and strength are key pointers while interpreting the correlation,
there are finer aspects of interpretation to correlations.
They are :
z range,
z outliers,
z reliability of variables, and
z linearity
The above are some of the important aspects which all obscure the interpretation
of correlation coefficient. Let’s discuss them one by one.

1.3.8 Restricted Range

It is expected the variables in correlation analysis are measured with full range. For
example, suppose we want to study the correlation between hours spent in studies
and marks. We are suppose to take students who have varying degree of hours of
studies, that is, we need to select students who have spent very little time in studies
to the once who have spent great deal of time in studies. Then we will be able to
obtain true value of the correlation coefficient.
But suppose we take a very restricted range then the value of the correlation is likely
to reduce. Look at the following examples the figure 1.9a and 1.9b.
The figure 1.9a is based on a complete range.
The figure 1.9b is based on the data of students who have studied for longer
durations.
The scatter shows that when the range was full, the correlation coefficient was
showing positive and high correlation. When the range was restricted, the correlation
has reduced drastically.
You can think of some such examples. Suppose, a sports teacher selects 10 students
from a group of 100 students on basis of selection criterion, that is their athletic
performance.
The actual performance of these ten selected students in the game was correlated
with the selection criterion. A very low correlation was obtained between selection
criterion and actual game performance. This would naturally mean that the selection
criterion is not related with actual game performance. Is it true..? Why so…?
If you look at the edata, you will realise that the range of the scores on selection
criterion is extremely restricted (because these ten students were only high scorers)
and hence the relationship is weak. So note that whenever you interpret correlations,
the range of the variables is large. Otherwise the interpretations will not be valid.
22
 r = .96 r = .50
Product Moment
50 50 Coefficient of Correlation

40 48

30 46

20 44

MARKS
10 42
marks
MARKS

0 40
0 1 2 3 4 5 6 7 8 4.5 5.0 5.5 6.0 6.5 7.0 7.5

hours studied Hours Studied

Fig. 1.9a: Scatter showing full range Fig. 1.9b: Scatter with restricted
on both variables range on hours studied

1.4 UNRELIABILITY OF MEASUREMENT

Psychological research involves the use of scales and tests. One of the psychometric
property psychological instruments should poses is reliability. Reliability refers to
consistency of a measurement. If the instrument is consistent, then the test has high
reliability. But at times one of the variable or both the variables may have lower
reliability. In this case, the correlation between two less reliable variable reduces.
Generally, while interpreting the correlation, the reliability is assumed to be high. The
general interpretations of correlations are not valid if the reliability is low. This reduction
in the correlation can be adjusted for the reliability of the psychological test. More
advanced procedures are available in the books of psychological testing and statistics.
They involve calculating disattenuated correlations. Correlation between two variables
that have less than perfect reliability is adjusted for unreliability. This is called as
disattenuated correlation. If both variables were perfectly reliable then correlation
between them is disattenuated correlation.

1.4.1 Outliers
Outliers are extreme score on one of the variables or both the variables. The presence
of outliers has deterring impact on the correlation value. The strength and degree of
the correlation are affected by the presence of outlier. Suppose you want to compute
correlation between height and weight. They are known to correlate positively. Look
at the figure below. One of the scores has low score on weight and high score on
height (probably, some anorexia patient).
Figure 1.10. Impact of an outlier observation on correlation. Without the outlier, the
correlation is 0.95. The presence of an outlier has drastically reduced a correlation
coefficient to 0.45.

r = +.45
70

60

50

40

30
Weight

20
3.0 3.5 4.0 4.5 5.0 5.5 6.0 6.5

Height
23
Correlation and Regression 1.4.2 Curvilinearity
We have already discussed the issue of linearity of the relationship. The Pearson’s
product moment correlation is appropriate if the relationship between two variables
is linear. The relationships are curvilinear then other techniques need to be used. If
the degree of cruviliniarity is not very high, high score on both the variable go
together, low scores go together, but the pattern is not linear then the useful option
is Spearman’s rho.

1.5 USING RAW SCORE METHOD FOR

CALCULATING r
The method which we have learned to compute the correlation coefficient is called
as deviation scores formula. Now we shall learn another method to calculate Pearson’s
correlation coefficient. It is called as raw score method. First we will understand
how the two formulas are similar. Then we will solve a numerical example for the raw
score method. We have learned following formula for calculating r.

1.5.1 Formulas for Raw Score

We have already learnt following formula of correlation (eq. 1.8). This is a deviation
score formula.

r=
∑ ( X − X )(Y − Y )
nS X SY
The denominator of correlation formula can be written as

∑ ( X − X ) (Y − Y )
2 2
(eq. 1.10)

Which is

( SS X SSY ) (eq. 1.11)

We have already learnt that

(∑ X )2
SS X = ∑ ( X − X ) = ∑ X −
2 2
(eq. 1.12)
n
and

(∑ Y ) 2
SSY = ∑ (Y − Y ) 2 = ∑ Y 2 − (eq. 1.13)
n
The numerator of the correlation formula can be written as

(∑ X )(∑ Y )
∑ ( X − X )(Y − Y ) = ∑ XY − n
(eq. 1.14)

24
So r can be calculated by following formula which is a raw score formula: Product Moment
Coefficient of Correlation

∑ XY − ∑ n ∑
( X )( Y )
r= (eq. 1.15)
(SS X SSY )

1.5.2 Solved Numerical for Raw Score Formula

We shall solve the same numerical example by using the formulas shown above.
Table 3 shows how to calculate Pearson’s r by using raw score formula.
Table 1.3: Table showing the calculation of r by using raw score formula.

Subject BHS BDI X2 Y2 XY

(X) (Y)
1 11 13 100 676 260
2 13 16 64 529 184
3 16 14 81 529 207
4 9 10 169 676 338
5 6 8 121 576 264
6 17 16 196 900 420
7 7 9 256 729 432
8 12 12 144 729 324
9 5 7 225 841 435
10 14 15 144 625 300
Summation 110 120 1366 1540 1437
X = 11 Y = 12

SS X = ∑ X 2 − ( ∑ X ) 2 / n = 1366 – (110) /10 = 156

2

SSY = ∑ Y 2 − (∑ Y ) 2 / n = 1540 – (120)2 / 10 = 100

∑ ( X − X )(Y − Y ) = ∑ XY − ∑ n ∑
( X )( Y )
=117

∑ XY − ∑ n ∑
( X )( Y )
=0.937
r=
(SS X SSY )
Readers might find one of the methods easier. There is nothing special about the
methods. One should be able to correctly compute the value of correlation.

1.6 LET US SUM UP

In this unit we started with definition and meaning of correlation and followed it up
with how the correlation could be depicted in graphical form and scatter diagram.
We then learnt about linear and non-linear and curvilinear relationship amongst
variables. We also learnt that the direction of relationship could be either in the
positive or in the negative direction. It can also be no correlation in that it can be
a zero correlation. Then we learnt the methods of measurement of correlation and
25
Correlation and Regression learnt how to use the formula for calculating the Pearson’s r, that Pearson’s Product
Moment Coefficient of Correlation. We then discussed about the building blocks of
correlation which included variance and co variance. We also learnt how to test
the level of significance of a particular coefficient of correlation calculated by us.
Then we learnt about the interpretation of the correlation coefficient and also its
ramifications. Then we looked into the unreliability of a correlation and the causes
for the same such as the inclusion of outliers etc.

1.7 UNIT END QUESTIONS

1) Problem:
Plot scatter diagram for the following data. Compute Pearson’s correlation between
x and y. Write the null hypothesis stating that population correlation is zero. Test the
significance of the correlation coefficient.

X Y
12 20
13 22
15 28
17 31
11 22
9 24
8 18
10 21
11 23
7 16

2) Plot scatter for following example. The data was collected on Perceived stress
and anxiety on 10 subjects. Compute the Pearson’s correlation between them State
the null hypothesis. Test the null hypothesis using this hypothesis. Do the similar
exercise after deleting a pair that clearly looks an outlier observation.

Perceived Anxiety
stress
9 12
8 11
7 9
4 5
8 9
4 6
6 8
14 2
7 11
11 9
9 11

26
3) Data showing scores on time taken to complete 200 meters race and duration Product Moment
Coefficient of Correlation
of practice for 5 runners. Plot the scatter. Compute mean, variance, SD, and
covariance. Compute correlation coefficient. Write the null hypothesis.

Time taken Duration of

(in Seconds) Practice
(in months)
31 11
32 14
36 9
26 15
38 7

4) Data showing scores on dissatisfaction with work and scores on irritability

measured by standardised test for thirteen individuals. Plot the scatter. Compute
mean, variance, SD, and covariance. Compute correlation coefficient. Write the null
hypothesis stating no relationship. Test the significance at 0.05 level of significance.

Dissatisfaction Irritability
with work scores
12 5
16 7
19 9
27 13
30 16
25 11
22 6
26 14
11 7
17 9
19 14
21 18
23 19

5) Check whether the following statements are true or false.

1) Positive correlation means as X increases Y decreases. True/False

2) Negative correlation means as X decreases Y decreases. True/False
3) Generally, in a scatter, lower scores on X are paired True/False
with lower scores on Y for negative correlation.
4) – 1.00 ≤ Pearson’s correlation ≤ + 1.00 True/False
5) Generally, in a scatter, lower scores on X are paired True/False
with higher scores on Y in positive correlation.
6) The scatter diagram cannot indicate the direction of the True/False
relationship.
7) Percentage of shared variance by X and Y can be True/False
obtained by squaring the value of correlation.

Answers: 1) = False, 2) = False, 3) = False, 4) = True, 5) = False, 6) = False,

7) = True
27
Correlation and Regression Answer in brief.
6) What is correlation coefficient?
7) What is the range of correlation coefficient?
8) Is correlation coefficient a percentage?
9) How to calculate common variance from correlation coefficient?
10) What is the percentage of variance shared by X and Y if the rxy = 0.77?
11) What is the percentage of variance shared by X and Y if the rxy = – 0.56?
Answers:
A number expressing the relationship between two variables.
The range of correlation coefficient is from –1.00 to + 1.00.
No. Correlation is not a percentage. But it can be converted into percentage of
variance shared.
Common variance is calculated from correlation coefficient by using a formula: rxy2
× 100.
59.29%
31.36%

1.8 SUGGESTED READINGS

Aron, A., Aron, E. N., Coups, E.J. (2007). Statistics for Psychology. Delhi: Pearson
Education.
Minium, E. W., King, B. M., & Bear, G. (2001). Statistical Reasoning in Psychology
and Education. Singapore: John-Wiley.
Guilford, J. P., & Fructore, B. (1978). Fundamental Statistics for Psychology and
Education. N.Y.: McGraw-Hill.
Wilcoxon, R. R. (1996). Statistics for Social Sciences. San Diego: Academic Press.

28