5.3 Calculation of Correlation Coefficient

The correlation coefficient will be in the interval \([-1,1]\).

W-1100-0PPNpne.4.eeooea4rrsgkfefeliitaccnivtictteevo p naeroerfa fresg cialiaoriratitr ctivivreocoeelron carrroctielerooalarrrntteleioialanottnioinon

Figure 39: The range of patterns a correlation coefficient describes, from perfect positive to perfect negative.
5.3.1 Pearson Product-Moment Correlation Coefficient

\begin {align*} \text {(a).}\hspace {1cm} r&=\frac {\sum (X-\overline {X})(Y-\overline {Y})}{\sqrt {\sum (X-\overline {X})^2}\hspace {0.1cm}.\hspace {0.1cm}\sqrt {\sum (Y-\overline {Y})^2}}\\\\\\ \text {(b).}\hspace {1cm} r&=\frac {n\sum XY-\sum X\sum Y}{\sqrt {n\sum X^2-\Bigg (\sum X\Bigg )^2}\sqrt {n\sum Y^2-\Bigg (\sum Y\Bigg )^2}}\\ \end {align*}

Example 5.3. Compute the product-moment correlation coefficient for the examination marks used in the scatter diagram above.

Solution. Build the three extra columns and total everything.

\(X\) \(Y\) \(XY\) \(X^2\) \(Y^2\)
42 31 1302 1764 961
84 83 6972 7056 6889
50 42 2100 2500 1764
42 60 2520 1764 3600
33 28 924 1089 784
50 63 3150 2500 3969
69 59 4071 4761 3481
81 92 7452 6561 8464
80 73 5840 6400 5329
35 40 1400 1225 1600
566 571 35731 35620 36841

\[r=\frac {n\sum XY-\sum X\sum Y} {\sqrt {n\sum X^2-\left (\sum X\right )^2}\ \sqrt {n\sum Y^2-\left (\sum Y\right )^2}} =\frac {10(35731)-566(571)}{\sqrt {10(35620)-566^2}\ \sqrt {10(36841)-571^2}}=0.876.\] It is convenient to record the three quantities that every later formula needs: \[S_{XX}=3584.4,\qquad S_{YY}=4236.9,\qquad S_{XY}=3412.4,\] so that \(r=S_{XY}/\sqrt {S_{XX}S_{YY}}\) and, below, \(\hat {b}=S_{XY}/S_{XX}\).

5.3.2 Spearman’s Rank Correlation Coefficient

In this case we do not have two continuous variables which are normally distributed but two discrete variables in the form of ranking \[r=1-\frac {6\sum d^2}{n(n^2-1)}\] If there is an agreement between the two ranks, then \(r=1\). \(d=\) difference between the two ranks.

Example 5.4. Two judges at a competition awarded the following marks to candidates

Judge A 5.8 5.5 5.9 4.9 5.9 5.6 5.0
Judge B 5.5 5.4 5.8 5.3 5.7 5.7 5.7

1.
Calculate the Spearmans rank correlation coefficient.
2.
Do the judges agree on the order in which they place the candidate.

Solution.

Judge A Rank Judge B Rank \(d\) \(d^2\)
5.8 3 5.5 5 \(-2\) 4
5.5 5 5.4 6 \(-1\) 1
5.9 3/2 5.8 1 0.5 0.25
4.9 7 5.3 7 0 0
5.9 3/2 5.7 3 \(-1.5\) 2.25
5.6 4 5.7 3 1 1
5.0 6 5.7 3 3 9
17.5

\begin {align*} r&=1-\frac {6\sum d^2}{n(n^2-1)}=1-\frac {6\times 17.5}{7(7^2-1)}\\\\ \implies \hspace {0.5cm} r&=0.6875 \end {align*}

There is a positive fair correlation between the judges. So we have that \(r_s=0.6875\)

Now we solve for the second part.
Set the hypothesis
\(H_0\): The is no correlation between the judges \(r_s=0\)
\(H_1:\) There is an association \(r_s\neq 0\)

\(|r_c|=\) critical value from the table, \(n=7\), \(|r_c|=0.75\)
\[r_s=0.6875\] Since \(r_s = 0.6875 < |r_c| = 0.75\), we fail to reject \(H_0\) at this level.

Be careful how that is reported. It does not show the judges disagree. The observed \(r_s = 0.6875\) is a fairly strong positive association – it is simply not strong enough, with only 7 items, to rule out chance at this significance level. A test on so few observations has little power, and the honest conclusion is that there is insufficient evidence of agreement, not evidence of its absence.

Questions on this section

Stuck on something here? Ask below and it stays attached to this topic.