Skills Companion 3

Interpretation of Numerical Values

The numerical value of the correlation coefficient, rs, ranges between -1 and +1. The
correlation coefficient is the number indicating the how the scores are relating.
r
s
correlation coefficient
!n general,
r
s
" # implies positive agreement among ranks
r
s
$ # implies negative agreement %or agreement in the reverse direction&
r
s
# implies no agreement
Closer r
s
is to 1, better is the agreement while rs closer to -1 indicates strong agreement in
the reverse direction.
Assigning Ranks
!n order to compute Spearman 'ank Correlation Coefficient, it is necessar( that the data be
ranked. There are a few issues here.
Suppose that scores of the )udges %out of 1# were as follows&*
Contestant No. 1 + 3 , - . / 0 1 1#
Score by Judge A - 1 3 0 . / , 0 , .
Score by Judge B / 0 . / 0 - 1# . - 0
'anks are assigned separatel( for the two )udges either starting from the highest or from the
lowest score. 2ere, the highest score given b( 3udge 4 is 1.
!f we begin from the highest score, we assign rank 1 to contestant + corresponding to the
score of 1.
The second highest score is 0 but two competitors have been awarded the score of 0. !n this
case both the competitors are assigned a common rank which is the arithmetic mean of
ranks + and 3. !n this wa(, scores of 3udge 4 can be converted into ranks.
Similarl(, ranks are assigned to the scores awarded b( 3udge 5 and then differences
between ranks for each contestant are used to evaluate rs. 6or the above e7ample, ranks are
as follows.
Contestant No. 1 + 3 , - . / 0 1 1#
Ranks of scores by Judge A / 1 1# +.- -.- , 0.- +.- 0.- -.-
Ranks of scores by Judge B -.- 3 /.- -.- 3 1.- 1 /.- 1.- 3
89C
Skills Companion 3
Spearman Rank Correlation Coefficient tries to assess the relationship beteen ranks
ithout making any assumptions about the nature of their relationship.
!ence it is a non"parametric measure " a feature hich has contributed to its
popularity and ide spread use.
something said as a warning, caution, or qualifcation
Ad#antages and Ca#eats
$ther measures of correlation are parametric in the sense of being based on possible
relationship of a parameteri%ed form& such as a linear relationship.
Another ad#antage ith this measure is that it is much easier to use since it does not
matter hich ay e rank the data& ascending or descending. 'e may assign rank (
to the smallest #alue or the largest #alue& pro#ided e do the same thing for both sets
of data. Can be mentioned as strength.
)he only re*uirement is that data should be ranked or at least con#erted into ranks.
Spearman 'ank Correlation 4 measure of 'ank Correlation :roup 3
The Spearman Correlation
Spearman;s correlation is designed to measure the relationship between variables
measured on an ordinal scale of measurement.
Similar to 8earson;s Correlation, however it uses ranks as opposed to actual values.
4ssumptions
The data is a bivariate random variable.
The measurement scale is at least ordinal.
< i , = i is independent of < ) , = ) where i > )
Ad#antages
?ess sensitive to bias due to the effect of outliers
Can be used to reduce the weight of outliers %large distances get treated as a one-
rank difference&
@oes not reAuire assumption of normalit(.
Bhen the intervals between data points are problematic, it is advisable to stud( the
rankings rather than the actual values.
+isad#antages
Calculations ma( become tedious. 4dditionall( ties are important and must be
factored into computation.
89C
Skills Companion 3
Steps in Calculating Spearman;s 'ho
Convert the observed values to ranks %accounting for ties&
6ind the difference between the ranks, sAuare them and sum the sAuared differences.
Set up h(pothesis, carr( out test and conclude based on findings.
Steps in Calculating Spearman;s 'ho
!f the null is re)ected then calculate the Spearman correlation coefficient to measure
the strength of the relationship between the variables.
2(pothesis* !
%Two-Tailed&
2 o * There is no correlation between the <s and the =s.
%there is mutual independence between the <s and the =s&
2 1 * There is a correlation between the <s and the =s.
%there is mutual dependence between the <s and the =s&
2(pothesis* !!
5. %Cne-Tailed - ?ower&
2 o * There is no correlation between the <s and the =s.
%there is mutual independence between the <s and the =s&
2 1 * There is a negative correlation between the <s and the =s.
2(pothesis* !!!
C. %Cne-Tailed - Dpper&
2 o * There is no correlation between the <s and the =s.
%there is mutual independence between the <s and the =s&
2 1 * There is a positive correlation between the <s and the =s.
Test Statistic
6or small samples %E $ ,#&*
T F d i + F G'%< i & - '%= i &H +
6or large samples*
%'e)ect using the appropriate I critical value&
Test Statistic
!n the case of a large sample*
89C
Skills Companion 3
@ecision 'ules
Two-tailed*
'e)ect 2 # if TJ S KL+ or T " S 1- KL+ .
@o not re)ect otherwise.
5. Cne-tailed - ?ower*
'e)ect 2 # if T " S 1- K .
@o not re)ect otherwise.
C. Cne-tailed- Dpper*
'e)ect 2 # if TJ S K .
@o not re)ect otherwise.
!n the case of few ties %less than -M of the sample&*
Bhere d i is the difference in the ranks of each pair and E is the number of pairs
Spearman;s 'ho
Spearman;s 'ho
!f there are numerous ties*
Spearman;s 'ho
4ssumes values between -1 and +1
-1 J N J # J N J +1
8erfectl( Eegative Correlation 8erfectl( 8ositive Correlation
O7ample 1
The !CC rankings for Cne @a( !nternational %C@!& and Test matches for nine teams
are shown below.
Test whether there is correlation between the ranks
Team Test 'ank C@! 'ank 4ustralia 1 1 !ndia + 3 South 4frica 3 + Sri ?anka , / Ongland -
. 8akistan . , Eew Iealand / - Best !ndies 0 0 5angladesh 1 1
O7ample 1 4nswer* Team Test 'ank C@! 'ank d d + 4ustralia 1 1 # # !ndia + 3 1 1
South 4frica 3 + 1 1 Sri ?anka , / 3 1 Ongland - . 1 1 8akistan . , + , Eew Iealand / - +
, Best !ndies 0 0 # # 5angladesh 1 1 # # Total +#
O7ample +
4 composite rating is given b( e7ecutives to each college graduate )oining a plastic
manufacturing firm. The e7ecutive ratings represent the future potential of the
college graduate. The graduates then enter an in-plant training programme and are
given another composite rating. The e7ecutive ratings and the in-plant ratings are as
follows*
4t the -M level of significance, determine if there is a positive correlation
between the variables
89C
Skills Companion 3
6ind the rank correlation coefficient if the null is re)ected
:raduate O7ecutive rating %<& Training rating %=& 4 0 , 5 1# , C 1 , @ , 3 O 1+ . 6 11 1 :
11 1 2 / . ! 0 . 3 13 1 P 1# - ? 1+ 1
'hat are the limitations,disad#antages of spearman-s rank.
4s SpearmanQs rank onl( uses rank, it is not affected b( significant variations in readings.
4s long as the order remains the same, the coefficient will sta( the same. 4s with an(
comparison, the possibilit( of chance will have to be evaluated to ensure that the +
Auantities are actuall( connected.
Spearmans Rank Correlation Coefficient (
s
r
)
This is a method to measure the correlation between 2 variables recorded from the
same sample. It works on a scale from 1 to +1: It is used to test the strength of a
relationship between two sets of data, providing ou with a numerical value.
!or e"ample if ou draw a scatter graph ou can see b ee if there is a
relationship, but ou will probabl not be able to clearl assess the strength of the
relationship as man points ma be some distance from the line of best fit.
#nce an answer to the $pearman%s &ank formula has been reached ou can test
its significance this means the likelihood of our results occurring b chance.
This is a very important part of the process and must not be left out.
$pearman%s rank is used to test a hpothesis. Therefore at the beginning of the
calculation ou must state our hypothesis and our null hypothesis 'this is
simpl the inverse of the hpothesis(
e.g. )pothesis &iver discharge increases downstream
*ull )pothesis: &iver discharges decreases downstream
When can you use Spearmans Rank
The test can be used with an set of raw data or percentages, but it is onl suitable
if all of the following criteria appl.
There are two data sets which ou believe ma or ma not be related e.g.
velocit and hdraulic radius
+t least 1, pairs of data are available to be used.
The test is limited to -, pairs more than this makes the e"ercise unwield.
89C
Skills Companion 3
The !ormula is"
n n
d
r
s

=

3
+
.
1
r
s #
Spearmans Rank Coefficient
d # The difference bet$een the t$o ranks d
2
= The difference s%uared
& # ('reek letter sigma) meaning the sum of
N # The number of variables
Site (ndependen
t )ariable
(*)
+ependent
)ariable
(y)
Rank
*
Rank
y
d d
,
The results can be interpreted as follo$s
-. means /erfect 0egative Correlation 1 as one set of data increases the
other decreases
-2 means 0o Correlation 1 you $ould accept the null hypothesis
3. means /erfect /ositive Correlation 1 as one set of data increases the
other set increases
.hen organising and ranking our data begin with the independent variable e.g.
The distance downstream, and rank from highest to lowest. !ollow this b the
dependent variable e.g. bedload si/e.
$trengths .eaknesses
It gives ou ob0ective data
It enables ou to demonstrate a
clear relationship between two
sets of data
1ou can state whether the
relationship is significant or if
our results were 0ust a fluke.
It is less sensitive to anomalies in
data as each piece is ranked
large differences could onl be
one rank different
It does not tell ou whether there
is one causal link 'i.e. that one
change leads to a change in
another(, 0ust that a relationship
e"ists.
Too man 2tied ranks% can affect
the validit of the te"t.
It could be sub0ect to human error
e.g. inaccurate calculations.
Testing Significance
+4 04T 567)6 T8(S 9(T 4:T
89C
Skills Companion 3
)aving completed our calculations ou must now analse our result, and test its
statistical significance.
3 If our final result is not between 41 and +1 ou have gone wrong in our
calculations and need to check them.
.hat does m result mean5
41 , +1
6erfect negative 0o relationship 6erfect positive
relationship relationship
7ocate our result along this continuum, and make a statement about the strength
of the relationship ou have identified.
1ou now need to do another test to check whether our results could have
occurred b chance i.e. how significant is the result5
This means ou have to compare our result with a table of critical values.
8efore ou can use a critical values table ou must calculate degrees of freedom.
The formula for this is n 2 'i.e. The number of values ou have 42( Therefore for
a table with 12 values our degrees of freedom would be 1,.
9ritical value of $pearman%s rank
9orrelation coefficient r
s
$ignificance 7evel
:egrees of !reedom ,.,; ';<( ,.,1 '1<(
= ,.>> 1.,,
; ,.>- ,.?@
@ ,.>, ,.?1
A ,.AA ,.>A
> ,.A2 ,.>=
? ,.@> ,.>,
1, ,.@= ,.AA
11 ,.@, ,.A=
12 ,.;A ,.A1
1; ,.;, ,.@;
2, ,.=A ,.;?
2; ,.== ,.;=
-, ,.-? ,.=>
=, ,.-; ,.=-
;, ,.-1 ,.->
89C
Skills Companion 3
1ou can also use a significance graph to calculate significance.
What do these results mean
?;< significance level ';.,p(
This means that the relationship ma have occurred b chance ; times out of 1,,.
This is a critical level of significance to achieve if the relationship is to be stated as
significant with an acceptable degree of reliabilit.
??< significance level '1.,p(
This means that the relationship ma have occurred b chance onl 1 time out of
1,,. This level if achieved provides a more reliable relationship than the ?;<
level.
89C
Skills Companion 3
??.?< significance level ',.1p(
This level indicates a highl significant relationship. There is onl 1 chance out of
1,,, that the association ma have occurred b chance.
If our e"am Buestion asks about a geographical e"planation ou must go back to
geographical theor. C.g. There is a strong negative relationship between distance
downstream and bedload si/e. The geographical relationship for this is that over
distance downstream erosion occurs.
89C