CORRELATI

ONANDREGRESSI
ONANALYSI
S
BIVARIATEDISTRI ON,
BUTI CORRELATI
ON:
Sof arwehav econfinedour selv
est ouni vari
atedist
ri
buti
ons,i.e.
,thedi st
ri
butions
i
nv ol
v i
ngonl yonev ar
iabl
e.Wemay ,howev er,comeacrosscertai
nser i
eswher eeach
i
tem oft heseri
esmayassumet hevaluesoft woormor evari
ables.Forexampl e,i
fwe
measur etheheightsandwei ghtsofacer t
aingr oupofpersons,weshel lgetwhati s
knownasBi vari
ateDistr
ibuti
on-onev ar
iabler el
ati
ngtoheightandanot hervari
able
relat
ingt oweight.

Cor
rel
ati
onAnal
ysi
s:

Inabivari
at edistr
ibuti
onwemayi nter
estedtofindoutifther
ei sanycorrel
ati
onorco-
vari
ati
onbet weent het wov ari
abl
esunderst udy.Ifthechangeinonev ar
iableaffect
sa
changeint heotherv ar
iabl
e, t
hev ari
ablesar
esaidt obecorrel
ated.
Correl
ati
onAnal y
sisisagr oupofst ati
sti
calt
echniquesusedtomeasur ethestrengthof
theassociationbetweent wov ari
ables.

AScatt
erDiagr
am i
sachartthatportr
aystherel
ationshi
pbetweenthetwovari
ables.
TheDependentVar
iabl
eist
hev ar
iablebei
ngpredictedorest
imated.
TheI
ndependentVari
abl
eprovi
dest hebasi
sforest i
mati
on.Iti
sthepredi
ctorvar
iabl
e.

TypesofCor rel
ation
Correlat
ioncanbecl assif
iedinvar
iousway s.
Posit
iveandNegat ivecor r
elati
on–Posi t
ivecorr
elat
ioni sanassoci
ati
onwherei
ncr
easein
onev ariabl
er esul
tsi nt
oani ncr
easeint heotherv ariabl
ewhil
einnegati
vecorr
elat
ion,
i
ncreasei nonev ar
iableresult
sintosi
multaneousdecr easeint
heother
.

Simple,par
ti
alandmult
ipl
ecor
relat
ions–Si
mpl eistherelat
ionshipbetweenonl
ytwo
vari
ableswhil
emult
ipl
eisther
elati
onshi
pbetweenmor ethant wov ari
abl
es.Par
ti
al
correl
ati
on,istherelati
onshi
pinwhichmor
e
than2

Cor
rel
ati
on&Regr
essi
onAnal
ysi
s
Howwi l
lyoudetermi netherel
ationship?
Correl
ation analysis det
ermines the relati
on bet
ween two quantit
ies known as
vari
ables,‘x’and‘ y
.’Correl
ati
oni sobser vedwhen,atthetimeoft hestudy ,auni
t
changei nxisretali
atedbyanequi valentchangeiny.I
fanincreaseinxresult
sinthe
corr
espondi ngincreasei ny(andvi cever sa),t
heyareconsideredtobeposi ti
vel
y
corr
elated.Ifanincreaseinxr esul
tsinadecr easei
ny( andvi
ceversa),i
t’
sacaseof
negati
vecor rel
ati
on.

Thefoll
owi
ngexampleswil
lil
lustrat
eacoupl
eofmet
hodst
hatar
ecommonl
yused
forf
indi
ngt
hecorr
elati
oncoef
f i
cient
.

Ascat t
erdiagram i
salsoknownasacor relati
onchar torascat terpl
ot.Below i
sa
si
mpl eexampl e.Thegoali st of i
ndanyr elat
ionbet weent heamountofr evenue
generatedbycustomerandt hei
rsatisfact
ionwi thacompany .Eachdotrepresentsa
customer.Thehor i
zontalandver t
icalpositi
onsi ndicat
et heamountofmoneyspent
byeachcust omer(x)andt helevelofcustomersat i
sfact
ion(y)respect
ivel
y.

Scat
terdi
agr
am ofcust
omersat
isf
act
ionv
srev
enue.

Theextenttowhicht
hedotsli
eonastr
aightl
ineindicat
esthestr
engthofthe
rel
ati
on.Thefi
gur
eaboveshowsa‘
moder
atecorrel
ati
on.’Thedat
apoint
sforma

2
l
ineardi
str
ibuti
onpatter
nsoy oucanassumet hatthecust
omer s’sat
isfact
ionand
r
evenuesaresomehowr elat
ed.Werethedotsspr
eadrandomly
,onecoul dsurelysay
t
hatnor el
ationexi
sted.Ifastr
aightl
inecanbedrawnbyf oll
owingthepat tern,i
t
i
mpl i
esastrongcor
relat
ion.

Fi
gureA Fi
gureB
Posi
ti
veLi
nearr
elat
ionshi
p Negati
veLi
nearr
elat
ionshi
p

Fi
gureC Fi
gureD
Norelat
ionshi
p Posi
ti
veCur
vil
i
nearr
elat
ionshi
p

Fi
gureE
Negati
veCur
vil
i
nearr
elat
ionshi
p

2
ThePearsonproduct
-momentcorrelat
ioncoef
fici
enti
sanothermethod.Itmaybe
thebestwaytomeasuretheassociati
onbetweenconti
nuousvari
ables.I
tstat
es
boththestr
engthoft
heassociat
ionandt hedi
recti
onoftherel
ati
onship.

KarlPearson’sCor r
elat
ionCoef fi
cient
:Asameasureofi nt
ensi
tyordegreeoflinear
rel
ati
onshipbet weent wov ar
iabl
es,KarlPear
son(
1867-1936)devel
opedaf or
mul a
cal
led correl
ati
on Coefficient
.Wecal cul
atet
hecoef
fi
cientofcorr
elat
ion f
rom t he
fol
lowi
ngf ormulas
Cov( xy) σxy
r= =
Var(x)
×Var( ) σx×σy
y

Her
e,σxy= {
1
n
x-x)
∑(
̅ ̅
(y-y),
2
}
1
n
̅2
x-x)
andσx= ∑( { }
∑ (x-x)(y-y)
̅ ̅
r=
∑(x-x)×∑(y-y)
2 2
̅ ̅

̅ ̅
r=
∑xy-n×x×y
2 2
̅ ̅
∑x-n×x )×(∑y-n×y )
(
2 2

Li
mit
sforCor
rel
ati
onCoef
fi
cient
:
Thenumer i
calv
alueoft
hePear
soncor
rel
ati
oncoef
fi
cient(
r)r
angesf
rom
+1to-1:i
.e.
,-1≤r≤+1

r>0i
ndi
cat
esaposi
ti
veli
nearr elati
onshipbet
weent
het
wov
ari
abl
es
r<0i
ndi
cat
esanegat
ivel
inearr elat
ionshi
p
r=0i
ndi
cat
est
hatnoli
nearrelationshipexi
sts

I
fthev alueisnear±1, aperfectcorr
elati
oni sobserv
ed:whenx
i
ncreases, ytendstoincreaseordecreaseaswel l,orvi
ceversa.
I
ftherv alueliesbetween±0. 70andlesst han±1, t
hecor r
elat
ionis
consideredtobest rong.
Val
uesl yingbetween±0. 30andl essthan±0. 70mar kamoder at
e
correlat
ion.
Whenri sbel ow±0. 30,
thereisapoororweakcor rel
ation.

3
Positi
veCorrelati
on:Ifthetwovari
ablesdevi
atei
nsamedi r
ect
ion
Negat i
veCorrelati
on:Ifthetwovar
iablesdev
iat
einopposi
tedi
rect
ion
PerfectCorr
elat i
on:I
ft hedevi
ati
oninonev ari
abl
eisfoll
owedbyacorr
espondi
ngand
proporti
onaldev i
ati
oni ntheot
her.

Example:
DanIreland,thestudentbodypr esi
dentatTol edoSt ateUni
versi
ty,i
sconcernedabout
thecostt ost udentsoft ext
books.Hebel ievest hereisar elat
ionshi
pbet weenthe
numberofpagesi nthetextandtheselli
ngpriceoft hebook.Topr ovi
deinsi
ghtint
othe
probl
em hesel ectsasampl eofeightt
extbookscur rent
lyonsal
ei nthebookstor
e.Draw
ascatterdiagram.Comput ethecorrel
ati
oncoef fi
cient.
Book Page Price($)
I
ntotoHistory 500 84
BasicAl
gebra 700 75
I
ntotoPsy c 800 99
I
ntotoSociology 600 72
Bus.Mgt 400 69
I
ntrotoBiol
ogy 500 81
Fund.ofJazz 600 63
Pri
nc.ofNursing 800 93

Sol
uti
on:

[
Todr aw ascat t
erdi agram,youhavetopl otthev alueoft hevari
abl
eyagai nsteach
valueofx.Her eno.ofpagespl ott
edinhorizontalaxisandt hesell
i
ngpriceofbooksare
plott
edinv ert
icalaxis.Inthescatt
erdi
agram itisshownt hatthesel
li
ngpriceofbooks
arelesserforlessno.ofpageofbooksandhi gherforl ar
gerno.ofpagesofbooks.This
meanst hechangeofpr i
ceofbooksandtheno.ofpagesofbooksar einsamedirect
ion.
So, t
hereisaposi t
iver el
ati
onbetweenboth.]

4
Tabl
efort
hecal
cul
ati
onofcor
rel
ati
oncoef
fi
cient
:
Book Page Pri
ce(
$)
X Y XY X2 Y2
I
ntotoHistory 500 84 42,
000 250,
000 7,
056
BasicAl
gebra 700 75 52,
500 490,
000 5,
625
I
ntotoPsy c 800 99 79,
200 640,
000 9,
801
I
ntotoSociology 600 72 43,
200 360,
000 5,
184
Bus.Mgt 400 69 27,
600 160,
000 4,
761
I
ntrotoBiol
ogy 500 81 40,
500 250,
000 6,
561
Fund.ofJazz 600 63 37,
800 360,
000 3,
969
Pri
nc.ofNursing 800 93 74,
400 640,
000 8,
649
Total 4,
900 636 397,
200 3,
150,
000 51,606
n(
ΣXY)
-(
ΣX)
(ΣY)
r=
[ ΣX)] n(
[ ΣY) ]
2 2 2
n(ΣX) -
( (
-ΣY)
2

8(
397,
200)
-(
4,900)
(636)
=
[
8(3,
150,
000-
( 900)]
4,
2
[
8(51,
606)
-636)]
(
2

=0.614
Thecorr
elat
ionbet
weenthenumberofpagesandthesell
i
ngpr
iceoft
hebooki
s0.
614.
Thi
sindi
catesamoderat
eassoci
ati
onbetweenthevar
iabl
e.

UsesofCorrel
ationAnal y
sis
 Iti
susedt ogivethesizeanddirecti
onofassociati
onbetweenv ariabl
es
 Iti
susedt ominimizetherangeofuncer tai
ntyi
nforecasti
ng
 Iti
susedt opresenttheaveragerel
ationshi
pbetweenanyt wov ari
ablesthr
ought
he
coeff
ici
ent.
 Iti
susedf ordeci
sionmakingi nthefi
eldofscienceandphilosophy
 Inthefi
eldofnature,i
tisusedinobser vi
ngthemultipl
ici
tyoftheinter
-r
elat
ed

REGRESSI
ONANALYSI
S
REGRESSI
ON:Thet ermr egressionl i
ter
all
ymeans“ steppingbackt owardst heav er
age”.I
t
wasf i
rstusedbyBr it
ishBi omet r
ici
an,Sir,Franchi
sGal ton( 1822-1911).
Regressionanalysisisamat hemat i
calmeasur eoft heav eragerelati
onshipbet ween
twoormor ev ar
iablesinter msoft heor i
ginalunit
soft hedat a.
I
nr egressi
onanal ysisther ear etwot y
pesofv ariables.Thev ari
ablewhosev al
ueis
i
nfluencedori st obepr edictediscal l
eddependentv ar i
ableandt hev ari
ablewhich
i
nfluencesthevaluesori st ousef orpredicti
oniscal l
edi ndependentv ar
iables.

5
**Inregr essi onanal y
sisi ndependentv ariablei salsoknownasr egressor,orpredictor
,
orexpl anat orywhi lethedependentv ariablei salsoknownasr egressedorexpl ained
variable.
At echniquef orfi
ttingast raightl i
net hroughasetofpoi ntsinsuchawayt hatthesum
oft hesquar edv erti
caldi st ancesf r
om t henpoi ntstothel i
nei smi ni
mizedi st he
met hodofl eastsquar es.Inr egressionanal y
sisweuset heindependentv ari
able(X)t o
estimat et hedependentv ariable( Y).
̂2
Thel eastsquar escr it
erioni susedt odet erminet heequati
on.Thati st heter
m ∑(Y- Y)
i
smi nimi zed.
̂ ̂
Ther egr essi onequat i
on: Y =a+bX, wher eY i stheaveragepr edict
edv al
ueofYf or
anyX.
ai stheY- i
nter
cept .Itist heest i
mat edYv alue whenX=0
bi sthesl opeoft heline, ort heav eragechangei nYforeachchangeofoneuni tinX
thel eastsquar espr i
ncipl eisusedt oobt ainaandb.
Thel eastsquar espr i
nciplei susedt oobt ainaandb.Theequat ionst odetermineaand
bar e:
Thecoef ficientofr egressi onofyonx( i.
e., ydependonx)i s
Cov( xy) σxy
byx= =
Var (x) σ2 x

Her
e,σxy=
1
n
{ ̅ ̅
x-x)
∑( (y-y),
2 1
n
} ̅2
x-x)
andσx= ∑( { }

or
,byx=
( )( )
̅ ̅
∑x-x y-y
,byx
Or
SP(xy) Sxy Sum oft
= =
heproduct(xy
)
( )
∑ x
x
̅2
-
x) S2
SS( x
Sum ofsquares(x)

̅ ̅
,b =
or y
x
∑xy-n×x×y 2
̅
∑x-n×x )
(
2

̅ ̅
a= y-
byx×x

Dist
inct
ionbet
weenCor relat
ionandRegression:
Correlation Reressi
on
Meaning Cor relation i s a st at
isti
cal Regression descri
bes how an
measur ewhi chdet er
minesaco- i ndependent variable is
rel
ationshi porassociati
onoft wo numer i
call
y r el
ated t o a
vari
abl es. dependentvari
bale.
Main Correlation analy
sis l
ets Regression analysis hel
ps
purpose experiment er know the determine a functional

6
 associ at i
onort heabsenceoft he relati
onshi p bet ween t
eo
relationshi p bet ween two var iabl es so as t o est i
mat e
variabl es. unknownvar iabl ewi tht hehel pof
known var iabl e(s) and make
futurepr ojectiononevent s.
Obj ecti
ve To f ind a numer i
aclval ue t hat To est imat e t he val ue of a
expr esses t
he rel
ationshi p random var iableont hebasi sof
bet weent hevar iables. theval ueofaf ixedvar iabl e.
Usage Repr esent sthel i
nearr el
ationshi p Fitst hebestl ineandest i
mat es
bet weent wovar iabl es. one var i
abl e on t he basi s of
anot hervar i
abl e.
Nat ure of Thevar i
abl esar enotdesi gnat ed Concept of dependent and
the asdependentori ndependent . independentvar i
abl eismat terof
variables fact.Her eonedependentvar iable
is expl ained by one or mor e
independentvar iable.
indication Cor relation coef f i
cienti ndicat es Regr essi on coef f i
cienti ndicat es
theext entt owhi cht wovar iables thei mpactofoneuni tchangei n
movet oget her. theknownvar iabl e(Independent )
on t he est imat ed ( dependent )
variabl e.
Range Itrangesf r
om -1. 00t o+1. 00 In r egressi on anal ysis t he
coef fi
ci ent can t ake any r eal
value,
i.
e.,-∞<b<∞
Nat ure of I tissy mmet r
ical.i.e.,r
xy=r yx Itisnotsy mmet rical,i.e.
, byx≠bxy
the
coef fi
cient
Effect of Cor relation coeffient i
s Regr essi on coeffient i
s
shifti
ng independentofbot h ori
gin and i ndependent of or i
gin but
scal e and scal eoft hemeasur ement s depends on t he scal e of t he
origin measur ement

Examples:
Devel
opar egressionequati
onfortheinf
or mationgiveninprevi
ousexamplethatcanbe
usedtoestimatetheselli
ngpri
cebasedont henumberofpages.
8(
397,200) -
(4,
900)(636)
b= 2 =.05143
8(3,
150,000)-(4,
900)
636 4,900
a= -0.05143 =48.0
8 8
̂
Theregressi
onequat i
onis: Y=48. 0+. 05143X
TheequationcrossestheY-axi
sat$48.Abookwi t
hnopageswoul dcost$48.
Theslopeofthelineis.05143.Eachadditionpagecost saboutanickel
.

7
[
Thesignofthebval
ueandthesignofrwil
lal
waysbethesame.
]
Wecanuset her
egressi
onequati
ontoestimat
evaluesofY. Theest
imat
edsel
l
ing
pr
iceofan800pagebookis$89.
14,foundby:

PROBLEMSONCORRELATI ONANDREGRESSI ON:

1.Fr om t he f ollowi ng dat a,cal culate coef f i
cientofcor r
elati
on bet ween the
percentagey ieldonsecur iti
esandwhol esal epr iceindicesf orcertainyears:
Year 1982 1983 1984 1985 1986 1987 1988
%y i
eldonsecur i
ties 5.0 5.1 5.2 4.9 4.8 5.
3 5.
4
I
ndex no.of whol esal e 140 138 126 132 140 135 132
pr i
ces
Alsocal culatet het wor egressionl i
nes.Est imat eper centagey ieldonsecur i
ti
es
wheni ndexno.ofwhol esalepr icesi s150.Al soest imat eindexno.ofwhol esal
e
pri
ceswheny ieldonsecur i
ti
esi s6. 0.
Solution:
[
Hi nts:Establisht wor egr essi onlines; y=a1+b1x, andx=a2+b2y
̅ ̅
∑xy -n× x× y ̅ ̅
Her e,b1orbyx= 2 anda = y - byx× x
2 ̅ 1
∑x- n×x
̅ ̅
∑xy n× x× y
- ̅ ̅
Simi l
arly,b2orbxy= 2 anda2= x- byxy×y
2 ̅
∑y- n×y
Afterput ti
ngt heval ueofa’ sandb’ sy ouwi llobt ai ntwoequat i
ons.
NowYouhavet oest i
mat et heper cent agey ieldonsecur i
tieswheni ndexnumberof
whol esalepr i
cei sgi ven.Her ei tist hequest ionwhi chequat ionwi llbeusedt o
estimat et his.Consi dert hevar iables,whi chonehavet obet akenasi ndependent
andwhi choneasdependent .Her e,y ouar egi vent heval ueofwhol esalepriceindex
andy ouhavet opr edi ctt heval ueofper cent agey ieldonsecur it
ies.Sincet hevalue
ofwhol esalepr icei ndexi sgi ven,so,y ouhavet ot akeper cent agey i
eldonsecur i
ti
es
asdependentvar i
abl e.I fper cent agey i
el donsecur itiesismar kedast hevar i
bleY,
youhavet ouset heequat ionyonx, y=a1+b1x.
Nextonei sr ever seoft hisr equi rement .]

2.Foll
owing datar el
ates to adver
ti
sing expenditur
e( i
nl akh t
aka)and t
hei
r
corr
espondingsales(i
ncroresoftaka):
Adv er
ti
sing 10 12 15 23 20
Expendit
ure
Sales 14 17 13 25 21
Esti
mat ei)thesalescorrespondi
ngt oadvert
isingexpendit
ureoftaka30l
akh
and

8
 ii
)t headver
ti
singexpendit
ureforasalestar
getoftk.35cror
es.
[Soluti
on:SameasEx#1. ]
3.From thef oll
owingdat
acomput ecoef f
ici
entofcorr
elat
ionbetweenXandY.
X-Ser
ies Y–Seri
es
No.ofi t
ems 15 15
Average 25 18
Sum of squar es of deviati
on from 136 138
mean
Sum oftheproductofdev i
ati
onofXandY- seri
esfr
om thei
rrespecti
veAMs’i
s122
[
Solut
ion:Hints:Hereyouaregiven,n=15,
̅ ̅
x=25,y=18, ( )
̅2
∑x-x =136,

∑(
y
̅2 ( )( )
̅ ̅
-y)=138,and∑x-x y
-y =122

Usef
oll
owi
ngf
ormul
atodet
ermi
net
hev
alueofr

r=
( )( )
̅ ̅
∑x-x y-y
]
( ) ( )
2
̅ ̅2
∑x-x ×∑y-y

4.Fi ndoutt heregressionequationshowingt her egressi

onofcapaci t
yutil
izat
ion
onpr oducti
onfrom thef ol
l
owingdata:
Aver
age SD
Producti
on( i
nLakhuni ts) 35. 6 10.
5
Capacityuti
li
zati
on( i
n%) 84. 8 8.
5
r=0.62
Esti
mat etheproductionwhent hecapaci
tyut i
li
zationis70per cent.
[
Solution:
Hints.Hereyouar egiven,
̅ ̅
n=15, x=35. 6,y=84. 8,σx=10.5,
σy=8. 5,andr=0. 62
Youhavet oesti
mat eproductionwhencapacityuti
lizati
oni sgi
v enas70per cent.Todo
i
tyouhav etodev el
opregressionequati
onofpr oductiononcapaci tyutil
i
zation,
i.
e.,you
hav
etodev eloptheequat i
on,x=a2+b2y .

Usethef or
mula
Cov(xy) σxy
byx= = , t
hev
alueofσxywi
l
lfoundbyusi
ngt
hev
alueofr
.]
Varx) σ2
( x

5.Coeffi
ci
entofcorrel
ati
onbetweentwo var
iablesX andY is0.
32.Thei
rCo-
var
ianceis7.
86.Thevar
ianceofXi
s10.Fi
ndt heSDofYser
ies.

6.TheGener
alManagerofKi
ranEnt
erpr
ises-anent
erpr
isedeal
i
ngi
nthesal
esof

9
 readymademen’ swear s–i stoyingwiththei deaofincr easinghi ssalest oTk.
80,000.Onchecki ngt her ecordsofsal esdur i
ngthel ast10y ear s,itwasf ound
thattheannualsal esproceedsandadv erti
singexpenditurewashi ghlycorrel
ated
tot heextentof0.8.Itwasf urthernot
edt hattheannualav er
agesal eshavebeen
Tk.45, 000andannualav erageexpendi tur
eTk.30, 000,wi thav arianceof1600
and626i nannualaveragesal esandannual averageexpendi turer especti
vely.
Int hev i
ew oftheabov epi cture,how muchexpendi t
ur eonadv erti
sementy ou
woul dsuggestthegener alsal esmanageroft heenterprisetoi ncurt omeethi s
targetofsales.
7.For10obser vati
onsonpr ice( X)andsuppl y(Y)oft hef oll
owi ngdat awer e
obtained.
∑X=130,∑Y=220,∑X =2288,∑Y=5506and∑XY=3467
2 2

Obtai
ntheli
neofregr
essi
onofYonXandXonY,andest
imat
ethesuppl
ywhen
pr
iceis16uni
ts.

8.Fi
ndCoeff
ici
entofCorr
elat
ionforthedistr
ibuti
oninwhi
chSDofXis3.
0uni
ts,
SDofYis1.4uni
tsandthecoef
fici
entofregressi
onofYonXi
s0.
28.

10