4.

Measur esofVar iati

on( Dispersi
on),
4.1Mot i
vati
ngexampl es
4.2Obj ecti
vesofmeasur esofvari
at i
on
4.3Measur esofDi spersion(Vari
ation)
4.3.1 RangeandRel ativ
eRange
4.3.2 Var i
ance,St andardDev i
ati
onandCoef f
ici
entofVar
iat
ion
4.3.3 Standar dScor es
4.1Introduction
Considert hefoll
owi ngtwoset sofscores:
Set1:40, 50,60, 60,40, 50
Set2:0, 100,25, 75,80, 20
Boththeseset shav ethesamemean( 50),butt
hesecondseti
salotmor
ewi
del
ydi
sper
sed
("
scatt
er ed")t
hant hefirst.

Set1 Set2

Measureofcent ral t
endencyal onedoesnotadequat el
ydescri
beasetofobser vat
ionunl ess
al
lobservat i
onsar ethesame.Soweneedsomeaddi ti
onali
nformationli
ke
1)Theext entt owhi chtheitemsi napar t
iculardistr
ibut
ionarescattersaroundthecent ral
tendencyi .
e.measur eofdi sper si
on.
2)Thedi rectionofscat t
erednesswhet hermor ei
temsar eatt
achedt owardshigheror
lowerv aluesi .e.measur eofskewness.
3)Theext entt owhi chthedist r
ibut i
onismor epeakedormor ef l
attopedthanthenor mal
distr
ibut i
oni.e.measur eofkur tosi
s.
Definit
ion:
 Thescat t
erorspr eadofi temsofadi stributi
oni sknownasdi spersionorv ari
ation.In
otherwor dst hedegr eetowhi chnumer ical dat
atendt ospreadaboutanav eragev alueis
call
eddi spersionorv ari
ationoft hedata.
 Measur esofdi spersi
onsar est ati
sti
calmeasur eswhi chprovideway sofmeasur ingt he
extentinwhi chdat aaredi sper sedorspr eadout .

4.
2 Obj
ect
ivesofmeasur
ingv
ari
ati
on

 Todet erminet her eli

abil
it
yofanav eragebypointingoutashowfaran
av
eragei srepresent ati
veoft heent ir
edat a.
 Todet er
mi nethenat ureandcauseofv ari
ati
oninordertocont
rol
thev ar
iati
on
it
self.
 Enablecompar isonoft woormor edi stribut
ionwithregar
dtothei
rvari
abili
ty.
 Measur i
ngv ariabil
ityisofgr eatimpor tancet oot
herstati
sti
cal
analy
sis.E.g.
,iti
s
thebasisofst at i
sti
cal quali
tycont r
ol.
Agoodmeasur eofv ari
at i
onposses:
 I
tshouldbeeasyt ocomput eandunder stand.
 I
tshouldbebasedonal lobser v
ations.
 I
tshouldbeUni quel
ydef i
ned

1
  I
tshouldbecapabl eoff ur theral gebr ai
ct r
eat ment .
 I
tshouldbeasl i
ttleasaf fect edbyext r
emev alues.
Absoluteandr elat ivemeasur es
Measuresofdi sper sionmaybeei therabsol uteorr elati
v e
1.Absol utemeasur esofdi sper sion( AMD) :Absol utemeasur eisexpr essedi nt heSI
unitinwhi cht heor igi
nal dat aar egivensuchaski l
ograms, t
oneset c.These
measur esar esui tablef orcompar ingt hev ariabili
tyintwodi str
ibutionshav i
ng
vari
abl esexpr essedi nt hesameuni tsandoft hesameav eragingsi ze.These
measur esar enotsui tabl ef orcompar ingt hev ariabi
lit
yi ntwodi str
ibutionshav ing
vari
abl esexpr essedi ndi fferentuni ts.
2.Relativemeasur esofdi sper si on(RMD) :whenonedesi rest ocompar et hedi sper si
on
intwoset sofdat a, howev ercompar i
ngt het woAMDsmayl eadt of all
aci ousr esult
s.
Itmaybet hatthet wov ar iablesi nvolvedar emeasur edi ndiffer
entuni ts.Forexampl e,
wemaywi sht oknow, foracer t
ainpopul at i
on, whetherser um chol ester ol l
ev els,
measur edi nmi lli
gramsper100ml ,aremor evar i
ablethanbodywei ght, measur ein
kil
ograms.
Furthermor e,althought hesameuni tofmeasur ementi sused, thet woMCT( means)
maybequi tedi fferent.I fwecompar etheAMDofwei ght soff i
rstgr adechi ldrenwi th
theAMDoft hewei ghtsofhi ghschool freshmen, wemayf indt hatt hel atterAMDi s
numer i
callylargert hant hef or mer,becauset hewei ghtst hemsel v
esar el arger ,not
becauset heAMDi slar ger .
Whati sneededi nsi tuat ionl iket hesei sameasur eofrelativevar i
at i
onr athert han
absolut evar i
ation.I tist her at iosofabsol utedi spersiont oanappr opr i
at eav erage
suchasco-ef ficientofSt andar dDev iati
onorCo- eff
icientofMeanDev iation.

4.3 Ty pesofMeasur esofDi sper si

on
Variousmeasur esofdi sper
sionsar einuse.Themostcommonl yusedmeasur esof
dispersionsar e:
Absol utemeasur e Relati
vemeasur es
Range Relati
ver ange
Var i
ance Coef f
icientofv ar
iati
on
St andar ddev i
ation Standar dscores
4.3.1 TheRange( R)
Ther angei st helargestscoremi nusthesmallestscore.I tisaqui ckanddirt
ymeasur eof
vari
abilit
y ,alt
houghwhenat estisgivenbackt ostudentst heyveryoftenwishtoknowt he
rangeofscor es.Becauset her angei sgreat
lyaffectedbyext remescor es,i
tmaygi vea
distor
tedpi ctureofthescores.
Eg.Supposet hatwehav ethedi stri
buti
onofthey i
elds( kgperpl ot)oftwopaddyv ar
ieti
es
from 5pl otseach.Thedi str
ibuti
onmaybeasf oll
ows

Itcanbeseent hatthemeany i
eldforbothv ar
ieti
esis42kgbutcannotsayt hatt
he
performancesoft het wov ari
eti
esaresame.Ther eisgreateruni
for
mityofyiel
dsinthef i
rst
vari
etywher easther eismor evar
iabil
i
tyinthey i
eldsofthesecondv ari
ety
.Thef i
rstvar
iety
maybepr efer
redsi nceitismoreconsistentinyieldperfor
mance.
Formt heabov eexampl eiti
sobviousthatameasur eofcent r
alt
endencyaloneisnot
suffi
cienttodescribeaf requencydist
ribut
ion.Inadditi
ontoi tweshoul
dhav eameasur eof
scatternessofobser vati
ons.

2
Rangeisthedif
fer
enceoft
hemaxi
mum andmi
nimum i
tem.
Iti
sthecr
udesmeasur
esof
var
iat
ion
Forungroupeddat
a:

Forgr
oupeddat
a:

Relat
iveRange(
RR)
I
tisal
sosomet i
mescal
l
edcoef
fi
cientofr
angeandgi
venby
:
Forungr
oupeddat
a:

Forgr
oupeddat
a:

Mer i
tsandDemer it
sofr ange
Mer it
s:
 I tisr i
gidlydefined.
 I tiseasyt ocalculateandsi mpletounderst
and.
Demer it
s:
 I tisnotbasedonal lobservat
ion.
 I tishi ghlyaff
ect edbyext remeobser v
ati
ons.
 I tisaf fectedbyf luctuat
ioninsampl i
ng.
 I tisnotl iabl
et of urt
heralgebrai
ctreatment
.
 I tcannotbecomput edinthecaseofopenenddi st
ri
buti
on.
 I tisv erysensitivetot hesizeofthesample.
Exampl e1: They i
elds( kgperplot)ofacot t
onvari
etyf
rom f
ivepl
otsar
e8,
9,8,
10and11.
Fi
ndt her ange?
Solut
ion
L=11, S=8.
Range=L–S=11-8=3
Example2:
Heightplants Numberof
(i
n) Plant
s
Lesst han 59.5 0
Lesst han 62.5 5
Lesst han 65.5 23
Lesst han 68.5 65
Lesst han 71.5 92
Lesst han 74.5 100
R=74.
5-56.5=18

Meri
tsandDemer
it
sofRange
Meri
ts
1.I
tissi
mpl
etounder
stand.
3
 2.Itiseasyt ocal culat e.
3.Incer taintypesofpr oblemsl i
kequal i
tycont r
ol,weatherforecasts,shar
epri
ceanal
ysi
s,
etc.,
rangei smostwi delyused.
Demer i
ts
1.Itisv erymuchaf fect edbyt heext r
emei t
ems.
2.Itisbasedononl yt woext remeobser v
ations.
3.Itcannotbecal cul atedf rom open- endcl assinterv
als.
4.Itisnotsui t
abl eformat hemat icaltreatment .
5.Itisav eryrarelyusedmeasur e.
2.I
ftherangeandr elativ er angeofaser iesar e4and0. 25r especti
v el
y.Thenwhatist
heval
ue
of:
a)Smal lestobser vation
b)Lar gestobser vation
4.3.2 TheVar i
ance
Popul ati
onVar i
ance
Ifwedi v i
det hev ariationbyt henumberofv al
uesint hepopulation,wegetsomethi
ngcal
led
thepopul at
ionv ar i
ance.Thi sv ariancei sthe"averagesquar eddev i
ati
onf r
om t
hemean".

Sampl eVariance
Onewoul dexpectt hesampl evar
iancet o si
mplybet hepopulat
ionv ari
ancewi t
ht he
populati
onmeanr epl
acedbyt hesamplemean.Howev er,
oneofthemajorusesofst at
ist
ics
i
st o estimat ethe correspondi
ng parameter.Thi
sf ormul
a has the probl
em thatt he
est
imat edvalueisn'
tthesameast hepar ameter
.Tocounter
actt
his,t
hesum oft hesquares
ofthedev i
ationsisdi
videdbyonelessthanthesamplesize.

I
.e.Thesampl
evar
iance,
denot ofasetofnobser
edby , vedv
alueshav
ingamean i
s
thesum oft hesquar eddev i
ati
onsdi v
idedby .
Thefollowingstepsar eusedt ocalculatethesampl evar
iance:
1.Findt heari
thmet icmean.
2.Findt hediff
erencebet weeneachobser vati
onandthemean.
3.Squar ethesedi ff
erences.
4.Sum t hesquar eddiffer
ences.
5.Sincet hedatai sasampl e,div
idet henumber( fr
om step4abov
e)bythenumberof
obser vat
ionsmi nusone, i
.e.,
n-1(wher eni sequalt
othenumberofobser
vati
onsi
nthe
dat aset)
.
Forthecaseoff r
equencydi str
ibuti
oni tisexpressedas:

Weusual
l
yuset
hef
oll
owi
ngshor
tcutf
ormul
a.

4
 StandardDevi
ati
on
Thereisaprobl
em withvari
ances.Recal
lthatt
hedevi
ati
onswer
esquar
ed.Thatmeanst
hat
theunit
swerealsosquared.Togettheunitsbackt
hesameastheor
igi
naldat
aval
ues,t
he
squarerootmustbetaken.

Exampl
es:

Exampl
e

5
Exampl
e

6
Coef f
icientofVariati
on( C.V)
TheSt andarddev iat
ionisanabsol utemeasur eofdispersion.I
tisexpressedi nter
msofuni tsi
n
whicht heor igi
nalfi
guresar ecollectedandst at
ed.Thest andarddeviati
onofhei ghtsofpl ants
cannotbecompar edwitht hestandar ddevi
at i
onofwei ghtsofthegrains,asbot hareexpressed
i
ndi fferentunits,i.
ehei ghtsincent imeterandwei ghtsinkilogr
ams.Ther efor
et hestandard
deviati
onmustbeconv ertedint
oar el
ativ
emeasur eofdispersi
onforthepurposeofcompar ison.
Ther elati
vemeasur eisknownast hecoef fi
cientofvari
ation.
•Isdefinedast heratioofstandar ddevi
ationtothemeanusual lyexpressedasper cents.

•Thedistr
ibut
ionhav i
ngl essC.Vissaidtobelessv
ariabl
eormor econsi
stent
.
Examples:Consi derthemeasur ementonyiel
dandplantheightofapaddyv ar
iet
y.The
meanandst andar ddeviati
onforyiel
dare50kgand10kgr especti
vel
y.Themeanand
standar
ddev i
ati
onf orplantheightar
e55am and5cm r espectiv
ely.
Herethemeasur ement sforyi
eldandplanthei
ghtarei
ndifferentunit
s.Hencethe

7
 v
ari
abi
l
iti
escanbecompar
edonl
ybyusi
ngcoef
fi
cientofv
ari
ati
on.

.
2.Ameteorologisti
nter
estedi
ntheconsist
encyoftemperat
uresi
nthreeci
ti
esdur
ingagi
venweek
col
l
ect edthef ol
l
owingdata.Thet
emper atur
esf
orthefiv
edaysoftheweekint
hethr
eecit
ies
were
Ci t
y1 25 24 23 26 17
Cit
y2 22 21 24 22 20
Cit
y3 32 27 35 24 28
Whi chcityhavethemostconsi
stenttemper
atur
e,basedonthesedata?
(Exercise)

4.
3.3 St
andar
dScor
es(
Z-scor
es)

 IfXisameasur
ementf
rom adi
str
ibut
ionwi
thmeanXandst
andar
ddev
iat
ionS,
then
i
tsval
uei
nstandar
duni
tsi
s

 Zgi vesthedevi
ationsf
rom themeani nuni
tsofstandarddev i
ati
on.
 Zgi vesthenumberofstandarddeviat
ionaparti
cularobservati
onli
eaboveorbel
ow
t
hemean.
 I ti
susedtocompar etwoobser v
ati
onscomi ngfr
om differ
entgroups.
Examples:
1.Twosect i
onsweregivenint
roducti
ontostati
sti
csexami nat
ions.Thefol
lowi
ng
i
nformati
onwasgiven.
Value Secti
on1 Secti
on2
Mean 78 90
StudentA fr
om sect
ion1
Stand.dev
iat
ion 6 5
scored90and st
udentB
fr
om secti
on2scored95.Rel
ati
vel
yspeaki
ngwhoper
for
medbet
ter
?
Soluti
ons:
Calcul
atethest
andar
dscoreofbot
hstudents.

 StudentAper
for
medbetterr
elat
ivetohi
ssecti
onbecausethescor
eofstudentAist
wo
standar
ddevi
ati
onabovethemeanscoreofhissecti
onwhile,t
hescor
eofstudentBis
onlyonest
andar
ddevi
ati
onabovet hemeanscoreofhi
ssecti
on.

8
2.Twogr oupsofpeopl eweret
r ai
nedt oper f
orm acertai
ntaskandt estedtof
indoutwhich
gr
oupi sfastert ol
earnthetask.Forthet wogr oupsthefoll
owi nginformat
ionwasgiv
en:
Value Gr oupone Gr oupt wo
Mean 10. 4min 11. 9mi n
Stand. dev
. 1.2mi n 1.3mi n
Relat
ivel
yspeaki ng:
a)Whi chgr oupismoreconsistentini tsperf
ormance
b)Supposeaper sonAf r
om gr ouponet ake9.2minuteswhi lepersonBfrom Gr
oup
twot ake9. 3minut
es,whowasf ast erinperf
ormingthet ask?Why ?

Sol
uti
ons:
a)Usecoef
fi
ci
entofv
ari
ati
on.

Si
nceC.
V2<C.
V1,
group2i
smor
econsi
stent
.

b)Cal
cul
atet
hest
andar
dscor
eofAandB

 Chil
dBisfast
erbecausethet i
metakenbychildBistwost
andarddev
iat
ionshor
terthan
theaver
ageti
met akenbygr oup2while,theti
metakenbychi
ldAisonlyonestandard
devi
ati
onshor
tert
hant heaveraget
imet akenbygr
oup1.