PublishedonStatistics(https://onlinecourses.science.psu.

edu/statprogram)
Home>4.0ChiSquareTests

4.0ChiSquareTests
ChiSquareTestofIndependence
Doyourememberhowtotesttheindependenceoftwocategoricalvariables?ThistestisperformedbyusingaChisquare
testofindependence.
Recallthatwecansummarizetwocategoricalvariableswithinatwowaytable,alsocalledarccontingencytable,wherer
=numberofrows,c=numberofcolumns.OurquestionofinterestisArethetwovariablesindependent?Thisquestionis
setupusingthefollowinghypothesisstatements:
NullHypothesis:Thetwocategoricalvariablesareindependent.
AlternativeHypothesis:Thetwocategoricalvariablesaredependent.
Thechisquareteststatisticiscalculatedbyusingtheformula:

= (O E) /E

whereOrepresentstheobservedfrequency.Eistheexpectedfrequencyunderthenullhypothesisandcomputedby:

row total column total

E =
sample size

Wewillcomparethevalueoftheteststatistictothecriticalvalueof2 withdegreeoffreedom=(r1)(c1),andrejectthe
nullhypothesisif2 > 2 .

Example
Isgenderindependentofeducationlevel?Arandomsampleof395peopleweresurveyedandeachpersonwasaskedto
reportthehighesteducationleveltheyobtained.Thedatathatresultedfromthesurveyissummarizedinthefollowingtable:

HighSchool

Bachelors

Masters

Ph.d.

Total

Female

60

54

46

41

201

Male

40

44

53

57

194

100

98

99

98

395

Total

Question:Aregenderandeducationleveldependentat5%levelofsignificance?Inotherwords,giventhedatacollected
above,istherearelationshipbetweenthegenderofanindividualandthelevelofeducationthattheyhaveobtained?
Here'sthetableofexpectedcounts:

HighSchool

Bachelors

Masters

Ph.d.

Total

Female

50.886

49.868

50.377

49.868

201

Male

49.114

48.132

48.623

48.132

194

100

98

99

98

395

Total
So,workingthisout,2

= (60 50.886) /50.886 + + (57 48.132) /48.132 = 8.006

Thecriticalvalueof2 with3degreeoffreedomis7.815.Since8.006>7.815,thereforewerejectthenullhypothesisand

concludethattheeducationleveldependsongenderata5%levelofsignificance.
UsingMinitab
WecanenterthedataintoMinitabandrequestthatthe'Chisquaretest'beconductedfortheabovehypotheses.The
Minitaboutputforthisexampleisshownbelow:

TheChisquaretestofindependencevaluethatMinitabcalculatedis8.006,whichisthesameaswecalculatedabove.
TheChisquaretestforindependenceisanimportantmethodfordeterminingifthereisarelationshipbetweenvariables
wherethechancethatsomethingfallsintoaparticularcategorydependsonwhetherthevariablefallsintoanothercategory
comesintoplay.Thisrelationshipofindependence/dependenceisimportanttobeabletounderstandanduse.

ChiSquareGoodnessofFitTests
Doyourememberhowtousethechisquaregoodnessoffittesttotestwhetherrandomcategoricalvariablesfollowa
particularprobabilitydistribution?Let'stakealookatanexample:

Example
SupposethePennStatestudentpopulationis20%PAresidentand80%nonPAresident.Then,ifasampleof100students
yields16PAresidentand84nonPAresident,how'good'dothedata'fit'theassumedprobabilitymodelof20%PAresident
and80%nonPAresident?

Wecanusethechisquaregoodnessoffitstatistictotestthe
hypothesesstatements:
NullHypothesis:Pr

= 0.2

AlternativeHypothesis:Pr

0.2

Workingthisoutweget,
2

(16 20)
=

(84 80)
+ +

20

= 1
80

Thecriticalvalueof2 with1degreeoffreedomis3.84.Since1<
3.84,wecannotrejectthenullhypothesis.Thereisnotenough
evidencetoconcludethatthedatadon'tfittheassumedprobability
modelat5%levelofsignificance.Inotherwords,thestudentsthatwere
randomlyselectedinthisexampledidresembletheprobabilitydistributionthatwasspecified.
SourceURL:https://onlinecourses.science.psu.edu/statprogram/node/158