Upload
luis-valens
View
214
Download
0
Embed Size (px)
Citation preview
8/12/2019 Hypotesis Test
1/10
Hypothesis Test
Setting up and testing hypotheses is an essential part of statisticalinference. In order to formulate such a test, usually some theory has beenput forward, either because it is believed to be true or because it is to be
used as a basis for argument, but has not been proved, for example,claiming that a new drug is better than the current drug for treatment ofthe same symptoms.
In each problem considered, the question of interest is simplified into twocompeting claims / hypotheses between which we have a choice; the nullhypothesis, denoted H0, against the alternative hypothesis, denoted H1.These two competing claims / hypotheses are not however treated on anequal basis: special consideration is given to the null hypothesis.
We have two common situations:
1. The experiment has been carried out in an attempt to disprove orreject a particular hypothesis, the null hypothesis, thus we give thatone priority so it cannot be rejected unless the evidence against it issufficiently strong. For example,H0: there is no difference in taste between coke and diet cokeagainstH1: there is a difference.
2. If one of the two hypotheses is 'simpler' we give it priority so that amore 'complicated' theory is not adopted unless there is sufficientevidence against the simpler one. For example, it is 'simpler' toclaim that there is no difference in flavour between coke and dietcoke than it is to say that there is a difference.
The hypotheses are often statements about population parameters likeexpected value and variance; for example H0might be that the expected
value of the height of ten year old boys in the Scottish population is notdifferent from that of ten year old girls. A hypothesis might also be astatement about the distributional form of a characteristic of interest, forexample that the height of ten year old boys is normally distributed withinthe Scottish population.
8/12/2019 Hypotesis Test
2/10
8/12/2019 Hypotesis Test
3/10
Thetermnot r
If wenullevidsugg
A sidistri
Exa1. H0
2. H0
See
A copopu
Exa1. X
2. X
See
H1: the n
inal conclof the nject H0".
concludeypothesince agaiests that
imple H
ple hypobution co
ples: X ~ Bi(1
: X ~ N(5,
lso com
omposi
posite hlation dist
plesBi(100,
N(0, )
lso simpl
ype I Er
ew drug i
usion oncll hypoth
e never
"Do not ris true, itst H0in fhe altern
pothesi
hesis is apletely.
0,1/2), i.
20), i.e.
osite hyp
e Hypot
pothesisribution c
) and H1:
and H1:
e hypoth
or
better th
e the testsis. Weconclude
ject H0",only sugvour oftive hypo
hypothe
. p is spe
and a
othesis.
esis
is a hypompletely
p > 0.5
unspeci
sis.
an the cu
has beenither "Rej"Reject
this doesests that
1. Rejectithesis ma
is which
cified
e specifi
hesis whi.
fied
rrent drug
carried oect H0in1", or eve
not necethere is ng the nulybe true.
pecifies t
d
ch does
, on aver
ut is alwaavour ofn "Accept
sarily meot sufficiel hypothe
he popul
ot specif
ge.
ys given i1" or "DoH1".
an that thntis then,
tion
the
n
e
8/12/2019 Hypotesis Test
4/10
In arejec
Forthat t
A typdiffeThetest:
A typimpothererejecprob
The
If weas thhypo
Forthe s
A typ
In anot r
ypothesited when
xample, ihe new d
H0: there I errorent effectollowing t
Truth
e I error irtant to afore adjuting the nbility of aP(type I
xact pro
do not ree samplethesis (es
ny givenmaller th
e I error
ype II Eypothesijected w
test, a tit is in fac
n a clinicaug is no
is no diffould occ
s when inable give
Rejec
0 Type I
1 Right d
often cooid, thanted so thll hypoth
type I ererror) = si
ability of
ect the nmay notpecially if
set of datrisk of o
an also b
rortest, a ten it is in
pe I errort true; tha
l trial of aetter, on
rence ber if we cfact therea summ
Decisio
t H0 Do
rror Ri
cision Ty
nsidereda type IIt there issis wron
or can begnificanc
a type II e
ll hypothe big enothe truth i
, type I ae, the hig
e referred
pe II errofact false
occurs wt is, H0is
new drugaverage,
tween thncluded twas nory of pos
n't reject
ht decision
pe II Error
o be morrror. Thea guaranly; this p
preciselylevel =
rror is ge
sis, it maugh to ids very clo
nd type IIher the ri
to as an
occurs. For exa
hen the nrongly r
, the nullhan the
two drughat the tifferencesible res
0
serious,hypothesteed 'low'obabilitycompute
erally un
y still be fntify these to hyp
errors ark of the
error of th
hen theple, in a
ull hypothjected.
hypothesiurrent dr
s on avero drugs pbetweenlts of any
and theris test proprobabilitis never 0d as
known.
alse (a tyalsenessothesis).
inverselther.
e first kin
ull hypotclinical tri
esis is
s might bg; i.e.
age.roducedhem.hypothes
fore morcedure isy of. This
e II errorof the nul
related;
.
esis H0, ial of a ne
is
)l
w
8/12/2019 Hypotesis Test
5/10
drug,aver
A typprod
drug
A typ
The
by
A typ
ComSee
A teused
our h
Themod
Theof ththe n
Theleveltwo-
the nullge, thanH0: ther
e II errorced the
on aver
e II error
robabilit
and writt
P(type II
e II error
pare typelso pow
est Stati
t statisticto decide
ypothesis
hoice ofl and the
ritical V
ritical valtest statiull hypoth
ritical valat whichided.
ypothesithe curreis no diffould oc
ame effe
ge, when
is frequen
of a type
n
error) =
can also
I error.r.
stic
is a quanwhether
test.
test stathypothes
lue(s)
ue(s) forstic in a sesis is rej
ue for anhe test is
might bet drug; i.rence be
ur if it wat, i.e. the
in fact th
tly due to
II error is
e referre
ity calculor not the
istic will des under
hypotheample isected.
hypothecarried o
that the.tween thconclud
re is no di
y produc
sample s
generall
to as an
ted fromnull hypo
epend onquestion.
is test isompared
is test det, and w
ew drug i
two drugd that th
fference
ed differe
izes bein
unknow
error of t
our sampthesis sh
the assu
a threshoto deter
pends onether the
is no bett
s on avertwo druetween t
nt ones.
too smal
, but is s
e secon
le of data.uld be re
ed prob
ld to whicine whet
the signiftest is on
r, on
age.se two
l.
mbolised
kind.
Its valueected in
bility
h the valuer or not
icancee-sided o
is
e
8/12/2019 Hypotesis Test
6/10
See
Thestatiis, thonethe omemmem
SeeSee
Theof wr
It is tto thsignihypoinad
The
Usu
lso critic
ritical R
ritical retic for whe sampleegion (thther will nber of theber of the
lso criticlso test
ignifica
ignificanongly reje
he probaconseqicance lethesis anertently
ignificanSignifica
lly, the si
P-Value
l region.
gion
ion CR, oich the nuspace for
critical rot. So, if tcritical recritical re
l value.tatistic.
ce Level
e level ofcting the
ility of a tences ofel as smto preveaking fal
e level isnce Level
nificance
r rejectioll hypothethe test sgion) will
he obsergion, wegion then
a statistiull hypot
pe I errosuch an ell as pos
nt, as fare claims.
usually d= P(type
level is c
region Rsis is rejetatistic islead us ted valueoncludewe concl
al hypothhesis H0, i
and is srror. Thatible in ors possibl
noted byI error) =
hosen to
R, is a sected in aartitionereject th
of the tesReject Hde "Do n
esis test iif it is in f
t by the iis, we wer to pro
e, the inv
e 0.05 (
t of valueypothesiinto two
e null hypstatistic i"; if it is not reject
s a fixedct true.
vestigatnt to maktect the nestigator
r equival
of the tetest. Th
regions;othesis Hs aot a
0".
robabilit
r in relatie thellrom
ntly, 5%)
stt
0,
n
.
8/12/2019 Hypotesis Test
7/10
Theprobextretrue.
It is ttrue.
It isrejecsigniThatlevel
Smal
smalindicrathe
The
rejeccorre
In otcoma typ
Thewant
robabilitbility of g
me than t
he proba
qual to tht the nullicance leis, if the, this woul
l p-value
ler it is, thtes the s
r than si
Power
ower of
t the nullct decisio
er wordsitting a t
e II error
Power =
aximuma test to
ne-side
value (p-etting a vhat obser
ility of wr
e significypothesiel of ourull hypotd be repo
suggest
e more ctrength ofply concl
statistic
ypothesin.
, the powpe II errorom 1, us
1 - P(typ
power a tave high
Test
value) oflue of thed by ch
ngly reje
nce level. The p-v
test and, iesis werrted as "p
hat the n
nvincing ievidenceding "Re
l hypothe
when it i
r of a hyr. It is calually expr
II error)
est can hpower, cl
a statistictest stati
ance alon
cting the
of the tealue is cof it is smato be rej< 0.05".
ll hypoth
s the rejefor say, rect H0' or
sis test m
s actually
othesis tulated b
essed as:
ve is 1, tose to 1.
l hypothstic as exe, if the n
ull hypot
t for whicmparedller, the rcted at t
esis is unl
ction of tejecting t"Do not r
easures t
false - th
st is thesubtracti
he minim
sis test itreme asull hypoth
esis if it i
h we wouith the acsult is sie 5% sig
likely to b
e null hye null hy
eject H0".
he test's
at is, to m
robabilitng the pr
m is 0. I
theor moreesis H0, i
s in fact
ld only jutualnificant.ficance
true. Th
othesis. Iothesis
bility to
ake a
of notbability o
eally we
t
0,
f
8/12/2019 Hypotesis Test
8/10
A onwhicof th
In ot
lesscritic
A on
Thethe ptest.
Exa
Supaver
agai
EithePresalterbe le
theyYetleadi
agai
Herein awoulless
-sided tewe canprobabil
er words
han the cl value o
-sided te
hoice beurpose of
ple
ose wege, 50 mH0: =stH1: 50ative hyant to te
nce it wo, on aver
ber of maypothesi-sided te
50n be saidcould rejrage num50.
othesis thesis, H0
or a one-
st, or the
as a on
nd a twoprior rea
ufacturercould se
othesest the nullld be usege, in a
tches incould be
st:
about thct the nulber of ma
est in whiare locat
ided test
set of va
-tailed te
-sided tessons for
claim tht up the f
ould leahypothesiful to knoox (no o
box or mtested a
averagel hypothetches in
ch the vald entirely
is the set
lues grea
t of signi
t is detersing a on
at there allowing h
to a oneis againstw if theree would
ore).ainst the
numberis in ourbox is lik
ues forin one ta
of values
er than th
icance.
ined bye-sided
e, onpothese
-sided testhe firstis likely tomplain i
same null
f matcheest, weely to be
il
e
t.
f
,
8/12/2019 Hypotesis Test
9/10
A twwhicprob
In ot
lessa se
A tw
Theby thtest.
Exa
Supaver
agai
EithePresalterbe le
theyYetleadi
agai
Herein awoulless
-sided tewe canbility dist
er words
han a firsond critic
-sided te
hoice bee purpos
ple
ose wege, 50 mH0: =stH1: 50ative hyant to te
nce it wo, on aver
ber of maypothesi-sided te
50n be saidcould rejrage num50.
othesis thesis, H0
or a two-
test and
as a two
est and aor prior
ufacturercould se
othesest the nullld be usege, in a
tches incould be
st:
about thct the nulber of ma
st in whiare locat
ided test
the set o
-tailed te
two-sideeasons f
claim tht up the f
ould leahypothesiful to knoox (no o
box or mtested a
averagel hypothetches in
h the vald in both
is the set
values g
t of signifi
test is dr using a
at there allowing h
to a oneis againstw if theree would
ore).ainst the
numberis in ourbox is lik
es fortails of th
of values
reater tha
cance.
terminedone-side
e, onpothese
-sided testhe firstis likely tomplain i
same null
f matcheest, weely to be
n
t.
f
,
8/12/2019 Hypotesis Test
10/10
A onmea
from
The
Thatunkn
Thishypo
A twmeainde
Whevaria
The
Thatpopualter
samplewhere t
an underl
ull hypotH0: =
is, the saown varia
null hypottheses, dH1: isH1: >H1: H1: 1