Douglas A Lind, William G Marchal and
Douglas A Lind, William G Marchal and
Samuel A Wathen. Statistical techniques
Samuel A Wathen. Statistical techniques
in Business and Economics
in Business and Economics
Chapter 1, 11 Chapter 1, 11
!"pothesis #est
!"pothesis #est
1 1What is a !"pothesis$
What is a !"pothesis$
%
%
What is a Hypothesis?What is a Hypothesis?
AA !"pothesis!"pothesis is a statement a&out the 'alue o( a is a statement a&out the 'alue o( a populationpopulation
parameter de'eloped (or the purpose o( testing. E)amples parameter de'eloped (or the purpose o( testing. E)amples o( h"potheses made a&out a population parameter are* o( h"potheses made a&out a population parameter are*
#he mean monthl" income (or s"stems anal"sts is + #he mean monthl" income (or s"stems anal"sts is +,-%.,-%.
What is Hypothesis Testing?What is Hypothesis Testing?
!"pothesis testing is a procedure, &ased on sample!"pothesis testing is a procedure, &ased on sample
e'idence and pro&a&ilit" theor", used to determine /hether e'idence and pro&a&ilit" theor", used to determine /hether the h"pothesis is a reasona&le statement and should not &e the h"pothesis is a reasona&le statement and should not &e re0ected, or is unreasona&le and should &e
What is a !"pothesis$
What is a !"pothesis$
%
%
What is a Hypothesis?What is a Hypothesis?
AA !"pothesis!"pothesis is a statement a&out the 'alue o( a is a statement a&out the 'alue o( a populationpopulation
parameter de'eloped (or the purpose o( testing. E)amples parameter de'eloped (or the purpose o( testing. E)amples o( h"potheses made a&out a population parameter are* o( h"potheses made a&out a population parameter are*
#he mean monthl" income (or s"stems anal"sts is + #he mean monthl" income (or s"stems anal"sts is +,-%.,-%.
What is Hypothesis Testing?What is Hypothesis Testing?
!"pothesis testing is a procedure, &ased on sample!"pothesis testing is a procedure, &ased on sample
e'idence and pro&a&ilit" theor", used to determine /hether e'idence and pro&a&ilit" theor", used to determine /hether the h"pothesis is a reasona&le statement and should not &e the h"pothesis is a reasona&le statement and should not &e re0ected, or is unreasona&le and should &e
!"pothesis #esting Steps
!"pothesis #esting Steps
mportant #hings to 2emem&er
mportant #hings to 2emem&er
a&out !
a&out !
and !
and !
113
3
!!* null h"pothesis and !* null h"pothesis and !11* alternate h"pothesis* alternate h"pothesis
!! and ! and !11 are mutuall" e)clusi'e and collecti'el" are mutuall" e)clusi'e and collecti'el"
e)hausti'e e)hausti'e
!! is al/a"s presumed to &e true is al/a"s presumed to &e true
!!11 has the &urden o( proo( has the &urden o( proo(
A random sample 4A random sample 4nn5 is used to 65 is used to 6reject Hreject H77
( /e conclude 8do not re0ect !( /e conclude 8do not re0ect !8, this does not necessaril"8, this does not necessaril"
mean that the null h"pothesis
mean that the null h"pothesis is true, it onl" suggests thatis true, it onl" suggests that there is not su9cient e'idence to re0ect !
there is not su9cient e'idence to re0ect !: re0ecting the: re0ecting the
null h"pothesis then, suggests that the alternati'e null h"pothesis then, suggests that the alternati'e h"pothesis ma" &e true.
h"pothesis ma" &e true.
Equality is always part of HEquality is always part of H00 (e.g. “=” , “≥” , (e.g. “=” , “≥” ,
“”!.
“”!.
!o/ to Set ;p a Claim as
!o/ to Set ;p a Claim as
!"pothesis
!"pothesis
n actual practice, the status quo is set
n actual practice, the status quo is set
up as !
up as !
( the claim is 6&oast(ul7 the claim is set
( the claim is 6&oast(ul7 the claim is set
up as !
up as !
1.1.2emem&er, !
2emem&er, !
11has the &urden
has the &urden
o( proo(
o( proo(
n pro&lem sol'ing, loo< (or
n pro&lem sol'ing, loo< (or
'ey wor$s
'ey wor$s
and con'ert them into s"m&ols. Some
and con'ert them into s"m&ols. Some
<e" /ords include* 6
<e" /ords include* 6improved, better
improved, better
than, as efective as, diferent rom,
than, as efective as, diferent rom, has
has
changed
=arts o( a Distri&ution in
!"pothesis #esting
->ne?tail 's. #/o?tail #est
!"pothesis Setups (or #esting a
Mean 4
µ
5
!"pothesis Setups (or #esting a
=roportion 4
π
5
E)ample
1
amesto/n Steel Compan" manu(actures
and assem&les des<s and other o9ce
equipment at se'eral plants in /estern e/
or< State. #he /ee<l" production o( the
Model A% des< at the Fredonia =lant
(ollo/s the normal pro&a&ilit" distri&ution
/ith a mean o( % and a standard de'iation
o( 1-. 2ecentl", &ecause o( mar<et
e)pansion, ne/ production methods ha'e
&een introduced and ne/ emplo"ees hired.
#he 'ice president o( manu(acturing /ould
li<e to in'estigate /hether there has &een a
change in the /ee<l" production o( the
=ro&lem
11
tep &) tate the null hypothesis an$ the alternate hypothesis.
!* µ %
!1* µ H %
(note) 'eywor$ in the pro*le+ “has hange$”! tep -) elet the leel of signi/ane.
= 0.0& as state$ in the pro*le+ tep 1) elet the test statisti.
=ro&lem
1% 58 . 2 not is 55 . 1 50 / 16 200 5 . 203 / 2 / 01 . 2 / 2 / > > − > − > Z Z n X Z Z α α σ µtep 2) 3or+ulate the $eision rule.
4e5et H0 if 676 % 7
Step 5: Make a decision and interpret the result.
Because 1.55 does not fall in the rejection region, H0 is not rejected. We conclude that the population mean is not different from 00. So ! e !ould report to the "ice president of manufacturing that the sample e"idence does not sho! that the production rate at the #redonia $lant has changed from 00 per !eek.
8- #"pe o( Errors in !"pothesis #esting
1
#"pe Error ?
DeKned as the pro&a&ilit" o( re0ecting the
null h"pothesis /hen it is actuall" true.
#his is denoted &" the Gree< letter 6α7
Also <no/n as the signiKcance le'el o( a test #"pe Error*
DeKned as the pro&a&ilit" o( 6accepting7 the
null h"pothesis /hen it is actuall" (alse.
#est Error
p
?alue in !"pothesis #esting
1
p9:;<E is the pro&a&ilit" o( o&ser'ing a
sample 'alue as e)treme as, or more e)treme than, the 'alue o&ser'ed, gi'en that the null h"pothesis is true.
n testing a h"pothesis, /e can also compare
the p?'alue to /ith the signiKcance le'el 4α5.
( the p?'alue N signiKcance le'el, H is
p?alue in !"pothesis #esting ?
E)ample
1-2ecall the last pro&lem /here the h"pothesis and decision rules /ere set up as*
!* µ O %
!1* µ P %
2e0ect ! i( I P Iα
/here I 1. and Iα %. 2e0ect ! i( p?'alue N α
.-- is not N .1 Conclude* Fail to re0ect !
What does it mean /hen p?'alue N
α
$
1@
4a5 .1, /e ha'e some e'idence that H is not true.
4&5 ., /e ha'e strong e'idence that H is not true.
4c5 .1, /e ha'e 'er" strong e'idence that H is not true.
4d5 .1, /e ha'e e)tremel" strong e'idence that H is not true.
#esting (or the =opulation Mean* =opulation
Standard De'iation ;n<no/n
1
When the population standard de'iation 4J5 is
un<no/n, the sample standard de'iation 4s5 is used in its place
#he t ?distri&ution is used as test statistic, /hich
#esting (or the =opulation Mean* =opulation
Standard De'iation ;n<no/n ? E)ample
1
#he McFarland nsurance Compan" Claims
Department reports the mean cost to process a claim is +-. An industr" comparison sho/ed this amount to &e larger than most other
insurance companies, so the compan"
instituted cost?cutting measures. #o e'aluate the eQect o( the cost?cutting measures, the
Super'isor o( the Claims Department selected a random sample o( %- claims processed last month. #he sample in(ormation is reported &elo/.
At the .1 signiKcance le'el is it reasona&le a
#esting (or a =opulation Mean /ith a
Rno/n =opulation Standard De'iation? E)ample
%
tep &) tate the null hypothesis an$ the alternate hypothesis.
!* µ +-
!1* µ N +-
(note) 'eywor$ in the pro*le+ “now less than”!
tep -) elet the leel of signi/ane. = 0.0& as state$ in the pro*le+ tep 1) elet the test statisti.
#esting (or a =opulation Mean /ith a
Rno/n =opulation Standard De'iation? E)ample
%1
tep 2) 3or+ulate the $eision rule.
2e0ect ! i( t N ?tα,n?1
Step 5: Make a decision and interpret the result.
Because -1.818 does not fall in the rejection region, H0 is not rejected at the . 01 significance level. We have not demonstrated that the cost-cutting
measures reduced the mean cost per claim to less than $60. he difference of $!."8 #$"6.% - $60& 'et(een the sample mean and the population mean could 'e due to sampling error.
#esting (or a =opulation Mean /ith an ;n<no/n =opulation Standard De'iation? E)ample
%%
#he current rate (or producing amp (uses at
ear" Electric Co. is % per hour. A ne/ machine has &een purchased and installed
that, according to the supplier, /ill increase the production rate. A sample o( 1 randoml"
selected hours (rom last month re'ealed the mean hourl" production on the ne/ machine /as %- units, /ith a sample standard
de'iation o( - per hour.
At the . signiKcance le'el can ear"
#esting (or a =opulation Mean /ith a
Rno/n =opulation Standard De'iation?
E)ample continued
%
Step 1* State the null and the alternate h"pothesis.
H* T O %: H1* T P %
Step %* Select the le'el o( signiKcance. t is ..
Step * Find a test statistic. ;se the t distri&ution &ecause the population standard de'iation is
#esting (or a =opulation Mean /ith a
Rno/n =opulation Standard De'iation?
E)ample continued
%3
Step 3* State the decision rule.
#here are 1 V 1 degrees o( (reedom. #he null h"pothesis is re0ected i( t P 1..
Step * Ma<e a decision and interpret the results. #he null h"pothesis is re0ected. #he mean num&er
produced is more than % per hour.
162 . 3 10 6 250 256 = − = − = n s X t µ
#ests Concerning =roportion
%
A =roportion is the (raction or percentage that
indicates the part o( the population or sample ha'ing a particular trait o( interest.
#he sample proportion is denoted &" p and is
(ound &" x/n
Assumptions in #esting a =opulation
=roportion using the U?Distri&ution
%-A random sample is chosen (rom the population.
t is assumed that the &inomial assumptions discussed
in Chapter - are met*
1. the sample data collected are the result o( counts:
%. the outcome o( an e)periment is classiKed into one o(
t/o mutuall" e)clusi'e categoriesa 6success7 or a 6(ailure7:
. the pro&a&ilit" o( a success is the same (or each trial:
and 435 the trials are independent
#he test /e /ill conduct shortl" is appropriate /hen
&oth nπ and n41? π 5 are at least .
When the a&o'e conditions are met, the normal
distri&ution can &e used as an appro)imation to the &inomial distri&ution
#est Statistic (or #esting a Single
=opulation =roportion
%@ n p z)
1
(
π π π − − = )ample proportion H*pothesi+ed population proportion )ample si+e#est Statistic (or #esting a Single
=opulation =roportion ? E)ample
%
Suppose prior elections in a certain state
indicated it is necessar" (or a candidate (or go'ernor to recei'e at least percent o( the 'ote in the northern section o( the state to &e elected. #he incum&ent go'ernor is interested in assessing his chances o( returning to o9ce and plans to conduct a sur'e" o( %,
registered 'oters in the northern section o( the state. ;sing the h"pothesis?testing procedure, assess the go'ernorXs chances o( reelection.
#est Statistic (or #esting a Single
=opulation =roportion ? E)ample
%
tep &) tate the null hypothesis an$ the alternate hypothesis.
!* π .
!1* π N .
(note) 'eywor$ in the pro*le+ “at least ”!
tep -) elet the leel of signi/ane. = 0.0& as state$ in the pro*le+ tep 1) elet the test statisti.
;se I?distri&ution since the assumptions are met and nπ and n41?π5
#esting (or a =opulation =roportion ? E)ample
tep 2) 3or+ulate the $eision rule.
4e5et H0 if Z #97
Step 5: Make a decision and interpret the result.
he computed value of z#%.80& is in the rejection region, so the null h*pothesis is rejected at the .0" level. he difference of %." percentage points 'et(een the sample percent #." percent& and the h*pothesi+ed population percent #80& is statisticall* significant. he
evidence at this point does not support the claim that the incum'ent governor (ill return to the governors mansion for another four *ears.
=ro&lem Y
1
According to a recent sur'e", Americans
get a mean o( @ hours o( sleep per night. A
random sample o( students at West
irginia ;ni'ersit" re'ealed the mean
num&er o( hours slept last night /as
-hours and 3 minutes. #he standard
de'iation o( the sample /as . hours. s it
reasona&le to conclude that students at
West irginia sleep less than the t"pical
American$ Compute the p?'alue.
=ro&lem Y
3-%
n the "ear %- the mean (are to Z" (rom
Charlotte, orth Carolina, to Seattle,
Washington, on a discount tic<et /as +%-@.
A random sample o( round?trip (ares on this
route last month gi'es in the (ollo/ing
ta&le. At the .1 signiKcance le'el can /e
conclude that the mean (are has increased$
What is the p?'alue$
=ro&lem Y @
#he ;.S. presidentXs call (or designing and
&uilding a missile de(ense s"stem that
ignores restrictions o( the Anti?Ballistic
Missile De(ense S"stem treat" 4ABM5 is
supported &" 3 o( the respondents in a
nation/ide poll o( 1% adults. s it
reasona&le to conclude that the nation is
e'enl" di'ided on the issue$ ;se the .
signiKcance le'el.
Testing Hypothesis a*out >ierene *etween Two @opulation Aeans (Two a+ple Tests of Hypothesis! •ull !"pothesis* !* 4T1 ? T%5D or T1 T% •Alternate !"pothesis !1* 4T1 ? T%5PD 4T1 ? T%5ND [T1 P T% T1 N T%\ ..one?tailed test Alternate !"pothesis !1* 4T1 ? T%5 HD [T1 H T% \ .. t/o?tailed test • #est Statistics* 2 2 2 1 2 1 2 1 2 2 2 1 2 1 0 2 1 ) ( ) ( n n x x z or n n D x x z σ σ σ σ + − = + − − =
4replace s /hen J not a'aila&le5
•2e0ection 2egion U P U] or U N ?U] 4one tailed test5
U P U]^% or U N ?U]^% 4t/o tailed test5
Mar" o FitUpatric is the ice =resident (or ursing Ser'ices at St. Lu<eXs Memorial !ospital. 2ecentl" she noticed in the 0o& posting (or nurses that those that are unionised seem to oQer higher /ages. She decided to in'estigate and gathered the (ollo/ing sample in(ormation. Group Mean Wage Sample Standard De'iation Sampl e SiUe ;nion +%.@ +%.% 3 onunion +1. +1. 3
Would it &e reasona&le (or her to conclude that there is signiKcant diQerence in earning &et/een union and non?union nurses$ ;se the . 1 signiKcance le'el.
E)ample
A manpower-development statistician is asked to
determine whether the horl! wa"es o# semiskilled workers are the same in two cities. $he reslts o# the srve! are presented in the #ollowin" ta%le&
'orl! a"e ate
-Solution
Step 1:
$his is a two-tailed test. $he h!pothesis is stated %elow. $he si"ni#icance level is 0.05 ("iven)
'0& *1 + *2 verss '1& *1 , *2
Step 2:
ince this is a test o# the means and the de"rees o# #reedom (n1 n2 - 2) is in e/cess o# 30 a test is appropriate. $he critical
vales are 1.6 (#rom a ta%le).
Step 3:
Solution
tep 4& ketch the distri%tion locate the critical vales
and the test statistic.
tep 5& ecide ince the test statistic vales lies within
the rejection re"ion then there is s##icient statistical evidence %ased on this sample to re7ect '0.
$he test #or a di##erence %etween parameters does not have
to %e ero it can %e non-ero. 8or e/ample&
'0& *1 - *2 9 0.10 verss '1& *1 - *2 : 0.10
@ro*le+) B91
#/o research la&oratories ha'e
independentl" produced drugs that
pro'ide relie( to arthritis suQerers. #he
Krst drug /as tested on a group o(
arthritis suQerers and produced an
a'erage o( . hours o( relie(, and a
sample standard de'iation o( 1. hours.
#he second drug /as tested on
arthritis suQerers, producing an a'erage
o( @. hours o( relie(, and a sample
standard de'iation o( %.1 hours. At the
. le'el o( signiKcance, does the
second drug pro'ide a signiKcantl"
@ro*le+) B9C
ot/ithstanding the Equal =a" Act o(
1-, in 1 it still appeared that men
earned more than /omen in similar 0o&s.
A random sample o( male machine
tool operators (ound a mean hourl"
/age o( +11., and the sample
standard de'iation /as +1.3. A random
sample o( 3 (emale machine?tool
operators (ound their mean /age to &e
+.3%,
and
the
sample
standard
de'iation /as +1.1. >n the &asis o(
these samples, is it reasona&le to
conclude 4at a .15 that the male
operators are earning o'er +%. more
per hour than the (emale operators$
$estin" '!pothesis a%ot i##erence
%etween $wo ;oplation ;roportion
1.ull !"pothesis* !* 4p1 ? p%5D or p1 p% %.Alternate !"pothesis !1* 4p1 ? p%5PD 4p1 ? p%5ND [p1 P p% p1 N p%\ ..one?tailed test Alternate !"pothesis!1* 4p1 ? p%5 HD [p1 H p% \ .. t/o?tailed test . #est Statistics*
2e0ection 2egion U P U] or U N ?U] 4one tailed test5
U P U]^% or U N ?U]^% 4t/o tailed test5
$estin" '!pothesis a%ot i##erence %etween
$wo ;oplation ;roportion& ;ooled
When population parameters ‘p’ and ‘q’ are unknown: .Dull !"pothesis* !* p1Gp% [p1?
p%G\
.Alternate !"pothesis !1* p1Pp% p1Np% ..one?tailed test
Alternate !"pothesis !1* p1 Hp% t/o?tailed test
.=ooled proportion* 2 1 2 1 < n n x x p + + = . #est Statistics* = 1 1 > < < ) < < ( 2 1 2 1 n n q p p p z + − =
.2e0ection 2egion U PU]or U N?U] 4one tailed test5
U PU]^% or U N?U]^%4t/o tailed test5
E)ample*
1. Accordin" to a report %! the American ?ancer ociet! more men
than women smoke and twice as man! smokers die prematrel!
than nonsmokers. @n random samples o# 200 males and 200 #emales 62 o# the males and 54 o# the #emales were smokers. @s there
s##icient evidence to conclde that the proportion o# male smokers hi"her #rom the proportion o# #emale smokers when + .01B
2. A #inancial anal!st wants to compare the trnover rates in percent
#or shares o# oil related stocks verss other stocks. he selected 32 oil-related stocks and 4 other stocks. $he mean trnover o# oil
related stocks is 31.4 percent and the standard deviation 5.1 percent. 8or the other stocks the mean rate was compted to %e 34. percent and the standard deviation 6.C percent. @s there a si"ni#icant
di##erence in the trnover rates o# the two t!pes o# stockB
@ro*le+)
B9--A coal?Kred po/er plant is considering t/o
diQerent s"stems (or pollution a&atement. #he Krst s"stem has reduced the emission o(
pollutants to accepta&le le'els - percent o( the time, as determined (rom % air samples. #he second, more e)pensi'e s"stem has
reduced the emission o( pollutants to ac cepta&le le'els @- percent o( the time, as determined (rom % air samples. ( the e)pensi'e s"stem is signiKcantl" more eQecti'e than the ine)pensi'e s"stem in
reducing pollutants to accepta&le le'els, then the management o( the po/er plant /ill install the e)pensi'e s"stem. Which s"stem /ill &e installed i( management uses a signiKcance
@ro*le+) B9-1
A group o( clinical ph"sicians is per(orming tests on patients to
determine the eQecti'eness o( a ne/ antih"pertensi'e drug.
=atients /ith high &lood pressure /ere randoml" chosen and then randoml" assigned to either the control group 4/hich recei'ed a /ell?esta&lished antih"pertensi'e5 or the treatment group 4/hich recei'ed the ne/ drug5. #he doctors noted the percentage o(
patients /hose &lood pressure /as reduced to a normal le'el /ithin 1 "ear. At the .1 le'el o( signiKcance, test appropriate h"potheses to determine /hether the ne/ drug is signiKcantl"
more eQecti'e than the older drug in reducing high &lood pressure.
Droup @roportion That
+proe$ Fu+*er of @atients #reatment .3 1% Control .- 1 3
Test for $ierene *etween
Aeans) s+all sa+ple siGe
For small samples siUes, /e must estimate
a 8pooled8 estimate 4a.<.a. a /eighted
a'erage5 o( the 'ariances (or the t/o
groups. #his estimate is*
and then, the estimated standard error is*
3- #ests (or DiQerence Bet/een #/o Means* Small Sample SiUe
.Dull !"pothesis !* T1GT%
.Alternati'e !"pothesis !1* T1 PT% or T1 NT%
4one tailed test5
!1* T1 H T% 4t/o?tailed test5 .=ooled Estimator* ( 1) ( 1) ) 1 ( ) 1 ( 2 1 2 2 2 2 1 1 2 − + − − + − = n n s n s n s p . #est Statistics* 2 1 < 2 1 x x x x t − − = σ 2 1 2 1 1 1 n n s x x t p + − =
)
2
(
1 + 2 − =n
n
df
.2e0ection 2egion t Pt] or t N?t]4one tailed test5
t Pt]^% or t N?t]^%4t/o tailed test5
E)ample*
A compan! wishes to test when the sensitivit! achieved %!
a new pro"ram is si"ni#icantl! hi"her than achieved nder the le"ac! pro"ram. $he #ollowin" in#ormation is
availa%le #rom test reslts.
ensitivit!
ean
)tandard
/eviation
)ample
)i+e
roposed
%
1"
1%
..2.
8
1
1"
3Solution
Step 1:
This is a one-tailed test. The hypothesis is stated below. The significance level is 0.05 (given) H0: μ1 ≤ μ ve!s"s H1: μ1 # μ
$tep :
$ince this is a test of the %eans and neithe! n1 o! n is in e&cess of '0 a t test is app!op!iate. The c!itical val"e is 1.0* (f!o% a t table with 5 deg!ees of f!eedo%). $tep ':
Solution
tep 4&
ketch the distri%tion locate the critical vales and the
test statistic.
tep 5&
ecide ince the test statistic vales lies within the
retention re"ion then there is no s##icient statistical evidence %ased on this sample to re7ect '0.
@ro*le+) B9