Professional Documents
Culture Documents
1. A random sample of 1,562 undergraduates enrolled in marketing courses was asked to respond on a scale from one (strongl disagree! to se"en (strongl agree! to t#e proposition$ %Ad"ertising #elps raise our standard of li"ing.% &#e sample mean response was '.2( and t#e sample standard de"iation was 1.)2. &est at t#e 1* le"el, against a two+sided alternati"e, t#e null # pot#esis t#at t#e population mean is '.,. -ormulate . pot#eses$ H,$ / ' HA$ ' 0reate 1ecision 2ule$ 2e3ect H, if 0alculate &est Statistic$ z 1ecision$ 2e3ect H, at t#e 1* le"el.
= '.2( ' 1.)2 / 1562
= 6.,6
2. A random sample of (6 percentage c#anges in promised pension benefits of single emplo er plans after t#e establis#ment of t#e Pension 7enefit 8uarantee 0orporation was obser"ed. &#e sample mean percentage c#ange was ,.,(6 and t#e sample standard de"iation was ,.2,1. -ind and interpret t#e p+"alue of a test of t#e null # pot#esis t#at t#e population mean percentage c#ange is ,, against a two+sided alternati"e. -ormulate . pot#eses$ H,$ / , HA$ , 0alculate &est Statistic$ z p+"alue$ p+"alue
= ,.,(6 ,.2,1 / (6
= ).)6)
= 2[ P ( z > ).)6)) ]
= 2( ,.,,,)59 ) = ,.,,,(1(
&#is null # pot#esis can be re3ected at an "alue of (significance le"el! greater t#an ,.,(1(*. :ote$ since t#e z "alue is too large to look up in t#e standard normal table, we need to use ;<cel to calculate t#is p+"alue. (=se t#e function$ =2*(1-NORMSDIST(3.38))) >n an e<am, ou could 3ust sa t#at t#e p+"alue is appro<imatel e?ual to @ero.
Aanagerial Statistics
1),
Prof. Buran
). >n t#e basis of a random sample, t#e null # pot#esis H,$ / , is tested against t#e alternati"e HA$ 4 , and t#e null # pot#esis is not re3ected at t#e 5* significance le"el. (a! 1oes t#is necessaril impl t#at , is contained in t#e 95* confidence inter"al for C
:o. Since t#is is a one+tailed test, it is possible for X to be "er low (wa down in t#e lower tail! and still not re3ect t#e null # pot#esis. =nder t#is scenario, t#e # pot#esi@ed true mean , mig#t not be contained in t#e confidence inter"al for . Df t#e # pot#esis test were two+sided, t#e answer would #a"e been es. (b! 1oes t#is necessaril impl t#at , is contained in t#e 95* confidence inter"al for , if t#e obser"ed sample mean is bigger t#an ,C
Ees. &#e second sentence confines X to t#e upper tail, between , and , F 1.6'5 standard errors. &#e 95* confidence inter"al for goes 1.96 standard errors on eit#er side of X , and t#erefore must contain ,.
Aanagerial Statistics
1)1
Prof. Buran
'. A beer distributor claims t#at a new displa , featuring a life+si@e picture of a well+known at#lete, will increase product sales in supermarkets b an a"erage of 5, cases in a week. -or a random sample of 2, supermarkets, t#e a"erage sales increase was '1.) cases and t#e sample standard de"iation was 12.2 cases. &est at t#e 5* le"el t#e null # pot#esis t#at t#e population mean sales increase is at least 5, cases, stating an assumption ou make. -ormulate . pot#eses$ H,$ = 5, HA$ < 5, 0reate 1ecision 2ule$ :ote t#at t(19, ,.,5! / 1.(29 2e3ect H, if t, 5 +1.(29, 0alculate &est Statistic$ t, 1ecision$ 2e3ect H, at t#e 5* le"el.
= '1.) 5, 12.2 / 2,
= ).169
Aanagerial Statistics
1)2
Prof. Buran
5. >f a sample of )61 owners of retail ser"ice and business firms t#at #ad gone into bankruptc , 1,5 reported #a"ing no professional assistance prior to opening t#e business. &est t#e null # pot#esis t#at at most 25* of all members of t#is population #ad no professional assistance before opening t#e business. -ormulate . pot#eses$ H,$ p / ,.25 HA$ p 4 ,.25 0alculate &est Statistic$ z
= ,.29,9 ,.25 ( ,.25 )( ,.(5 ) )61
= 1.(9
p+"alue$ p+"alue
= P( z > 1.(9 )
= ,.,)6(
&#is null # pot#esis can be re3ected at an "alue of (significance le"el! greater t#an ).6(*.
Aanagerial Statistics
1))
Prof. Buran
6. Dn a random sample of 16, business sc#ool graduates, se"ent +two sample members indicated some measure of agreement wit# t#e statement$ %A reputation for et#ical conduct is less important for a managerGs c#ances for promotion t#an a reputation for making mone for t#e firm.% &est t#e null # pot#esis t#at one+#alf of all business sc#ool graduates would agree wit# t#is statement against a two+sided alternati"e. -ind and interpret t#e p+"alue of t#e test.
(2 H= = ,.'5 :ote t#at p 16,
= 1.26
p+"alue$ p+"alue
= 2[ P( z > 1.26 ) ]
= 2 ( ,.1,)6 )
= ,.2,(6
&#is null # pot#esis can be re3ected at an "alue of (significance le"el! greater t#an 2,.(6*.
Aanagerial Statistics
1)'
Prof. Buran
(. &#e AA&I;S procedure was designed to measure attitudes toward women as managers. .ig# scores indicate negati"e attitudes and low scores indicate positi"e attitudes. Dndependent random samples were taken of 151 male A7A students and 1,6 female A7A students. -or t#e former group, t#e sample mean and standard de"iation AA&I;S scores were 65.6 and 19.), w#ile t#e corresponding figures for t#e latter group were (1.5 and 12.2. &est t#e null # pot#esis t#at t#e two population means are e?ual against t#e alternati"e t#at t#e true mean AA&I;S score is #ig#er for male t#an for female A7A students. -ormulate . pot#eses$ H,$ X Y HA$ X Y 0alculate &est Statistic$ z
=
/, 4,
( 65.6 (1.5) ,
19.) 2 12.2 2 + 151 1,6
= (.29
p+"alue$ Se"en standard de"iations awa from t#e # pot#esi@ed meanJ Ie can re3ect t#is null # pot#esis at an reasonable le"el.
Aanagerial Statistics
1)5
Prof. Buran
6. Dn 196,, a random sample of 1,556 people was asked to respond to t#e statement$ %0apitalism must be altered before an significant impro"ements in #uman welfare can be reali@ed.% >f t#ese sample members, )6.'* agreed wit# t#e statement. I#en t#e same statement was presented to a random sample of 1,1,6 people in 1969, 52.,* agreed. &est t#e null # pot#esis t#at t#e population proportions agreeing wit# t#is statement were t#e same in t#e two ears, against t#e alternati"e t#at a #ig#er proportion agreed in 1969. :ote t#at p , = ,.)6'
1556 11,6 + ,.52 1556 + 11,6 1556 + 11,6
/, 5,
= 6.9(
p+"alue$ >nce again, t#is z+"alue is so far awa from t#e # pot#esi@ed mean t#at we can re3ect t#is null # pot#esis at an reasonable le"el.
Aanagerial Statistics
1)6
Prof. Buran
9. >f a random sample of )61 in"estment+grade corporate bonds, 191 #ad sinking funds. >f an independent random sample of 166 speculati"e+grade corporate bonds, 1'5 #ad sinking funds. &est against a two+sided alternati"e t#e null # pot#esis t#at t#e two population proportions are e?ual. :ote t#at p , =
191 + 1'5 = ,.61' )61 + 166
/, ,
( ,.5,1 ,.6() ) ,
,.61'( ,.)66 ) )61 + 166 ( )61)( 166 )
= 6.216
Aanagerial Statistics
1)(
Prof. Buran
1,. A wine producer claims t#at t#e proportion of its customers w#o cannot distinguis# its product from fro@en grape 3uice is at most ,.1,. &#e producer decides to test t#is null # pot#esis against t#e alternati"e t#at t#e true proportion is greater t#an ,.1,. &#e decision rule adopted is to re3ect t#e null # pot#esis if t#e sample proportion t#at cannot distinguis# between t#ese two fla"ors e<ceeds ,.1'. (a! Df a random sample of 1,, customers is c#osen, w#at is t#e probabilit of a & pe D error, using t#is decision ruleC
= P ( z > 1.)) )
= ,.,916
Aanagerial Statistics
1)6
Prof. Buran
(b! Df a random sample of ',, customers is selected, w#at is t#e probabilit of a & pe D error, using t#is decision ruleC ;<plain, in words and grap#icall , w# our answer differs from t#at in part (a!.
= P ( z > 2.6( )
= ,.,,)6
Iit# more data, our standard error is smaller. &#ere is muc# less c#ance t#at a sample of ',,, taken from t#e # pot#esi@ed distribution, would #a"e a sample proportion greater t#an ,.1'.
Aanagerial Statistics
1)9
Prof. Buran
(c! Suppose t#at t#e true proportion of customers w#o cannot distinguis# between t#ese fla"ors is ,.2,. Df a random sample of 1,, customers is selected, w#at is t#e probabilit of a & pe DD errorC
Aanagerial Statistics
1',
Prof. Buran
(d! Suppose t#at instead of t#e gi"en decision rule, it is decided to re3ect t#e null # pot#esis if t#e sample proportion of customers w#o cannot distinguis# between t#e two fla"ors e<ceeds ,.16. A random sample of 1,, customers is selected. (i! Iit#out doing t#e calculations, state w#et#er t#e probabilit of a & pe D error will be #ig#er t#an, lower t#an, or t#e same as t#at in part (a!. Kower. Ie are, in effect, mo"ing t#e critical "alue fart#er to t#e rig#t (,.16 is fart#er t#an ,.1' from t#e # pot#esi@ed proportion of ,.1,!. &#e tail of t#e normal distribution be ond t#is new cut+off point will be smaller, and so will t#e probabilit of a & pe D error.
Aanagerial Statistics
1'1
Prof. Buran
(ii! Df t#e true proportion is ,.2,, will t#e probabilit of a & pe DD error be #ig#er t#an, lower t#an, or t#e same as t#at in part (c!C .ig#er. &#is new cut+off point (,.16! is closer to t#e true proportion (,.2,! t#an t#e pre"ious cut+ off point (,.1'!. As t#e cut+off point gets closer to t#e true proportion, it becomes more difficult to s#ow t#at t#e # pot#esi@ed proportion (,.1,! is false. >ur probabilit of accepting a false null # pot#esis increases as our cut+off point approac#es t#e true mean.
Aanagerial Statistics
1'2
Prof. Buran
11. State w#et#er eac# of t#e following is true or false. (a! &#e significance le"el of a test is t#e probabilit t#at t#e null # pot#esis is false. -alse. &#e significance (! is t#e probabilit of a sample falling into t#e re3ection region, gi"en t#at t#e null # pot#esis is true. (b! A & pe D error occurs w#en a true null # pot#esis is re3ected. &rue. (c! A null # pot#esis is re3ected at t#e ,.,25 le"el, but is accepted at t#e ,.,1 le"el. &#is means t#at t#e p+"alue of t#e test is between ,.,1 and ,.,25. &rue. (d! &#e power of a test is t#e probabilit of accepting a null # pot#esis t#at is true. -alse. &#e power (1 + ! is t#e probabilit of correctl re3ecting a false null # pot#esis. (e! Df a null # pot#esis is re3ected against an alternati"e at t#e 5* le"el, t#en using t#e same data, it must be re3ected against t#at alternati"e at t#e 1* le"el. -alse. 2ecall t#at we re3ect t#e null # pot#esis w#en t#e p+"alue is smaller t#an alp#a. :ow, if we re3ect at t#e 5* alp#a le"el, t#e p+"alue is known to be less t#an 5*. &#at does not necessaril mean t#at it is less t#an 1*. (f! Df a null # pot#esis is re3ected against an alternati"e at t#e 1* le"el, t#en using t#e same data it must be re3ected against t#at alternati"e at t#e 5* le"el. &rue. (g! &#e p+"alue of a test is t#e probabilit t#at t#e null # pot#esis is true. -alse. &#e p+"alue is based on t#e assumption t#at t#e null # pot#esis is true. 8i"en t#at assumption, t#e p+"alue is t#e probabilit of finding a sample as far (or fart#er! awa from t#e null # pot#esi@ed "alue as t#e sample being anal @ed.
Aanagerial Statistics
1')
Prof. Buran
12. Supporters claim t#at a new windmill can generate an a"erage of at least 6,, kilowatts of power per da . 1ail power generation for t#e windmill is assumed to be normall distributed wit# a standard de"iation of 12, kilowatts. A random sample of 1,, da s is taken to test t#is claim against t#e alternati"e # pot#esis t#at t#e true mean is less t#an 6,, kilowatts. &#e claim will be accepted if t#e sample mean is ((6 kilowatts or more and re3ected ot#erwise. (a! I#at is t#e probabilit of a & pe D error using t#e decision rule if t#e population mean is in fact 6,, kilowatts per da C
= ,.,226
(b! I#at is t#e probabilit of a & pe DD error using t#is decision rule if t#e population mean is in fact (', kilowatts per da C
= ,.,,1)
(c! Suppose t#at t#e same decision rule is used, but wit# a sample of 2,, da s rat#er t#an 1,, da s. (i! Iould t#e "alue of be larger t#an, smaller t#an, or t#e same as t#at found in (a!C Smaller. (ii! Iould t#e "alue of be larger t#an, smaller t#an, or t#e same as t#at found in (b!C Smaller.
Aanagerial Statistics
1''
Prof. Buran
(d! Suppose t#at a sample of 1,, obser"ations was taken but t#at t#e decision rule was c#anged so t#at t#e claim would be accepted if t#e sample mean was at least (65 kilowatts. (i! Iould t#e "alue of be larger t#an, smaller t#an, or t#e same as t#at found in (a!C Smaller. (ii!Iould t#e "alue of be larger t#an, smaller t#an, or t#e same as t#at found in (b!C Karger.
Aanagerial Statistics
1'5
Prof. Buran
1). Df a ,.,5 le"el of significance is used in a two+tailed # pot#esis test, w#at will ou decide if t#e computed "alue of t#e test statistic is F2.21C >ur decision rule will be to re3ect t#e null # pot#esis if t#e absolute "alue of t#e test statistic is greater t#an 1.96. Ie are told t#at t#e computed "alue of t#e test statistic z, is F2.21, so we decide to re3ect t#e null # pot#esis. &#e ,.,5 le"el of significance means we are prepared to take a 5* risk of a & pe D error. Since t#is is a two+tailed test, t#is corresponds to a 2.5* probabilit in eac# of t#e two tails. &#erefore, eac# of our two cut+off "alues oug#t to be far enoug# from t#e null+# pot#esi@ed parameter t#at t#e correspond to 5,* + 2.5* / '(.5* probabilit . Kook in t#e middle of t#e @+table for a "alue near ,.'(5, and we find t#at it corresponds to appro<imatel 1.96 standard de"iations from t#e mean. So our decision rule will be to re3ect t#e null # pot#esis if t#e test statistic is less t#an +1.96, or if it is greater t#an F1.96. An alternati"e met#od for getting t#e critical "alue 1.96 is to use t#e ;<cel function =NORMSINV(0.975).
Aanagerial Statistics
1'6
Prof. Buran
1'. Df a ,.1, le"el of significance is used in a two+tailed # pot#esis test, w#at will be our decision rule in terms of @ for re3ecting a null # pot#esis t#at t#e population mean is 5,,C Dn t#is case, eac# tail must contain 5*, so we look in t#e middle of t#e z+table for a "alue near ,.'5, and we find t#at it corresponds to appro<imatel 1.6'5 standard de"iations from t#e mean. Eou can also get t#e 1.6'5 critical "alue using t#e ;<cel function =NORMSINV(0.95). >ur test statistic will be$
z,
= X 5,, s n
>ur decision rule will be to re3ect H, if t#e test statistic is less t#an +1.6'5, or if it is greater t#an F1.6'5.
15. Df a ,.,1 le"el of significance were used in a two+tailed # pot#esis test, w#at would be our decision rule in terms of @ for re3ecting H,$ =12.5 C Dn t#is case, eac# tail must contain ,.,,5 so we look in t#e middle of t#e z+table for a "alue near ,.'95, and we find t#at it corresponds to appro<imatel 2.5(5 standard de"iations from t#e mean. Eou can also get t#e 2.5(5 critical "alue using t#e ;<cel function =NORMSINV(0.995). >ur test statistic will be$
z,
= X 12.5 s n
>ur decision rule will be to re3ect t#e H, if t#e test statistic is less t#an +2.5(5, or if it is greater t#an F2.5(5.
Aanagerial Statistics
1'(
Prof. Buran
16. I#at would be our decision in L15 abo"e if t#e computed "alue of t#e test statistic @ were +2.61C +2.61 is less t#an +2.5(5, so we decide to re3ect H,.
1(. Suppose t#e director of manufacturing at a clot#ing factor needs to determine w#et#er a new mac#ine is producing a particular t pe of clot# according to t#e manufacturerGs specifications, w#ic# indicate t#at t#e clot# s#ould #a"e a mean breaking strengt# of (, pounds and a standard de"iation of ).5 pounds. A sample of '9 pieces re"eals a sample mean of 69.1 pounds. (a! State t#e null and alternati"e # pot#eses. H,$ HA$ (b!
= (, (,
Ds t#ere e"idence t#at clot# from t#is mac#ine #as an a"erage breaking strengt# t#at is different from t#e manufacturerGs specificationsC (=se a .,5 le"el of significance.!
X , . s n
&est Statistic$ z , =
1ecision 2ule$ 2e3ect H, if t#e test statistic is greater t#an 1.96 or less t#an +1.96.
z,
= X , s n 69.1 (, = ).5 '9
= 1 .6
Ie do not re3ect. &#ere is insufficient e"idence against H, to re3ect it at t#e ,.,5 le"el.
:otice in t#e c#art abo"e #ow t#e test statistic z, lies in t#e non+re3ection region.
Aanagerial Statistics
1'6
Prof. Buran
(c!
I#at will our answer be in (b! if t#e standard de"iation is specified as 1.(5 poundsC
z,
= 69.1 (, 1.(5 '9
= ).6
(d!
= 2.,
2e3ect.
Aanagerial Statistics
1'9
Prof. Buran
16. A potential entrepreneur is considering t#e purc#ase of a coin+operated laundr . &#e present owner claims t#at o"er t#e past 5 ears t#e a"erage dail re"enue #as been M6(5. &#e bu er would like to find out if t#e true a"erage dail re"enue is different from M6(5. A sample of ), selected da s re"eals a dail a"erage re"enue of M625, wit# a standard de"iation of M(5. (a! State t#e null and alternati"e # pot#eses. H,$ HA$ (b!
= 6(5 6(5
Ds t#ere e"idence t#at t#e claim of t#e present owner is not "alidC (=se a .,1 le"el of significance.!
X , . s n
&est Statistic$ z , =
1ecision 2ule$ 2e3ect H, if t#e test statistic is greater t#an 2.5(5 or less t#an +2.5(5.
z,
= X , s n 625 6(5 = (5 ),
= ).65
Ie re3ect H,. &#ere is sufficientl strong e"idence against H, to re3ect it at t#e ,.,1 le"el. Ie conclude t#at t#e true mean is different from M6(5.
Aanagerial Statistics
15,
Prof. Buran
(c!
I#at will our answer be in (b! if t#e standard de"iation is now M1,,C
z,
= 625 6(5 1,, ),
= 2.('
(d!
= 1.6)
Aanagerial Statistics
151
Prof. Buran
19. A&As must be stocked wit# enoug# cas# to satisf customers making wit#drawals o"er an entire weekend. >n t#e ot#er #and, if too muc# cas# is unnecessaril kept in t#e A&As, t#e bank is forgoing t#e opportunit of in"esting t#e mone and earning interest. Suppose t#at at a particular branc# t#e e<pected (i.e., population! a"erage amount of mone wit#drawn from A&A mac#ines per customer transaction o"er t#e weekend is M16, wit# an e<pected (i.e., population! standard de"iation of M),. (a! State t#e null and alternati"e # pot#eses. H,$ HA$
=16, 16,
N&#is is a two+tailed test because t#e language of t#e ?uestion suggests t#at we are interested to know w#et#er t#e true mean is different from (eit#er greater t#an or less t#an! t#e null+ # pot#esi@ed mean of M16,.O (b! Df a random sample of )6 customer transactions is e<amined and it is obser"ed t#at t#e sample mean wit#drawal is M1(2, is t#ere e"idence to belie"e t#at t#e true a"erage wit#drawal is no longer M16,C (=se a ,.,5 le"el of significance.!
X , . s n
&est Statistic$ z , =
1ecision 2ule$ 2e3ect H, if t#e test statistic is greater t#an 1.96 or less t#an +1.96.
z,
= X , s n 1(2 16, = ), )6
= 2 .'
Ie re3ect H,. &#ere is sufficient e"idence against H, to re3ect it at t#e ,.,5 le"el.
Aanagerial Statistics
152
Prof. Buran
(c!
I#at will our answer be in (b! if t#e standard de"iation is reall M2'C
z,
= 1(2 16, 2' )6
= ).,
(d!
:ow t#e critical "alues are +2.5(5 and F2.5(5. &#e test statisticGs "alue of 2.' is in t#is inter"al, so we do not re3ect H,.
Aanagerial Statistics
15)
Prof. Buran
2,. Suppose t#at in a two+tailed # pot#esis test ou compute t#e "alue of t#e test statistic as F2.,,. I#at is t#e p+"alueC Ie look up 2.,, in t#e standard normal table and see t#at it corresponds to a probabilit of ,.'((2. &#erefore, t#e area under t#e normal cur"e outside of 2 standard de"iations from t#e mean is ,.5 + ,.'((2 / ,.,226 in eac# tail. Since t#is is a two+tailed test, we need to include bot# tails of t#e distribution in our p+"alue$ 2 P ,.,226 / ,.,'56. Alternati"el , we could use t#e ;<cel function$ =2*(1-NORMSDIST(2))
21. Suppose t#at in a two+tailed # pot#esis test ou compute t#e "alue of t#e test statistic as +1.)6. I#at is t#e p+"alueC Ie look up 1.)6 in t#e standard normal table and see t#at it corresponds to a probabilit of ,.'162. &#erefore, t#e area under t#e normal cur"e more t#an 1.)6 standard de"iations from t#e mean is ,.5 + ,.'162 / ,.,6)6 in eac# tail. Since t#is is a two+tailed test, our p+"alue is 2 P ,.,6)6 / ,.16(6. Alternati"el , we could use one of t#ese ;<cel functions$ =2*(1-NORMSDIST(1.38)) =2*(NORMSDIST(-1.38))
22. Dn L21 abo"e, w#at is our statistical decision if ou test t#e null # pot#esis at t#e ,.,1 le"el of significanceC Ie can answer b e<amining t#e relations#ip between alp#a (,.,1 in t#is instance! and t#e p+ "alue (,.16(6 in t#is instance!. 2elations#ip between and p+"alue p+"alue 5 p+"alue 4 Dn t#is problem t#e p+"alue 4 , so we do not re3ect H,. 1ecision 2e3ect H, 1o :ot 2e3ect H,
Aanagerial Statistics
15'
Prof. Buran
2). =se t#e same information from Problem 1(. (a! 0ompute t#e p+"alue and interpret its meaning.
= 1.6
Kooking at t#e standard normal table, we see t#at t#is @ "alue corresponds to a probabilit of ,.'6'1. &#is implies a two+tailed p+"alue of 2P(,.5 Q ,.'6'1! / ,.,(16. (:ote t#at t#ere is no implied direction to our testR we do not #a"e a one+sided alternati"e # pot#esis.! Ie could also use$ =2*(1-NORMSDIST(1.8)) (b! I#at is our statistical decision if ou test t#e null # pot#esis at t#e ,.,5 le"el of significanceC
&#e decision rule is to re3ect H, if t#e p+"alue is less t#an ,.,5. >ur p+"alue is greater t#an ,.,5, so we do not re3ect t#e null # pot#esis. (c! Ds t#ere e"idence t#at t#e mac#ine is not meeting t#e manufacturerGs specifications for a"erage breaking strengt#C
:o, not at t#e ,.,5 le"el of significance. (d! 0ompare our conclusions #ere wit# t#ose of (b! in L1( abo"e.
>ur conclusion is identical to t#e one in L19. .owe"er, instead of using t#e four+step classical # pot#esis testing procedure, we #a"e gone directl to t#e p+"alue.
Aanagerial Statistics
155
Prof. Buran
2'. =se t#e same information from L19 abo"e. (a! 0ompute t#e p+"alue and interpret its meaning.
z,
= X , s n 1(2 16, = ), )6
= 2 .'
2.' standard de"iations corresponds to a probabilit of ,.'916, so our two+tailed p+"alue is 2P(,.5 Q ,.'916! / ,.,16'. Ie could also use$ =2*(1-NORMSDIST(2.4)) (b! I#at is our statistical decision if ou test t#e null # pot#esis at t#e ,.,5 le"el of significanceC
>ur p+"alue is smaller t#an ,.,5, so we re3ect t#e null # pot#esis. (c! Ds t#ere e"idence to belie"e t#at t#e true a"erage wit#drawal is no longer M16,C
Ees. &#e obser"ed sample mean of M1(2 would be e<tremel unlikel if t#e true population mean were reall M16,. Ie take t#is to be strong e"idence t#at t#e true population mean is not M16,.
(d!
>ur conclusion is identical to t#e one in L19 (b!. .owe"er, instead of using t#e four+step classical # pot#esis testing procedure, we #a"e gone directl to t#e p+"alue.
Aanagerial Statistics
156
Prof. Buran
25. &#e 8len Valle Steel 0ompan manufactures steel bars. Df t#e production process is working properl , it turns out steel bars wit# an a"erage lengt# of at least 2.6 feet wit# a standard de"iation of ,.2, foot (as determined from engineering specifications on t#e production e?uipment in"ol"ed!. Konger steel bars can be used or altered, but s#orter bars must be scrapped. A sample of 25 bars is selected from t#e production line. &#e sample indicates an a"erage lengt# of 2.() feet. &#e compan wis#es to determine w#et#er t#e production e?uipment needs an immediate ad3ustment. (a! State t#e null and alternati"e # pot#eses.
-rom t#e ?uestion, we can infer t#at we are not concerned wit# bars t#at are too long, because t#e can be s#ortened and still be usable. &#is implies a one+tailed test, wit# t#e re3ection region in t#e lower tail. (Ie will o"erlook t#e ob"ious wastefulness of #a"ing to re+cut all of t#e bars t#at come out too long.! H,$ HA$
= 2.6 < 2 .6
Aanagerial Statistics
15(
Prof. Buran
(b!
Df t#e compan wis#es to test t#e # pot#esis at t#e ,.,5 le"el of significance, w#at decision would it make using t#e classical approac# to # pot#esis testingC ;<press our decision rule in terms of t#e a"erage bar lengt# in feet.
&#e critical t+"alue corresponding to a lower tail probabilit of ,.,5 (wit# 2' degrees of freedom! is Q1.(11. Ie want to find a critical "alue for t#e sample mean, in terms of bar lengt#, below w#ic# we will re3ect t#e null # pot#esis. 1.(11 standard errors below t#e # pot#esi@ed mean can be translated$
1.(11
1.(11
1.(11( ,.,' )
X P , s n P X 2.6 = ,.2, 25
X P 2.6 ( ,.,' ) ,.,' = X P 2.6 =X P =X P
&#e critical "alue is 2.()16. Df t#e obser"ed sample mean is less t#an 2.()16, t#en our decision rule will be to re3ect t#e null # pot#esis. &#is decision rule is e?ui"alent to sa ing t#at if t#e obser"ed test statistic t, is less t#an Q1.(11, t#en our decision rule will be to re3ect t#e null # pot#esis &#e sample a"erage is 2.() feet, so we re3ect t#e null # pot#esis and conclude t#at t#e true mean is less t#an 2.6 feet.
Aanagerial Statistics
156
Prof. Buran
(c!
Df t#e compan wis#es to test t#e # pot#esis at t#e ,.,5 le"el of significance, w#at decision would it make using t#e p+"alue approac# to # pot#esis testingC
Dn a procedure similar to t#at used in Problem L22, we can answer b e<amining t#e relations#ip between alp#a (,.,5 in t#is instance! and t#e p+"alue. 2elations#ip between and p+"alue p+"alue 5 p+"alue 4
X ,
t,
= 1.(5 Dn t#e @+table, 1.(5 standard de"iations corresponds to a probabilit of ,.'599, so our one+tailed p+"alue is appro<imatel ,.5 Q ,.'599 / ,.,',1. Since t#is "alue is smaller t#an ,.,5, we re3ect t#e null # pot#esis, and conclude t#at t#e true population mean is less t#an 2.6 feet. :ote t#at t#is is merel an estimate of t#e true p+"alue, using z as an appro<imation of t. A more precise p+"alue calculation can be done wit# t#e ;<cel function$
TDIST(Test Stat, n-1, Tai s) = TDIST(!"S(-1.75),25-1,1) = 0.04#45
(d!
Df t#e null # pot#esis were true and t#e true population mean were reall 2.6 feet, t#e probabilit of obser"ing a sample mean of 2.() feet or smaller would be onl '.6'*. -or # pot#esis testing purposes, we take t#is to be strong e"idence against t#e null # pot#esis. (e! 0ompare our conclusions in (b! and (c!.
&#e conclusions are identical. Dn bot# cases we conclude t#at t#e true population mean is less t#an 2.6 feet.
Aanagerial Statistics
159
Prof. Buran
26. &#e director of manufacturing at a clot#ing factor needs to determine w#et#er a new mac#ine is producing a particular t pe of clot# according to t#e manufacturerGs specifications, w#ic# indicate t#at t#e clot# s#ould #a"e a mean breaking strengt# of (, pounds and a standard de"iation of ).5 pounds. &#e director is concerned t#at if t#e mean breaking strengt# is actuall less t#an (, pounds, t#e compan will face too man lawsuits. A sample of '9 pieces re"eals a sample mean of 69.1 pounds. Ds t#ere e"idence t#at clot# from t#is mac#ine #as an a"erage breaking strengt# t#at is less t#an t#e manufacturerGs specificationsC (a! State t#e null and alternati"e # pot#eses.
Dn contrast to Problems L1( and L2) abo"e, t#e language #ere suggests t#at we are interested specificall in w#et#er t#e population mean is less t#an (and not merel different from! t#e # pot#esi@ed "alue of (, pounds. H,$ HA$ (b!
= (, < (,
At t#e ,.,5 le"el of significance, is t#ere e"idence t#at t#e mean breaking strengt# is less t#an (, poundsC ;<press our decision rule in terms of a"erage breaking strengt# in pounds.
>ur critical "alue will be t#e @+"alue corresponding to a lower tail probabilit of ,.,5, namel Q 1.6'5.
1.6'5
X P , s n
1.6'5
1.6'5( ,.5)
X P (, = ).5 '9
X P (, ( ,.5) ,.5 = X P (, =
=XP =XP
&#e critical "alue is 69.1((5 pounds. Df t#e obser"ed sample mean is less t#an 69.1((5, t#en we will re3ect t#e null # pot#esis. Dn t#is case, t#e sample mean of 69.1 pounds is less t#an t#e critical "alue. &#erefore we re3ect t#e null # pot#esis and conclude t#at t#e true mean strengt# is less t#an (, pounds.
Aanagerial Statistics
16,
Prof. Buran
(c!
At t#e ,.,5 le"el of significance, using t#e p+"alue approac# to # pot#esis testing, is t#ere e"idence t#at t#e mean breaking strengt# is less t#an (, poundsC
z,
= X , s n 69.1 (, = ).5 '9
= 1.6
Kooking at t#e standard normal table, we see t#at t#is @ "alue corresponds to a probabilit of ,.'6'1. &#is implies a one+tailed p+"alue of ,.5 Q ,.'6'1 / ,.,)59. :ote t#at t#e p+"alue of ,.,)59 is less t#an our acceptable le"el of & pe D error risk of ,.,5. &#erefore we re3ect t#e null # pot#esis and conclude t#at t#e true mean breaking strengt# is less t#an (, pounds. (d! Dnterpret t#e meaning of t#e p+"alue in t#is problem.
Df t#e null # pot#esis were true and t#e true population mean were reall (, pounds, t#e probabilit of obser"ing a sample mean of 69.1 pounds feet or less would be onl ).59*. -or # pot#esis testing purposes, we take t#is to be strong e"idence against t#e null # pot#esis. (e! 0ompare our conclusions in (b! and (c!.
Aanagerial Statistics
161
Prof. Buran
2(. A manufacturer of salad dressings uses mac#ines to dispense li?uid ingredients into bottles t#at mo"e along a filling line. &#e mac#ine t#at dispenses dressings is working properl w#en 6 ounces are dispensed. &#e standard de"iation of t#e process is ,.15 ounce. A sample of 5, bottles is selected periodicall , and t#e filling line is stopped if t#ere is e"idence t#at t#e a"erage amount dispensed is actuall less t#an 6 ounces. Sup+pose t#at t#e a"erage amount dispensed in a particular sample of 5, bottles is (.96) ounces. (a! State t#e null and alternati"e # pot#eses. H,$ HA$ (b!
=6 <6
At t#e ,.,5 le"el of significance, is t#ere e"idence t#at t#e a"erage amount dispensed is less t#an 6 ouncesC ;<press our decision rule in terms of t#e a"erage amount of li?uid dispensed.
X P , = s n
= X P 6 ,.15 5,
1.6'5
1.6'5( ,.,2121) =
X P 6 ( ,.,2121) ,.,2121 = X P 6
=XP =XP
&#e critical "alue is (.9651. Df t#e obser"ed sample mean is less t#an (.9651, t#en we will re3ect t#e null # pot#esis. As it #appens, t#e sample mean is (.96) ounces, so we do not re3ect t#e null # pot#esis. &#e true mean is not significantl less t#an 6 ounces.
Aanagerial Statistics
162
Prof. Buran
(c!
At t#e ,.,5 le"el of significance, using t#e p+"alue approac# to # pot#esis testing, is t#ere e"idence t#at t#e a"erage amount dispensed is less t#an 6 ouncesC
= ,.6,
Kooking at t#e standard normal table, we see t#at t#is @ "alue corresponds to a probabilit of ,.2661. &#is implies a one+tailed p+"alue of ,.5 Q ,.2661 / ,.2119. (d! Dnterpret t#e meaning of t#e p+"alue in t#is problem.
Df t#e null # pot#esis were true and t#e true population mean were reall 6 ounces, t#e probabilit of obser"ing a sample mean of (.96) ounces or smaller would be 21.19*. -or # pot#esis testing purposes, we take t#is to be weak e"idence against t#e null # pot#esis. &#is sample mean would not reall be "er unusual if t#e null # pot#esis were true. (e! 0ompare our conclusions in (b! and (c!.
Aanagerial Statistics
16)
Prof. Buran
26. &#e polic of a particular bank branc# is t#at its A&As must be stocked wit# enoug# cas# to satisf customers making wit#drawals o"er an entire weekend. 0ustomer goodwill depends on suc# ser"ices meeting customer needs. At t#is branc# t#e e<pected (i.e., population! a"erage amount of mone wit#drawn from A&A mac#ines per customer o"er t#e weekend is M16, wit# an e<pected (i.e., population! standard de"iation of M),. Suppose t#at a random sample of )6 customer transactions is e<amined and it is obser"ed t#at t#e sample mean wit#drawal is M1(2. (a! State t#e null and alternati"e # pot#eses. &#is ?uestion is similar to L19 and L2' abo"e, but t#ere is a subtle difference in t#e language t#at suggests an upper tail test instead of a two+tailed test. :ote t#at management seems to be concerned onl wit# t#e possibilit t#at t#e true mean is greater t#an t#e null+# pot#esi@ed "alue of M16,R in t#e pre"ious problems t#ere was some concern about t#e mean being eit#er greater t#an or less t#an M16,. IeSll use an upper tail test in t#is case. H,$ HA$
=16, >16,
(b! At t#e ,.,5 le"el of significance, is t#ere e"idence to belie"e t#at t#e true a"erage wit#drawal is greater t#an M16,C ;<press our decision rule in terms of a"erage dollar "alue per wit#drawal. &#e critical @+"alue corresponding to an upper tail probabilit of ,.,5 is 1.6'5.
1.6'5
X P , = s n
=
=
1.6'5
1.6'5(5)
X P 16, ), )6
X P 16, ( 5) 5 = X P 16,
=XP
=XP
&#e critical "alue is 166.225. Df t#e obser"ed sample mean is greater t#an 166.225, t#en we will re3ect t#e null # pot#esis. Dn t#is case, our sample mean of 1(2 is greater t#en t#e critical "alue, so we re3ect t#e null # pot#esis and conclude t#at t#e true mean is greater t#an M16,.
Aanagerial Statistics
16'
Prof. Buran
(c! At t#e ,.,5 le"el of significance, using t#e p+"alue approac# to # pot#esis testing, is t#ere e"idence to belie"e t#at t#e true a"erage wit#drawal is greater t#an M16,C >ur test statistic is$
z,
= X , s n 1(2 16, = ), )6
= 2 .'
Kooking at t#e standard normal table, we see t#at t#is @ "alue corresponds to a probabilit of ,.'916. &#is implies a one+tailed p+"alue of ,.5 Q ,.'916 / ,.,,62. (d! Dnterpret t#e meaning of t#e p+"alue in t#is problem. Df t#e null # pot#esis were true and t#e true population mean were reall M16,, t#e probabilit of obser"ing a sample mean of M1(2 or greater would be onl ,.62*. -or # pot#esis testing purposes, we take t#is to be strong e"idence against t#e null # pot#esis. (e! 0ompare our conclusions in (b! and (c!. &#e conclusions are t#e same.
Aanagerial Statistics
165
Prof. Buran
29. Df, in a sample of si@e n = 16 selected from an underl ing normal population, t#e sample mean is X =56 and t#e sample standard de"iation is s = 12 , w#at is t#e "alue of t#e t+test statistic if we are testing t#e null # pot#esis H, t#at = 5, C >ur test statistic is$
t,
X , s n 56 5, = 12 16 =
= 2.,
),. Dn L29 abo"e, #ow man degrees of freedom would t#ere be in t#e one+sample t+testC &#e number of degrees of freedom is n Q 1 / 16 Q 1 / 15
Aanagerial Statistics
166
Prof. Buran
)1. Dn L29 and L), abo"e, w#at are t#e critical "alues from t#e t+table if t#e le"el of significance is c#osen to be ,.,5 and t#e alternati"e # pot#esis HA is as follows$ (a! 5, C &#e critical t+"alue corresponding to a two+tailed probabilit of ,.,5 (,.,25 in eac# tail! is T2.1)1. Df t#e absolute "alue of t#e obser"ed test statistic is greater t#an 2.1)1, t#en we will re3ect t#e null # pot#esis. (0ompare t#is "alue to t#e more familiar z+"alue of 1.96 associated wit# t#e same probabilit .! Ie could also get t#e critical "alue of 2.1)1 using t#e ;<cel function$ =TINV(0.05,15)
Aanagerial Statistics
16(
Prof. Buran
(b! > 5, C &#e critical t+"alue corresponding to an upper tail probabilit of ,.,5 is T1.(5). Df t#e absolute "alue of t#e obser"ed test statistic is greater t#an 1.(5), t#en we will re3ect t#e null # pot#esis. (0ompare t#is "alue to t#e more familiar z+"alue of 1.6'5 associated wit# t#e same probabilit .! Ie could also get t#e critical "alue of 1.(5) using t#e ;<cel function$ =TINV(2*(0.05),15)
Aanagerial Statistics
166
Prof. Buran
)2. Dn L29, L),, and L)1 abo"e, w#at is our statistical decision if our alternati"e # pot#esis HA is as follows$ (a!
5, C
2ecall t#at t#e critical t+"alue for t#is two+tailed test is T2.1)1 (in ot#er words, if t#e absolute "alue of t#e obser"ed test statistic is greater t#an 2.1)1, t#en we will re3ect t#e null # pot#esis!. .ere, our test statistic is 2.,, so we do not re3ect t#e null # pot#esis. Ie conclude t#at t#e true mean is not significantl different from 5,.
(b!
> 5, C
:ow we #a"e an upper tail test, in w#ic# t#e entire 5* re3ection region is in t#e upper tail. &#e critical "alue is now 1.(5) (from )1b abo"e!. Since our test statistic is greater t#an t#e critical "alue (i.e. 2., 4 1.(5)! we re3ect t#e null # pot#esis and conclude t#at t#e true mean is significantl greater t#an 5,.
Aanagerial Statistics
169
Prof. Buran
)). &#e manager of t#e credit department for an oil compan would like to determine w#et#er t#e a"erage mont#l balance of credit card #olders is e?ual to M(5. An auditor selects a random sample of 1,, accounts and finds t#at t#e a"erage owed is M6).', wit# a sample standard de"iation of M2).65. (a! =sing t#e ,.,5 le"el of significance, s#ould t#e auditor conclude t#at t#ere is e"idence t#e a"erage balance is different from M(5C H,$ HA$
= (5 (5
>ur decision rule is to re3ect t#e null # pot#esis if t#e absolute "alue of t#e test statistic is greater t#an 1.96. >ur test statistic is$
z,
X , s n 6).' (5 = 2).65 1,, =
= ).55
Since ).55 is greater t#an 1.96, we re3ect t#e null # pot#esis and conclude t#at t#e true population mean is significantl different from M(5. Kooking at t#e standard normal table, we see t#at t#is @ "alue is off t#e c#art. Ie conclude t#at t#e p+"alue is appro<imatel @ero.
Aanagerial Statistics
1(,
Prof. Buran
(b!
= 2.25
&#is test statistic is still in t#e re3ection regionR we conclude again t#at t#e true mean is different from M(5. Kooking at t#e standard normal table, we see t#at t#is @ "alue corresponds to a probabilit of ,.'6(6. &#is implies a two+tailed p+"alue of$ 2P(,.5 Q ,.'6(6! / ,.,2''.
Aanagerial Statistics
1(1
Prof. Buran
(c!
=1.61
&#is test statistic is no longer in t#e re3ection region. Ie conclude t#is time t#at t#e true mean is not significantl different from M(5. Kooking at t#e standard normal table, we see t#at t#is @ "alue corresponds to a probabilit of ,.''6). &#is implies a two+tailed p+"alue of 2P(,.5 Q ,.''6)! / ,.1,('.
Aanagerial Statistics
1(2
Prof. Buran
)'. A manufacturer of detergent claims t#at t#e mean weig#t of a particular bo< of detergent is ).25 pounds. A random sample of 6' bo<es re"eals a sample a"erage of ).2)6 pounds and a sample standard de"iation of ,.11( pound. (a! =sing t#e ,.,1 le"el of significance, is t#ere e"idence t#at t#e a"erage weig#t of t#e bo<es is different from ).25 poundsC H,$ HA$ >ur test statistic is$
z,
= X , s n ).2)6 ).25 = ,.11( 6'
=).25 ).25
= ,.62
&#is test statistic is wit#in t#e non+re3ect region. Ie conclude t#at t#e true population a"erage is not significantl different from ).25. Kooking at t#e standard normal table, we see t#at t#is @ "alue corresponds to a probabilit of ,.29)9. &#is implies a two+tailed p+"alue of 2P(,.5 Q ,.29)9! / ,.'122.
Aanagerial Statistics
1()
Prof. Buran
(b!
= 1.92
&#e absolute "alue of t#is statistic is less t#an t#e critical "alue, so we do not re3ect t#e null # pot#esis. &#e true mean is not significantl different from ).25. Kooking at t#e standard normal table, we see t#at t#is @ "alue corresponds to a probabilit of ,.'(26. &#is implies a two+tailed p+"alue of 2P(,.5 Q ,.'(26! / ,.,5'6.
Aanagerial Statistics
1('
Prof. Buran
(c!
= 2.6(
&#e absolute "alue of t#is statistic is greater t#an our critical "alue, so we re3ect t#e null # pot#esis and conclude t#at t#e true population mean is different from ).25. Kooking at t#e standard normal table, we see t#at t#is @ "alue corresponds to a probabilit of ,.'962. &#is implies a two+tailed p+"alue of ,.5 Q ,.'962 / ,.,,(6.
Aanagerial Statistics
1(5
Prof. Buran
)5. A manufacturer of plastics wants to e"aluate t#e durabilit of rectangularl molded plastic blocks t#at are to be used in furniture. A random sample of 5, suc# blocks is e<amined, and t#e #ardness measurements (in 7rinell units! are recorded as follows$
26).6 269.1 2(9.) 255.) 2().6 2().) 25,.1 252.) 26(.2 26(.( 2(6.6 ),1.6 2(1.( 255.) 26).1 2)6.( 269.2 2)5., 261., 26,.9 ))'.9 2',.6 )1).2 ),2.1 2('.6 ),2.6 26(.5 2((.6 256.) 2((.' 2)9.9 2(9.) 2').6 2))., 2(6.9 25'.6 226.' 295.5 19'.' 259.5 261.9 265.2 2'9.) 291.9 262., 2(,.' 265.9 226.( 26).( 26).5
(a! =sing t#e ,.,5 le"el of significance, is t#ere e"idence t#at t#e a"erage #ardness of t#e plastic blocks e<ceeds 26, (in 7rinell units!C H,$ HA$
= 26, > 26,
Ie #a"e an upper tail test wit# alp#a / ,.,5, so our critical "alue of z is 1.6'5. >ur test statistic is$
z,
= X , s n 26(.6 26, = 2'.2, 5,
= 2.2)
&#is test statistic is in t#e re3ect region. Ie conclude t#at t#e true population a"erage is significantl greater t#an 26,.
Aanagerial Statistics
1(6
Prof. Buran
= STDEVP( B8 :K1 2 ) = COUNT( B8 :K1 2 ) = AVERAGE( B8 :K1 2 ) = ( B4 - B1 ) / ( B2 / SQRT( B3 ) ) = 1 - N ORM SDI ST( B5 )
(b!
Ie assume t#at t#ese 5, data form a representati"e sample of t#e population, suc# t#at X is a good estimate of and t#at s is a good estimate of . Ie also assume, based on t#e 0entral Kimit &#eorem, t#at t#e distribution of sample means is normal. (c! -ind t#e p+"alue and interpret its meaning. Kooking at t#e standard normal table, we see t#at t#is @ "alue corresponds to a probabilit of ,.'6(1. &#is implies a one+tailed p+"alue of ,.5 Q ,.'6(1 / ,.,129. Df t#e null # pot#esis were true and t#e true population mean were reall 26, 7rinell units, t#e probabilit of obser"ing a sample mean of 26(.6 units or greater would be onl 1.26* (note t#at we get a slig#tl more precise estimate of t#e p+"alue from ;<cel t#an from t#e z+table!. -or # pot#esis testing purposes, we take t#is to be strong e"idence against t#e null # pot#esis. (d! I#at will be our answer in (a! if t#e first data "alue is 2)).6 instead of 26).6C Ie c#ange cell 76 in t#e spreads#eet to 2)).6, and t#e sample mean and sample standard de"iation update t#emsel"es. . >ur test statistic is now$
z,
X , s n 266.6 26, = 2'.5 5, =
=1.91
Ie still re3ect t#e null # pot#esis. Kooking at t#e standard normal table, we see t#at t#is @ "alue corresponds to a probabilit of ,.'(19. &#is implies a one+tailed p+"alue of ,.5 Q ,.'(19 / ,.,261.
Aanagerial Statistics
1((
Prof. Buran
)6. Df in a random sample of ',, items, 66 are found to be defecti"e, w#at is t#e sample proportion of defecti"e itemsC
H p
n defective
)(. Dn L )6 abo"e, if it is # pot#esi@ed t#at 2,* of t#e items in t#e population are defecti"e, w#at is t#e "alue of t#e z-test statisticC >ur test statistic is$
z,
p , (1 p , ) n
,.22 ,.2,
H p, p
)6. Dn L)6 and L)( abo"e, suppose ou are testing t#e null # pot#esis H,$ p =,.2, against t#e two+tailed alternati"e # pot#esis HA$ p ,.2, and ou c#oose t#e le"el of significance to be ,.,5. I#at is our statistical decisionC &#e decision rule will be to re3ect t#e null # pot#esis if t#e absolute "alue of t#e test statistic is greater t#an 1.96. >ur test statistic is 1.,,, so we do not re3ect t#e null # pot#esis. Ie conclude t#at t#e true proportion is not significantl different from 2,*. Kooking at t#e standard normal table, we see t#at t#is @ "alue corresponds to a probabilit of ,.)'1). &#is implies a two+tailed p+"alue of 2P(,.5 Q ,.)'1)! / ,.)1('.
Aanagerial Statistics
1(6
Prof. Buran
)9. A tele"ision manufacturer claims in its warrant t#at in t#e past not more t#an 1,* of its tele"ision sets needed an repair during t#eir first 2 ears of operation. &o test t#e "alidit of t#is claim, a go"ernment testing agenc selects a sample of 1,, sets and finds t#at 1' sets re?uired some repair wit#in t#eir first 2 ears of operation. =sing t#e ,.,1 le"el of significance, (a! is t#e manufacturerGs claim "alid or is t#ere e"idence t#at t#e claim is not "alidC H,$ HA$
p =,.1, p > ,.1,
Ie will re3ect t#e null # pot#esis if our test statistic is greater t#an 2.)). >ur test statistic is$
z,
p , (1 p , ) n
,.1' ,.1,
H p, p
Ie conclude t#at t#ere is insufficient e"idence against t#e manufacturerSs claimR we cannot re3ect t#e null # pot#esis at t#e 1* significance le"el. (b! compute t#e p+"alue and interpret its meaning. Kooking at t#e standard normal table, we see t#at t#is @ "alue corresponds to a probabilit of ,.',62. &#is implies a one+tailed p+"alue of ,.5 Q ,.',62 / ,.,916. Df t#e null # pot#esis were true and t#e true population proportion were reall 1,*, t#e probabilit of obser"ing a sample proportion of 1'* or greater would be 9.16*. -or # pot#esis testing purposes, we take t#is to be weak e"idence against t#e null # pot#esis. Dn ot#er words, t#e e"idence is insufficient for us to re3ect t#e null # pot#esis.
Aanagerial Statistics
1(9
Prof. Buran
(c! I#at is our answer in (a! if 16 sets re?uired some repairC :ow our test statistic is$
z,
p , (1 p , ) n
,.16 ,.1,
H p, p
:ow, we re3ect t#e null # pot#esis and conclude t#at t#e true proportion is greater t#an 1,*. Kooking at t#e standard normal table, we see t#at t#is @ "alue corresponds to a probabilit of ,.'962. &#is implies a one+tailed p+"alue of ,.5 Q ,.'962 / ,.,,)6. .ereSs a close+up "iew of t#e e<treme upper tail of t#e distribution of sample proportions, based on t#e assumption t#at t#e null # pot#esis is true. &#e s#aded area represents our alp#a U a 1* probabilit . &#e area to t#e rig#t of t#e test statistic is onl ,.)6*.
Aanagerial Statistics
16,
Prof. Buran
',. &#e personnel director of a large insurance compan is interested in reducing t#e turno"er rate of data processing clerks in t#e first ears of emplo ment. Past records indicate t#at 25* of all new #ires in t#is area are no longer emplo ed at t#e end of 1 ear. ;<tensi"e new training approac#es are implemented for a sample of 15, new data processing clerks. At t#e end of a 1+ ear period, 29 of t#ese 15, indi"iduals are no longer emplo ed. (a! At t#e ,.,1 le"el of significance, is t#ere e"idence t#at t#e proportion of data processing clerks w#o #a"e gone t#roug# t#e new training and are no longer emplo ed is less t#an ,.25C H,$ HA$
p = ,.25 p < ,.25
Ie will re3ect t#e null # pot#esis if our test statistic is less t#an +2.)). >ur sample proportion is 29V15, / ,.19)). &#e test statistic$
z,
p , (1 p , ) n
,.19)) ,.25
H p, p
&#is @+"alue is in t#e non+re3ect region. &#erefore we do not re3ect t#e null # pot#esisR t#e true population turno"er rate is not significantl less t#an 25*.
Aanagerial Statistics
161
Prof. Buran
(b!
-rom t#e standard normal table, we see t#at our @ "alue corresponds to a probabilit of ,.''52. &#is implies a one+tailed p+"alue of ,.5 Q ,.''52 / ,.,5'6. Df t#e null # pot#esis were true and t#e true population proportion were reall 25*, t#e probabilit of obser"ing a sample proportion of 19.))* or smaller would be 5.'6*. &#is is a larger c#ance of & pe D ;rror t#an we are prepared to accept. (Ie want t#e c#ance of a & pe D ;rror to be no more t#an 1*.! (c! I#at is our answer to (a! if 22 of t#e indi"iduals are no longer emplo edC
:ow t#e statistics c#ange, as follows$ :umber :o Konger ;mplo ed Sample Proportion &est Statistic 0onclusion p+Value 7efore 29 ,.19)) +1.6, 1o :ot 2e3ect ,.,5'6 After 22 ,.1'6( +2.92 2e3ect ,.,,16
Aanagerial Statistics
162
Prof. Buran
'1. &#e marketing branc# of t#e Ae<ican &ourist 7ureau would like to increase t#e proportion of tourists w#o purc#ase sil"er 3ewelr w#ile "acationing in Ae<ico from its present estimated "alue of ,.',. &oward t#is end, promotional literature describing bot# t#e beaut and "alue of t#e 3ewelr is prepared and distributed to all passengers on airplanes arri"ing at a certain seaside resort during a 1+week period. A sample of 5,, passengers returning at t#e end of t#e 1+week period is randoml selected, and 22( of t#ese passengers indicate t#at t#e purc#ased sil"er 3ewelr . (a! At t#e ,.,5 le"el of significance, is t#ere e"idence t#at t#e proportion #as increased abo"e t#e pre"ious "alue of ,.',C H,$ HA$
p =,.', p >,.',
Ie will re3ect t#e null # pot#esis if t#e test statistic is greater t#an 1.6'5.
z,
p , (1 p , ) n
H p, p
&#is @+"alue is in t#e re3ection region. &#erefore we re3ect t#e null # pot#esisR t#e population proportion of tourists w#o purc#ase sil"er 3ewelr #as increased.
(b!
>ur test statistic corresponds to a probabilit of ,.'9)1, w#ic# implies a one+tailed p+"alue of ,.,,69. Df t#e null # pot#esis were true and t#e true population proportion were reall ,.',, t#e probabilit of obser"ing a sample proportion of ,.'5' or greater would be onl ,.69*. -or # pot#esis testing purposes, we take t#is to be strong e"idence against t#e null # pot#esis.
Aanagerial Statistics
16)
Prof. Buran
(c!
I#at is our answer to (a! if 21) passengers indicate t#at t#e purc#ased sil"er 3ewelr C
:ow t#e sample proportion is ,.'26, t#e test statistic is 1.19, and t#e p+"alue is ,.11(,. &#is time, we do not re3ect t#e null # pot#esis.
Aanagerial Statistics
16'
Prof. Buran
'2. &#e operations manager at a lig#t bulb factor wants to determine if t#ere is an difference in t#e a"erage life e<pectanc of bulbs manufactured on two different t pes of mac#ines. &#e process standard de"iation of Aac#ine D is 11, #ours and of Aac#ine DD is 125 #ours. A random sample of 25 lig#t bulbs obtained from Aac#ine D indicates a sample mean of )(5 #ours, and a similar sample of 25 from Aac#ine DD indicates a sample mean of )62 #ours. (a! =sing t#e ,.,5 le"el of significance, is t#ere an e"idence of a difference in t#e a"erage life of bulbs produced b t#e two t pes of mac#inesC H,$ HA$
X Y = , X Y ,
IeSll re3ect t#e null # pot#esis if t#e absolute "alue of our test statistic is greater t#an t(2', ,.,25! / 2.,6'. >ur test statistic is$
t,
(X Y ) ,
2 sX s2 + Y nX nY ( )(5 )62 ) ,
&#is test statistic is in t#e non+re3ect range. Ie conclude t#at t#ere is no significant difference between t#e a"erage life e<pectancies of bulbs from t#e two mac#ines.
Aanagerial Statistics
165
Prof. Buran
(b!
Kooking at t#e standard normal table, we see t#at t#is t "alue corresponds to a probabilit of about ,.151( (using z as an appro<imation of t!. &#is implies a two+tailed p+"alue of 2P(,.5 Q ,.151(! / ,.696). Df t#e null # pot#esis were true and t#ere were no difference between t#e two population means, t#e probabilit of obser"ing two samples wit# a difference of at least ()(5 Q )62! / 1) #ours would be 69.6)*. Ie take t#is to be weak e"idence against t#e null # pot#esis. (Dn fact, it is mildl surprising t#at t#ese two sample means are so close to eac# ot#erJ! (:ote t#at t#ere is a more complicated "ersion of t#is test, called t#e pooled+"ariance t+test, w#ic# is more statisticall "alid but be ond t#e scope of t#is course. Dt is demonstrated in Problem '9 below. Dn t#is case, t#e ot#er met#od would not ield a different conclusion.!
Aanagerial Statistics
166
Prof. Buran
'). &#e purc#asing director for an industrial parts factor is in"estigating t#e possibilit of purc#asing a new t pe of milling mac#ine. S#e determines t#at t#e new mac#ine will be boug#t if t#ere is e"idence t#at t#e parts produced #a"e a #ig#er a"erage breaking strengt# t#an t#ose from t#e old mac#ine. A sample of 1,, parts taken from t#e old mac#ine indicated a sample mean of 65 kilograms and a sample standard de"iation of 1, kilograms, w#ereas a similar sample of 1,, from t#e new mac#ine indicated a sample mean of (2 kilograms and a sample standard de"iation of 9 kilograms. (a! =sing t#e ,.,1 le"el of significance, is t#ere e"idence t#at t#e purc#asing director s#ould bu t#e new mac#ineC H,$ HA$
X Y = , X Y < ,
&#is is a lower tailed test, because we want to see if t#ere is e"idence t#at t#e parts produced on t#e new mac#ine #a"e a #ig#er a"erage breaking strengt# t#an t#ose from t#e old mac#ine. Ie let t#e random "ariable W represent t#e breaking strengt# of parts produced on t#e old mac#ine and E represent t#e corresponding measurements from t#e new mac#ine. Df t#e difference X Y turns out to be significantl less t#an @ero, t#en it suggests t#at t#e parts from t#e new mac#ine are in fact stronger. >ur critical @+"alue will be Q2.)). Df t#e test statistic is lower t#an t#at, t#en we will re3ect t#e null # pot#esis. >ur test statistic is$
z,
(X Y ) ,
2 sX s2 + Y nX nY ( 65 (2 ) ,
Ie re3ect t#e null # pot#esis and conclude t#at t#e parts from t#e new mac#ine #a"e a #ig#er a"erage breaking strengt#. (Dn ot#er words, we will recommend bu ing t#e new mac#ine.!
Aanagerial Statistics
16(
Prof. Buran
(b!
Kooking at t#e standard normal table, we see t#at t#is @ "alue is off t#e c#art.. &#is implies a one+ tailed p+"alue of appro<imatel @ero. =sing t#is ;<cel function, we can calculate it to be appro<imatel ,.,,,,,,1$ =1-NORMSDIST(!"S(-5.2031)) Df t#e null # pot#esis were true and t#e true difference between t#e population means were reall @ero, t#e e"ent of obser"ing two samples wit# means t#is far apart would be nearl impossible. -or # pot#esis testing purposes, we take t#is to be strong e"idence against t#e null # pot#esis.
Aanagerial Statistics
166
Prof. Buran
''. Dn intaglio printing, a design or figure is car"ed beneat# t#e surface of #ard metal or stone. Suppose t#at an e<periment is designed to compare differences in a"erage surface #ardness of steel plates used in intaglio printing (measured in indentation numbers! based on two different surface conditions U untreated "ersus treated b lig#t polis#ing wit# emer paper. Dn t#e e<periment ', steel plates are randoml assigned, 2, t#at are untreated and 2, t#at are lig#tl polis#ed. =se a ,.,5 le"el of significance to determine w#et#er t#ere is e"idence of a significant treatment effect (i.e., a significant difference in a"erage surface #ardness between t#e untreated and t#e polis#ed steel plates! if t#e sample mean for t#e untreated plates is 16).' and t#e sample mean for t#e polis#ed plates is 156.9. &#e sample standard de"iation in surface #ardness (in indentation numbers! was 1,.2 for t#e untreated surfaces and 6.' for t#e lig#tl polis#ed surfaces. H,$ HA$
X Y = , X Y ,
&#is is a two+tailed test, because t#ere is no t#eori@ed direction to t#e difference between t#e two means. =sing t#e most conser"ati"e test (19 degrees of freedom!, our critical t+"alue will be 2.,9)R if t#e absolute "alue of our test statistic is greater t#an 2.,9), t#en we will re3ect t#e null # pot#esis. >ur test statistic is$
t,
(X Y ) ,
2 sX s2 + Y nX nY ( 16).' 156.9 ) ,
&#is is greater t#an t#e critical "alue, so we re3ect t#e null # pot#esis and conclude t#at t#ere is a significant difference in a"erage surface #ardness between plates t#at are untreated and t#ose t#at #a"e been treated wit# emer paper. Ie can appro<imate t#e p+"alue for t#is test using z instead of t. Kooking at t#e standard normal table, we see t#at t#is z "alue corresponds to a probabilit of ,.'92,. &#is implies a two+tailed p+ "alue of 2P(,.5 Q ,.'92,! / ,.,16,.
Aanagerial Statistics
169
Prof. Buran
'5. A real estate agenc wants to compare t#e appraised "alues of single+famil #omes in two :assau 0ount , :ew Eork, communities. A sample of 6, listings in -armingdale and 99 listings in Ke"ittown ields t#e following results (in t#ousands of dollars!$
X
S n
(a! At t#e ,.,5 le"el of significance, is t#ere e"idence of a difference in t#e a"erage appraised "alues for single+famil #omes in t#e two :assau 0ount communitiesC H,$ HA$
X Y = , X Y ,
&#is is a two+tailed test, because t#ere is no t#eori@ed direction to t#e difference between t#e two means. >ur critical @+"alue will be 1.96R if t#e absolute "alue of our test statistic is greater t#an 1.96, t#en we will re3ect t#e null # pot#esis. >ur test statistic is$
z,
(X Y ) ,
2 sX s2 + Y nX nY ( 191.)) 1(2.)' ) ,
&#is is greater t#an t#e critical "alue, so we re3ect t#e null # pot#esis and conclude t#at t#ere is a significant difference in a"erage appraised real estate "alues between t#ese two communities. &#e p+"alue is appro<imatel @ero. (b! 1o ou t#ink an of t#e assumptions needed in (a! #a"e been "iolatedC ;<plain. -or t#is t pe of test, we assume t#at t#e two population "ariances are e?ual (in fact, t#e null # pot#esis assumes t#at t#ere is reall onl one population!. .ere t#ere is some e"idence t#at t#e two townsS appraisals #a"e significantl different standard de"iations.1
&#ere is a "ersion of t#e F+test t#at can be used to test t#e null # pot#esis t#at two populations #a"e t#e same "ariance, but t#at test is be ond t#e scope of t#is course. &#e test is built into ;<cel (&ools Q 1ata Anal sis Q F+&est &wo Sample for Variances!, but it re?uires all of t#e raw data, not 3ust summar statistics as we #a"e #ere. Aanagerial Statistics 19, Prof. Buran
'6. &#e manager of a nationall known real estate agenc #as 3ust completed a training session on appraisals for two newl #ired agents. &o e"aluate t#e effecti"eness of #is training, t#e manager wis#es to determine w#et#er t#ere is an difference in t#e appraised "alues placed on #ouses b t#ese two different indi"iduals. A sample of 12 #ouses is selected b t#e manager, and eac# agent is assigned t#e task of placing an appraised "alue (in t#ousands of dollars! on t#e 12 #ouses. &#e results are as follows$
.>=S; 1 2 ) ' 5 6 ( 6 9 1, 11 12 A8;:& 1 161., 1(9.9 16)., 216., 21)., 1(5., 21(.9 151., 16'.9 192.5 225., 1((.5 A8;:& 2 162., 16,., 161.5 215., 216.5 1(5., 219.5 15,., 165.5 195., 222.( 1(6.,
(a! At t#e ,.,5 le"el of significance, is t#ere e"idence of a difference in t#e a"erage appraised "alues gi"en b t#e two agentsC &#is is a two+tailed test, because t#ere is no t#eori@ed direction to t#e difference between t#e two means. As it #appens, we #a"e a good situation for using t#e matc#ed pairs met#od, because eac# #ouse was appraised b eac# agent. Ket t#e true population difference between t#ese two means be represented b
D = X Y
H,$ HA$
D = , D ,
Ie #a"e a small+sample test, so weSll get our critical "alue from t#e t+table. Ie #a"e two tails, alp#a / ,.,5 (,.,25 in eac# tail!, and 11 degrees of freedom, so t#e critical t+"alue will be 2.2,1. Df t#e absolute "alue of our test statistic is greater t#an 2.2,1, t#en we will re3ect t#e null # pot#esis.
Aanagerial Statistics
191
Prof. Buran
= ,.),
Ie do not re3ect t#e null # pot#esis. &#ere is no significant difference between t#e a"erage appraisals made b t#ese two agents. (b! I#at assumption is necessar to perform t#is testC Ie #a"e assumed t#at t#e underl ing distributions of appraisals from eac# agent are normal, and t#at t#e #a"e e?ual "ariances. Ie also assume t#at t#ese twel"e #ouses are representati"e of t#e population of interest. (c! -ind t#e p+"alue in (a! and interpret its meaning. &#ere are t#ree approac#es to getting a p+"alue for t#is problem. -irst, we can use t#e z+table as an appro<imation, knowing t#at it will alwa s under+estimate t#e p+"alue for a t+test. Dn t#e z+ table we get a "alue of ,.11(9, w#ic# implies a two+tailed p+"alue of 2P(,.5 Q ,.11(9! / ,.(6'2 Anot#er wa to go is to use a special t+table, suc# as #ttp$VVwww.columbia.eduVd311'Vt+table2.doc =sing t#at table, we get 2P(,.)6'9! / ,.(696. -inall , we could use t#e ;<cel function =TDIST(0.3,11,2). &#at also ields ,.(696. (&#is is not surprising, because t#at is e<actl #ow t#e special t+table was created.! Actuall , our test statistic is more like +,.),)195, and a more precise estimate of t#e p+"alue is ,.(6('. Dn an case, all t#ree of t#ese "arious p+"alues are fairl close toget#er, and t#e all carr t#e same implication. Df t#e null # pot#esis were true and t#ere reall were no difference between t#e population means, t#en t#e e"ent of #a"ing two sample means t#is far apart or fart#er would be (6.('*. &#is is "er weak e"idence against t#e null # pot#esis. Dn fact, it is mildl surprising to see t#e difference so close to @ero.
Aanagerial Statistics
192
Prof. Buran
:otes$ (1! Dn cell 7' we #a"e made it eas to c#ange from a one+tail to a two+tail test, simpl b c#anging t#e "alue in cell 72 from a one to a two. (2! Dn cell ;) we need to use t#e absolute "alue function to deal wit# t#e negati"e test statistic.
'(. A few engineering students decide to see w#et#er cars t#at supposedl do not need #ig#+ octane gasoline get more miles per gallon using regular or #ig#+octane gas. &#e test se"eral cars (under similar road surface, weat#er, and ot#er dri"ing conditions!, using bot# t pes of gas in eac# car at different times. &#e mileage for eac# gas t pe for eac# car is as follows$
#!$ %&'( 2egular .ig#+octane 1 15 16 2 2) 21 3 21 25 4 )5 )' 5 '2 '( !" 6 26 ), 7 19 19 8 )2 2( 9 )1 )' 10 2' 2,
(a!
Ds t#ere an e"idence of a difference in t#e a"erage gasoline mileage between regular and #ig#+octane gasC (=se / ,.,5.!
&#is is a two+tailed test, because t#ere is no t#eori@ed direction to t#e difference between t#e two means. >nce again, we #a"e a good situation for using t#e matc#ed pairs met#od, because eac# car was tested wit# eac# t pe of fuel. Ket t#e true population difference between t#ese two means be represented b
D = X Y
H,$ HA$
D = , D ,
Ie #a"e two tails, alp#a / ,.,5 (,.,25 in eac# tail!, and 9 degrees of freedom, so t#e critical t+ "alue will be 2.262. Df t#e absolute "alue of our test statistic is greater t#an 2.262, t#en we will re3ect t#e null # pot#esis.
Aanagerial Statistics
19)
Prof. Buran
= ,.'596
Ie do not re3ect t#e null # pot#esis. &#ere is no significant difference between t#e a"erage gas mileage ac#ie"ed using t#e two t pes of fuel. (b! I#at assumption is necessar to perform t#is testC
Ie #a"e assumed t#at t#e underl ing distributions of gas mileage results from eac# t pe of fuel are normal, and t#at t#e #a"e e?ual "ariances. Ie also assume t#at t#ese ten cars are representati"e of t#e population of interest. Dn particular, t#e assumption of normalit mig#t be unreasonable #ere, especiall in t#e case of t#e #ig#+octane sample2$
%&'()*+,-:.M&/0,*0.A12&0304.5&(2.%&*2.O1(,60.F70/
6
F+097061:
0 0 10 20 30 40 50 60
M&/0'.80+.G,//)6
&#ere are more sop#isticated statistical # pot#esis tests to see w#et#er data #a"e come from a normal distribution (or an ot#er distribution, for t#at matter!. >ne suc# test is t#e c#i+s?uare test, w#ic# is be ond t#e scope of t#is class. -or our purposes, we can use a #istogram to perform an Xe eballY test to see if it is reasonable to assume t#at t#ese data came from a normal distribution. Aanagerial Statistics 19' Prof. Buran
(c!
=sing t#e ;<cel function =TDIST(0.459#,9,2), we get a p+"alue of ,.656(. Df t#e null # pot#esis were true and t#ere reall were no difference between t#e population means, t#en t#e e"ent of #a"ing two sample means t#is far apart or fart#er would be 65.6(*. &#is is "er weak e"idence against t#e null # pot#esis.
'6. Dn order to measure t#e effect of a storewide sales campaign on nonsale items, t#e researc# director of a national supermarket c#ain took a random sample of 1) pairs of stores t#at were matc#ed according to a"erage weekl sales "olume. >ne store of eac# pair (t#e e<perimental group! was e<posed to t#e sales campaign, and t#e ot#er member of t#e pair (t#e control group! was not. &#e following data indicate t#e results o"er a weekl period$
S&>2; 1 2 ) ' 5 6 ( 6 9 1, 11 12 1) ID&. SAK;S 0AAPAD8: 6(.2 59.' 6,.1 '(.6 9(.6 )6.' 5(.) (5.2 9'.( 6'.) )1.( '9.) 5'., ID&.>=& SAK;S 0AAPAD8: 65.) 5'.( 61.) )9.6 92.5 )(.9 52.' 69.9 69., 56.' ))., '1.( 5).6
(a!
At t#e ,.,5 le"el of significance, can t#e researc# director conclude t#at t#ere is e"idence t#e sales campaign #as increased t#e a"erage sales of nonsale itemsC
&#is is a one+tailed test, because it is proposed t#at per#aps t#e sales campaign #as some positi"e effect on sales. &#ere is some ?uestion #ere as to w#at is t#e proper statistical test. >ne possibilit would be to use t#e independent samples met#od, on t#e assumption t#at t#ere is no logical one+to+one relations#ip between t#e members of t#e two groups. (i.e. Store 1 in t#e campaign group #as no more in common wit# Store 1 in t#e control group t#an it #as wit# an ot#er store in t#e control group!. Alternati"el , we could use t#e mat)hed pairs met#od, arguing t#at t#e 1) pairs of stores were matc#ed according to a"erage weekl sales "olume, and t#erefore do represent a one+to+one relations#ip. IeSll do bot# met#ods #ere, to illustrate an differences in conclusions t#at mig#t result from our c#oice of met#ods.
Aanagerial Statistics
195
Prof. Buran
-irst, t#e independent samples met#od$ Ket t#e true population mean sales wit# t#e campaign be represented b W, and t#e population mean sales wit#out t#e campaign be represented b E. H,$ X Y = , HA$ X Y > , Ie #a"e a small+sample test, so weSll get our critical "alue from t#e t+table. Ie #a"e one tail, alp#a / ,.,5, and 12 degrees of freedom, so t#e critical t+"alue will be 1.(62. Df t#e "alue of our test statistic is greater t#an 1.(62, t#en we will re3ect t#e null # pot#esis. >ur test statistic is$
t,
=
( X Y ) ,
( 62.65 59.19) ,
2 2 sX sY + nX nY
Ie do not re3ect t#e null # pot#esis. &#e campaign made no significant difference in sales.
Aanagerial Statistics
196
Prof. Buran
(b!
Ie #a"e assumed t#at t#e underl ing distributions of sales in eac# population are normal, and t#at t#e #a"e e?ual "ariances. Ie also assume t#at t#ese sets of t#irteen stores are representati"e of t#e population of interest. .ereSs a #istogram of t#e more skewed of t#e two distributions, t#e X:o 0ampaignY sample$
%&'()*+,-:.S,/0'.D&'(+&;7(&)6.(N).C,-8,&*6)
7 6 5
F+097061:
DtSs #ard to be confident t#at t#is sample came from a normal distribution, alt#oug# it isnSt as ob"iousl non+normal as t#e pre"ious e<ample in Problem '(. (c! -ind t#e p+"alue in (a! and interpret its meaning.
Kooking at t#e standard normal table, we see t#at t#is t+"alue is corresponds to a probabilit of ,.16,6 in t#e standard normal table. &#is implies a one+tailed p+"alue of at least ,.5 Q ,.16,6 / ,.)192. Anot#er approac# to estimating t#e p+"alue would be to use a special t+table, suc# as #ttp$VVwww.columbia.eduVZd311'Vt+table2.doc, w#ic# ields a p+"alue of somew#ere between ,.)1)1 and ,.)'61. A t#ird, more precise met#od is to use t#e ;<cel function =TDIST(0.4714,12,1), w#ic# gi"es us ,.)229. Df t#e null # pot#esis were true and t#e difference between t#e true population means were reall ,, t#e probabilit of obser"ing two sample means wit# a difference t#is large or larger would be )2.29*. &#is is weak e"idence against t#e null # pot#esis.
Aanagerial Statistics
19(
Prof. Buran
:ow, t#e mat)hed pairs met#od$ Ket t#e true difference between t#e population means be represented b 1. H,$ D = , HA$ D > , &#e critical t+"alue will still be 1.(62. Df t#e "alue of our test statistic is greater t#an 1.(62, t#en we will re3ect t#e null # pot#esis. >ur test statistic is$
t,
&#is time we do re3ect t#e null # pot#esis, and conclude t#at t#e campaign actuall did #a"e a significant positi"e effect on sales.
Aanagerial Statistics
196
Prof. Buran
(b!
Ie #a"e assumed t#at t#e underl ing distribution of sample mean differences ( D ! is normal, and t#at t#e pairs of stores are representati"e of t#e population of interest. &#e assumption of normalit is not well supported b t#e #istogram (see below!.
%&'()*+,-:.D&<<0+0610'.B0(5006.S,/0'.,(.P,&+'.)<.S()+0'.
5
F+097061:
0 '2 0 2 4 6 8 10 12
S,-8/0.A30+,*0.D&<<0+0610
&#e most critical assumption #ere is t#at t#e pairs reall are pairs, in t#e sense t#at an difference between t#e two storesS sales in eac# pair can be attributed to t#e difference between t#e campaign and no campaign. (c! -ind t#e p+"alue and interpret its meaning.
Kooking at t#e standard normal table (or t#e special t+table for t#at matter!, we see t#at t#is t+ "alue of '.1' is so large t#at it is off t#e table. Ie conclude t#at t#e p+"alue is "er close to @ero. &#e ;<cel function =TDIST(4.14,12,1) gi"es us a one+tailed p+"alue of ,.,,,(. Df t#e null # pot#esis were true and t#e difference between t#e true population means were reall ,, t#e probabilit of obser"ing sample differences wit# an a"erage t#is large or larger would onl be ,.,(*. &#is is "er strong e"idence against t#e null # pot#esis. Dn comparing t#e results from t#e two met#ods, notice #ow t#e superior statistical power of t#e matc#ed pairs tec#ni?ue allows us to make inferences t#at were not possible w#en we used t#e independent samples met#od.
Aanagerial Statistics
199
Prof. Buran
'9. &#e 2 [ A department store #as two c#arge plans a"ailable for its credit+account customers. &#e management of t#e store wis#es to collect information about eac# plan and to stud t#e differences between t#e two plans. Dt is interested in t#e a"erage mont#l balance. 2andom samples of 25 accounts of Plan A and 5, accounts of Plan 7 are selected wit# t#e following results$ PKA: A
n A = 25
A =M(5
PKA: 7
n B = 5,
B =M11,
s A = M15
s B = M1'.1'
=se statistical inference (confidence inter"als or tests of # pot#eses! to draw conclusions about eac# of t#e following$ :ote$ =se a le"el of significance of ,.,1 (or use 99* confidence! t#roug#out. (a! A"erage mont#l balance of all Plan 7 accounts. 0onfidence Dnter"al
= B z sB nB
1'.1' 5,
=11, 2.5(5
>r (b!
(1,'.65,
=11, 5.15
115.15)
Ds t#ere e"idence t#at t#e a"erage mont#l balance of Plan A accounts is different from M1,5C H,$ HA$
A = 1,5 A 1,5
1ecision rule$ 7ased on t#e two+tailed critical "alues in t#e t+table, weSll re3ect t#e null # pot#esis if t#e absolute "alue of t#e test statistic t, is greater t#an 2.(9(. >ur test statistic is$
t,
=
=
A , sA nA
(5 1,5 15 25
= 1,
Ie definitel re3ect t#is null # pot#esis. &#ere is almost no c#ance t#at we would see a sample mean t#is far awa from M1,5 if t#e true mean were reall M1,5. Ie conclude t#at t#e true a"erage mont#l balance of Plan A accounts is significantl different from M1,5.
Aanagerial Statistics
2,,
Prof. Buran
(c!
Ds t#ere e"idence of a difference in t#e a"erage mont#l balances between Plan A and Plan 7C
Ket t#e true population mean mont#l balance for Plan A be represented b A, and t#e population mean mont#l balance for Plan 7 be represented b 7. H,$ A B = , HA$ A B , &#is is an ambiguous situation, because we are used to #a"ing large samples w#en we perform t#e independent samples test for t#e difference between two means. .ere we #a"e on sample t#at is not ?uite large enoug# b statistical con"ention (Plan ASs sample #as onl 25 obser"ations!. IeSll use t#is problem to illustrate two possible testing met#ods$ t#e familiar independent samples test, and t#e more correct pooled-varian)e t-test. -irst, t#e independent samples test. Dt isnSt clear #ere w#et#er t#e critical "alue oug#t to come from t#e t+table or t#e z+table, because one sample #as a XlargeY si@e and t#e ot#er is XsmallY. Ie will go wit# t#e more conser"ati"e approac# #ere and get our critical "alue from t#e t+table. Ie #a"e two tails, alp#a / ,.,1, and at least 2' degrees of freedom, so t#e most conser"ati"e critical t+"alue will be 2.(9(. Df t#e absolute "alue of our test statistic is greater t#an 2.(9(, t#en we will re3ect t#e null # pot#esis. >ur test statistic is$
t,
( A B) ,
( (5 11,) ,
s2 s2 A + B nA nB
15 2 1'.1' 2 + 25 5, = 9.(1
Ie re3ect t#e null # pot#esis. &#ere is a significant difference between t#e two population means. (d! 1etermine t#e p+"alues in (b! and (c! and interpret t#eir meaning.
Dn bot# cases t#e p+"alue is appro<imatel @ero. Df t#e null # pot#eses were true, t#ere would be "irtuall no c#ance of seeing samples w#ose means were t#is far apart.
Aanagerial Statistics
2,1
Prof. Buran
(e!
>n t#e basis of t#e results of (a!, (b!, (c!, and (d!, w#at will ou tell t#e management about t#e two plansC
&#ere is "er compelling e"idence to indicate t#at t#ere is a significant difference between t#e a"erage mont#l balances for t#e two plans. D would tell t#em t#at Plan 7 customers carr significantl #ig#er mont#l balances t#an Plan A customers. (:ote t#at tec#nicall we ne"er actuall tested w#et#er Plan 7 balances were greater t#an Plan A balances U we onl asked w#et#er t#e were different from eac# ot#er U but t#at is an ob"ious conclusion from our anal sis.!
Aanagerial Statistics
2,2
Prof. Buran
A B = , A B ,
Ie will get our critical "alue from t#e t+table. Ie #a"e two tails, / ,.,1, and nW F nE + 2 / () degrees of freedom, so t#e appropriate critical t+"alue will be t ( 2 ,n A +n B 2 ) , w#ic# we can get using ;<cel$ TINV(0.01,(25$50)-2) = 2.#45 Df t#e absolute "alue of our test statistic is greater t#an 2.6'5, t#en we will re3ect t#e null # pot#esis. (Eou could make a reasonable argument for using t#e analogous z+"alue of 2.5(5 #ere, but weSll stick wit# t.! >ur test statistic is$
t,
=
( A B) ,
1 2 1 Sp n + n B A
2 ( n A 1) s 2 1 A + ( n B 1) s B 1 n + n ( n 1) + ( n 1) A B B A
( A B) ,
( (5 11, ) , ( 25 1 ) 15 2 + ( 5, 1) 1'.1' 2 1 1 + ( 25 1) + ( 5, 1) 25 5,
)5 ).5)' = 9.9,
>ur conclusion #ere is prett muc# identical to t#at using t#e ot#er met#od. :ote, #owe"er, t#at we #a"e a more powerful test, statisticall speaking. &#e difference #ere is minimal, because (a! t#e difference between t#ese sample means was in fact ?uite large, and (b! t#e samples were bot# fairl large, e"en if one of t#em was less t#an ),. Df t#ese two conditions were not true, t#en t#e pooled "ariance test mig#t actuall allow for re3ecting H, w#en t#e standard test mig#t not.
Aanagerial Statistics
2,)
Prof. Buran
5,. &#e manager of computer operations of a large compan wis#es to stud computer usage of two departments wit#in t#e compan , t#e accounting department and t#e researc# department. A random sample of fi"e 3obs from t#e accounting department in t#e last week and si< 3obs from t#e researc# department in t#e last week are selected, and t#e processing time (in seconds! for eac# 3ob is recorded.
1epartment Accounting 2esearc# 9 ' Processing &ime (Dn Seconds! ) 6 ( 12 1) 1, 9 9
0#oosing a le"el of significance of ,.,5 (or 95* confidence!, use statistical inference (confidence inter"als or tests of # pot#eses! to draw conclusions about eac# of t#e following$ (a! A"erage processing time for all 3obs in t#e accounting department. 0onfidence Dnter"al
= X t ( n 1, 2 ) s n ).2(1 = (.6 2.((6 5
= (.6 '.,6
>r ().(', 11.66,! (b! Ds t#ere e"idence t#at t#e a"erage processing time in t#e researc# department is greater t#an 6 secondsC H,$ HA$
=6 >6
IeSll #a"e an upper+tail test, wit# a critical "alue of t(n+1, ! / t(5, ,.,5! / 2.,15. >ur decision rule will be to re3ect t#e null # pot#esis if t#e test statistic is greater t#an 2.,15. >ur test statistic is$
t,
= Y , s n 6.5 6 = ).1'6 6
=1.95
Ie cannot re3ect t#e null # pot#esisR t#ere is insufficient e"idence to conclude t#at t#e mean is greater t#an 6 seconds.
Aanagerial Statistics
2,'
Prof. Buran
(c!
Ds t#ere e"idence of a difference in t#e mean processing time between t#e accounting department and t#e researc# departmentC H,$ HA$
X Y = , X Y ,
&#is situation re?uires t#e pooled "ariance t+test, because we #a"e suc# small samples and t#ere is no basis for a matc#ed+pairs test. Ie will get our critical "alue from t#e t+table. Ie #a"e two tails, alp#a / ,.,5, and nW F nE + 2 / 9 degrees of freedom, so t#e appropriate critical t+"alue will be 2.262. Df t#e absolute "alue of our test statistic is greater t#an 2.262, t#en we will re3ect t#e null # pot#esis. >ur test statistic is$
t,
=
(X Y ) ,
1 2 1 Sp n + n Y X
2 2 ( n x 1) s X + ( n Y 1) s Y ( n 1) + ( n 1) x Y
(X Y ) ,
1 1 + n n Y X
Ie cannot re3ect t#e null # pot#esis. &#ere is no significant difference between t#e two population means. (d! I#at must ou assume to do in (c!C
Ie #a"e assumed t#at t#e two populations not onl #a"e t#e same mean (on t#e basis of t#e null # pot#esis!, but t#at t#e also #a"e t#e same "ariance. Dn t#is case, t#e common "ariance assumption mig#t be reasonable, considering t#at t#e two sample standard de"iations are similar.
Aanagerial Statistics
2,5
Prof. Buran
(e!
1etermine t#e p+"alues in (b! and (c!, and interpret t#eir meanings.
-or (b!, using t#e ;<cel function =TDIST(1.95, 5, 1), we see t#at t#e 1+tailed p+"alue is appro<imatel ,.,55. Df t#e null # pot#esis were true and t#e true population mean were reall 6 seconds, t#e probabilit of obser"ing a sample mean of 6.5 seconds or greater would be 5.5*. -or # pot#esis testing purposes, we take t#is to be weak e"idence against t#e null # pot#esisR it is slig#tl more risk of a & pe D error t#an we are prepared to accept. -or (c!, using t#e ;<cel function =TDIST(!"S(-0.3#1),((5$#)-2),2), we see t#at t#e 2+tailed p+"alue is appro<imatel ,.(26'. Df t#is null # pot#esis were true and t#e difference between t#e true population means were reall @ero, t#e probabilit of obser"ing two sample means w#ose difference was ,.)61 seconds or greater would be (2.6'*. -or # pot#esis testing purposes, we take t#is to be "er weak e"idence against t#e null # pot#esisR it is not at all surprising to see t#at t#e two sample means are t#is far apart, e"en if t#e come from t#e same underl ing distribution.
Aanagerial Statistics
2,6
Prof. Buran