business statistics (bk/iba) tutorial 3 full...

19
BS 1 Tutorial 3 Business Statistics (BK/IBA) Tutorial 3 โ€“ Full solutions Instruction In a tutorial session of 2 hours, we will obviously not be able to discuss all questions. Therefore, the following procedure applies: โ€ข we expect students to prepare all exercises in advance; โ€ข we will discuss only a selection of exercises; โ€ข exercises that were not discussed during class are nevertheless part of the course; โ€ข students can indicate their wish list of exercises to be discussed during the session; โ€ข teachers may invite students to answer questions, orally or on the blackboard. We further understand that your time is limited, and in particular that your time between lecture and tutorial may be limited. In case you have no time to prepare everything, we kindly advise you to give priority to the exercises that are indicated with the icon. This does not mean that the other questions are not relevant! 5A : estimates, confidence intervals, and tests Q1 (based on Doane & Seward, 4/E, 8.57) The weights of 20 oranges (in ounces) are shown below. (Data are from a project by statistics student Julie Gillman.) 5.50 6.25 6.25 6.50 6.50 7.00 7.00 7.00 7.50 7.50 7.75 8.00 8.00 8.50 8.50 9.00 9.00 9.25 10.00 10.50 a. Check that ฬ… = 7.7750 and = 1.3325. b. Construct a 95 percent confidence interval for the population standard deviation. Note: Scale was only accurate to the nearest 1/4 ounce. Sol (โˆ’1) 2 2 < 2 < (โˆ’1) 2 2 so (20โˆ’1)1.3325 2 32.852 < 2 < (20โˆ’1)1.3325 2 8.907 so 1.0269 < 2 < 3.788 Extra Take the square root of the lower and upper CI values given to get the CI for the standard deviation of the population. Q2 (Doane & Seward, 4/E, Minicase 9.63) A sample of size = 19 has variance 2 = 1.96. At = .05 in a right-tailed test, does this sample contradict the hypothesis that 2 = 1.21? Sol (i) 0 : 2 โ‰ค 1.21; 1 : 2 > 1.21; = 5% (ii) Sample statistic: 2 ; reject for large values (iii) Distribution test statistic under 0 : (โˆ’1) 2 2 ~ 2 ( โˆ’ 1) Requirements: population must be normally distributed (iv) Calculated test statistic: 2 = 29.16 Critical value: 2 (18; 0.05) = 28.87 Q1 1.0269 < 2 < 3.788 Q2 reject 0 (there is reason to believe that the variance is larger than 1.21)

Upload: votuong

Post on 17-Sep-2018

213 views

Category:

Documents


0 download

TRANSCRIPT

BS 1 Tutorial 3

Business Statistics (BK/IBA)

Tutorial 3 โ€“ Full solutions

Instruction

In a tutorial session of 2 hours, we will obviously not be able to discuss all questions. Therefore, the

following procedure applies:

โ€ข we expect students to prepare all exercises in advance;

โ€ข we will discuss only a selection of exercises;

โ€ข exercises that were not discussed during class are nevertheless part of the course;

โ€ข students can indicate their wish list of exercises to be discussed during the session;

โ€ข teachers may invite students to answer questions, orally or on the blackboard.

We further understand that your time is limited, and in particular that your time between lecture and

tutorial may be limited. In case you have no time to prepare everything, we kindly advise you to give

priority to the exercises that are indicated with the icon. This does not mean that the other questions

are not relevant!

5A ๐ˆ๐Ÿ: estimates, confidence intervals, and tests

Q1 (based on Doane & Seward, 4/E, 8.57)

The weights of 20 oranges (in ounces) are shown below.

(Data are from a project by statistics student Julie Gillman.)

5.50 6.25 6.25 6.50 6.50 7.00 7.00 7.00 7.50 7.50

7.75 8.00 8.00 8.50 8.50 9.00 9.00 9.25 10.00 10.50

a. Check that ๏ฟฝ๏ฟฝ = 7.7750 and ๐‘  = 1.3325.

b. Construct a 95 percent confidence interval for the population standard deviation. Note: Scale

was only accurate to the nearest 1/4 ounce.

Sol (๐‘›โˆ’1)๐‘ 2

๐œ’๐‘ข2 < ๐œŽ2 <

(๐‘›โˆ’1)๐‘ 2

๐œ’๐‘™2 so

(20โˆ’1)1.33252

32.852< ๐œŽ2 <

(20โˆ’1)1.33252

8.907 so 1.0269 < ๐œŽ2 < 3.788

Extra Take the square root of the lower and upper CI values given to get the CI for the standard

deviation of the population.

Q2 (Doane & Seward, 4/E, Minicase 9.63)

A sample of size ๐‘› = 19 has variance ๐‘ 2 = 1.96. At ๐›ผ = .05 in a right-tailed test, does this

sample contradict the hypothesis that ๐œŽ2 = 1.21?

Sol (i) ๐ป0: ๐œŽ2 โ‰ค 1.21; ๐ป1: ๐œŽ2 > 1.21; ๐›ผ = 5%

(ii) Sample statistic: ๐‘†2; reject for large values

(iii) Distribution test statistic under ๐ป0: (๐‘›โˆ’1)๐‘†2

๐œŽ2 ~๐œ’2(๐‘› โˆ’ 1)

Requirements: population must be normally distributed

(iv) Calculated test statistic: ๐œ’๐‘๐‘Ž๐‘™๐‘2 = 29.16

Critical value: ๐œ’๐‘๐‘Ÿ๐‘–๐‘ก2 (18; 0.05) = 28.87

Q1 1.0269<๐œŽ2

<3.788

Q2 reject ๐ป0 (there is reason to believe that the variance is larger than 1.21)

BS 2 Tutorial 3

(v) Decision: Reject the null hypothesis because ๐œ’๐‘๐‘Ž๐‘™๐‘

2 > ๐œ’๐‘๐‘Ÿ๐‘–๐‘ก2 and conclude there is reason to

believe that the variance is larger than 1.21.

Extra The requirement of a normally distributed population holds for any size, also for ๐‘› โ‰ฅ 30!

The formulation of the question is a bit awkward: there is an =-sign, suggesting two-sided, but

the sentence contains the word โ€œright-tailedโ€.

5B Median: non-parametric tests

Q1 (based on Doane & Seward, 4/E, Minicase 10.3)

The table below shows the results of a weight-loss contest sponsored by a local newspaper.

Participants came from all over the city, and were encouraged to compete over a 1-month

period.

Obs Name After (pounds) Before

(pounds)

Difference

1 Mickey 203.8 218.3 โ€“14.5

2 Teresa 179.3 189.3 โ€“10.0

3 Gary 211.3 226.3 โ€“15.0

4 Bradford 158.3 169.3 โ€“11.0

5 Diane 170.3 179.3 โ€“9.0

6 Elaine 174.8 183.3 โ€“8.5

7 Kim 164.8 175.8 โ€“11.0

8 Cathy 154.3 162.8 โ€“8.5

9 Abby 171.8 178.8 โ€“7.0

10 William 337.3 359.8 โ€“22.5

11 Margaret 175.3 182.3 โ€“7.0

12 Tom 198.8 211.3 โ€“12.5

At ๐›ผ = .01, can we โ€˜proveโ€™ the claim that the mean weight loss is more than 8 pounds? See the

SPSS output below. Do the test assuming that the differences are normally distributed.

BS 3 Tutorial 3

Sol a. Consider the vector of differences ๐ท. Five-step procedure:

(i) ๐ป0: ๐œ‡๐ท โ‰ฅ โˆ’8; ๐ป1: ๐œ‡๐ท < โˆ’8; ๐›ผ = 0.01

(ii) Sample statistic ๏ฟฝ๏ฟฝ; reject for small values

(iii) Under ๐ป0, ๐‘ก =๏ฟฝ๏ฟฝโˆ’๐œ‡๐ท

๐‘†๐ท/โˆš๐‘›~๐‘ก๐‘›โˆ’1 = ๐‘ก11; it was given what we could assume that the population

(of differences) is normally distributed

(iv) Calculated test statistic: โˆ’11.375โˆ’(โˆ’8)

1.2630= โˆ’2.6722; critical values: โˆ’2.718

(v) Decision: do not reject ๐ป0 because ๐‘ก๐‘๐‘Ž๐‘™๐‘ > ๐‘ก๐‘๐‘Ÿ๐‘–๐‘ก and conclude that there is no reason to

doubt the hypothesis that ๐œ‡๐ท โ‰ฅ โˆ’8.

Extra Note that the standard deviation of the differences is very much smaller than the standard

deviations of the individual variables!

Q2 (based on Doane & Seward, 4/E, Minicase 10.3)

We repeat the previous question, now assuming that the differences are symmetrically

distributed. Use the table and SPSS output below.

Q1 do not reject ๐ป0 (there is no reason to doubt the hypothesis that ๐œ‡๐ทโ‰ฅโˆ’8.)

BS 4 Tutorial 3

Legend: Column 1: data, Column 2: sorted data, Column 3: data โˆ’(โˆ’8), Column 4: remove

zeros, Column 5: assign ranks to absolute values and give signs (note: 1 and 2 are equal 1.5,

3 and 4 and 5 are equal 4), Column 6: add positive ranks, Column 7: count signs for sign

test.

Sol Problem: ๏ฟฝ๏ฟฝ is no longer normally distributed because of the small sample size. Idea: subtract

โˆ’8 from all differences. Then replace observations by ranks of absolute values and add sign

according to larger or smaller.

(i) ๐ป0: ๐œ‡๐ท โ‰ฅ โˆ’8; ๐ป1: ๐œ‡๐ท < โˆ’8 (mean ๐œ‡ or median ๐‘€, but symmetry is given); ๐›ผ = 0.01

(ii) Sample statistic: ๐‘Š (Sum of positive Ranks); reject for small values (make graph under

๐ป1!)

(iii) Distribution test statistic under ๐ป0: ๐‘Š~? . Use Signed-Ranks table for critical values from

distribution!

Requirements: Differences are symmetrically distributed

(iv) Calculated test statistic: ๐‘Š๐‘๐‘Ž๐‘™๐‘ = 8

Critical values: 10 (the smaller one! One-tailed test)

(v) Decision: do reject ๐ป0 (but it is close), because ๐‘Š๐‘๐‘Ž๐‘™๐‘ = 8 โ‰ค ๐‘Š๐‘๐‘Ÿ๐‘–๐‘ก = 10. Conclude that

there is no evidence that the mean difference is larger than โˆ’8.

Extra Note: if no table available or if we state that normal approximation must be used:

H0: ME = -8 TRUE TRUE TRUE

TRUE TRUE

xi xi Xi - -8 Xi - -8 Rank(|.|) Sign Test

-14,5 -22,5 -14,5 -14,5 -12 -12 -1

-10 -15 -7 -7 -11 -11 -1

-15 -14,5 -6,5 -6,5 -10 -10 -1

-11 -12,5 -4,5 -4,5 -9 -9 -1

-9 -11 -3 -3 -7,5 -7,5 -1

-8,5 -11 -3 -3 -7,5 -7,5 -1

-11 -10 -2 -2 -6 -6 -1

-8,5 -9 -1 -1 -4 -4 -1

-7 -8,5 -0,5 -0,5 -1,5 -1,5 -1

-22,5 -8,5 -0,5 -0,5 -1,5 -1,5 -1

-7 -7 1 1 4 4 1

-12,5 -7 1 1 4 4 1

8 70 2 10

Signed Ranks

Test

Xi - ME Rank|..|

Signed Ranks

Sign |..|

N Mean Rank Sum of Ranks

After - Negative Ranks 10 a 7,00 70,00Before Positive Ranks 2 b 4,00 8,00

Ties 0 c

Total 12 a. After < Before

b. After > Before

c. After = Before

Ranks

After-

Before

Z -2,432 a

Asymp. Sig. (2-tailed) ,015

a. Based on positive ranks

b. Wilcoxon Signed Ranks Test

Test Statisticsb

Q2 do not reject ๐ป0 (there is no evidence that the mean difference is larger than โˆ’8.)

BS 5 Tutorial 3

(iii) Distribution (standardized) test statistic approximately under ๐ป0: ๐‘Š~๐‘(๐œ‡๐‘Š, ๐œŽ๐‘Š2 ) where

๐œ‡๐‘Š =๐‘›(๐‘›+1)

4= 39 and ๐œŽ๐‘Š

2 =๐‘›(๐‘›+1)(2๐‘›+1)

24= 162.5

Requirements: Differences are symmetrically distributed

(iv) Calculated test statistic: ๐‘ง๐‘๐‘Ž๐‘™๐‘ =8โˆ’39

โˆš162.5= โˆ’2.4318; etc.

Q3 (based on Doane & Seward, 4/E, Minicase 10.3)

We repeat the previous two questions, now not assuming that the differences are symmetrically

distributed, and testing ๐ป0: ๐‘€๐‘‘ โ‰ฅ โˆ’8. Use the table at Q3 and the SPSS output below.

Sol Problem: ๐‘Š can no longer be used because of lack of symmetry. Idea: subtract โˆ’8 from all

differences. Then replace observations by plus or minus sign according to larger or smaller.

(i) ๐ป0: ๐‘€๐ท โ‰ฅ โˆ’8; ๐ป1: ๐‘€๐ท < โˆ’8; ๐›ผ = 0.01

(ii) Sample statistic: ๐‘‹ (= # plusses); reject for small values (make graph!)

(iii) Distribution test statistic under ๐ป0: ๐‘‹~๐ต๐‘–๐‘›(12,0.5).

(iv) Calculated test statistic: ๐‘‹๐‘๐‘Ž๐‘™๐‘ = 2

๐‘-value of this statistical problem: ๐‘ƒ๐œ‹=0.5(๐‘‹ โ‰ค 2) = 0.0193

(v) Decision: do not reject ๐ป0, because ๐‘-value > ๐›ผ = 0.01; conclude that there is no evidence

that the median difference is larger than โˆ’8.

6A Two ๐s or medians: comparisons

Q1 Shipments of meat, meat by-products, and other ingredients are mixed together in several filling

lines at a pet food factory. After the ingredients are thoroughly mixed, the pet food is placed in

eight-ounce cans. Descriptive statistics concerning fill weights from the two production Lines,

from two independent samples are given in the following table.

Assuming that the population variances are equal, at the 0.05 level of significance, is there

evidence of a difference between the mean weight of cans filled on the two lines?

Sol a. Use five steps with the equal-variance ๐‘ก-test.

(i) ๐ป0: ๐œ‡๐ด = ๐œ‡๐ต (where Populations: 1 = Line A, 2 = Line B) or ๐ป0: ๐œ‡๐ด = ๐œ‡๐ต; ๐ป1: ๐œ‡๐ด โ‰  ๐œ‡๐ต (๐›ผ =0.05)

(ii) Sample statistic:๐‘‹1 โˆ’ ๐‘‹2

. Reject for large and small values.

(iii) Distribution under ๐ป0: ๐‘ก =(๐‘‹1 โˆ’๐‘‹2 )โˆ’(๐œ‡1โˆ’๐œ‡2)

โˆš๐‘†๐‘2(

1

๐‘›1+

1

๐‘›2)

~๐‘ก๐‘›1+๐‘›2โˆ’2 if we assume that population 1 is

normally distributed and population 2 is symmetrically distributed (โ€˜equal variancesโ€™ is given).

(iv) ๐‘ก๐‘๐‘Ž๐‘™๐‘ =(๐‘ฅ1 โˆ’๐‘ฅ2 )โˆ’(๐œ‡1โˆ’๐œ‡2)

โˆš๐‘ ๐‘2(

1

๐‘›1+

1

๐‘›2)

=(8.005โˆ’7.997)โˆ’0

โˆš7.26ร—10โˆ’5(1

11+

1

16)

=0.008

0.003337= 2.3972

because ๐‘ ๐‘ƒ2 =

(๐‘›1โˆ’1)๐‘ 12+(๐‘›2โˆ’1)๐‘ 2

2

(๐‘›1โˆ’1)+(๐‘›2โˆ’1)=

10โ‹…(0.012)2+15โ‹…(0.005)2

10+15= 7.26 ร— 10โˆ’5

๐‘ก๐‘๐‘Ÿ๐‘–๐‘ก(25) = ยฑ2.0595

Q3 do not reject ๐ป0 (there is no evidence that the median difference is larger than โˆ’8.)

Q1 Reject ๐ป0 (mean weight from line A is larger)

BS 6 Tutorial 3

(v) Decision rule: If ๐‘ก๐‘๐‘Ž๐‘™๐‘ < โˆ’2.0595 or ๐‘ก๐‘๐‘Ž๐‘™๐‘ > 2.0595, reject ๐ป0.

Since ๐‘ก๐‘๐‘Ž๐‘™๐‘ = 2.3972 > 2.0595 reject ๐ป0.

There is sufficient evidence of a difference in the mean weight of cans filled on the two lines.

Practical conclusion: mean weight from line A is larger (even if we have a two-sided test: reject

๐ป0 and give 1-sided practical conclusion).

Q2 The same problem as before, but now not assuming that the population variances are equal.

Sol Similar to Q1, except:

(iii) Distribution under ๐ป0: ๐‘ก =(๐‘‹1 โˆ’๐‘‹2 )โˆ’(๐œ‡1โˆ’๐œ‡2)

โˆš๐‘†1

2

๐‘›1+

๐‘†22

๐‘›2

~๐‘ก๐‘‘๐‘“

Now, ๐‘‘๐‘“ =(

๐‘ 12

๐‘›1+

๐‘ 22

๐‘›2)

2

(๐‘ 1

2

๐‘›1)

2

๐‘›1โˆ’1+

(๐‘ 2

2

๐‘›2)

2

๐‘›2โˆ’1

= 12.41, so use ๐‘‘๐‘“ = 12

Requirements: assume that population 1 is normally distributed (๐‘›1 < 15) and that population

2 is symmetrically distributed (๐‘›2 = 16 > 15). (We make no further assumptions about the

variances)

(iv) Calculations:

๐‘ก๐‘๐‘Ž๐‘™๐‘ =(๐‘ฅ1 โˆ’๐‘ฅ2 )โˆ’(๐œ‡1โˆ’๐œ‡2)

โˆš๐‘ 1

2

๐‘›1+

๐‘ 22

๐‘›2

=0.008

0.003828= 2.0899

๐‘ก๐‘๐‘Ÿ๐‘–๐‘ก(12) = ยฑ2.1788

(v) Decision rule: use the approximation ๐‘‘๐‘“ for the degrees of freedom in the ๐‘ก-distribution.

Since ๐‘ก๐‘๐‘Ž๐‘™๐‘ = 2.0899 < 2.1788 do not reject ๐ป0.

There is not sufficient evidence of a difference in the average weight of cans filled on the two

lines.

Q2 Do not reject ๐ป0 (there is not sufficient evidence of a difference in the average weight of

cans filled on the two lines)

BS 7 Tutorial 3

Extra Students have to be able to do this by hand and also from computer (SPSS) output!

Q3 Compare the results of the two previous questions.

Extra N.B. Equality of variances can be tested too (see later!).

It is essential that students are able to make this exercise with and without (SPSS) computer

output. (They should at least do the calculations once only from the table in the output!). In

exam papers we do not often ask to compute the degrees of freedom from the samples.

Q4 Same data as in Q1.

a. Assuming that the population variances are equal, find a 90% confidence interval for ๐œ‡๐ด โˆ’๐œ‡๐ต.

b. Not assuming that the population variances are equal, find a 90% confidence interval for

๐œ‡๐ต โˆ’ ๐œ‡๐ด.

Sol a. Use: pooled variance

(๐‘ฅ๐ด โˆ’ ๐‘ฅ๐ต ) ยฑ ๐‘ก๐‘‘๐‘“;0.05๐‘ ๐‘‹๐ด โˆ’๐‘‹๐ต = (8.005 โˆ’ 7.997) ยฑ 1.708 ร— 0.003337 โ†’ [0.002299,0.1370]

b. Use: separate variance

(๐‘ฅ๐ต โˆ’ ๐‘ฅ๐ด ) ยฑ ๐‘ก๐‘‘๐‘“;0.05๐‘ ๐‘‹๐ด โˆ’๐‘‹๐ต = (7.997 โˆ’ 8.005) ยฑ 1.782 ร— 0.003828 โ†’

[โˆ’0.01482, โˆ’0.001177]

Q5 (based on Doane & Seward, 4/E, 10.6)

Are womenโ€™s feet getting bigger? Retailers in the last 20 years have had to increase their stock

of larger sizes. Wal-Mart Stores, Inc., and Payless ShoeSource, Inc., have been aggressive in

stocking larger sizes, and Nordstromโ€™s reports that its larger sizes typically sell out first.

Assuming equal variances, at ๐›ผ = .05, do these random shoe size samples of 12 randomly

chosen women in each age group show that womenโ€™s shoe sizes are different for women born

in 1960 and women born in 1980? (See The Wall Street Journal, July 17, 2004.)

Born in 1980:

8 7.5 8.5 8.5 8 7.5 9.5 7.5 8 8 8.5 9

Born in 1960:

8.5 7.5 8 8 7.5 7.5 7.5 8 7 8 7 8

You may use the output from SPSS

Q3 The results from Q1 and Q2 are different. The results obtained from Q2 are perhaps more

reliable because the sample variances from both samples suggest (??) that the two

population variances are not likely to be equal (why not always use the separate variance

test? It has less power if the true variances are equal).

Q4 a. [0.002299,0.1370 b. [โˆ’0.01482,โˆ’0.001177]

BS 8 Tutorial 3

Sol Use the 5-steps procedure (including the assumptions)

Problem is two-sided with ๐›ผ = 0.05 (although it was originally formulates as a one-sided

problem with ๐›ผ = 0.025)

(i) ๐ป0: ๐œ‡1 = ๐œ‡2 (where Populations: 1 = 1980, 2 = 1960) or ๐ป0: ๐œ‡1980 โˆ’ ๐œ‡1960 = 0 (Mean shoe

size is the same.)

๐ป1: ๐œ‡_1 โ‰  ๐œ‡2 (๐›ผ = 0.05)

(ii) Sample statistic: ๐‘‹1 โˆ’ ๐‘‹2

; reject for large and small values.

(iii) Distribution under ๐ป0: ๐‘ก =(๐‘‹1 โˆ’๐‘‹2 )โˆ’(๐œ‡1โˆ’๐œ‡2)

โˆš๐‘†๐‘ƒ2(

1

๐‘›1+

1

๐‘›2)

~๐‘ก๐‘›1+๐‘›2โˆ’2 = ๐‘ก22 (see output)

Requirements: both populations should be normally distributed (both ๐‘› < 15)

(iv) Computations:

๐‘ ๐‘ƒ2 =

(๐‘›1โˆ’1)๐‘ 12+(๐‘›2โˆ’1)๐‘ 2

2

(๐‘›1โˆ’1)+(๐‘›2โˆ’1)=

11(0.6201)2+11(0.4502)2

11+11= 0.2936

๐‘ ๐‘ƒ2 (

1

๐‘›1+

1

๐‘›2) = 0.2936 ร— (

1

12+

1

12) = 0.04893 = (0.2212)2

๐‘ก๐‘๐‘Ž๐‘™๐‘ =(8.208โˆ’7.708)โˆ’0

0.2212= 2.260

๐‘-value = 0.0340 (see output)

(v) ๐‘-value = 0.0340 < 5%, so reject ๐ป0.

Decision: Since ๐‘ก๐‘๐‘Ž๐‘™๐‘ = 2.260 is outside the lower and upper critical bound of โˆ’2.074 and

2.074, do reject ๐ป0. There is enough evidence to conclude that the mean shoe size is different.

One-sided โ€˜Post Hoc conclusionโ€™: Shoe Size has increased.

Q6 (based on Doane & Seward, 4/E, 16.B-2)

Q5 Reject ๐ป0 (there is enough evidence to conclude that the mean shoe size has increased)

BS 9 Tutorial 3

Below are data for two different regions, showing the number of days that kidney transplant

patients had to wait before a donor was found (๐‘›๐ธ = 6 patients, ๐‘›๐‘Š = 8 patients).

East: 109 248 85 107 28 67

West: 137 93 52 191 236 205 92 133

Do not assume a normal distribution of waiting times.

Use Table 16.B1 to test the hypothesis of equal medians at ๐›ผ = .05 Show the steps in your

analysis.

Sol Replace data by ranks: combine samples, rank them, and put ranks back in original sample.

(i) ๐ป0: ๐‘€๐ธ = ๐‘€๐‘Š; ๐ป1: ๐‘€๐ธ โ‰  ๐‘€๐‘Š (๐›ผ = 5%)

(ii) Sample statistic: ๐‘‡๐ธ (sum of ranks from sample smallest sample, so from East); reject for

large and small values

(iii) Distribution test statistic under ๐ป0: directly from โ€˜Wilcoxonโ€™ table

Requirements: both distributions have similar shape

(iv) Calculated test statistic: The ranks for (smallest) sample (so sample ๐ธ) are 1, 3, 4, 7, 8 and

14, respectively; ๐‘‡๐ธ = 37 (so ๐‘‡๐‘Š = 105 โˆ’ 37 = 68).

Critical values: ๐‘›๐ธ = 6, ๐‘›๐‘Š = 8, ๐‘‡๐ธ(๐‘๐‘Ÿ๐‘–๐‘ก, ๐ฟ) = 29, ๐‘‡๐ธ(๐‘๐‘Ÿ๐‘–๐‘ก, ๐‘…) = 61

(v) Decision: do not reject ๐ป0 because ๐‘‡๐ธ not in critical region; conclude there is no reason to

doubt the equality of the medians (or the means, because of similar shape of both distributions)

Q7 (based on Doane & Seward, 4/E, 16.B-2)

Use the data from Q6 and ๐›ผ = 5%.

a. Test ๐ป0: ๐‘€๐‘Š โ‰ค ๐‘€๐ธ against ๐ป1: ๐‘€๐‘Š > ๐‘€๐ธ (where ๐‘€ is median), using the tables on the

website

b. Use the normal approximation to answer the same question. Is your conclusion the same?

Sol a. ๐ป0: ๐‘€๐‘Š โ‰ค ๐‘€๐ธ vs. ๐ป1: ๐‘€๐‘Š > ๐‘€๐ธ, but we prefer to write ๐ป0: ๐‘€๐ธ โ‰ฅ ๐‘€๐‘Š vs. ๐ป1: ๐‘€๐ธ < ๐‘€๐‘Š

(smallest sample first and take statistic ๐ธ1). Reject for small values of ๐‘‡๐ธ

(iv) Critical values: ๐‘›๐ธ = 6, ๐‘›๐‘Š = 8, ๐‘‡๐ธ(๐‘๐‘Ÿ๐‘–๐‘ก) = 31

(v) Decision: do not reject ๐ป0 because ๐‘‡๐ธ > 31

b. Note: meant is to use the large sample distribution of ๐‘‡๐ธ (small sample, so normal

approximation for ๐‘‡๐ธ is not OK).

(ii) Sample statistic ๐‘‡๐ธ; reject for small values

(iii) Distribution test statistic under ๐ป0: approximately normal, see formula sheet

๐‘ =๐‘‡๐ธโˆ’๐œ‡๐‘‡๐ธ

๐œŽ๐‘‡๐ธ

=๐‘‡๐ธโˆ’

๐‘›๐ธ(๐‘›+1)

2

โˆš๐‘›๐ธ๐‘›๐‘Š(๐‘›+1)

12

where ๐‘› = ๐‘›๐ธ + ๐‘›๐‘Š

Requirements: both distributions have similar shape. But check on ๐‘›๐ธ and ๐‘›๐‘Š (both should be

more than 10) fails

(iv) Calculated test statistic:

๐‘‡๐ธ = 37 (so ๐‘‡๐‘Š = 68). Further ๐œ‡๐‘‡๐ธ= 45 and ๐œŽ๐‘‡๐ธ

= 7.7460; ๐‘ง๐‘๐‘Ž๐‘™๐‘ =37โˆ’45

7.7460= โˆ’1.0328

Critical values: ๐‘ง๐‘๐‘Ÿ๐‘–๐‘ก = โˆ’1.645

Q6 do not reject ๐ป0 (no reason to doubt the equality of the medians or means)

Q7 do not reject ๐ป0 (no reason to doubt that ๐‘€๐‘Šโ‰ค๐‘€๐ธ)

BS 10 Tutorial 3

๐‘-value of this statistical problem: 0.1508

(v) Do not reject ๐ป0, and conclude that there is no reason to doubt that ๐‘€๐‘Š โ‰ค ๐‘€๐ธ

Q8 (based on Doane & Seward, 4/E, 16.B-1)

A trucking company wants to compare the number of miles driven by two delivery truck drivers

in one week on different days (๐‘›1 = 5 days, ๐‘›2 = 7 days). Do not assume that distances driven

are normally distributed.

Driver 1: 128 102 78 40 76

Driver 2: 97 158 112 112 216 316 112

a. Use Table 16.B1 to test the hypothesis of equal medians at ๐›ผ = .05. Show the steps in your

analysis.

b. Perform a large-sample test using a normal approximation for the distribution of ๐‘‡1. Is your

conclusion the same?

Sol Replace data by ranks: combine samples, rank them, and put ranks back in original sample.

a. Five steps:

(i) ๐ป0: ๐‘€1 = ๐‘€2; ๐ป1: ๐‘€1 โ‰  ๐‘€2 (๐›ผ = 5%)

(ii) Sample statistic: ๐‘‡1 (=sum of ranks from sample โ€˜Driver 1โ€™); reject for large and small

values

Use 16.9 from Doane 4th edition.

(iii) Distribution test statistic under ๐ป0: directly from โ€˜Wilcoxonโ€™ table

Requirements: both distributions have similar shape

(iv) Calculated test statistic: The ranks for (smallest) sample 1 are 1, 2, 3, 5 and 9, respectively;

๐‘‡1 = 20 (so ๐‘‡2 = 78 โˆ’ 20 = 58).

Critical values: ๐‘›1 = 5, ๐‘›2 = 7, ๐‘‡1(๐‘๐‘Ÿ๐‘–๐‘ก, ๐ฟ) = 20 and ๐‘‡1(๐‘๐‘Ÿ๐‘–๐‘ก,๐‘…) = 45

(v) Decision: reject ๐ป0 because ๐‘‡1 in the critical region (on the border) and conclude (but be

careful) that there is reason to doubt the equality of the medians or the means (if assumption of

similar shape of both distributions is reasonable)

Driver 1 Driver 2

128 9

102 5

78 3

40 1

76 2

97 4

158 10

112 7

112 7

216 11

316 12

112 7

20 58

Wilcoxon - Mann/Whitney Test

n sum of ranks

5 20 Driver 1

7 58 Group 2

12 78 total

32,500 expected value

6,158 standard deviation

-2,030 z

,0424 p-value (two-tailed)

Q8 reject ๐ป0 (there is reason to doubt the equality of the medians or means)

BS 11 Tutorial 3

b. Note: meant is to use the large sample distribution of ๐‘‡1 (or to use ๐‘‡1 โˆ’ ๐‘‡2, which we do not

recommend) (note: variance is unknown, so normal approximation for ๐‘‡1 is not OK).

(ii) Use ๐‘‡1

(iii) Distribution test statistic under ๐ป0: approximately normal, see formula sheet

๐‘ =๐‘‡1โˆ’๐œ‡๐‘‡1

๐œŽ๐‘‡1

=๐‘‡1โˆ’

๐‘›1(๐‘›+1)

2

โˆš๐‘›1๐‘›2(๐‘›+1)

12

where ๐‘› = ๐‘›1 + ๐‘›2

Requirements: both distributions have similar shape. But check on ๐‘›1 and ๐‘›2 (both should be

more than 10) fails

(iv) Calculated test statistic: ๐‘‡1 = 20 (so ๐‘‡2 = 58). Further, ๐œ‡๐‘‡1= 32.5 and ๐œŽ๐‘‡1

= 6.158 and

๐‘ง๐‘๐‘Ž๐‘™๐‘ =20โˆ’32.5

6.158= โˆ’2.030

Critical values: ๐‘ง๐‘๐‘Ÿ๐‘–๐‘ก = ยฑ1.96

๐‘-value of this statistical problem: 2 ร— 0.02118 = 0.04236

Extra The alternative approach, following chapter 16, not recommended, but given for completeness

(ii) Use ๐‘‡1 โˆ’ ๐‘‡2

(iii) Distribution test statistic under ๐ป0: approximately normal, see formula sheet

๐‘ =๐‘‡1 โˆ’๐‘‡2 โˆ’0

(๐‘›1+๐‘›2)โˆš๐‘›1+๐‘›2+1

12๐‘›1๐‘›2

Requirements: both distributions have similar shape. But check on ๐‘›1 and ๐‘›2 (both should be

more than 10) fails

(iv) Calculated test statistic: ๐‘‡1 =20

5= 4 and ๐‘‡2

=58

7= 8.2587 ๐œŽ๐‘‡1 โˆ’๐‘‡2 = 2.111195

๐‘ง๐‘๐‘Ž๐‘™๐‘ =โˆ’4.2857โˆ’0

2.111195= โˆ’2.030

Critical values: ๐‘ง๐‘๐‘Ÿ๐‘–๐‘ก = ยฑ1.96

๐‘-value of this statistical problem: 2 ร— 0.02118 = 0.04236

Q9 (based on Doane & Seward, 4/E, 16.7)

Bob and Tom are โ€œpaper investors.โ€ They each โ€œbuyโ€ stocks they think will rise in value and

โ€œholdโ€ them for a year. At the end of the year, they compare their stocksโ€™ appreciation (percent).

Bobโ€™s Portfolio (10 stocks):

7.0 2.5 6.2 4.4 4.2 8.5 10.0 6.4 3.6 7.6

Tomโ€™s Portfolio (12 stocks):

5.2 0.4 2.6 โ€“0.2 4.0 5.2 8.6 4.3 3.0 0.0 8.6 7.5

a. At ๐›ผ = .05, is there a difference in the medians (assume these are samples of Bobโ€™s and

Tomโ€™s stock-picking skills). Use the SPSS output below

b. Now test ๐ป0: ๐‘€1 โ‰ค ๐‘€2 against ๐ป1: ๐‘€1 > ๐‘€2 (๐›ผ = 5%; Bob=1 and Tom=2) using SPSS

output.

c. Perform a two-tailed parametric ๐‘ก test for two independent sample means by using the SPSS

output. Do you get the same decision?

BS 12 Tutorial 3

Sol Replace data by ranks: combine samples, rank them, and put ranks back in original sample.

Note: no table available.

Do the test using the computer output (new for them) and only the one-sided (extra) question.

a. Five steps:

(i) ๐ป0: ๐‘€1 = ๐‘€2; ๐ป1: ๐‘€1 โ‰  ๐‘€2; ๐›ผ = 5% (Bob=1, Tom=2)

(ii) Sample statistic: ๐‘‡1 (sum of ranks from sample โ€˜Bobโ€™); reject for large and small values

(iii) Distribution test statistic under ๐ป0: approximately normally distributed (parameters

depending on choice in step (ii)

Requirements: both distributions have similar shape. Check on ๐‘›1 and ๐‘›2: both at least 10, so

approximation should be OK

(iv) Calculated test statistic: ๐‘ง๐‘๐‘Ž๐‘™๐‘ = โˆ’1.320

Reported ๐‘-value: 0.187; ๐‘-value of this statistical problem: 0.187

Q9 do not reject ๐ป0 (no reason to doubt the equality of the medians or means)

BS 13 Tutorial 3

(v) Decision: do not reject ๐ป0 because ๐‘-value > 5% and conclude there is no reason to doubt

the equality of the medians or the means (!!!!, because of similar shape of both distributions).

b. To test 1-sided hypothesis: look at mean ranks!

(i) ๐ป0: ๐‘€1 โ‰ค ๐‘€2; ๐ป1: ๐‘€1 > ๐‘€2 (๐›ผ = 5%)

(iv) Calculated test statistic: ๐‘ง๐‘๐‘Ž๐‘™๐‘ = โˆ’1.320, but sign is meaningless (!!!)

Because MeanRanks(1) > MeanRanks(2) in sample, we are close to rejection region (whatever

statistic we might have chosen).

So ๐‘-value = 0.5 ร— ๐‘-twosided =0.187

2= 0.093

c. Steps that change:

(iii) Distribution test statistic under ๐ป0: ๐‘ก~๐‘ก20

Requirements: both distributions are normal, equal variance (latter is reasonable assumption in

Leveneโ€™s test: ๐‘-value = 39.6%)

(v) Decision: do not reject ๐ป0 because ๐‘-value = 0.121 > 5% and conclude there is no reason

to doubt the equality of the means

6B Two ๐ˆ๐Ÿs: comparisons

Q1 (Doane & Seward, 4/E, 10.39)

A manufacturing process drills holes in sheet metal that are supposed to be . 5000 cm in

diameter. Before and after a new drill press is installed, the hole diameter is carefully measured

(in cm) for 12 randomly chosen parts. At ๐›ผ = .05, do these independent random samples prove

that the new process has smaller variance? Show the hypotheses, decision rule, and test statistic.

Sol (i) ๐ป0: ๐œŽ1

2 โ‰ค ๐œŽ22; ๐ป1: ๐œŽ1

2 > ๐œŽ22 (where 1=Old, 2=New) (โ€˜Oldโ€™ in numerator for easy critical

value)

(ii) Test Statistic: ๐น =๐‘†1

2

๐‘†22. Reject for large values

(iii) Under ๐ป0: ๐น~๐น11;11

Requirement: both populations normal

(iv) Test statistic: ๐น๐‘๐‘Ž๐‘™๐‘ =๐‘ 1

2

๐‘ 22 =

3.183ร—10โˆ’5

3.265ร—10โˆ’6 = 9.748 (see output below)

Critical values: ๐น๐‘๐‘Ÿ๐‘–๐‘ก(11; 11; 0.05) = 2.82

Q1 reject ๐ป0 (the new drill has a significantly smaller variance)

BS 14 Tutorial 3

(v) Decision: Since ๐น๐‘๐‘Ž๐‘™๐‘ = 9.748 is above the critical bound of 2.82 do reject ๐ป0. There is

enough evidence to conclude that the new drill has reduced variance.

Compare the two Excel outputs, one with the smallest in the numerator (so ๐น๐‘๐‘Ž๐‘™๐‘ < 1), the other

with the largest in the numerator (so ๐น๐‘๐‘Ž๐‘™๐‘ > 1).

Extra Excel produces the following two tables, with left 1=new, 2=old, and right 1=old, 2=new. There

is an error in Excel in the second table: the โ€œ<=โ€ should be โ€œ>=โ€.

Q2 (Doane & Seward, 4/E, 10.40)

Examine the data below showing the weights (in pounds) of randomly selected checked bags

for an airlineโ€™s flights on the same day.

a. At ๐›ผ = .05, is the mean weight of an international bag greater? Show the hypotheses,

decision rule, and test statistic.

b. At ๐›ผ = .05, is the variance greater for bags on an international flight? Show the hypotheses,

decision rule, and test statistic.

Use the output below:

F-Test Two-Sample for Variances

New Drill Old Drill

Mean 0,5002167 0,5000167

Variance 3,265E-06 3,183E-05

Observations 12 12

df 11 11

F 0,1025849

P(F<=f) one-tail 0,0003546

F Critical one-tail 0,3548704

F-Test Two-Sample for Variances

Old Drill New Drill

Mean 0,5000167 0,5002167

Variance 3,183E-05 3,265E-06

Observations 12 12

df 11 11

F 9,7480278

P(F<=f) one-tail 0,0003546

F Critical one-tail 2,8179305

BS 15 Tutorial 3

Sol a. First test the equality of variances in order to choose between equal and unequal variance ๐‘ก-

test. The ๐‘-value is 0.000887, so we reject the hypothesis of equal variance, and start using the

๐‘ก-test not assuming equal variances.

Note: as an exercise for the ๐น-test, we do this below for a full 5-step procedure. At the exam,

this is not needed when we ask for testing the equality of two means.

(i) ๐ป0:๐œŽ1

2

๐œŽ22 = 1; ๐ป1:

๐œŽ12

๐œŽ22 โ‰  1; ๐›ผ = 0.05 (with 1=international, 2=domestic)

(ii) Test Statistic: ๐น =๐‘†1

2

๐‘†22; reject for small and large values

(iii) Under ๐ป0: ๐น~๐น9;14

Requirement: both populations normal.

(iv) Test statistic: ๐น๐‘๐‘Ž๐‘™๐‘ =๐‘ 1

2

๐‘ 22 =

141.3778

20.98095= 6.738 (see output above)

Critical values: ๐น๐‘๐‘Ÿ๐‘–๐‘ก,๐‘… = ๐น9;14;0.025 = 3.21 and ๐น๐‘๐‘Ÿ๐‘–๐‘ก,๐ฟ = ๐น9;14;0.975 = โ‹ฏ (It is not necessary to

find ๐น๐‘๐‘Ÿ๐‘–๐‘ก,๐ฟ.

t-Test: Two-Sample Assuming Equal Variances

International Domestic

Mean 48,6 36,13333

Variance 141,3778 20,98095

Observations 10 15

Pooled Variance 68,09275

Hypothesized Mean Difference 0

df 23

t Stat 3,700629

P(T<=t) one-tail 0,00059

t Critical one-tail 1,713872

P(T<=t) two-tail 0,001179

t Critical two-tail 2,068658

t-Test: Two-Sample Assuming Unequal Variances

International Domestic

Mean 48,6 36,13333

Variance 141,3778 20,98095

Observations 10 15

Hypothesized Mean Difference 0

df 11

t Stat 3,162814

P(T<=t) one-tail 0,004517

t Critical one-tail 1,795885

P(T<=t) two-tail 0,009034

t Critical two-tail 2,200985

F-Test Two-Sample for Variances

International Domestic

Mean 48,6 36,13333

Variance 141,3778 20,98095

Observations 10 15

df 9 14

F 6,738387

P(F<=f) one-tail 0,000887

F Critical one-tail 2,645791

Q2 a. Reject ๐ป0 (the mean of international bag weight is greater than domestic bag weight)

b. Reject ๐ป0 (the variance of international bag weight is greater than domestic bag weight.)

BS 16 Tutorial 3

(v) Decision: Since ๐น๐‘๐‘Ž๐‘™๐‘ = 6.738 is above the critical bound of 3.21 do reject ๐ป0. There is

enough evidence to conclude that the variances are not equal.

So we have to use the โ€˜separate variance ๐‘ก-test for the ยต-problem.

(Post-hoc: the variance for international is larger)

Now we test the question from a.

(i) ๐ป0: ๐œ‡1 โˆ’ ๐œ‡2 โ‰ค 0; ๐ป1: ๐œ‡1 โˆ’ ๐œ‡2 > 0 (๐›ผ = 0.05)

(ii) Test statistic: ๐‘‹1 โˆ’ ๐‘‹2

; reject for large values

(iii) Under ๐ป0: ๐‘ก =(๐‘‹1 โˆ’๐‘‹2 )โˆ’(๐œ‡1โˆ’๐œ‡2)

โˆš๐‘†1

2

๐‘›1+

๐‘†22

๐‘›2

~๐‘ก๐‘‘๐‘“ where ๐‘‘๐‘“ =(

๐‘ 12

๐‘›1+

๐‘ 22

๐‘›2)

2

(๐‘ 1

2

๐‘›1)

2

๐‘›1โˆ’1+

(๐‘ 2

2

๐‘›2)

2

๐‘›2โˆ’1

(from output: ๐‘‘๐‘“ = 11)

Requirements: assume that population 1 is normally distributed (๐‘›1 < 15) and population 2 is

symmetrically distributed (๐‘›2 = 15 โ‰ฅ 15).

(iv) Calculations:

๐‘ก๐‘๐‘Ž๐‘™๐‘ =(๐‘ฅ1 โˆ’๐‘ฅ2 )โˆ’(๐œ‡1โˆ’๐œ‡2)

โˆš๐‘ 1

2

๐‘›1+

๐‘ 22

๐‘›2

=(48.6โˆ’36.13333)โˆ’0

โˆš141.3778

10+

20.98095

15

=12.46667

3.941638= 3.162814

๐‘ก๐‘๐‘Ÿ๐‘–๐‘ก = ๐‘ก11;0.05 = 1.796 (from table; from output: 1.795885)

(Note: ๐‘ก๐‘๐‘Ž๐‘™๐‘ bot drawn to scale)

(v) Since ๐‘ก๐‘๐‘Ž๐‘™๐‘ = 3.163 > 1.796 or (from output) ๐‘-value = ๐‘ƒ(๐‘ก11 > 3.162814) =0.004517 < 0.05, reject ๐ป0. The mean of international bag weight is greater than domestic bag

weight.

b. Now test variances

(i) ๐ป0:๐œŽ1

2

๐œŽ22 โ‰ค 1; ๐ป1:

๐œŽ12

๐œŽ22 > 1; (๐›ผ = 0.05)

(ii) Test Statistic: ๐น =๐‘†1

2

๐‘†22; reject for large values

(iii) Under ๐ป0: ๐น~๐น9;14

Requirements: both populations normal.

(iv) Test statistic: ๐น๐‘๐‘Ž๐‘™๐‘ =๐‘ 1

2

๐‘ 22 =

141.3778

20.98095= 6.738 (see output above)

BS 17 Tutorial 3

Critical value: ๐น๐‘๐‘Ÿ๐‘–๐‘ก;๐‘… = ๐น9;14;0.05 = 2.65; ๐น๐‘๐‘Ÿ๐‘–๐‘ก;๐ฟ not needed

(v) Decision rule: Reject ๐ป0 if ๐น๐‘๐‘Ž๐‘™๐‘ โ‰ฅ 2.65 and ๐น๐‘๐‘Ž๐‘™๐‘ = 6.7387 so we reject the null

hypothesis. The variance of international bag weight is greater than domestic bag weight.

Note: From output: ๐‘-value = ๐‘ƒ(๐น9;14 โ‰ฅ 6.7387) = 0.000887 < 0.05

Q3 (based on Doane & Seward, 4/E, 10.6)

At the tests for comparing two ๐œ‡s above, we analyzed the change of foot size of women. Now

test the equality of variance at ๐›ผ = 5%.

a. Use the usual variance test, with a full calculation.

b. Use the SPSS output, without a full calculation.

Data are repeated below:

Born in 1980:

8 7.5 8.5 8.5 8 7.5 9.5 7.5 8 8 8.5 9

Born in 1960:

8.5 7.5 8 8 7.5 7.5 7.5 8 7 8 7 8

Sol On the basis of the usual variance test

(i) ๐ป0: ๐œŽ12 = ๐œŽ2

2; ๐ป1: ๐œŽ12 โ‰  ๐œŽ2

2 (where Populations: 1 = 1980, 2 = 1960); ๐›ผ = 0.05

(ii) Sample statistic: ๐น =๐‘†1

2

๐‘†22; reject for large and for small small values

(iii) Distribution under ๐ป0: ๐น =๐‘ 1

2

๐‘ 22 ~๐น11;11

Requirements: Both populations normally distributed

(iv) Computations: ๐น๐‘๐‘Ž๐‘™๐‘ =(0.6201)2

(0.4502)2 = 1.8972

๐น๐‘๐‘Ÿ๐‘–๐‘ก;๐‘… = ๐น11,11;0.025 = 3.47 (not in table) and ๐น๐‘๐‘Ÿ๐‘–๐‘ก;๐ฟ < 1

Note: from table ๐น๐‘๐‘Ÿ๐‘–๐‘ก between 3.53 and 3.43.

Q2 Do not reject ๐ป0 (the two variances do not differ significantly)

BS 18 Tutorial 3

(v) Do not reject ๐ป0. because ๐น๐‘๐‘Ž๐‘™๐‘ not in rejection region. Variances do not differ significantly.

Using SPSS output, you can use the Leveneโ€™s test instead:

(i) identical

(ii) Sample statistic: Leveneโ€™s ๐น; (reject for large values)

(iii) Distribution under ๐ป0: ๐น~? ? ? (now see SPSS output). This test is a two-sided but one-

tailed test!

Requirements: ???

(iv) Computations: ๐น๐‘๐‘Ž๐‘™๐‘ = 0.993 (see output) and ๐‘-value 0.3300

(v) Do not reject ๐ป0 because ๐‘-value larger than 5%.

Old exam questions

Q1 23 March 2016, Q3a

It has been suggested that students perform differently in the afternoon compared to the

morning. To investigate this phenomenon, a number of tests have been made: 12 randomly

chosen students did an exam in the morning, and 12 others did an exam in the afternoon. The

results of the exams are on a scale from 1.0 (low) to 10.0 (high). One group of researchers

proposes to compare the mean score in the morning to the mean score in the afternoon. Perform

this test at ๐›ผ = 5%, using the 5-step procedure.

Sol This is a case of comparing the means of two independent samples.

(i) ๐ป0: ๐œ‡1 = ๐œ‡2, ๐ป1: ๐œ‡1 โ‰  ๐œ‡2, ๐›ผ = 0.05 (where 1 codes for morning and 2 for afternoon)

(ii) Sample statistic: ๐‘‹1 โˆ’ ๐‘‹2

, reject for small and for large values

(iii) null distribution: ๐‘‹1 โˆ’๐‘‹2

๐‘†๐‘‹1 โˆ’๐‘‹2 ~๐‘ก22, provided both populations are normally distributed (which

we will have to assume) and the two populations have the same variance (which looks quite

plausible, given Leveneโ€™s test with ๐‘-value = 0.976)

(iv) ๐‘ก๐‘๐‘Ž๐‘™๐‘ = 1.410, ๐‘ก๐‘๐‘Ÿ๐‘–๐‘ก = ยฑ2.074, ๐‘-value = 0.172

(v) do not reject ๐ป0, there is no evidence for concluding that the population means are different

Q2 22 May 2017, Q3d

Electricity supply is a vital element for nearly every part of the economy. Supply must be stable:

voltage should be 230 volt and variations must be small. We measure at 21 random times in

The Netherlands (NL) and at 26 random times in Germany (DE) voltage and find the following

descriptive statistics:

Q1 Do not reject ๐ป0 (there is no evidence for concluding that the population means are different)

BS 19 Tutorial 3

Use the data above to test if the assumption ๐œŽ๐‘๐ฟ

2 = ๐œŽ๐ท๐ธ2 is reasonable, at ๐›ผ = 5%. Calculate the

value of the test statistic, as well as the critical value(s). Make assumptions and state

requirements where needed and/or check requirements where possible.

Sol The standardized test statistic for comparing two variances is ๐น =๐‘†๐‘๐ฟ

2

๐‘†๐ท๐ธ2 . Its observed value is

1.306 (or 0.766 in case you define ๐น =๐‘†๐ท๐ธ

2

๐‘†๐‘๐ฟ2 ).

Under the null hypothesis, this test statistic is distributed as ๐น๐‘›๐‘๐ฟโˆ’1,๐‘›๐ท๐ธโˆ’1 (or ๐น๐‘›๐ท๐ธโˆ’1,๐‘›๐‘๐ฟโˆ’1).

The test is two-sided. Critical values are thus ๐น๐‘๐‘Ÿ๐‘–๐‘ก,๐‘ข๐‘๐‘๐‘’๐‘Ÿ = ๐น20,25;0.025 = 2.30 and

๐น๐‘๐‘Ÿ๐‘–๐‘ก,๐‘™๐‘œ๐‘ค๐‘’๐‘Ÿ =1

๐น25,20;0.025=

1

2.40= 0.42. (Or if you used the other definition of ๐น: 2.40 and

1

2.30=

0.43)

The requirement is that both populations (NL and DE) are normal; this seems reasonable given

the estimated values of skewness and kurtosis for both NL and DE (should be between โˆ’1 and

1).

Q2 ๐น๐‘๐‘Ž๐‘™๐‘=1.306; ๐น๐‘๐‘Ÿ๐‘–๐‘ก=2.30 and 0.42; both populations must be normal, which is OK.