s s lewis 2/15/051 confidence intervals and maximum errors by sidney s. lewis for baltimore section,...

28
S S Lewis 2/15/05 1 Confidence Intervals and Maximum Errors By Sidney S. Lewis For Baltimore Section, ASQ February 15, 2005

Upload: sydney-hensley

Post on 02-Jan-2016

219 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: S S Lewis 2/15/051 Confidence Intervals and Maximum Errors By Sidney S. Lewis For Baltimore Section, ASQ February 15, 2005

S S Lewis 2/15/05 1

Confidence Intervals and Maximum Errors

By Sidney S. Lewis

For

Baltimore Section, ASQ

February 15, 2005

Page 2: S S Lewis 2/15/051 Confidence Intervals and Maximum Errors By Sidney S. Lewis For Baltimore Section, ASQ February 15, 2005

S S Lewis 2/15/05 2

CONFIDENCE INTERVAL

A confidence interval expresses our belief, or confidence, that the interval we construct from the data will contain the mean, µ, (for example) of the population from which the data were drawn.

Confidence Intervals can be computed on any population parameter: µ, , p’, c’, and even on complex parameters, such as Cp and Cpk.

812.1

Page 3: S S Lewis 2/15/051 Confidence Intervals and Maximum Errors By Sidney S. Lewis For Baltimore Section, ASQ February 15, 2005

S S Lewis 2/15/05 3

CONFIDENCE INTERVAL EXAMPLE

• Example of a confidence Interval (C.I.) on the population mean µ:

• Statement: The interval 38.0 - 42.0 contains µ with 90% confidence.

• Alternatively, there is a 5% chance that the C.I. falls entirely below µ (µ above 42.0), and likewise, a 5% chance that the C.I. is entirely above µ (µ below 38.0).

812.1

Page 4: S S Lewis 2/15/051 Confidence Intervals and Maximum Errors By Sidney S. Lewis For Baltimore Section, ASQ February 15, 2005

S S Lewis 2/15/05 4

Example

Pollsters report that 55% of a sample of 1005 members of the voting population support Proposition A.

The Margin of Error is 3.1%

There is an implied risk of being wrong, usually 5%

Calcs.: %15.3n/5.*5.96.1ME

812.8

Page 5: S S Lewis 2/15/051 Confidence Intervals and Maximum Errors By Sidney S. Lewis For Baltimore Section, ASQ February 15, 2005

S S Lewis 2/15/05 5

POPULATIONPROCESS

SAMPLESAMPLE

PROCESS

POPULATION vs. SAMPLES

810

Page 6: S S Lewis 2/15/051 Confidence Intervals and Maximum Errors By Sidney S. Lewis For Baltimore Section, ASQ February 15, 2005

S S Lewis 2/15/05 6

Distribution of X-Bar

37 38 39 40 41 42 43

0

0.4

/2 /2

= 2n = 4(xbar) = 1 = 5%

810.05

Page 7: S S Lewis 2/15/051 Confidence Intervals and Maximum Errors By Sidney S. Lewis For Baltimore Section, ASQ February 15, 2005

S S Lewis 2/15/05 7

Logic:

is known to be 2.0

A sample of 4 yields X-bar = 42.0.

Question? What values of are probable,

with 95% Confidence ( = 5%).

Solution: Xbar = /n = 1.0

Z/2 = 1.96, or about 2

ME = Z/2 Xbar = ± 2.0

or from 40.0 to 44.0

Maximum Error

37 38 39 40 41 42 43 44 45 46 47

2?

Xbar

MAXIMUM ERROR

810.15

Page 8: S S Lewis 2/15/051 Confidence Intervals and Maximum Errors By Sidney S. Lewis For Baltimore Section, ASQ February 15, 2005

S S Lewis 2/15/05 8

MAXIMUM ERROR of the MEAN

The Maximum Error (ME) is the largest expecteddeviation of the sample mean from thepopulation mean, with the stated confidencelevel, 1-

ME = z /2 σ/n if σ is known,

= t /2 s/n if σ is unknown.

ME = z /2 σ/n if σ is known,

= t /2 s/n if σ is unknown.

812.3

Page 9: S S Lewis 2/15/051 Confidence Intervals and Maximum Errors By Sidney S. Lewis For Baltimore Section, ASQ February 15, 2005

S S Lewis 2/15/05 9

C. I. on the MEAN

Calculation of the C.I. on the mean µ typically uses either the population standard deviation, , if known, or if not, the sample standard deviation, s.

If X-bar is the sample mean, then a 1- confidence interval on µ is:

C.I. =X-bar ± Maximum Error (ME)

= X-bar ± z/2 /n if is known, or

= X-bar ± ta/2s/n if is unknown.812.2

Page 10: S S Lewis 2/15/051 Confidence Intervals and Maximum Errors By Sidney S. Lewis For Baltimore Section, ASQ February 15, 2005

S S Lewis 2/15/05 10

Diameters of 3/4" HR BarsDiameters of 3/4" HR bars X-bar R

0.7466 0.7457 0.7524 0.7495 0.7489 0.7486 0.0067

0.7496 0.7549 0.7542 0.7566 0.7493 0.7529 0.0073

0.7563 0.7436 0.7475 0.7525 0.7492 0.7498 0.0127

0.7491 0.7508 0.7512 0.7482 0.7520 0.7502 0.0038

0.7498 0.7552 0.7508 0.7477 0.7453 0.7497 0.0099

0.7508 0.7480 0.7498 0.7526 0.7532 0.7509 0.0052

0.7498 0.7491 0.7507 0.7514 0.7527 0.7507 0.0037

0.7526 0.7521 0.7496 0.7507 0.7533 0.7516 0.0037

0.7520 0.7470 0.7550 0.7517 0.7404 0.7492 0.0146

0.7463 0.7554 0.7483 0.7507 0.7474 0.7496 0.0091

0.7537 0.7520 0.7501 0.7522 0.7524 0.7521 0.0036

0.7476 0.7535 0.7542 0.7548 0.7515 0.7523 0.0072

0.7516 0.7442 0.7499 0.7509 0.7472 0.7488 0.0074

940

Page 11: S S Lewis 2/15/051 Confidence Intervals and Maximum Errors By Sidney S. Lewis For Baltimore Section, ASQ February 15, 2005

S S Lewis 2/15/05 11

X-Bar Chart of 3/4" HR Bar Diameters

0.744

0.746

0.748

0.75

0.752

0.754

0.756

0 10 20 30 40

Sample No.

Avg

. Dia

met

er

840.1

Page 12: S S Lewis 2/15/051 Confidence Intervals and Maximum Errors By Sidney S. Lewis For Baltimore Section, ASQ February 15, 2005

S S Lewis 2/15/05 12

R Chart of 3/4" HR Bars

0

0.005

0.01

0.015

0.02

0 10 20 30 40

Sample No.

Ran

ge

840.1

Page 13: S S Lewis 2/15/051 Confidence Intervals and Maximum Errors By Sidney S. Lewis For Baltimore Section, ASQ February 15, 2005

S S Lewis 2/15/05 13

DATA STATISTICS

X-bar R s

0.7486 0.0067 0.0026

0.7529 0.0073 0.0033

0.7498 0.0127 0.0048

0.7502 0.0038 0.0016

0.7497 0.0099 0.0037

0.7509 0.0052 0.0021

0.7507 0.0037 0.0014

0.7516 0.0037 0.0015

0.7492 0.0146 0.0057

0.7496 0.0091 0.0036

0.7521 0.0036 0.0013

840.01

Page 14: S S Lewis 2/15/051 Confidence Intervals and Maximum Errors By Sidney S. Lewis For Baltimore Section, ASQ February 15, 2005

S S Lewis 2/15/05 14

3/4" HR Bars – MEANS and CONFIDENCE INTERVALS

z(.90) = 1.645 t(4,.90) = 2.132

Sample X-bar s LCI(z) UCI(z) LCI(t) UCI(t)

1 0.7486 0.00263 0.7464 0.7508 0.7461 0.7511

2 0.7529 0.00328 0.7507 0.7551 0.7498 0.7561

3 0.7498 0.00484 0.7476 0.7520 0.7452 0.7544

4 0.7502 0.00157 0.7480 0.7525 0.7488 0.7517

5 0.7497 0.00371 0.7475 0.7519 0.7462 0.7533

6 0.7509 0.00209 0.7487 0.7531 0.7489 0.7529

7 0.7507 0.00142 0.7485 0.7529 0.7494 0.7521

8 0.7516 0.00148 0.7494 0.7539 0.7502 0.7531

9 0.7492 0.00568 0.7470 0.7514 0.7438 0.7546

10 0.7496 0.00360 0.7474 0.7518 0.7462 0.7530

11 0.7521 0.00129 0.7499 0.7543 0.7509 0.7533840.41

Page 15: S S Lewis 2/15/051 Confidence Intervals and Maximum Errors By Sidney S. Lewis For Baltimore Section, ASQ February 15, 2005

S S Lewis 2/15/05 15

90% Confidence Limits using

0.744

0.746

0.748

0.750

0.752

0.754

0.756

0 5 10 15 20 25 30 35 40

Sample No.

Con

fiden

ce In

terv

als

840.5

Page 16: S S Lewis 2/15/051 Confidence Intervals and Maximum Errors By Sidney S. Lewis For Baltimore Section, ASQ February 15, 2005

S S Lewis 2/15/05 16

90% Confidence Limits using s

0.742

0.744

0.746

0.748

0.750

0.752

0.754

0.756

0.758

0 5 10 15 20 25 30 35 40

Sample No.

Con

fiden

ce In

terv

als

840.6

Page 17: S S Lewis 2/15/051 Confidence Intervals and Maximum Errors By Sidney S. Lewis For Baltimore Section, ASQ February 15, 2005

S S Lewis 2/15/05 17

FACTORS AFFECTING THE WIDTH OF A CONFIDENCE INTERVAL

C I X z n. . // 2

C I X t s n. . // 2

812.4

Factors: or s, n,

Page 18: S S Lewis 2/15/051 Confidence Intervals and Maximum Errors By Sidney S. Lewis For Baltimore Section, ASQ February 15, 2005

S S Lewis 2/15/05 18

FACTORS AFFECTING THE WIDTH OF A CONFIDENCE INTERVAL

or s: ...................Width increases as or s increases;

Sample size, n .....Width decreases as n increases;

C. I. is proportional to 1/n

Confidence level 1 - , or risk :

Width increases as confidence increases, or as risk decreases.

812.4

Page 19: S S Lewis 2/15/051 Confidence Intervals and Maximum Errors By Sidney S. Lewis For Baltimore Section, ASQ February 15, 2005

S S Lewis 2/15/05 19

CONFIDENCE INTERVALS ON

Small samples (n<30):

Upper C.I. s df / /22

Low er C.I. s df 1 22

/ /

C Is

z n. .

//

1 22

Large samples (n>30):

812.7

Page 20: S S Lewis 2/15/051 Confidence Intervals and Maximum Errors By Sidney S. Lewis For Baltimore Section, ASQ February 15, 2005

S S Lewis 2/15/05 20

CONFIDENCE INTERVALS ON p’

Large samples (np>5):

C I p z p p n

M E z p p n

. . / ,

/

/

/

2

2

1

1

812.7

Page 21: S S Lewis 2/15/051 Confidence Intervals and Maximum Errors By Sidney S. Lewis For Baltimore Section, ASQ February 15, 2005

S S Lewis 2/15/05 21

CONFIDENCE INTERVAL ON p’, SMALL n

Upper C.I. on p' Lower C.I. on p'

c P’ n c P’ n

27.9% 1 5.0% 50 82.7% 1 1.5% 50

3.38% 1 10.0% 50 91.1% 1 1.0% 50

5.32% 1 9.0% 50 97.4% 1 0.5% 50

4.25% 1 9.5% 50 96.4% 1 0.6% 50

4.87% 1 9.2% 50 95.2% 1 0.7% 50

5.00% 1 9.14% 50 95.0% 1 0.72% 50

Excel Function: =BINOMDIST(c,n,p’,1)861

90% C.I. on p’ for n=50, c=1

Page 22: S S Lewis 2/15/051 Confidence Intervals and Maximum Errors By Sidney S. Lewis For Baltimore Section, ASQ February 15, 2005

S S Lewis 2/15/05 22

STATISTICAL CALCULATION EXAMPLE A

TECHNIQUE: 1-SAMPLE TEST OF THE MEAN, SIGMA UNKNOWN: t-TEST

SUBJECT:Machinability: Increased by a New Practice?

GOAL: Determine whether the average machinability of steel made using a new practice in the Melt Shop can increase the machinability, with 95% CONFIDENCE (5% ).

HISTORIC DATA: The recent past average machinability is 85.0 = 0 ; = 12.2 .

DATA: Machinability data of steel made with the new practice are

91 99 83 87 98 94 86 92 85 81890.53

Page 23: S S Lewis 2/15/051 Confidence Intervals and Maximum Errors By Sidney S. Lewis For Baltimore Section, ASQ February 15, 2005

S S Lewis 2/15/05 23

STATISTICAL CALCULATION EXAMPLE A

Calcs.: X-bar = 89.6; s = 6.196; n = 10;

Maximum Error, ME = tdf) * SX-bar

= 1.833 * 1.957 = 3.6 units

90% C.I. = X-bar ± ME

= 89.6 ± 3.6 = 86.0 to 93.2 .

That means that the machinability should increase by at least 1.0 units, and may increase by 8 units.

890.53

Page 24: S S Lewis 2/15/051 Confidence Intervals and Maximum Errors By Sidney S. Lewis For Baltimore Section, ASQ February 15, 2005

S S Lewis 2/15/05 24

STATISTICAL CALCULATION EXAMPLE B

TECHNIQUE: 1-SAMPLE TEST OF PROPORTIONS – Z-TEST

SUBJECT:Cap leakers – reduction trial

GOAL: Determine whether the rates of leaking caps are lower if a new cap design is used, with 95% CONFIDENCE (= 5%).

HISTORIC DATA: Cap leaker rate = 1.2% = p’.

DATA: A trial using 2000 caps of a new design found 18 leaking caps. p = 0.90%.

890.83

Page 25: S S Lewis 2/15/051 Confidence Intervals and Maximum Errors By Sidney S. Lewis For Baltimore Section, ASQ February 15, 2005

S S Lewis 2/15/05 25

STATISTICAL CALCULATION EXAMPLE B

FORMULAS: where p is in percent.

CALCS: p = 18/2000 = 0.90%; = p0 – p = 1.2 - 0.90 = 0.30%

Maximum Error, ME = Z.05 * p = 1.645 * 0.243 = 0.400%

90% C.I. (2-tail) on the difference = ( - 0) ± ME = (0.90 – 1.20) ± 0.400 = +0.10% to -0.70%;

90% C.I. on p’: p ± ME = 0.90 ± 0.40 = 0.5% to 1.3%

zp p

a ndp p

npp

0 0 0 0100

;( )

,

890.83

p 12100 12 2000 0 243%. ( . ) / .

Page 26: S S Lewis 2/15/051 Confidence Intervals and Maximum Errors By Sidney S. Lewis For Baltimore Section, ASQ February 15, 2005

S S Lewis 2/15/05 26

CONCLUSION:

The long term leaker rate of the new caps may be 0.7%

lower than the old caps, but it may also be 0.1% higher,

which if true, says to avoid the new caps. Therefore the

data are insufficient to show, with 95% confidence, that

the new caps are definitely better, which confirms the

test of hypothesis.

STATISTICAL CALCULATION EXAMPLE B

890.83

Page 27: S S Lewis 2/15/051 Confidence Intervals and Maximum Errors By Sidney S. Lewis For Baltimore Section, ASQ February 15, 2005

S S Lewis 2/15/05 27

STATISTICAL CALCULATION EXAMPLE C

TECHNIQUE: 1-SAMPLE TEST OF A SAMPLE STANDARD DEVIATION–CHI-SQUARED TEST

SUBJECT: XYZ Digital Blood Pressure Monitor measures of systolic blood pressure

GOAL: To determine if the monitor has become more variable than when new.

HISTORIC DATA: Early evaluation of this monitor found the standard deviation to be 2.5 units.

DATA : Using the monitor, the systolic blood pressure of a patient was measured 7 times over a ten minute period. The patient sat quietly throughout the testing. The results were: 144, 147, 147, 149, 140, 140, 144, from which s = 3.51.

890.72

Page 28: S S Lewis 2/15/051 Confidence Intervals and Maximum Errors By Sidney S. Lewis For Baltimore Section, ASQ February 15, 2005

S S Lewis 2/15/05 28

STATISTICAL CALCULATION EXAMPLE C

CONFIDENCE INTERVAL: a 2-tail, 90% confidence interval will be calculated

The critical values of 2 are:

. .( ) . ; ( ) . .052

9526 12 59 6 1635

Upper C I s df. / . . / .052 3 51 12 59 6 5 08

Lower C I s / df 3.51 1.635 / 6 1.83.952

With 90% confidence, the true standard deviation lies between

1.83 and 5.05 units, which includes the earliest determined

standard deviation of 2.5.

890.72