Control Charts for Variation and Mean SECTION 14-2 1

Chapter 14

Statistical Process Control

14-2 Control Charts for Variation and Mean

1. No. Control charts are constructed entirely from the observed data, and they are totally

independent of any product specifications. The upper and lower control limits of control

charts, for example, are based on the actual behavior of the process under study, and not the

desired behavior. It is possible for the process of filling 12 oz cans of Coke to be well within

statistical control by consistently filling the cans with, say, 11.5 oz.

3. Random variation is the natural fluctuation due to chance alone that is inherent in every

process which does not function with perfect precision. Assignable variation is deviance from

perfect precision that can be attributed to an identifiable source – e.g., machinery that needs

adjustment.

5. a. Within statistical control.

b. Not applicable; none of the three criteria applies.

c. No; the variability is too large. Even though the process may be within control statistically,

it is not within control practically. For the amounts of cola in 12 oz cans to vary from 10 oz

to 13 oz is clearly unacceptable.

7. a. Not within statistical control.

b. Criterion 3 applies; there are 8 consecutive points below the centerline. For these data,

the criterion applies two times – as there are also 8 consecutive points above the

centerline.

c. No; there appears to have been a shift in the process after observation 8, as all the values

before then are below the centerline and all the observations after that are above the

centerline. It is usually important to identify the cause for such a shift.

Exercises 9-12 refer to the following table.

decade 00 01 02 03 04 05 06 07 08 09 R

1880 290.7291.0291.2291.4291.6291.9292.1292.3292.6292.9291.77 2.2

1890 293.2293.5293.8294.1294.3294.6294.9295.2295.5295.8294.49 2.6

1900 295.6295.3295.1294.8295.9296.9297.5298.1298.6299.2296.70 4.4

1910 299.4299.6299.9300.1300.3300.5300.7300.9301.1301.2300.37 1.8

1920 301.4301.6302.3302.9303.6304.2304.9305.5305.6305.8303.78 4.4

1930 305.9306.1306.2306.3306.5306.6306.8306.9307.1307.3306.57 1.4

1940 307.4307.6307.7307.9308.4308.9309.3309.8310.3310.8308.81 3.4

1950 311.3311.7312.2312.7313.2313.7314.3314.8315.3316.0313.52 4.7

1960 316.9317.6318.5319.0319.5320.1321.3322.1323.1324.6320.27 7.7

1970 325.7326.3327.5329.6330.3331.2332.2333.9335.5336.9330.91 11.2

1980 338.7340.0341.1342.8344.4345.9347.1349.0351.4352.9345.33 14.2

1990 354.2355.6356.4357.1358.9360.9362.6363.8366.6368.3360.44 14.1

2000 369.5371.0373.1375.6377.4379.6379.6381.2382.8384.4377.42 14.9

9. There are k=13 samples (decades) of n=10 observations (years) each.

= 4150.38/13 = 319.26

= 87.0/13 = 6.692

For the R chart: UCL = D4= 1.777(6.692) = 11.89

LCL = D3= 0.223(6.692) = 1.49

For the chart: UCL = + A2= 319.26 + 0.308(6.692) = 319.26 + 2.06 = 321.32

LCL = – A2= 319.26 – 0.308(6.692) = 319.26 – 2.06 = 317.20

11. The R chart is given below. The process variation is not within statistical control because

there is an upward trend, there are 8 consecutive points below the centerline, and there are

points above the upper control limit.

13. There are k=8 samples (years) of n=6 observations (bi-monthly readings) each.

= 22253.67/8 = 2781.71

= 13835/8 = 1729.375

For the R chart: UCL = D4= 2.004(1729.375) = 3465.67

LCL = D3= 0.000(1729.375) = 0

For the chart: UCL = + A2= 2781.71 + 0.483(1729.375)

= 2781.71 + 835.29 = 3617.00

LCL = – A2= 2781.71 – 0.483(1729.375)

= 2781.71 – 835.29 = 1946.42

15. The chart is given below. The process mean is within statistical control. None of the

three criteria for being out of statistical control apply.

17. There are k=8 samples (weeks) of n=5 observations (weekdays) each.

= 989.30/8 = 123.66

= 3.0/8 = 0.375

For the R chart: UCL = D4= 2.114(0.375) = 0.793

LCL = D3= 0.000(0.375) = 0

For the chart: UCL = + A2= 123.66 + 0.577(0.375)

= 123.66 + 0.22 = 123.88

LCL = – A2= 123.66 – 0.577(0.375)

= 123.66 – 0.22 = 123.45

19. The run chart is given below. There appears to be a pattern of long range cycles which suggest

that the process may not be within statistical control. Several consecutive day of lower voltage

are followed by several consecutive days of higher voltage, followed by several consecutive

days of lower voltage, followed by several consecutive days of higher voltage.

21. There are k=13 samples of size n=10 each.

= Σs/k = 1.56029/13 = 0.1200

UCL = B4= 1.716(0.1200) = 0.2060

LCL = B3= 0.284(0.1200) = 0.0341

This is very similar to the R chart given in the text, and both charts indicate that the process

variation is within statistical control. For relatively small n’s in the sub-samples (n=10 in this

example), there will be a very high correlation between the values of R and s – and so the two

charts will be almost identical, but with different labels on the vertical axis.

14-3 Control Charts for Attributes

1. No. Control charts determine whether a process is within statistical control relative to the

values it is actually producing, not relative to the values it is supposed to be producing.

NOTE: The text defines as the pooled estimate of the proportion of successes among all items sampled (i.e., the weighted mean of the k values for p) and gives

= (Σx)/(Σn)

= (total number of “successes” among all items samples)/(total number of items sampled).

When there are k samples of size n each, this reduces to = (Σp)/k. Since each of k values for p must be calculated to make the chart, and since this is consistent with the notation in the last

section that this manual uses = (Σp)/k as the primary formula.

3. There are k=5 samples of size n=100 each.

= (Σp)/k = 0.10/5 = 0.02

= = 0.014

UCL = + 3 = 0.02 + 3(0.014) = 0.02 + 0.042 = 0.062

LCL = – 3 = 0.02 – 3(0.014) = 0.02 – 0.042 = -0.022, adjusted to 0.00

Yes. The lower control limit was adjusted from -0.022 to 0.00 because negative values are not

possible for proportions. Since the process is not within statistical control when data fall

outside the control limits (i.e., not on the control limits), a process with the above control limits

cannot fail to be within statistical control by having a sample proportion that is too low.

Control Charts for Attributes SECTION 14-3 1

5. The process is within statistical control; none of the three criteria applies. In this case,

however, there are warning signs that suggest continuing monitoring of the process is in order

– two of the data points are dangerously close to the upper control limit, and the final n=7

consecutive data points below the centerline (along with the general pattern of the data) are

dangerously close to indicating the presence of long term cycles.

7. The process is not within statistical control – and in this case, all three criteria apply. The data

end with an upward trend. There is a data point above the upper control limit. There are at

least 8 consecutive data points above the centerline.

9. There are k=20 samples of n=10,000 each.

= (Σp)/k = (0.0020 + 0.0014 + 0.0022 +…+ 0.0020)/20 = 0.0400/20 = 0.00200

= = 0.0004468

UCL = + 3 = 0.00200 + 3(0.0004468) = 0.00200 + 0.00134 = 0.00334

LCL = – 3 = 0.00200 – 3(0.0004468) = 0.00200 – 0.00134 = 0.00066

The p chart is given below. The process is within statistical control. None of the three criteria

for being out of statistical control applies.

11. There are k=15 samples of n=1000 each.

= (Σp)/k = (0.601 + 0.625 + 0.619 +…+ 0.667)/15 = 9.523/15 = 0.6349

= = 0.015225

UCL = + 3 = 0.6349 + 3(0.015225) = 0.6349 + 0.0457 = 0.6805

LCL = – 3 = 0.6349 – 3(0.015225) = 0.6349 – 0.0457 = 0.5892

The p chart is given below. The process is within statistical control. None of the three criteria

for being out of statistical control applies. Whether or not the enrollments are high enough

depends on the desired values – but control charts determine whether a process is within

statistical control relative to the values it is actually producing, not relative to the values

desired.

13. There are k=23 samples of n=1000 each.

= (Σp)/k = (0.585 + 0.454 + 0.577 +…+ 0.514)/23 = 10.199/23 = 0.4434

= = 0.0157099

UCL = + 3 = 0.4434 + 3(0.0157099) = 0.4434 + 0.0471 = 0.4906

LCL = – 3 = 0.4434 – 3(0.0157099) = 0.4434 – 0.0471 = 0.3963

The p chart is given below. The process is not within statistical control because there is an

alternating trend, there are points above the upper control limit, and there are points below the

lower control limit. The alternating trend is caused by the fact that the odd-numbered samples

represent presidential election years and the even-numbered samples represent non-presidential

election years. In addition, there may be a long-term cycle: it appears that voting in both types

of elections declined, reached a low point, and is beginning to increase again.

15. There are k=20 samples of size n=10000 each. The example calculated = 0.001.

The np chart is literally the p chart multiplied by n. It is based on x=np instead of on p.

n∙ = 10000(0.001) = 10

[OR n∙ = n∙(Σp)/k = (Σnp)/k = (Σx)/k = (15+12+14+…+7)/20 = 200/20 = 10]

n∙ = = = 3.1607

UCL = n∙ + 3 = 10.00 + 3(3.1607) = 10.0 + 9.48 = 19.48

LCL = n∙ – 3 = 10.00 – 3(3.1607) = 10.0 – 0.48 = 0.52

The np control chart that follows is identical to the p control chart in the text except for the

labeling on the vertical axis. The values on the vertical axis of the np chart are n=10000 times

the values on the p chart – e.g., the np UCL is 10000(0.001948) = 19.48, the np center line is

10000(0.001) = 10, and the np LCL is 10000(0.000052) = 0.52.

Statistical Literacy and Critical Thinking

1. Statistical process control is the monitoring of data over time in order to detect trends or

problems that indicate the production process is out of statistical control and/or problems that

lead to the production of a product that does not meet prescribed specifications.

2. It is important to monitor the manufacturing process over time because conditions affecting the

product may occur over time – i.e., the hiring of new employees, the wearing out of the

machinery, etc. A possible adverse consequence of not monitoring the manufacturing process

would be the financial and/or public relations loss in dealing with products not meeting the

proper specifications leaving the plant and entering the market. In addition, the product could

become so far out of specification that only a temporary shutdown (and subsequent loss of

revenue) could remedy the problem.

3. In general, a process can go out of control because of a change in mean or a change in

variation. The fact that the mean is within statistical control does not indicate that the variation

is within statistical control, and vice-versa. For this reason, one should monitor a process with

both an chart and an R chart.

4. No. Control charts determine whether a process is within statistical control relative to the

values it is actually producing, not relative to the values it is supposed to be producing.

Chapter Quick Quiz

1. Process data are values in chronological sequence that measure a characteristic of a

controllable activity.

2. Random variation is the natural fluctuation due to chance alone that is inherent in every

process which does not function with perfect precision. Assignable variation is deviance from

perfect precision that can be attributed to an identifiable source – e.g., machinery that needs

adjustment.

3. The test states a process is out of statistical control if any of the following three criteria applies.

(1) There is a data pattern that is obviously not random.

(2) There is a data value above the upper control limit or below the lower control limit.

(3) There are 8 consecutive data points either all above or all below the centerline.

4. An R chart monitors the variation of a process, while an chart monitors the mean of a

process.

5. No. The process variation is not within statistical control because the R chart indicates a data

point that is above the upper control limit.

6. The R chart indicates that = 21.2. In this context, is the mean of the k=20 values for R

obtained from the 20 subsamples of size n.

7. No. The process mean is not within statistical control because the X-bar chart indicates that

there are at least 8 consecutive data points below the centerline.

8. The X-bar chart indicates that = 6.45. In this context, is the mean of the k=20 values for

obtained from the 20 subsamples of size n.

9. A p chart is a control chart to monitor the proportion of items having a particular attribute over

time. It is used to determine whether the corresponding process is within statistical control.

10. False. A process is out of control when it appears to be under the influence of some factor or

force that causes other than the usual expected random deviation. That factor or force could be

causing a smaller proportion of defects, or it could be acting in some way (e.g., increasing the

variation) that does not alter the overall proportion of defects.

Review Exercises 1

Review Exercises

1. The run chart is given below. There is not a pattern suggesting that the process is not within

statistical control.

2. There are k=3 samples of n=7 observations each. For the R chart,

= (78+77+31)/3 = 62.0

UCL = D4= 1.924(62.0) = 119.3

LCL = D3= 0.076(62.0) = 4.7

3. The process variation is within statistical control because none of the three criteria for being

out of statistical control applies.

4. There are k=3 samples of n=7 observations each. For the chart,

= (252.71+247.86+270.29)/3 = 256.95

= (78+77+31)/3 = 62.0

UCL = + A2= 256.95 + 0.419(62.0) = 256.95 + 25.98 = 282.93

LCL = – A2= 256.95 – 0.419(62.0) = 256.95 – 25.98 = 230.97

5.The process mean is within statistical control because none of the three criteria for being out of

statistical control applies.

6. There are k=15 samples of n=1,000,000 each.

= (Σp)/k = (0.000085 + 0.000087 + 0.000094 +…+ 0.000056)/15

= 0.001153/15 = 0.0000769

= = 0.000008767

UCL = + 3 = 0.0000769 + 3(0.000008767) = 0.0000769 + 0.0000263 = 0.0001032

LCL = – 3 = 0.0000769 – 3(0.000008767) = 0.0000769 – 0.0000263 = 0.0000506

The p chart is given below. The process is not within statistical control because there is a

downward trend, there are 8 consecutive points above the centerline. The control chart

suggests that the downward trend in the proportion of homicides may have ended and is

leveling off.

7. There are k=15 samples of n=100 each.

= (Σp)/k = (0.05 + 0.04 + 0.03 +…+ 0.01)/15 = 0.81/15 = 0.054

= = 0.0226

UCL = + 3 = 0.054 + 3(0.0226) = 0.054 + 0.068 = 0.122

LCL = – 3 = 0.054 – 3(0.0226) = 0.054 – 0.068 = -0.014, truncated at 0.000

The p chart is given below. The process is not within statistical control because there is a

pattern of increasing variation.

Cumulative Review Exercises 1

Cumulative Review Exercises

1. The following summary information applies to all parts of the exercise.

Let CO2 = x and temperature = y.

n = 10

Σx = 3774.2 = (Σx)/n = 3774.2/10 = 377.42

Σy = 146.36 = (Σy)/n = 146.36/10 = 14.636

Σxy = 55242.261n(Σxy) – (Σx)(Σy) = 30.698

Σx2 = 1424685.94n(Σx2) – (Σx)2 = 2273.76

Σy2 = 2142.2118n(Σy2) – (Σy)2 = 0.8684

r =

= 30.698/[]

= 0.6908

a. Ho: ρ = 0

H1: ρ ≠ 0

α = 0.05 and df = 8

C.V. t = ±tα/2 = ±t0.025 = ±2.306 [or r = ±0.632]

calculations:

tr = (r – μr)/sr

= (0.691 – 0)/

= 0.691/0.2556

= 2.703

P-value = 2∙tcdf(2.703,99,8) = 0.0270

conclusion:

Reject Ho; there is sufficient evidence to conclude that ρ ≠ 0 (in fact, that ρ > 0). Yes;

there is sufficient evidence to support the claim of a linear correlation between carbon

dioxide concentration and temperature.

b. No. Linear correlation measures association, not causation. And if there does happen to be

an cause-and-effect relationship between the two variables, there is nothing in the

mathematics that would allow a person to say which variable is the cause and which variable

is the effect.

c. b1 = [n(Σxy) – (Σx)(Σy)]/[n(Σx2) – (Σx)2] = 30.698/2273.76 = 0.0135

bo = – b1= 14.636 – 0.0135(377.42) = 9.540

= bo + b1x = 9.540 + 0.0135x

d. = 9.540 + 0.0135(290.7) = 13.46 °C

The predicted temperature does not appear to be very close to the actual temperature of

13.88 °C. But a carbon dioxide concentration of 290.7 is well outside the range of x values

used to determine the regression line. Since 290.7 is (290.7 – 377.42)/5.02633 = -17.253

standard deviations from the mean of the x scores used to determine the line, the regression

line should probably not be used for that prediction.

2. There are k=10 samples (weeks) of n=200 belts each.

= (Σp)/k = (0.010 + 0.015 + 0.005 +…+ 0.085)/10 = 0.335/10 = 0.0335

= = 0.01272

UCL = + 3 = 0.0335 + 3(0.01272) = 0.0335 + 0.0382 = 0.0717

LCL = – 3 = 0.0335 – 3(0.01272) = 0.0335 – 0.0382 = -0.0047, truncated at 0.0000

The p chart is given below. The process is not within statistical control because there is a

pattern of increasing variability and there are points above the upper control limit.

3. α = 0.05 and zα/2 = z0.025 = 1.96; = x/n = 67/2000 = 0.0335

0.03350.0079

0.0256 < p < 0.0414

We have 95% confidence that the interval from 0.0256 to 0.0414 contains the true proportion

of defective seat belts produced by this process.

4. original claim: p > 0.03

= x/n = 67/2000 = 0.0335

Ho: p = 0.03

H1: p > 0.03

α = 0.05

C.V. z = zα = z0.05 = 1.645

calculations:

z= (– μ)/σ

= (0.0335 – 0.03)/

= 0.0035/0.003814

= 0.92

P-value = P(z>0.92) = 1 – 0.8212 = 0.1788

conclusion:

Do not reject Ho; there is not sufficient evidence to support the claim that p > 0.03. There

is not sufficient evidence to support the claim that the proportion of defects is greater than

3%.

Cumulative Review Exercises 1

5. a. P(all 8 above) = P(A1 and A2 and A3 and…and A8)

= P(A1)∙P(A2)∙P(A3)∙…∙P(A8)

= (1/2)(1/2)(1/2)…(1/2)

= (1/2)8 = 1/256 or 0.0039

b. P(all 8 below) = P(B1 and B2 and B3 and…and B8)

= P(B1)∙P(B2)∙P(B3)∙…∙P(B8)

= (1/2)(1/2)(1/2)…(1/2)

= (1/2)8 = 1/256 or 0.0039

c. P(all above OR all below) = P(all above) + P(all below) – P(all above AND all below)

= 1/256 + 1/256 – 0

= 2/256 = 1/128 or 0.0078

6. normal distribution: μ = 6.0, σ = 1.0

a. P(x>7.0) = P(z>1.00)

= 1 – 0.8413

= 0.1587 or 15.87%

Yes; if the helmets were made to fit only males with head breadths less than 7.0 inches, too

many men would be excluded.

b. For the smallest 5%,For the largest 5%,

A = 0.0500 and z = -1.645 A = 0.9500 and z = 1.645

x = μ + zσx = μ + zσ

= 6.0 + (-1.645)(1) = 6.0 + (1.645)(1)

= 6.0 – 1.645 = 6.0 + 1.645

= 4.355 inches = 7.645 inches

The helmets will men with head breadths from 4.355 inches to 7.645 inches.

7. No. Since the sample is a voluntary response sample, which is not necessarily representative

of the general population, it is not appropriate to use these results to make any conclusion

about the general population

8. No; the values do not appear to come from a normal distribution. The histogram indicates a

distribution that is positively skewed and far from bell-shaped, and the normal probability plot

is not close to a straight line.

9. Listed in order, the values are as follows.

0 0 0 0 0 0 0 0 0 0 0 0 5 10 20 23 35 35 40 43 62 73 90 100 130 178 185 200 250 500

The n=30 have Σx = 1979 and Σx2 = 469535.

a. = (Σx)/n = 1979/30 = 66.0 cents

b. = (20+23)/2 = 21.5 cents

c. s2 = [n(Σx2) – (Σx)2]/[n(n-1)]

= [30(469535) – (1979)2]/[30(29)] = 10169609/870 = 11689.2

s = = 108.1 cents

10. original claim: p > 0.50

= x/n = 18/30 = 0.60

Ho: p = 0.50

H1: p > 0.50

α = 0.05

C.V. z = zα = z0.05 = 1.645

calculations:

z= (– μ)/σ

= (0.60 – 0.50)/

= 0.10/0.09128

= 1.10

P-value = P(z>1.10) = 1 – 0.8643 = 0.1357

conclusion:

Do not reject Ho; there is not sufficient evidence to support the claim that p > 0.50. There

is not sufficient evidence to support the claim that most students have change in their possession.

FINAL NOTE: Congratulations! You have completed statistics – the course that everybody likes to hate. I trust that this manual has helped to make the course a little more understandable – and that you leave the course with an appreciation of broad principles, and mot merely memories of manipulating formulas. I wish you well in your continued studies, and that you achieve your full potential wherever your journey of life may lead.