t

UV6122

os
AMORE FROZEN FOODS (A): MACARONI AND CHEESE FILL TARGETS

Teaching Note

rP
Synopsis

Tom Jenkins, manager of Quality Services at the Amore Frozen Food plant, had to set the
fill target for Amore’s eight-ounce frozen macaroni and cheese pie. A federally approved
weight-control program specified that unless five sample pies taken every 20 minutes had an

yo
average weight of more than eight ounces, the entire 20-minute batch had to be reworked or sold
as underweight. Amore’s mechanical filling device filled amounts that had a standard deviation
of 0.22 ounces. Although industry practice called for a fill target of 8.22 ounces (so that 85% of
the pies would be in compliance), the case provides enough information to make an economic
decision on the fill target—rather than rely on common practice.
op
The analysis of the case requires the use of the normal distribution to measure the
probability that the five sample pies will average eight ounces. As such, the case may be used to
introduce the distribution of a sample average.

Objectives
tC

This teaching note assumes that the case will be used at the beginning of a module on
sampling. In this role, the main objective for the case will be to motivate discovery of the
distribution of a sample average. The students should already know that the standard deviation of
the sum of several independent uncertain quantities is the square root of the sum of squared
standard deviations of the individual uncertain quantities. With that knowledge in place, it will
No

be a short step to develop the standard deviation of a sample average. In summary, the objectives
for the class session are

 To motivate and develop the standard deviation of a sample average

 To clarify the importance and relevance of the independence assumption
 To describe a simple statistical quality-control system
Do

This teaching note was prepared by Associate Professor Phillip E. Pfeifer. Copyright  1985 by the University of
Virginia Darden School Foundation, Charlottesville, VA. All rights reserved. To order copies, send an e-mail to
[email protected]. No part of this publication may be reproduced, stored in a retrieval system, used in a
spreadsheet, or transmitted in any form or by any means—electronic, mechanical, photocopying, recording, or
otherwise—without the permission of the Darden School Foundation. Rev. 1/07.

This Teaching Note is authorized for use only by SHAILAJA REGO, Narsee Monjee Institute of Management Studies (NMIMS) until Jun 2023. Copying or posting is an infringement of
copyright. [email protected] or 617.783.7860.
 t
-2- UV6122

os
Please note that an exhibit containing a table of normal probabilities was removed from
the case in 2005. This was done under the assumption that students would be able to calculate
the necessary probabilities using other means (such as Excel’s NORMDIST function). If
instructors want to use the earlier version of the case, they may request it from the author,
Professor Phil Pfeifer, at [email protected].

rP
Assignment Questions

1. If the fill target is lowered to 8.22 ounces, how many 20-minute batches will fail to meet
the FDA-approved standard because the five sample pies average less than eight ounces
per pie?

yo
2. Prepare a recommendation to Jenkins regarding lowering the target to 8.22 ounces.

Analysis

The following assumptions will be made to facilitate the analysis that follows:
op
1. The mechanical filler fills to target with a standard deviation of 0.22 ounces.
2. The five sample pies’ weights are independent.
3. The production rate is 1,000 dozen every 20 minutes.
tC

4. The monthly production is 60 20-minute batches, or 60,000 dozen pies. Yearly

production is 720 batches.
5. Rejected batches are first sold at the company thrift store and then donated to charity
once the thrift store demand has been filled.
6. The thrift store demand is 240 dozen (4 × 60) per month.
No

After the preliminary analysis, this note will examine the sensitivity of the results to these
assumptions.

At the 8.44 target

At this target, the five sample pies will have an average weight that is normal with mean
8.44 ounces and standard deviation of 0.0984 (0.22  5 ) . The probability that the sample
Do

average will be less than eight ounces is the probability that a standard normal is less than –4.47
(or –0.44 ÷ 0.0984). There is almost no chance of rejecting a batch. This situation is consistent
with having only one batch rejected during the first six months of 1984.

The cost of ingredients at the 8.44 target is $1,820 per batch (or $1.82 × 1,000). The
$1.82 ingredient cost per dozen is taken from case Exhibit 2. At 60 batches per month, this

This Teaching Note is authorized for use only by SHAILAJA REGO, Narsee Monjee Institute of Management Studies (NMIMS) until Jun 2023. Copying or posting is an infringement of
copyright. [email protected] or 617.783.7860.
 t
-3- UV6122

os
equals $1,310,400 per year. Because no batches will be rejected, total revenues will be $4,500
per batch ($4.50 × 1,000) or $3,240,000 per year.

At the 8.22 target

rP
The five-pie average weight at this target will have mean 8.22 ounces and a standard
deviation of 0.0984. The probability of rejecting a batch is the probability that the standard
normal is less than –2.236 (or –0.22 ÷ 0.0984). Using the normal probability function in Excel
we get a calculated probability of 0.0127, which is calculated as either

=NORMDIST[–2.236,0,1,TRUE] or
=NORMDIST[8,8.22,0.0984,TRUE]

yo
On a yearly basis, Amore could expect 9.14 rejected batches (0.0127 × 60 × 12). The
annual demand at the thrift store is 3.12 batches (60 × (52 ÷ 1,000)). Thus, on average (assuming
no stock-outs at the thrift store and ignoring the uncertainty in the number of rejected batches per
year), 3.12 batches will be sold for $3.60 per dozen, 6.02 batches (or 9.14 – 3.12) will be
donated to charity, and 710.86 batches (720 – 9.14) will be sold for $4.50 per dozen. Total yearly
op
revenues are then (710.86 × $4.50 × 1,000) + (3.12 × $3.60 × 1,000), or $3,210,102.

The yearly cost of ingredients will be 8.22 ÷ 8.44 of the $1,310,400 cost at the 8.44
target. This amounts to $1,276,560 per year or $1,773 per batch.

Comparison and recommendation

tC

Table 1 summarizes the relevant costs and revenues at the two fill targets.

Table 1. Costs and revenues by fill target.

8.22 8.44
No

Yearly Per Batch Yearly Per Batch

Material Cost $1,276,560 $1,773 $1,310,400 $1,820
Revenues $3,210,084 $4,458 $3,240,000 $4,500
Contribution $1,933,524 $2,685 $1,929,600 $2,680

The recommendation is to lower the target to 8.22. The ingredient cost savings of $47 per batch
($1.82 × 1,000 × (0.22 ÷ 8.44)) is greater than the $42 per batch expected lost revenue. Notice
that, if $4,500 is used as the cost of rejecting a batch (ignoring the thrift shop), the expected lost
Do

revenue per batch would be $57 (0.0127 × $4,500) and the 8.44 target is preferred. If only $900
is used as the lost revenue (assuming all rejected batches can be sold at the thrift store for $0.90
less per dozen), the expected lost revenue would be $11 (or 0.0127 × $900) and the 8.22 target is
preferred. As it is, our calculation has slightly understated lost revenues by assuming no queuing
problems at the thrift store.

This Teaching Note is authorized for use only by SHAILAJA REGO, Narsee Monjee Institute of Management Studies (NMIMS) until Jun 2023. Copying or posting is an infringement of
copyright. [email protected] or 617.783.7860.
 t
-4- UV6122

os
Data Analysis

Although the data in case Exhibit 3 were provided only to illustrate the weight-control
report, many students will rely on it to provide representative data on the sample averages of five
pies. To convert the individual observations to ounces, divide by 50 and add 8.44. Thus, the first

rP
pie listed as +10 weighed (10 ÷ 50) + 8.44, or 8.64 ounces. The first five-pie sample weighed
+38 (42.96 ounces), or 8.59 ounces per pie. The sum of the 15 sample averages was 127.392
ounces, and the grand overall average was 8.493 ounces. Nine of the 75 individual pies, 12%,
were underweight (here meaning less than 8.22 ounces). Assuming the normal distribution, 15%
would be expected.

Are the data consistent with the assumptions? Are the 75 observations representative of a

yo
normal distribution with mean 8.44 ounces and standard deviation of 0.22 ounces? The standard
deviation is calculated to be 0.205, fairly close to the assumed value. The overall sample mean of
8.493 is higher than expected. The t-statistic to measure the likelihood of seeing a sample
average of 8.493 if the mean is 8.44 is

8.493  8.44
t= = 2.2
op
0.205  75

which gives some evidence that the fill target was actually greater than 8.44.

A histogram of the 75 pie weights exhibits some skewing to the left (Figure 1). This
tC

makes sense because of the physical limit to how fully the tin can be filled. With only five pie
weights averaged, this skewing will also be present in the averages—producing a slightly higher
rejection rate than calculated using the normal distribution. Higher rejection rates favor the safer,
higher-fill target.
No
Do

This Teaching Note is authorized for use only by SHAILAJA REGO, Narsee Monjee Institute of Management Studies (NMIMS) until Jun 2023. Copying or posting is an infringement of
copyright. [email protected] or 617.783.7860.
 t
-5- UV6122

os
Figure 1. Histogram of pie weights.

rP
6

4
Frequency

yo
3

2
op
1

0
8 8.1 8.2 8.3 8.4 8.5 8.6 8.7 8.8 8.9 9

Ounces
tC

What can be said about the independence of the process? The plot of the 15 sample
averages in case Exhibit 3 appears to be smoother than expected from a random series. A too
smooth plot is an indication that the successive sample averages are closer to each other than
they should be if the process were random. Perhaps the process mean was slowly changing
(drifting) over the day. A slowly changing process mean (in contrast to a constant mean)
produces a smoother plot. To quickly test this idea, we count five runs in the 15-point plot (a run
No

being successive points on either side of the zero line). Five runs is fewer than expected, given
15 observations. This run count confirms our judgment that the series is smoother than expected.
Consulting an appropriate statistical table, however, tells us the result could easily be due to
chance.

As another check on the independence of the process, we compare the sample standard
deviation of the individual pie weights to the sample standard deviation of the 15 sample
averages. The sample standard deviation of the 75 individual pie weights was 0.205. If these
Do

individual pie weights were independent (if the process filled pies randomly), then the standard
deviation of an average of five pies should be 0.092—calculated as

(0.205  5 )

This Teaching Note is authorized for use only by SHAILAJA REGO, Narsee Monjee Institute of Management Studies (NMIMS) until Jun 2023. Copying or posting is an infringement of
copyright. [email protected] or 617.783.7860.
 t
-6- UV6122

os
Recall that this calculation requires independence. If we calculate the sample standard deviation
of the 15 five-pie averages, we get 0.074. The calculated standard deviation of the averages is
less than what we would expect if the individual pie weights were independent. We have
evidence that the individual pie weights were not independent. We have evidence that the five-
pie averages do not jump around as much as they should (if the individual pie weights were

rP
random). It appears the individual pie weights tend to bounce back and forth or oscillate (extra
heavy pies tend to be followed by extra light pies and vice versa). Perhaps there are differences
in the heads on the filling machine. Or perhaps an above-average deposit in one pie “cleans out”
the filling mechanism so that the next pie gets a below-average dose. No matter what the cause
of this oscillation, it works to Amore’s advantage. A smaller standard deviation of the five-pie
averages means fewer rejected batches. Note that this oscillation of individual pie weights will
affect (improve) the standard deviation of the sample averages only if the sample pies are taken

yo
in succession. If the sample pies are chosen randomly from the entire batch, the standard
deviation of five-pie average would be the higher 0.092. Amore benefits because the sample pies
are taken in succession.

Sensitivity
op
We will now examine the sensitivity of our conclusion—lower the fill target to 8.22
ounces—to the assumptions we have made.

The data suggest that, although the mechanical filler may be able to be on target over the
long run, the process mean may drift. A drifting mean works against lowering the fill target.
tC

To illustrate the effects of a drifting mean, consider a process with an overall mean of
8.22 ounces and standard deviation of 0.22 that arises because half the batches have a mean of
8.30 and half have a mean of 8.14. In order to produce an overall standard deviation of 0.22, the
conditional standard deviation must be 0.205.1 Batches with a mean of 8.30 have almost no
chance of being rejected (Z = –3.27). When the mean is 8.14, however, the standard deviation of
No

the sample mean of

0.0917  (0.205  5 )

implies a Z of –1.527 (–0.14 ÷ 0.0917). Thus, that mean gives a 0.063 probability of rejecting the
batch half of the time. The probability of rejecting a randomly selected batch is thus 0.032,
which is a much higher rejection rate than the 0.0127 calculated, assuming a constant mean of
8.22 with standard deviation of 0.22. The point is that a drifting mean results in more rejected
Do

batches than expected. At this higher rejection rate, we would recommend against lowering the
target.

1
Var(X) = E[X2] × E[X]2
0.222 = (½)(0.2052 + 8.302) + (½)(0.2052 + 8.142) × 8.222

This Teaching Note is authorized for use only by SHAILAJA REGO, Narsee Monjee Institute of Management Studies (NMIMS) until Jun 2023. Copying or posting is an infringement of
copyright. [email protected] or 617.783.7860.
 t
-7- UV6122

os
The foregoing analysis also highlights the importance of understanding just what the case
standard deviation of 0.22 represents. Does the 0.22 measure the capability of the filling
machine? In this interpretation, 0.22 is the standard deviation of pie weights produced when the
machine is under control and on target. The 0.22 is a measure of the best the machine can do. Or,
in contrast, does the 0.22 measure the overall performance of the pie-filling process? In this later

rP
interpretation, 0.22 is the long-run standard deviation of pie weights actually producible by the
filling machine (which may drift out of control) and the accompanying Amore weight-control
system (which monitors the filling machine and takes corrective action as needed). Our analysis
uses the latter interpretation.

The data also suggest that the process does not produce successive pie weights that are
independent. If the pie weights do oscillate, it is indeed a good idea (from Amore’s perspective)

yo
to sample pies successively. Oscillating weights reinforce the decision to lower the target.

If Amore lowers the production rate, this reinforces the decision to lower the target.
Under the current assumptions, all thrift store demand is being met. If fewer pies are produced,
fewer pies will be rejected, and fewer pies will be donated to charity. Likewise, if Amore could
increase thrift store demand, or find a better alternative than charity, we would be more likely to
op
recommend lowering the target.

If rejected batches are reworked rather than donated to charity, Amore may be able to
recoup the ingredient costs. Assume ingredient costs of (8.22 ÷ 8.44) × $1.82, or $l.75 per dozen.
Subtracting $0.70 in labor costs for rework (an arbitrary amount equal to 10 times the total direct
labor used to fill a dozen tins), results in $1.05 per dozen, or $l,050 per batch. This rework
tC

savings for the expected 6.02 batches per year donated to charity would increase expected
revenues at the 8.22 target by $6,321 per year (6.02 × $1,050), or $9 per batch. This possibility
reinforces the decision to lower the target to 8.22.

Pedagogy
No

The teaching plan described here is fairly directive and safe. We present the issues, then
we list the tasks necessary to make a recommendation. This method provides a framework for
the rest of the class, as each task is addressed in turn. The case could also be taught by starting
with a recommendation and then asking for supporting evidence. This less directive approach
will undoubtedly lead to discussions of the various pieces of necessary analysis. The plan
outlined here will provide guidance as to how to structure such a discussion.
Do

1. “What are the issues here?”

So as not to lose sight of the big picture, this approach starts with an identification of the
important issues in the case. Here the trade-off seems to be between ingredient costs and the risk
of rejected batches. If the potential for local fines for underweight pies is mentioned by students,

This Teaching Note is authorized for use only by SHAILAJA REGO, Narsee Monjee Institute of Management Studies (NMIMS) until Jun 2023. Copying or posting is an infringement of
copyright. [email protected] or 617.783.7860.
 t
-8- UV6122

os
it may be a good idea to ask how big a problem this has been lately. The case makes clear that
these fines are virtually nonexistent, now that the energy crisis is over.

In this initial discussion, expect some students to be confused about the difference
between FDA-approved weight-control system and Amore’s system. The FDA rules specify

rP
rejecting batches if the sample average is less than eight ounces. Amore’s own system requires
the operator to react if the sample average was less than 8.22 ounces. Try to shift the focus to the
FDA weight-control program. Here is where Amore runs the risk of costly, rejected batches. The
trade-off seems to be ingredient savings from reducing the fill target versus increased chance of
having to reject a batch because it fails the FDA rule.

yo
2. “What tasks or analyses need to be performed before we can make a recommendation?”

The responses here will provide the framework for the remaining class discussions. The
three major tasks are to discover:

a. Cost savings at the 8.22 target

b. Number (probability) of rejected batches
op
c. Cost of rejected batches

The instructor should try to classify any subtasks mentioned into one of these three categories. A
fourth category, the number of underweight pies, may be added if it is mentioned, but the
calculation of this number (or percentage) may also be looked at as a first step toward evaluating
tC

the number of rejected batches.

3. “What are the cost savings at 8.22?”

This calculation is the easiest piece of analysis. It should be clear that the relevant cost
savings come from saving ingredients only: (0.22 ÷ 8.44) × ($1.82 × 1,000) or $47 per batch.
No

Someone may suggest that Amore will need to produce more batches at an 8.22 target (to replace
rejected batches). Producing more batches means incurring more than just the $1.82 ingredient
cost. The suggestion is that the ingredient cost savings from lowering the target must be net of
the costs of rejected batches. Suggest that the cost of rejected batches be handled separately. If
we reduce the target, we save $47 per batch. Now we turn to the question of number of batches.

4. “At an 8.22 target, how many batches will be rejected?”

Do

This is the big question of the day. Plan to spend much of the class discussing it.

You might find students attacking this question by noting that at 8.22, 15% of the pies
will be less than eight ounces. If all five of the sample pies are underweight, then so is the
sample. This happens with probability (0.15)5 = 0.0000759. They might also reason that four out
of five underweight pies also means an underweight sample. This event has probability 0.00215,

This Teaching Note is authorized for use only by SHAILAJA REGO, Narsee Monjee Institute of Management Studies (NMIMS) until Jun 2023. Copying or posting is an infringement of
copyright. [email protected] or 617.783.7860.
 t
-9- UV6122

os
calculated from the binomial. Three out of five underweight pies have probability 0.02438 but
that does not necessarily imply an underweight average. This approach is worth listening to, but
the class will soon realize that it will not work.

You will find students who relied solely on the data in case Exhibit 3 to answer this

rP
question. Ask first for an outline of the approach used and discuss whether it is worth pursuing.
It is certainly worthwhile to consider using the 15 sample averages to measure the uncertainty in
sample averages at the new target. Go slowly through the conversion from 50ths of an ounce to
ounces. The 15 averages might be put in a sample CDF, or the t-distribution might be used. The
use of the t-distribution is especially enlightening, because it illustrates how to calculate the
probability of rejecting a batch once you know the mean and standard deviation of sample
averages. Both approaches are reasonable, but the point should eventually be made that the class

yo
is using only a small number of data points, when in fact Amore has a vast history of experience
with the filling machine. Could this question be answered without resorting to the small sample
of data in case Exhibit 3?

If the class has trouble “letting go” of case Exhibit 3, it may help to ask them to step back
and gain some perspective on the purpose of analyzing that data. Remember that Amore believes
op
(based on years of experience with the filling machine) that it fills pies to target (8.44 ounces in
case Exhibit 3) with a standard deviation of 0.22 ounces. The data in case Exhibit 3 may or may
not support that belief. And it is certainly worthwhile to examine how consistent the data are
with Amore’s beliefs. But we must not lose sight of the fact that the data in case Exhibit 3 are
from one day only. Certainly we should not ignore the case statements about the capability of the
filling process. Certainly we should explore the implications of Amore’s beliefs that the machine
tC

can fill pies to target with a standard deviation of 0.22 ounces.

At this point the instructor may need to become more directive. Asking for the
distribution of individual pie weights should elicit the fact that pie weights can be assumed to be
normally distributed with mean 8.22 and standard deviation 0.22.
No

Next we need to determine what specific question we want to answer. The proper
response is: we want to measure the chances that the average of five pie weights is less than
eight ounces. A series of questions should point the class in the right direction: Is that not the
same as the chances that the total of five pie weights is less than 40 ounces? What do we expect
for the sum of five pie weights? We expect 8.22 × 5, or 41.1 ounces. How much uncertainty is
there in this sum? What is the standard deviation of this sum? It is the square root of the sum of
the squared standard deviations of the individual pie weights, 0.492 ounces (see Table 2). How
do we go from total to sample average? We divide by five. What do we expect for the average of
Do

five pie weights? We expect 8.22 ounces, obviously. What is the standard deviation of this
average? It is one-fifth the standard deviation of the total, 0.492 ÷ 5 or 0.0984.

This Teaching Note is authorized for use only by SHAILAJA REGO, Narsee Monjee Institute of Management Studies (NMIMS) until Jun 2023. Copying or posting is an infringement of
copyright. [email protected] or 617.783.7860.
 t
-10- UV6122

os
Table 2. Summary of expected pie weights and their standard deviations.

Pie Number Expected Weight Std. Deviation

1 8.22 0.22
2 8.22 0.22

rP
3 8.22 0.22
4 8.22 0.22
5 8.22 0.22
2
Total 41.10 (5  0.22 ) = 0.492
Average 41.1 ÷ 5 = 8.22 0.492 ÷ 5 = 0.0984

yo
In general, if a sample of size n is taken from a population with mean  and standard
deviation σ, the mean of the sample average will also be  , and the standard deviation of the
sample average will be  / n .

The final question, of course, is what assumptions this equation requires. What has to be
true in order to get just the right blend of highs and lows canceling out? The uncertain quantities
op
must be independent. What does independence mean here? It means the filler mechanism fills
randomly. Our sample is not random, because pies are picked in succession. Given a mean and
standard deviation for sample averages, it should be clear how to calculate the probability of
rejecting a batch.

If the data in case Exhibit 3 have also been used to get the standard deviation of sample
tC

averages, you should ask why the results are not the same. The answer is that the standard
deviation of the 75 individual weights was slightly less than 0.22 and there was more canceling
out than expected based on an assumption of independence. The 75 points exhibit a negative first
lag autocorrelation, meaning that pie weights tended to oscillate. The sample averages thus have
a lower standard deviation than predicted, a fortunate characteristic because it means fewer
rejected batches. The major point of this discussion is to understand the importance of the
No

independence assumption when using the distribution of the sample mean.

Some students will be uneasy with the above analysis for some very good reasons they
may not be able to articulate. The analysis has assumed that the machine will always be on
target. And if it is always on target, we calculated the probability the FDA rules will reject a
batch. Students are uneasy with this, because they know that the rejected batch was not really a
bad batch. It was simply an unlucky batch. But the students know that the purpose of the FDA
rules is not to reject good, but unlucky, batches, but rather to catch and reject truly bad batches.
Do

The students are uncomfortable, because our analysis started with an assumption that there will
be no truly bad batches. Under this assumption, is there really a need for a weight-control
system? The purpose of the FDA rules is to catch bad batches, but our analysis precludes the
possibility of our producing a truly bad batch.

This Teaching Note is authorized for use only by SHAILAJA REGO, Narsee Monjee Institute of Management Studies (NMIMS) until Jun 2023. Copying or posting is an infringement of
copyright. [email protected] or 617.783.7860.
 t
-11- UV6122

os
The answer to these uneasy students is that these are indeed valid concerns. We would
really like to take the next step in our analysis and allow for the possibility that the machine
occasionally “slips” and produces truly bad batches. Ask them to think through such an analysis.
We would need to consider how often the process slips, how big the slips are, how predictable
they are, and how detectable and correctable they are. This relates directly to the previous

rP
discussion about the interpretation of the 0.22 standard deviation. Are the effects of machine
slips included in the 0.22 measure? If so, then the independence assumption is certainly violated.
In summary, our analysis has taken a first step. It has measured the probability of rejecting a
batch even if the machine is always on target. Our analysis does fall short of solving Amore’s
real problem, but it is a start.

yo
5. “What is the cost of a rejected batch?”

Addressing this final piece of the puzzle is important in closing the discussion. The cost
of a rejected batch could be $900 lost revenue if it is sent to the thrift store and $4,500 lost
revenue if donated to charity. Another approach would be to assume that rejected batches are
valued at their direct cost (about $2.51 per dozen or $2,510 per batch). The $900 and $4,500
figures apply if the plant is operating at capacity; otherwise the $2,510 figure is relevant. If the
op
plant is not at capacity, rejected batches could be scrapped (or donated to charity) and simply
“replaced” by one extra batch produced at a cost of $2,510. Which cost is used is not as
important as pushing toward a preliminary recommendation to close the discussion. As always,
the process is more important than the result.
tC
No
Do

This Teaching Note is authorized for use only by SHAILAJA REGO, Narsee Monjee Institute of Management Studies (NMIMS) until Jun 2023. Copying or posting is an infringement of
copyright. [email protected] or 617.783.7860.