Physical Inventory Using Sampling Methods

Author(s): Marion R. Bryson

Source: Journal of the Royal Statistical Society. Series C (Applied Statistics), Vol. 9, No. 3
(Nov., 1960), pp. 178-188
Published by: Wiley for the Royal Statistical Society
Stable URL: [Link]
Accessed: 26-06-2016 16:09 UTC

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
[Link]

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted
digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about
JSTOR, please contact support@[Link].

Royal Statistical Society, Wiley are collaborating with JSTOR to digitize, preserve and extend access to
Journal of the Royal Statistical Society. Series C (Applied Statistics)

This content downloaded from [Link] on Sun, 26 Jun 2016 16:09:27 UTC
All use subject to [Link]
 PHYSICAL INVENTORY USING SAMPLING METHODS
MARION R. BRYSON
Duke University, Durham, North Carolina

An integral part of any commercial activity is the periodic

counting of assets. In some larger establishments this is a
major undertaking involving many man-hours of labour. In
this paper Dr Bryson presents a method of continuous in-
ventory in which some quality control methods are used. The
basic idea is to use sampling to determine when and where
an inventory should be taken rather than use the customary
method of taking an annual inventory at a pre-set time. The
results of the application of this method at two large govern-
ment supply centres are described, and the method is com-
pared with the annual inventory method.

Introduction

Business, industry, and government are all faced with the task of
periodically taking a physical inventory of all goods on hand. The
most commonly used method of performing this physical inventory is
to make a complete count of all stocks once each period, usually
annually. All counts of stocks are made within a short period of time,
say within a week. At the close of such an inventory, it is assumed that
accurate knowledge of the quantity and value of all stocks on hand has
been obtained. Unfortunately, in most cases this may be a false
assumption.
In most of the larger and many of the smaller establishments a con-
tinuous record is kept of stocks. These records are altered as activity
changes the quantity on hand. If the records were completely accurate
it would, obviously, be unnecessary ever to make a physical count of
stocks. Many of the records are inaccurate for the following reasons:
the previous inventory was in error causing the record to reflect the
wrong balance from the beginning of the fiscal period; some receipts
and/or issues were made without correct alteration of the records;
items were lost; items were pilfered; records were lost; and other
record adjustments, such as price changes or reserve stock levels, were
incorrectly processed. Hence it is the purpose of the inventory to
correct the records which reflect wrong information.
The question which immediately arises is 'How accurate are the
records at the close of the inventory?' The personnel employed in the
inventory are relatively untrained in inventory methods since this
activity occurs during only one week out of the year. During this week
personnel are drawn from all parts of the establishment to perform
the counting and clerical duties incidental to the inventory. Many of
these people are not interested in inventory work, and if they are
178

This content downloaded from [Link] on Sun, 26 Jun 2016 16:09:27 UTC
All use subject to [Link]
 INVENTORY USING SAMPLING METHODS 179

doing satisfactory work at their normal job they know they will not be
dismissed or demoted if their poor inventory work is detected.
Because of the suspected inaccuracies in inventory and because
operational record errors occur and remain undetected for as long as
one year, an agency of the US Government sponsored a research
project to determine what could be done about improving inventory
methods.

Description of Supply Centres

The government agency mentioned above maintains supply centres

in various parts of the country. Each of these supply centres has in its
storage areas from 5 000 to 150 000 different types of item stored.
Hereafter the term 'item' will be used to denote a type of item and the
term 'piece' to denote an individual part. From zero to more than one
million pieces of each item may be on hand at any given time. The
storage areas are in the nature of huge stockrooms whose function it is
to receive and store incoming goods and issue and ship goods ordered
by their customers. The supply centre maintains records of stocks on
hand and the records department receives orders for shipments, pro-
cesses them, and instructs the stockroom to issue the stock. It also
processes all receipt vouchers and instructs the stockroom to store the
stock in an assigned location.
Each item has an assigned location or locations in the stockrooms.
When an order comes in to the records department, the records are
checked to determine whether the stock in this quantity is available
for issue. If it is not, the order is sent to a master records centre which
will re-order the stock from another storage centre. If the stock is
available, a shipping order is drawn up and sent to the stockroom.
This shipping order designates the item number, location, and quantity
to be shipped, among other facts. If the item is stored in more than
one location, one of the locations is designated as the master location.
The recorded balances are kept by item number and not by location;
hence the single balance kept by the records department reflects the
sum of the quantities in each storage location. The records department
adjusts its balance to reflect the issue. Receipts are processed in
essentially the same way.

Basic Plan of Operation

One such supply centre was selected by the government agency as
the site of the study. After nine months of preliminary work at this
centre, an additional centre was assigned for further study and testing.
Each of these two centres has approximately 100 000 items in its
stockrooms.
Various methods of quality control have been used from time to
time in the actual taking of a shut-down inventory. These have been
principally confined to quality control of the individual counter and
quality control of the areas of the stockroom after the counting was

This content downloaded from [Link] on Sun, 26 Jun 2016 16:09:27 UTC
All use subject to [Link]
 I 8o APPLIED STATISTICS

complete. These methods have not proved to be very successful in the

inventories taken by the agency which sponsored this research.
The basic concept of the present project is the continuous quality
control of the storage areas. Those areas which are deemed to be out
of control are completely inventoried. This serves the same purpose
as 100% inspection of a rejected lot of pieces produced on a production
line.
The stockrooms of a supply centre are stratified by physical location.
All lots are of approximately equal size. An attempt is made to store
the more active items together in one or more lots and to store the less
active items in different lots. In addition to the advantage this 'popu-
larity' storage has of placing the active items nearer the shipping
facilities, it also serves to increase the between-lot variance of the
error rates.

Quality Control Procedure

To initiate the quality control once the stratification is complete, a

sample of items in each lot is selected. A physical count of the number
of pieces of each of these items in the stockroom is made and this count
is compared with the recorded balance. If the count disagrees with the
balance by more than the 'leeway', the item is said to be 'discrepant'
or 'in error'. This leeway is established for the discrepancies so that
a minor discrepancy may be ignored. In this study an item is classed
as discrepant if and only if the difference between the count and the
balance is greater than 1 % of the balance, or the monetary value of
the discrepancy is greater than $1 (approximately 7s.).
Following the count, an estimate of the percentage of items in each
lot which are discrepant is made. The simple binomial estimator

lOOdi
Pi-=
ni
is used. Here

Pi = the estimated percent discrepant in the i-th lot; (hereafter

called 'estimated error rate').
di = number of discrepant items in the sample from the i-th
lot.
n= sample size in the i-th lots i.e. the number of items.

The variance of this estimator is the usual one:

var p P (100-Pi)
ni

Next, an acceptable limit for the estimated error rate is established.

In this project the limit is 10%.
Hereafter the words 'sampling' or 'sample' refer to the process of
taking a physical count on only a sample of items in a lot. The word

This content downloaded from [Link] on Sun, 26 Jun 2016 16:09:27 UTC
All use subject to [Link]
 INVENTORY USING SAMPLING METHODS I8I

'inventory' will refer to a complete count of every item in the lot and
the reconciliation of this count with its recorded balance.
The results of the sample are then observed. If more lots have an
estimated error rate of above 10% than can be inventoried by the
counting crew in a three-month period (quarter), only enough of the
highest error-rate lots are rejected to consume one quarter of inventory
time. The remainder of those above 10% are temporarily accepted.
In the first quarter of the year, those lots rejected on the initial
sample are inventoried. Those lots which were accepted, whether or
not their estimated error rates were above 10%, are resampled as
before during this quarter. From this group of resampled lots the
inventory load for the second quarter is chosen. It will be seen that
those lots inventoried in the first quarter cannot be inventoried again
during the second quarter since they were not included in the re-
sampled lots during the first quarter. They will be resampled during
the quarter following their inventory, i.e. during the second quarter
in this case, so cannot be rescheduled for inventory before the third
quarter.
If the lots whose estimated error rates are above 10% do not consti-
tute a full quarter of inventory work, either the acceptable limit can
be lowered or the size of the inventory crew can be reduced.

Operation of Initial Study

In October of 1958 the project was initiated at supply centre number
one (SC-1). For the first nine months the work was confined to
methods of stratification and methods of counting an item without
freezing activity on the item. A team of 15 men was assigned to work
on the project. The centre was stratified into 25 lots and two samples
were taken in each lot, the time between samples being about four
months.
A completely random sample of items in each lot was drawn. First,
for sampling purposes, every piece of a given item was considered as
being stored in the lot with the master location of the item. A cumu-
lative list of all possible storage locations was drawn up. This list is
not the same as that kept in the 'location file' from which issues are
drawn up. For example, a given rack may have as many as three
items stored on it, all of which have the same numbered location in
the location file. This rack was considered as three 'item locations'
even though the records showed it as only one location. It was because
of this that it was not possible to sample from the location file. This
would result in clusters of items in the sample, the cluster size varying
from one to twenty-four items.
The sample size was not rigidly fixed in advance. A sample of item
locations was drawn from the cumulative list, using random number
tables, and each of these locations was visited. If the location was the
master location of some item, this item was included in the sample.
If it was a reserve location or an empty location to which no stock
D

This content downloaded from [Link] on Sun, 26 Jun 2016 16:09:27 UTC
All use subject to [Link]
 782 APPLIED STATISTICS

number was assigned, the location was dropped from the sample. If
the final sample size was too small, a second sample was taken to
augment the first one. A sample size of between 200 and 300 was selected
in all but a few small lots. The average lot size was about 4000.
The count on the sample items was taken without affecting the
normal centre operation. In simplified form the steps followed were:

(1) Count all pieces of an item in the stockroom.

(2) Obtain the recorded balance for the item.
(3) Compare the two figures.
(4) If the figures agree, classify it as a non-discrepant item.
(5) If the figures disagree, they may disagree because
(a) A miscount has occurred.
(b) An activity has occurred in the record section but has not
yet occurred in the stockroom.
(c) The item is discrepant.
This item is set aside for two weeks.
(6) Two weeks after the first count, retrace steps 1-4 if the item is
in disagreement.
(7) If the two figures now agree, it is a non-discrepant item.
(8) If they disagree by the same quantity as they did on the first
count, the item is classed as a discrepant item. Points (a) and (b)
in step 5 are improbable in this case since the same miscount
is unlikely and no activity should be in process for two weeks.
(9) If the item disagrees by an amount different from the first
discrepancy, set it aside for another two weeks.
(10) A third count is taken on all items falling into step 9 and
steps 1-4 and 8 are followed.
(11) For items which are still not reconciled, an investigation is
conducted to determine their true nature.

Several mninor refinements can be and have been made in the above
procedure but basically it has been operated successfully using this
system.
The purpose of the second sampling was to estimate the error rates
which existed in each of the 25 lots. The error rates obtained in this
sample were then compared with the results of the annual shut-down
inventory which was taken immediately following the second sampling
phase.

Results of the Initial Study

Table I gives the lot sizes, sample sizes, and the error rates of both
the sample and the shut-down inventory.
From the table it will be noted that the difference between the
sample results and the findings of the inventory is significant at the
95% level for more than half of the lots. Since it was the purpose of
both the sample and the inventory to estimate the true error rate of

This content downloaded from [Link] on Sun, 26 Jun 2016 16:09:27 UTC
All use subject to [Link]
 INVENTORY USING SAMPLING METHODS I83

TABLE I

Error rates (%)

Sample Estimated Shut-down Difference

Lot No. Lot Size Sample Error Rate Inventory (Inv i
Sie from Sample Error Rate (n.-Sm.
1 7 818 310 61*9 52-3 -9*6t
2 7 191 291 57*7 53*0 -47
3 1 809 217 51*6 54-8 3-2
4 2 344 66 46-9 50*9 4*0
5 6 717 285 43*5 50*5 7.0*

6 3 228 220 41*8 32*9 - 8-9*

7 7 945 300 39*0 41-5 2-5
8 3 493 212 37*3 47 0 9.7*
9 3 676 207 37-2 31*2 -6*0
10 3 136 217 36*9 53.7 16*8t

11 1 908 207 35-3 33-3 -2-0

12 5 053 259 34*8 50.5 15*7t
13 7 271 287 33-1 47.5 14*4t
14 2 930 213 31-9 34*3 2*4
15 6 320 239 31-4 37-5 6-1*

16 2 885 234 29-5 41-0 11.5t

17 614 72 29-2 39*9 10.7*
18 3 084 222 27-5 37-4 9.9t
19 7 074 233 27 0 25-7 - 1-3
20 1 918 226 26-5 31-2 4-7

21 2 266 197 24-4 28-6 4-2

22 3 175 216 22-2 33.4 11*2t
23 6 248 260 21-9 25-4 3-5
24 1 401 192 20*3 25-5 5-2
25 2 872 242 16-5 25-6 9*1t

TOTAL 102 376 5 624

MEAN 34*7 40 7 6 0

* Difference significant at 95% confidence level

t Difference significant at 99% confidence level

the lot it appears that something is wrong. It could be one or both of

two things:

(1) the sample was drawn incorrectly, biasing the results;

(2) the sample does not estimate the percentage found by the in-
ventory but some other quantity; either or both may differ
from the true error rate.

Let us look at the first of these two possibilities. In order to check

the accuracy of the sampling, the inventory findings on the sample items
only were compared with the inventory findings on the entire lot. The
difference between these two figures represents sampling error only.
Table II gives the results of this comparison.
In this comparison only three of the twenty-five lots had differences
which were significant at the 95% level. This leads one to believe that
the sampling was without bias with the possible exception of lots 1
and 10. Since one of these is negative and the other is positive, and

This content downloaded from [Link] on Sun, 26 Jun 2016 16:09:27 UTC
All use subject to [Link]
I84 APPLIED STATISTICS

since there are equal numbers of plus and minus differences, no

consistent bias is evident.
Let us look now at the second possible cause of the large differences
shown in Table I.
Since the differences in error rates given by Inventory (Inventory
denotes this particular inventory) and the estimates given by the
sampling plan (hereafter referred to as SQC) are not due primarily
to the sampling technique of SQC, these differences must be due in
large part to mechanical mistakes by SQC, or mistakes by Inventory,
or change in discrepancy status of items from one count to the other.
The latter group is regarded as not significant. It becomes then a
question of determining the relative order of magnitude of the mistakes
made by SQC and those made by Inventory.
TABLE II
Error rates (%)

Lot No. Inventory Error Rate Inventory Error Rate Difference

on Sample Items on All Items in Lot (Lot - Sample)

1 61-3 52*3 - 9*Ot

2 55 5 53*0 -2*5
3 57*8 54*8 -3*0
4 47-8 50*9 3-1
5 50*7 50*5 -0*2

6 39-3 32*9 -6-4

7 46*2 41*5 -4.7
8 455 470 1-5
9 41-4 312 - 10.2*
10 43.3 53 7 10*4t

11 39-2 33-3 -6.2

12 46 5 50 5 3.9
13 43*8 47.5 3.7
14 32*7 34-3 16
15 355 37.5 20

16 40*6 0*4 41-0

17 40*6 0*4 41*0
18 34*7 37.4 2*7
19 29*3 25*7 -3*6
20 31-3 31-2 -0*1

21 28*6 28*6 0.0

22 29*2 33-4 4-2
23 30-4 254 -5*0
24 26-8 25*5 - 1*3
25 22*4 25 6 3*2

MEAN 39*9 40*7 0*8

* Difference significant at 95% confidence level

t Difference significant at 99% confidence level

Information on the mistakes of SQC and those of Inventory is

provided through a study of the items counted by both SQC and
Inventory. The results of this study follow.
1. Inventory and SQC agree on a total of 81 per cent of the
items, the non-discrepant items in agreement being 58 per cent, and
the discrepant items 23 per cent.

This content downloaded from [Link] on Sun, 26 Jun 2016 16:09:27 UTC
All use subject to [Link]
 INVENTORY USING SAMPLING METHODS I85

2. Inventory found a discrepancy, SQC found no discrepancy, on

7-75 per cent of the items. In a further study it was found that
Inventory mistakes account for 7 per cent of these, and the other
0 75 per cent are actual discrepancies which occurred after the
sampling count was completed.
3. SQC found a discrepancy, Inventory found no discrepancy, on
approximately 1P5 per cent of the items. Essentially all of these
differences result from SQC mistakes.
4. SQC and Inventory both found a discrepancy, but disagreed
about its magnitude, on 9 75 per cent of the items. By special
investigation on these items, it was found that SQC and Inventory
shared about equally in the mistakes, but that about 2% of the
mistakes were due to an actual change in the size of the discrepancy
between the time of the sample count and the time of the inventory
count.

From this analysis we arrive at the figures in Table III. In accord-

ance with this table, for example, the SQC technique results in 22
mistakes for every 1000 good items counted, and in 122 mistakes
for every 1000 discrepant items counted. The rate of mistakes in
general is 5-5 per cent of the time for SQC and 11 per cent of the time
for Inventory. The general conclusion is that overall the SQC tech-
nique makes only half as many mistakes as the Inventory technique.
TABLE III

Frequency of mistakes, Inventory compared with SQC

Non-discrepant Discrepant All

Items Items Items

SQC 2.2% 12.2% 5X5%

Inventory 10X4% 12-2% 11 *0%

This explains the differences shown in Table I. Of the thirteen

significant differences shown there, only two of them are negative and
in these two cases the difference is largely explained by sampling bias
or sampling error, as indicated in Table II.
There are two major reasons for the superior performance of the
SQC team. They are:

(1) The procedure of accepting no count unless it agrees with a

previous count or with the recorded balance virtually eliminates
the possibility of a miscount.
(2) The team is experienced in inventory and interested in the work.

Results of Second Study

After the first study at SC-1, the project was initiated at a second
supply centre (SC-2). It was also expanded to include the physical
inventory of the rejected lots. In these two centres, no shut-down
inventory is being taken.

This content downloaded from [Link] on Sun, 26 Jun 2016 16:09:27 UTC
All use subject to [Link]
 i86 APPLIED STATISTICS

After some study it was decided that a centre should be divided into
sixteen lots of equal size. If more than three lots had an estimated
error rate of more than 10%, the three lots showing the highest per-
centage error in any quarter would be completely inventoried the
following quarter. The principal advantage of this is that the work
load in each quarter is the same so that it is not then necessary to have
a variable-sized inventory team.
TABLE IV
Comparison of rankings of error rates of lots by diJfferent methods

Rank by Rank by Rank by

Single Count Multiple Count Inventory

1 1 4
2 2 3
3 3 1
4 4 5
5 5 6

6 6 18
7 7 10
8 8 9
9 11 17
10 9 19

11 10 2
12 12 7
13 17 12
14 16 11
15 13 8

16 14 15
17 15 13
18 18 14
19 20 20
20 19 12

21 21 21
22 23 25
23 22 16
24 24 24

25 1 25 1 23
In any given quarter thirteen lots are sampled and three lots are
inventoried so that each year an equivalent of three-quarters of the
lots in the entire centre is inventoried completely. Actually some lots
may be inventoried twice in a year so that more than one-quarter of
the centre may miss inventory entirely. Adjustments in the recorded
balances for discrepant items are made.
It might at first sight be thought paradoxical that the recommended
plan should include a proportion of complete inventory work, since
the earlier results indicated that more mistakes were made in a com-
plete inventory than in a sample count. However, it must be stressed
that the earlier comparison was between an experienced SQC team
and relatively untrained personnel carrying out the inventory. A
team specialising in inventory work would be expected to have a low
mistake rate. Furthermore, the periodic complete inventory allows
proper adjustment of erroneous balances to be made.

This content downloaded from [Link] on Sun, 26 Jun 2016 16:09:27 UTC
All use subject to [Link]
 INVENTORY USING SAMPLING METHODS I87

Since the purpose of the sampling is to rank the lots from most
erroneous to least erroneous, the question whether recounts were
necessary in the sampling arose. If only one count is taken on the
sample items and the item is classed as discrepant if its count does not
agree with its balance, ignoring the possibility of miscounts and
documents in flow, what effect would this have on the ranking? This
question was studied using the data on the 5624 items in the sample
at SC-1. Each item was classed as a discrepant or a non-discrepant
item on the basis of the first count only and the lots were ranked. The
ranking thus obtained was compared with the ranking using the
recount procedure. The results of this are presented in Table IV.
Spearman's rank correlation coefficient has the following values:

(1) single count v. multiple count r, = 0984

(2) single count v. inventory r, = 0 754
(3) multiple count v. inventory r, = 0 775
The single count agrees extremely well with the multiple count and
both are in reasonable agreement with the inventory findings.
On the basis of the foregoing results, it was decided that for the
sample items, only a single count would be taken. This saves con-
siderable time in the counting. No adjustments are made on the basis
of the sample findings.
During the first quarter, SC-2 was sampled, without stratification.
On the basis of this sample, the error rate in each section of each
stockroom was estimated. The sections were then grouped together
into lots, sections with similar estimated error rates being put into the
same lots. The three lots with the highest error rates were inventoried,
and the remaining lots were resampled in the following quarter. In
TABLE V
Results of study at SC-2

Lot No. Error Rate on Error Rate on Error Rate on

First Sample (%) Inventory (%) Second Sample (%)

1 40*0 295 5.9

2 39*8 32'6 5*5
3 34.9 30-1 4.9
4 26-7 24-4
5 25*6 30*1
6 24-0 18-8
7 20*6 24.2
8 19*5 30.1
9 19*4 12.0
10 14-3 11*5
1 1 12*7 18*8
12 12-0 13.2
13 11-8 10*0
14 11*7 10-4
15 11*1 14-3
16**

Mean 216 30 7 15-6

Mean of Lots
4-15 17*5 18*2

* No results available

This content downloaded from [Link] on Sun, 26 Jun 2016 16:09:27 UTC
All use subject to [Link]
 x88 APPLIED STATISTICS

the quarter following the inventory, the three inventoried lots were
resampled.
The results of the samples and the inventories at SC-2 are given in
Table V.
It will be noted that the inventory in the first three lots reduced the
error rate radically. It had an overall effect of reducing the centre-
wide error rate by six percentage points. It will be noted also that the
error rate in the lots which were not inventoried did not grow sig-
nificantly during the quarter.
There are no significant differences between the sample estimates in
the first and second quarters except for lot 8. In this instance, much
stock was moved into the lot during the quarter. The stock which was
moved in was believed to have a high error rate by the centre personnel.

Comparison of Inventories

The direct cost per item of the SQC inventory is about 70-80% of
the direct cost of the shut-down inventory. The saving in cost is the
result of more efficient operation of the SQC inventory and the lack
of necessity for training a large group of people for inventory. Other
advantages of the SQC inventory are
(1) No shut-down period is necessary, so that normal operation is
carried on throughout the year.
(2) At any given time the records have a higher degree of accuracy.
(3) A trained team is available for any special inventories which
may be necessary.
(4) Inefficient or error-making procedures are quickly identified
and eliminated.
(5) An incentive for more efficient centre operation is produced.
(6) State of preservation of stocks is constantly reviewed.

Acknowledgement

I wish to express my thanks to Professor R. F. Rinehart, who critically

read the manuscript and made many helpful suggestions.

This content downloaded from [Link] on Sun, 26 Jun 2016 16:09:27 UTC
All use subject to [Link]