Audit Sampling Manual
Audit Sampling Manual
Audit Sampling Manual
Sampling is the application of an audit procedure to less than 100% of the items within an account balance or
class of transactions for the purpose of evaluating some characteristic of all the items within the balance or
class of transactions.
Much of the information included in this manual was taken from the Statement on Auditing Standards No. 39
on Audit Sampling which provides guidance on the use of sampling in an audit of financial statements. This
information has been adapted to fit the circumstances most often encountered in tax auditing.
HISTORY
The Department has used sampling in its audit procedures for many years. That sampling, for the most part,
has been block sampling. That is, taking a period of time and testing 100% of the records during that time.
Until 1990, the Department's policy on sampling was to take 100% samples from three test months per year
selected for being the high, low and average months of the year. From 1990 forward the Department has
adopted other systematic or random sampling techniques.
Random sampling techniques are both convenient and accurate when performed properly, for these reasons
sampling is the rule rather than the exception in most audits performed by the Department. The convenience
and accuracy extends to taxpayers as well. Audits based on sampling have been challenged. When challenged
we have allowed the taxpayer to present detailed information to refute the results of the sample.
This occurs when the auditor determines that a type of receipt, deduction, exemption or other
item does not need to be tested.
Note: Even though 100% examination may be done where appropriate. It is not mandatory for any
particular taxpayers or tax programs. Sampling procedures discussed below may be more cost effective.
SAMPLING RISK
Overall tax audit risk is made up of the risk of inaccurate records and the risk of misapplication of the tax law.
Both of these risks are made up of two components as well.
1.
2.
Audit risk, in turn is made up of two components, the risk that a procedure is not effective and sampling risk.
Sampling risk is the probability that the sample results are not representative of the entire population. In
general, factors that may lessen sampling risk include:
1.
2.
3.
4.
5.
6.
7.
homogeneous populations respectively. Stratification of a population can reduce sample size in most cases.
Provisions for sampling units based on time periods is also provided.
In any sampling approach, the auditor must evaluate the population that is being tested, must determine if any
stratification should be done, must evaluate the cause of any exceptions and must apply the results from the
sample to the remaining portion of the population.
5.
6.
7.
8.
9.
Does the Sales Summary contain all invoices and is information recorded accurately?
How often are invoices voided without explanation?
Does the taxpayer record all supplies pulled from inventory in the inventory log?
Does the sales supervisor correctly batch sales by destination so that they can be recorded by
state in the summary journal?
What is the amount of deductions not supported by NTTC's?
What is the percentage of sales of services to the government relative to total sales?
What percentage of supplies are pulled from inventory held for sale each month?
What percentage of supplies purchased outside NM have not had tax paid or accrued?
Auditors should record the objective of the test within the audit narrative and/ or other workpapers. Not only
does that help the reviewer, but it also clarifies the objective for the auditor so that the proper audit procedure
and sampling application can be defined. If the auditor can create a one sentence question, like the ones above,
Page 3 of 16
to be answered, he or she is less likely to perform an inappropriate procedure or use the wrong sampling
application.
TYPE
APPLICATION
Attribute
Variables
Attribute
Attribute
Variables
OF
EXPLANATION
The auditor doesnt want to know
what percentage or how much, only
yes or no.
Here, the auditor wants to know a
specific amount.
Presumably, if there are supplies not
recorded, the auditor will not
accept the inventory record.
This is a yes or no question, which
will result in accepting or rejecting
the data in the summary journal.
Variables
Variables
Same as above
Variables
Sampling applications can also be classified by the type of audit procedure in which they are used.
Compliance tests are tests which determine whether controls are being complied with. The answer to a
compliance test is yes or no. Substantive tests are tests which determine the amount of some class of items.
Attribute sampling is most often used in compliance tests and variables sampling is most often used in
substantive tests.
Page 4 of 16
COMPLIANCE TESTS
Compliance tests are most often used by tax auditors to determine if controls which ensure the
accuracy of records are in place and working correctly. These tests can be performed directly on the
control feature itself or indirectly on the outcome of the control.
An example of a direct test would be a test to determine that invoices are pre-numbered, used in
sequence and accounted for by those issuing the invoices. Such a test would be helpful in assuring
the auditor that all invoices issued in a period are used or voided.
An example of an indirect test would be a tracing of a sample of invoices to a summary journal to
determine that the controls over recording invoices in the summary journal are working. In this case,
the controls themselves are not actually tested, but the results of those controls are examined, and the
question of whether the summary record is reliable will be answered yes or no.
Note that tax auditors do not use formal compliance testing as frequently as other types of auditors.
However, tax auditors do make judgments about the level of risk of incorrect records and the risk of
misapplication of the tax law. These are the types of judgments that can be backed up by compliance
tests.
The decision to test controls or the accuracy of records is based on auditor judgement and the
circumstances of the audit. The decision should be documented. Compliance testing may help to limit
the scope of the audit to areas of higher risk or point out problems with records that may have
otherwise appeared reliable.
The main reason for performing compliance tests is to reduce the amount of substantive tests that
need to be performed. Therefore, the decision of whether to perform compliance tests should weigh
the possible compliance tests against the possible substantive tests that could be performed to
determine which test will be most efficient and effective.
For instance, if an auditor decides that he can either test the taxpayer's summary records and use them
to perform the audit, or, rely on comparing reports to bank statements, then he or she should
determine which method will be more efficient. If a compliance test of the summary records is
performed and the records prove to be unreliable, then the auditor may still have to rely on bank
statements. However, it may be that using the summary records will be much more efficient than
using bank statements. Therefore, testing those records is worth the time needed and the risk that the
test results will be negative. Before relying on the summary records the auditor should perform a test
of transactions to determine the records are reliable.
SUBSTANTIVE TESTS
Substantive tests are used to determine the amount, usually the dollar amount, of a specific group of
items. If the auditor seeks to determine the amount of disallowed deductions, for instance, the result
of the sample will be a dollar figure of disallowed deductions found in the sample. The assumption is
that the same proportion of disallowed deductions will exist in the population. Therefore, the final
result of the test will be a dollar amount of disallowed deductions for the population which will be
used as a basis for assessment.
Often, samples can be designed to serve both compliance and substantive tests. When it is likely that records
will be needed for both types of applications, the auditor should strive to pull one sample. This is called dualpurpose testing.
the control is not documented, you will have to rely on direct observation of the control being performed, or
on indirect evidence.
If you are testing controls indirectly, you would look at the error which the control is intended to prevent and
would base your deviation on what defines that error. In the case above, the control in place is intended to
prevent salesmen from not charging tax on sales that should be taxed. A deviation would therefore be
defined as an invoice that did not have tax and should have.
If you are performing a substantive test, the item(s) you are picking up might not necessarily be thought of as
deviations. For instance, you may be trying to determine the average New Mexico inventory value over a
period for testing the CIT property factor. However, the same principle applies. You need to define which
items meet the criteria necessary to reach the objective of the test. In this example, that might be inventory
control log entries backed up by shipping and receiving reports.
Auditors should be careful not to include factors in the deviation, which do not affect the objective of the
test. For instance, in the first test described above, invoices where the customer name was misspelled would
not affect the objective of the test and should not be treated as deviations. On the other hand, auditors should
also be careful to include all factors, which may affect the objective of the test. For instance, if an invoice
contains the initials of a supervisor from another Department who is not familiar with the customers who
have Non Taxable Transaction Certificates (NTTCs) on file, the invoice should be picked up as a deviation,
even though it contains a supervisor's initials.
Some of the most common problems faced by auditors performing any kind of test come from not properly
defining the deviation, and finding out after the test has been performed that there were other conditions that
should have been considered.
6. What other matters are relevant to the sample results? Have conditions which might affect the
results changed in the remaining period?
DEFINE THE SAMPLING UNIT
A sampling unit is any of the individual elements constituting the population. The auditor should define the
sampling unit in light of what is being tested and the type of records kept by the taxpayer. A sampling unit may
be, for example, a document, an entry in a journal, a line item, or a single transaction.
It is possible to sample based on time period representations such as days, weeks, or months. The department
has set minimums of 30, 25, and 9 respectively. The days, weeks, or months should be randomly selected from
the entire audit period. Random sampling within the selected days, weeks, or months is encouraged.
Regardless of the type of time period selected the number of items of interest (invoices, line items,
transactions, etc) must be achieved for the population type. In most time period samples the items of interest
will exceed the minimum required.
CONSIDER THE COMPLETENESS OF THE POPULATION
The population is physically represented by some form of record. For instance, sales invoices, entries in a
sales journal or summary entries in a ledger may represent total sales. The auditor actually selects sampling
units from this physical representation and so must confirm that all sample units from this record are included
in the entire population. If the physical representation differs from the actual population, the auditor might
make erroneous conclusions about the population. A simple example of this is testing a depreciation schedule
where a page of the schedule is missing.
Therefore, the auditor should be careful to determine that the records used to draw the sample are complete and
reflect the actual population being tested. One means of doing this is comparing different records and
reconciling differences found.
Page 7 of 16
SYSTEMATIC SELECTION
For this method, the auditor determines a uniform interval by dividing the number of physical units in the
population by the sample size then rounding up. A starting point is randomly selected and each item after that
is selected at the uniform interval. If the population is arranged randomly, systematic selection is essentially
the same as random number selection. However, if the population is not randomly arranged, for instance, if
sales are listed by item, rather than in the order made, there may be problems with this method.
One way to ensure more randomness in a systematic sample is to re-compute the interval each time by use of a
random-number table. In this approach the auditor would select a list of random numbers. The first number
would be the starting point. The second number would tell the auditor the interval to count to the next item to
be selected. For instance, if the first two random numbers are 503 and 219, the auditor would select item 503
to start, then item 722 (503 + 219). In this approach, the auditor might have to go through the population more
than once to finish drawing all the items. The number of digits to be used for the random numbers should
make intervals that are large enough to go through the entire population a least once.
If the sample is drawn from the entire audit period, use an initial sample of 50 items. If the sample is
drawn from a block, or if the population is limited in some other way, the initial sample should be 100
items.
If one or more deviations are found in the sample, the auditor must either reject the item being tested, or
may expand the sample. If the auditor chooses to expand the sample, an additional number of items equal
to the initial sample should be tested.
If one or more additional deviations are found, the auditor must either reject the item being tested, or may
expand the sample to the appropriate variables sample size and use the results to estimate the amount of
error in the item being tested.
See the section on evaluating the sampling results to determine whether qualitative aspects of a deviation may
determine whether a deviation can be overlooked in an attribute sample.
VARIABLES SAMPLING (NUMBERS)
A critical question must be answered before the sample size for a variables sample can be computed. Is the
population relatively homogeneous?
Homogeneous populations can be tested using smaller size samples since there are fewer exceptional items to
skew the results. Non-homogeneous populations require larger size samples.
Homogeneity is the tendency of items in a population to be similar, or closer to the same dollar value. For
instance, a population containing sales of three kinds of mid-priced property will be far more homogeneous
than a population containing all sales of a Department store.
If a population is non-homogeneous, the auditor can reduce the sample size through stratification and
identifying individually significant items. See Step 7 on performing the sample for an explanation of how
stratification should be done.
The Audit Sampling Workpaper, AUD-21, can be used to document whether a population is homogeneous or
not. Whether or not this workpaper used, the auditor should document why the population was either
determined to be homogeneous or non-homogeneous.
Page 8 of 16
UNLIMITED
1 INDIVIDUAL
ITEM
1 INDIVIDUAL
ITEM
STRATIFIED
LIMITED
SELECTION
METHOD
2 DAY *
3 WEEK *
4 MONTH *
(Note: the number to the left of the selection method is the order of preference for selection methods. A 1
indicates the most preferred methods and a 4 indicates the least preferred)
*LIMITED Month, week, and day samples should examine two times the minimum required items of interest
for the population type being tested, whenever possible. Random sampling of items of interest within the time
period selected is encouraged. When using the month selection method the auditor must evaluate all months
that have unusual balances and determine the circumstances for variance before including the months in the
sample or in the extrapolation procedure. LIMITED SAMPLING IS ALWAYS OUR LAST CHOICE AND
THE REASON FOR ITS USE SHOULD BE ADEQUETLY DOCUMENTED.
If the auditor determines that the population to be sampled is diluted with transactions that are not of interest
then the sample size should be increased. In the case of a test of deductions where taxable and non-taxable
sales are commingled the auditor would first need to determine the percentage of non-taxable sales in the
population and then use this percentage to compute the increased sample size.
(non taxable sales total sales) x 100 = % of non-taxable sales
250 ( % of non-taxable sales) = sample size
offset debits it is important to remove both. When stratifying a population it is necessary to use absolute values
in order to keep credits and matching debits within the same strata. A credit will generally not produce a
reduction in the exception total (the numerator in the error calculation). One example of a credit that might
generate a reduction in a compensating tax exception total is the return of an item which had compensating tax
accrued and paid at the time of purchase. A reduction would be acceptable if a reversal of the accrual did not
take place and if the auditor has reasonable assurance that the return is not an unusual event.
UNUSED OR INAPPLICABLE DOCUMENTS: The auditor's consideration of unused or inapplicable
documents is the same. For example, a sequence of vouchers might include unused vouchers or an intentional
omission of certain numbers. If the auditor selects such a document, he should obtain reasonable assurance
that the voucher number actually represents an unused voucher and does not represent a deviation. The unused
voucher may then be replaced with an additional voucher.
ERRORS IN RANDOM-NUMBER SELECTION: In a situation where the auditor generates a random
number that is not part of the population, that number should be replaced with another random number which
is part of the population.
INABILITY TO EXAMINE SELECTED ITEMS: If an item selected is missing and it cannot be
determined what happened to the item, it should normally be considered a deviation.
STRATIFICATION
In order to make a population more homogeneous for variables sampling, the auditor can use stratification.
Described below are some ways to stratify a population. This list is not all-inclusive and auditors may find
other appropriate ways to stratify a population.
It is important to note that generally, stratification requires that the sample results be extrapolated to each strata
separately. In other words, if an auditor divides the population into two groups, sales of tangibles and sales of
services, to make both groups more homogeneous, then the results of the sample from the tangible group
should only be extrapolated to the total tangible sales and the results of the sample from the services group
should only be extrapolated to the service sales.
Where the sample is limited to specific time periods, the auditor will need to confirm that the data from the
entire population can be stratified before the resulting error rates can be applied.
Stratification is most often performed during computer assisted audits. Computer software applications can
easily segregate a population and provide subtotals. It is required that auditors contact the Computer Assisted
Audit Team when working with a high volume of transactions or with large data files.
A minimum of three errors per strata is necessary for extrapolation to strata population.
The following are some methods of stratification.
BY DOLLAR AMOUNT - This is the most common type of stratification used by auditors. The auditor needs
to identify the number of different ranges, and their dollar values, into which the population most usually falls.
For instance, if the auditor is testing sales of equipment and the taxpayer sells several low-priced, several
medium-priced and two high-priced models, the auditor may decide to make strata from $7,000 to $12,000,
$12,001 to $18,000 and $18,001 to $25,000. Remember, the results of the sample pulled from each strata
would be extrapolated only to the total of that strata.
BY NATURE OF THE ITEMS: An easy way to stratify a total population is to stratify based on the nature of
the transactions, such as sales of tangibles and sales of services.
Any relevant attribute can be used so long as stratifying by that attribute tends to make the population more
homogeneous in dollar amounts.
BY NATURE OF THE TEST - Another way to stratify a population is to divide the items into groups
according to the nature of the test to be performed. For instance, if the auditor is testing deductions for
Page 10 of 16
supporting documentation, the population could be divided into groups where an NTTC is required and where
one is not required.
The methods for stratifying described above may be used separately or in combination. For instance, the
auditor may stratify first by the nature of the items, then by dollar amount.
STRATIFICATION EXAMPLE:
Taxpayer A sells three types of computer equipment: laptops, desktops and network servers. The auditor
decides to stratify the total population of sales by type of computer equipment, since that tends to create more
homogeneous sub-populations. Since this is still a non-homogeneous, stratified population the sample size
computed from the Sampling Size Table is 100 per strata.
There were $2,000,000 total laptop sales during the period, $3,000,000 total desktop sales and $5,000,000
server sales. The auditor should allocate the total sample size as follows:
100 from laptop sales
100 from desktop sales
100 from network sales
The results were as follows:
5% under-reported laptop sales X $2,000,000 = $100,000
5% over-reported desktop sales X $3,000,000 = ($150,000)
10% under-reported network sales X $5,000,000 = $500,000
The total exception in this case would be the sum of the separately extrapolated sample results, the net amount
of $450,000 under reported.
(Note: this is another example of a credit reducing an exception total)
If the test is a variables sample and five or more errors are found, the auditor must extrapolate the results to the
remainder of the population. This procedure calculates the percentage of error (POE) found in the sample and
applies that result to the population tested. To calculate the POE, take the dollar value of the deviations (or
other sample result), divide by the dollar value of the total sample. Then multiply that POE times the dollar
value of the population.
$Deviations (or sample results) X $Population
$Total Sample
NOTE: If fewer than five deviations are found in a non-stratified sample the auditor should detail the
exceptions instead of extrapolating. If working with a stratified sample the auditor should detail the
errors in each strata sample that contains fewer than three errors.
There are three rules for extrapolating:
The numerator should be the sample representation of what the auditor is trying to determine about the
population (or strata).
The denominator should be the sample representation of the population (or strata).
The population should be complete and should not include items that do not represent the population as defined
by the test.
For instance, it would not be appropriate to draw a sample of deductions taken and test them for validity, then
divide the exceptions found by total deductions tested and apply that percentage to gross receipts reported. In
this example, gross receipts is not the population; deductions reported is the population.
CONSIDERING THE QUALITATIVE ASPECTS OF DEVIATIONS
In addition to evaluating the frequency of deviations, the auditor should consider the qualitative aspects of the
deviations. Qualitative characteristics of the nature and cause of the deviations are:
Whether the deviations are errors or irregularities
Whether the deviations are due to misunderstanding instructions or carelessness, and
The relationship of the deviations to other phases of the audit.
NON-RECURRING ERRORS are defined as errors that are caused by factors that do not affect the rest of
the population. An example of a non-recurring error is a sale that was never recorded due to the fact that the
salesman died right after closing the sale.
Errors that are determined to be non-recurring should be extracted from the results of the sample and from the
total sample. They should not be used to extrapolate sample results to the total population.
For example, the auditor knows that some sales allocated to Texas actually had New Mexico delivery
addresses. A sample is taken of Texas sales to determine what percentage of sales allocated to Texas should be
reallocated to New Mexico. The sample of 100 items, totaling $2,000 showed three items that should have
been included as New Mexico sales. One item was non-recurring and amounted to $100. The other two items
amounted to a $40 total. The rate of error in the population would be computed as follows:
$40/1,900 = 2.105%
The $100 error would not be used to extrapolate sample results, but would be picked up as a separate
exception.
NON-SYSTEMATIC ERRORS are errors that are caused by factors that may affect the rest of the population
but whose effects are not predictable. An example of a non-systematic error is an error caused by the
taxpayer's occasional use of salesmen to do accounting work.
Errors that are determined to be non-systematic should also be extracted from the sample results. However,
since such errors may be recurring in the population, the auditor should understand the cause of the errors and
try determine the total population error.
Page 12 of 16
For example, a sample test of $1,000 worth of sales reveals a systematic, recurring error of $100 and a nonsystematic error totaling $50 in the sample. Investigation of the non-systematic errors shows that these errors
generally occur at the end of the third and fourth quarters. The taxpayer explains that they are caused by
differences in estimates used in recording end of quarter sales for those quarters in order to prepare the
financial reports. Sometimes these estimates are reversed and the actual amounts are entered and sometimes
they are not. The $50 found is a net under-reported amount due to these estimates. In this case it may be
appropriate to examine all such entries to see which ones were reversed properly and which ones were not.
Alternatively, the auditor may conduct a separate sample of those estimates to determine the percentage
difference resulting from improper reversal. If either of these is done, then the results of the entire test would
be computed as shown in the following table.
Total dollar value of
the population being
tested
Plus
Total net difference
caused by estimates
not reversed (picked
up by detail test).
Minus
of
the
OR
Extrapolated
net
difference
based on sample of the estimates
and the difference found.
Times
Page 13 of 16
Page 14 of 16
OR
Population is Homogeneous
2. Calculate the
average value of
an item
3. Is the standard
deviation > Average$
Value?
Page 15 of 16
Population is Homogeneous
Population is NonHomogeneous
Population is Homogeneous
Page 16 of 16