Business taxpayers sometimes face surprisingly large sales and use tax audit assessments when the tax or taxable amount is projected from seemingly small and infrequent errors in the auditor's sample. Depending on how the sample is designed and weighted, large audit assessments can either reasonably reflect the taxpayer's actual facts or unintentionally be in error due to the sampling methodology itself. Many taxpayers are not equipped to evaluate the appropriateness of a jurisdiction's methodologies and projections and do not know whether or how to challenge them.
Sampling methodologies vary by state. Most states administer sales and use tax audits covering statewide and substate jurisdictions at the state level. While there are similarities, each taxing jurisdiction follows its own independent sampling methodology.
Types of Sampling
In general, state sampling methodologies fall into one of two categories: nonstatistical (judgment and block) and statistical (stratified random) sampling.
Judgment sampling: This involves the selection of categories or dollar-size groupings of transactions to audit. This approach may be considered somewhat arbitrary because the selection is not random and has the potential of achieving a biased (or unrepresentative) result. A biased sample can result in either an over- or underassessment of the final amount of tax due. More importantly, there is no clear way to measure the bias or uncertainty. If, however, a small number of records make up a substantial amount of the total dollars under audit, it may be more efficient to follow this approach.
Block sampling: This takes the selection process one step further and focuses the sample on one or multiple time periods (blocks) of transactions. This can work reasonably well or quite poorly, depending on factors such as seasonality. Tax assessments identified in the block sample are projected back to the rest of the time period under audit.
Statistical, stratified random sampling: This is the most widely used approach. Random numbers are assigned to each record—such as a transaction or an invoice—in the population; the population is stratified by some predetermined category or dollar size; records assigned with the lowest random number within each stratum are selected; and, after review, errors in taxability or tax are statistically projected back to the population.
Even within the stratified random sampling framework, there is some technical variation across states. For illustration, sampling methodologies published by California, New York, and Texas are summarized in the exhibit below. While some states publish guidance on other random-sampling techniques, for simplicity, the following examples focus on stratified random sampling, the most widely used variant.
California State Board of Equalization
The California State Board of Equalization (BOE) Sales and Use Tax Department publishes guidance on audit sampling in its Audit Manual. The manual recommends that the population be stratified by similar types of characteristics or by dollar size for greater efficiency. Depending on the size of the population, the recommended number of dollar-based strata is four to five. Sample sizes for each stratum are determined by a formula using variance within the stratum at an 80% confidence level and a given expected error rate. Both underpayments and overpayments are included in the evaluation of the sample before projection. A stratum must have a minimum of three errors before the results can be projected.
The BOE uses three estimators: mean (mean per unit), difference, and separate ratio (percentage of error). The separate ratio is the most frequently used estimator by the BOE (as well as by many other states). The estimator with the smallest standard error is selected.
New York State Department of Taxation and Finance
The New York State Department of Taxation and Finance publishes guidance for computer-assisted sales tax audits in Publication 132, Computer-Assisted Audits: Guidelines and Procedures for Sales Tax Audits. The department follows a cumulative of the square root of the frequency method, a common sample design approach, for determining the dollar values of strata boundaries. Sample sizes for each stratum are calculated using the Neyman allocation, which allocates sample items based on the variance and record count of each stratum relative to the total population. New York state uses the difference (mean of differences) estimator, which applies a weight—the population size divided by the sample size—to the dollar amount found in error in the sample for each stratum.
Texas Comptroller of Public Accounts, Audit Division
The Audit Division of the Texas Comptroller of Public Accounts refers to its sampling methodology as nonstatistical; however, Texas refers to multiple elements of statistical measures as described in its Sampling Manual. The auditor has leeway to stratify the population using either the absolute value or actual value; zero-amount items are stratified separately. For stratified samples, a minimum of 100 sample items are necessary for each stratum; unstratified samples require a minimum of 250 sample items. The auditor can request incremental increases to the sample size if necessary; however, in practice, the auditors may face technical limitations in their ability to expand the sample after an initial sample has been selected.
Representativeness is tested using a variation percentage—the difference between the sample and population averages divided by the higher average of the two—for each stratum. Each stratum should have a variation percentage within 8%. The manual requires use of the ratio (separate ratio) estimation method for projecting errors.
Sampling frequently is used by state and local tax authorities for sales and use tax audits. While stratified random sampling is the most widely used approach, each state has its own unique guidance for sample design, implementation, and extrapolation. It is beneficial for taxpayers to become familiar with sales and use tax audit sampling techniques used in jurisdictions where they operate. The state's sampling method will affect how the final audit assessment is ultimately determined.
Annette Smith is a partner with PricewaterhouseCoopers LLP, Washington National Tax Services, in Washington.
For additional information about these items, contact Ms. Smith at 202-414-1048 or firstname.lastname@example.org.
Unless otherwise noted, contributors are members of or associated with PricewaterhouseCoopers LLP.