For some time now, Board of Equalization Auditors have utilized statistical sampling methods in their audit procedure. Statistical sampling is used when testing resales, interstate commerce sales or any other sale claimed as exempt. This method replaces block sampling which was the common sampling method used by the Board of Equalization up until about 10 years ago. In the typical block sample the auditor would test 3 months (one in each year of the audit period) and project the result to the entire population. In a random selection, the sample is drawn throughout the audit period usually using random invoice numbers. If the records are not conducive to this type of selection, (for example, when invoices are not sequentially filed) a variety of other creative methods have been used. These methods include testing:

* Random days
* Every Nth sale with a random starting point (e.g. every 25th sale)
* Random sales journal pages
* Each sales invoice within one inch of each other with a random starting point.
* Random clusters (e.g. a group of 5 consecutive invoices becomes one sample unit).

Both block samples and statistical samples are still subject to sample error. The biggest difference is that the sample error can now be measured by employing statistical computations. The end result of this higher math is the auditor can establish a confidence level and a confidence interval. The problem is that no matter how bad the statistical results of the test are, the test is often treated as sacred.

For example, I was recently involved in a case that involved a statistical test of resales. This test resulted in a substantial tax liability. The problem was that about half the errors in the sample were attributable to one customer. It would seem that a more equitable result would be obtained if all sales to this customer were isolated on an actual basis. Excluding this recurring identifiable error from the sample seemed like a more direct and accurate approach. The audit staff, however, held to their position that the sample was a valid statistical sample and thus could not be tampered with. They indicated that if this were a block sample, they would be willing to take the approach I suggested.

In this case, the State has elevated statistical sampling to a level higher than its results. The result of this sample was 80% confidence at an interval of 34%. In other words, the state is 80% sure that they are no more than 34% wrong.

Perhaps we are seeing a new standard of excellence. Try telling your client the next time you do their taxes that your 80% confident their return is within 34% of being right.

If they point out obvious errors, don’t forget to tell them that you can not change the return since you scientifically measured your inefficiency.