Tools › Finance and Business › Auditing › Audit Sampling › Example
Sixty invoices billed about 8 % too high: ratio, difference and mean-per-unit estimates of the same sample
| Tool | Audit Sampling |
| Source | Original illustrative case (seeded synthetic data). Method: Cochran, W. G. (1977). Sampling Techniques, 3rd ed. John Wiley and Sons, New York (ratio estimator; cited, the formulas were checked against an independent numpy computation). |
| Level | Graduate |
Problem
A supplier's 1,500 invoices (book total 1,231,938.38) were all billed at a rate about 8 % too high. A simple random sample of 60 is audited. Estimate the total overbilling with the ratio, difference and mean-per-unit estimators and compare the two-sided 90 % intervals.
Inputs to Enter
- Task: project a sample (classical variables); estimator ratio; confidence level 90 %
- Population summary: 1,500 items, book total 1,231,938.38
- Sample: 60 rows of id, book amount, audited amount
Answers
- Ratio: R = sum of errors / sum of book amounts = 0.080233; estimate 0.080233 · 1,231,938.38 = 98,841.59; standard error 2,196.41; 90 % limits 95,171.18 to 102,512.00 (± 3.71 %)
- Difference: 93,042.75; standard error 9,304.50; limits 77,494.07 to 108,591.43 (± 16.71 %)
- Mean per unit: 165,318.13; standard error 101,434.17; limits −4,187.81 to 334,824.07
- The ratio estimator's standard error is 46 times smaller than the mean-per-unit one because each error is nearly proportional to its book amount
Notes
Open this case in the tool: /analysis-tools/business/audit/audit-sampling/?try=ratio-estimate. The estimator belongs in the sampling plan, chosen before the results are seen. The ratio estimator is the standard survey-sampling form and is not one of the RAT-STATS variable appraisals. Every number is synthetic.
More Examples for This Tool
- Overpayment audit of 100 claims in three strata: projected error and its 90 % lower limit
- RAT-STATS Example 1: unrestricted variable appraisal of 50 items from a universe of 10,000
- RAT-STATS Example 2: stratified variable appraisal, two strata of 5,200 and 3,500 with 25 sampled in each
- Monetary unit sample with two partial errors: basic precision, projected misstatement and the Stringer upper limit