Evaluating competing interventions for reducing school-bus accidents. An insurance firm asked for help in determining which among four hi-tech interventions would best reduce collisions. Crucial to this study was power analysis: determining how many buses would need to be equipped with each device, and for how long. With accidents occurring on average only once every 30,000 miles, even an intuitive sense of the impact of each device would require many thousands of miles for comparison; to find statistically significant results would require even more. The necessary sample size was calculable through creative use of power-analysis software and Monte Carlo simulation. Without this work the firm would have had only a 14% chance of reaching a definitive answer after all its technology investment; the analysis showed ways to increase that probability to 80%, potentially saving $700,000.
Surveying the US market for staff development training. A publishing firm offering staff training and instructional materials wanted to make the most of an online survey to be sent to 20,000 professionals. We helped their marketing director formulate questions so that research could obtain the most useful information, and we set up the survey using an online tool. There were dozens of analytic topics of interest to the firm; we designed an efficient set of multivariate statistical procedures that each allowed investigation into several topics simultaneously. These procedures saved time and money compared to one-by-one investigation, and they had the added advantage of providing control for potentially confounding variables. Thus the analysis succeeded both in being efficient and in isolating the differences that were of greatest interest.
Devising a sampling plan to reduce corporate auditing expenses. An international manufacturer of agricultural equipment had hired a financial services firm for advice on how to save money on a tax audit. Of many thousands of transactions possibly eligible for tax deductions, the firm needed to identify samples that could be audited, since checking every single transaction for possible deductions would have been completely cost-prohibitive. We used IRS guidelines, empirical characteristics of the body of transactions, and inferred characteristics (from a Monte Carlo simulation we designed) to help the firm make informed recommendations to their client on how to create the most advantageous samples.