By Guy Spriggs

Because of incomplete or partial data, it can be hard to calculate accurate approximations important to scientific work such as medical research. When human subjects quit clinical trials, what do researchers and statisticians do with incomplete results when trying to estimate survival probability?

Statistics professor Mai Zhou’s new book, “Empirical Likelihood Method in Survival Analysis,” aims to answer these questions by applying a new principle to data approximation dealing with duration.

“Empirical likelihood is a relatively new method. I’ve been fascinated by this method since the terminology was invented in 2001,” Zhou explained. Since then, he has been on leave twice to collaborate with partners and been awarded two research grants from the National Science Foundation for work leading toward his book.

As Zhou explains, likelihood is an old method for statistics. However, likelihood requires the assumption of a model, which can affect calculations by forcing them into a preexisting pattern. “There are so many distribution families in statistics, so if you assume a bell-shaped curve or some other family, you may get a completely different answer,” he said.

But the empirical likelihood approach is a non-parametric method, meaning it doesn’t assume any specific model. “Empirical likelihood takes the advantage of the likelihood method but makes it so you don’t have to assume a particular shape,” said Zhou. “This method gets rid of the choice and adapts to whatever shape is underlying the data itself. It’s still providing an approximation, but it’s a better approximation.”

He not only believes that empirical likelihood offers better, more accurate calculations, but also says the method is easier because it relies on new computer power to do most of the work. “You can make variations without having to recalculate everything. It doesn’t have a very transparent formula, but you can let the computer do the dirty work,” he explained. An integral part of Zhou’s book is “emplik,” a software package he helped develop over the last 15 years which is now offered to the public.

Now that his book has been published, Zhou hopes the approach he outlines will be adopted as the default method by the SAS Institute in future versions of their industry-standard analytics software suite.