Workers Comp Algorithms – How Far Have They Progressed?
The panel discussion yesterday with Frank Pennachio covered one hot topic – Workers Comp Algorithms.
I do agree that almost everything in existence today has an algorithm attached to it. All one has to do is pick up their smartphone. You are generating hundreds of algorithmic functions by using your phone.
Workers Comp algorithms are no different. Every time an injury is reported, a policy is placed, or even when a premium audit is performed an algorithm takes over the calculations. Let us step back to first define algorithms – you may be surprised with this one.
A finite set of unambiguous instructions that, given some set of initial conditions, can be performed in a prescribed sequence to achieve a certain goal and that has a recognizable set of end conditions. (c) American Heritage Dictionary
The Wikipedia and Merriam Webster Dictionary (c) definition – In mathematics and computer science, an algorithm is a finite sequence of well-defined, computer-implementable instructions, typically to solve a class of problems or to perform a computation. Algorithms are always unambiguous and are used as specifications for performing calculations, data processing, automated reasoning, and other tasks.
The oldest algorithm is from 4,500 years ago by the Ancient Babylonians. The algorithm was called the Euclidian algorithm. I will not bore you with the details.
Check out the thumbnail of the Euclidian algorithm on the right side. If you click on it, it will resolve to a much larger graphic. The main things to remember are the words Entry and Input A, B.
Workers Comp Algorithms vs. Predictive Modeling
Are predictive modeling (analytics) and algorithms the same for Workers Compensation? Yes, they could be viewed as the same thing.
Frank brought up a great point that I think needs to be mentioned – there are 42 million claims that are digitized to some degree.
We could build workers comp predictive models with all that data. The largest single owner of workers comp data, from talking with data analysts, is Liberty Mutual. I have spoken with more than one of their analysts on the size of the bundled data sitting on servers somewhere – one word – humongous. By the way, their data analysts were impressive young men and women.
Simplest Algorithm Shows Major Modeling Unknown
Going back to the oldest algorithm quickly shows a major question mark. What data will be fed into any predictive model or algorithm. Remember, Input A, B from above? Who inputs those numbers? What do they decide the inputs should be for the world’s oldest model?
Workers Comp Algorithms = Same Model Problems
The inputs to the model would be almost the most crucial part. Check out the next section for the most crucial. The difficulty with workers’ compensation is that each state has a different set of rules. Would you need 50+ sets of algorithms/models with one for each state? (Probably so).
Take a look at the NCCI rating manuals, there are so many individual states that have modified each rule that the inputs to each model would have to almost be state-specific.
(Additional Rules: AZ, CO, CT, CT(A/R), DC, FL, GA, IL, KS, KY, LA, MT, NE, NV, NV(A/R), OR, OR(A/R), TN(A/R), TX, UT, VA, WV)
Now take the independent bureaus into account (15 of them) and you have a model that has many changes to the main Experience Mod calculation algorithm.
If one looks at the rating formula (an algorithm) in any of the rating bureaus, the number of state exceptions can be somewhat shocking.
Main Hidden Problem With Predictive Analytics and Modeling
The one area that I have not covered in any article is a very simple one that is passed over when discussing taking numbers, modeling them, then making predictions with them. Who decides what goes into the model and even more important – who writes the algorithms?
Until the day that software/computers can generate their own algorithms and models, bias (as Frank P. astutely pointed out) will always enter into the numbers.
I will not rehash the articles I have written previously on predictive analytics or workers comp algorithms. Feel free to follow the above links for the predictive modeling articles.