-
What is silent evidence?
-
How can I look for it?
-
What benefit does it provide?
As technology advances, companies are continuing to get better and better at capturing, manipulating, and examining data. One of the many results of this improvement is that the ability to differentiate your approach to risk management from others is gradually getting smaller. To truly stand out from the “crowd”, one must start to think outside the box when considering the many facets of data analysis. In this blog, we’ll be discussing one such non-standard approach in the form of “silent evidence”.
So, what exactly is silent evidence? Per Nassim Nicholas Taleb’s book, The Black Swan, silent evidence refers to the human tendency to view history with a lens that filters out evidence differing from our preconceptions. More specifically, silent evidence is evidence that we don’t have access to either due to the type of data being collected or our own inability to recognize that it exists. For example, consider a situation in which you are measuring the skill level for baseball at a certain high school. Your study takes all volunteers for a class period and puts them through a number of tests to determine their skill level. In this case, silent evidence refers to the fact that you are only able to measure those students who, for whatever reason, felt the desire to volunteer for your test. Outside of your sample, there exists a not-insignificant amount of silent evidence in the form of students who simply stayed in their classes.
More specifically, consider a situation in which you are studying a database of workers compensation claims with the goal of obtaining predictors for the most severe types of claim. While your study does a good job of prediction based upon claims that already meet the criteria of being filed and being considered severe, it ignores the similarly not-insignificant amount of claims that have occurred but, for whatever reason, either have yet to be reported or are yet to reach the threshold of severity. Naturally, someone armed with the knowledge of silent evidence would not blindly accept this study’s results as fully indicative predictors. A similar situation occurs for companies whose loss runs are limited in terms of the fields being captured. In some cases, these data sets are streamlined to the point of excluding potentially important evidence.
Now that we’re aware of silent evidence, it’s clear that something must be done to alleviate the problem it poses – that of incomplete prediction. But what? To start with, simple awareness of the issue helps a great deal in terms of identifying silent evidence in all its potential forms. When completing these studies, or any form of analysis that utilizes data, consider the ways in which the data being considered is captured. Are there any flaws in the methodology? What are some possible scenarios that would result in a data point not being included in the set? A great deal of insight can be gathered by simply asking these questions and considering their potential impact on the analysis. Unfortunately, the classic “80/20” adage applies here. Asking these questions and listing out their answers covers 80% of the goal and requires only about 20% of the necessary work. Finding possible solutions to these issues and applying them fills out the remaining percentages.
Of course, it’s also possible that the issues presented don’t have solutions that can be feasibly applied. More often than not, this presents less of a problem than one would think. There exist many “unknowable” possibilities in the world, especially in industries centered around risk. Their probabilities and properties cannot be easily ascertained, but as mentioned above, simple awareness of their existence and impact on a study can go a long way in improving its soundness in dealing with “real world” issues.
This concept can also be applied to the realm of enterprise risk management. Many times, when companies undertake enterprise risk initiatives, risk are identified that have little associated historical data and have not explicitly been monitored or analyzed. Starting from scratch in qualification and quantification of these risks may include an investigation of the available data and silent evidence.
Silent evidence is only one of several ways we are blinded to the true form of probability, especially in the insurance realm. In future blog posts, we’ll look to cover these in more depth and hopefully solidify them into actionable plans. If you’re interested in discussing the concept of silent evidence, or would like to have SIGMA examine a particular project and collaborate with you in the search of silent evidence, feel free to contact us at either tlc@sigmaactuary.com or enoch@sigmaactuary.com.