How to Talk Statistics to Military Officers
The opportunity for statisticians to work with members of the United States military can be exciting. To bridge the cultural differences between the military and academics successfully, statisticians should be prepared to overcome objections and misconceptions as they arise.
Resistance to Statistical Methods
Members of the United States military are often hesitant when faced with math-intensive situations. This is because the central goal of military professional development is to make officers well rounded. Since mastery of a technical field is outside their personal experience, military officers can be uncomfortable with specialists. This can lead to two forms of initial resistance: “I don’t understand statistics” and “There are lies, damn lies, and statistics.”
The way to deal with officers who are not mathematically inclined is to get ahead of the issue. Just as no one would ever say, “I can’t catch a football because I don’t understand physics,” a strong math background isn’t needed to benefit from statistics.
The reality is that human beings are naturally statisticians. All of us draw conclusions from data on a regular basis. One story I like to tell goes like this: Imagine going back in time 8,000 years to the banks of the Nile River. Find an old grandmother and ask her which is more dangerous—working by the river where the crocodiles might attack you or working in the fields where the snakes might bite you?
She can answer that question simply by tapping into the natural pattern-seeking capabilities in the human brain. Statistics merely formalizes this innate human ability.
More difficult to deal with is the resistance that comes from people who distrust statistics. These people claim, “You can make statistics say anything you want.” That sentiment is well justified because liars do use statistics. Liars take advantage of anything people trust. But one of the reasons statistics is so central to formal research is it involves the objective use of mathematical techniques. That is, statistics is its own lie detector.
If someone tells a military officer something that doesn’t seem true, encourage the officer to ask for the data and have his own staff repeat the analysis. Liars never want to share their data. Similarly, to get people to believe results, be open in sharing the data on which the analysis is based.
Preference of Point Estimates
Some say the hallmark of a statistician is that she doesn’t just provide a point estimate, she provides a confidence interval. When working with military officers, a statistician might be surprised to learn that they often only want a point estimate and don’t understand the risks of not knowing the interval estimate.
This desire for a single, clear number comes from a professional discomfort with uncertainty. For a military officer, uncertainty is closely associated with mission failure. Any open discussion of the uncontrolled variation implicit in a confidence interval suggests one of two scenarios that worry military leaders: either there is something beyond his control that he doesn’t even know exists or, more troublesome, there is something within his control that he does not effectively manage.
The challenge is not to convert the military officer to the statistical view of matters. Instead, the statistician should talk about the confidence interval qualitatively (with the confidence interval prepared in case it is needed). One approach to presenting results is to say, “Rather than give the confidence interval, what I have provided is the most likely outcome in the range of possible outcomes.”
Statisticians might assume the phrase “most likely outcome” refers to the maximum likelihood estimate, but this phrase is meant to tap into a concept of military battlefield intelligence. When being briefed on possible enemy actions, military leaders are given two specific scenarios: the enemy’s most likely course of action and the enemy’s most dangerous course of action. This pairing of ideas can be extended to characterize the bounds of the confidence interval as the most dangerous courses of action.
It is also good to have a discussion about how sensitive the situation is to the accuracy of the point estimate. If the confidence interval is narrow in the context of operational concerns, it is okay to gloss over it, but if the confidence interval is wide in the context of operational concerns, the risk associated with the estimate has to be addressed.
As a final note, if the estimate is going to be presented in written form, the statistician should lobby strongly to have the confidence interval included. PowerPoint slides, in particular, have a tendency to be re-used out of context. The risk is that a confidence interval that was not operationally relevant for its original purpose may be very relevant in a different context.
Extrapolating Beyond the Range of the Data
Some military officers will overestimate the ability of statistical methods to address their problems. This is particularly true when a large volume of data exists, but the data are not adequate to address the problem.
There are two specific military situations in which one must be mindful of this issue. The first is when asked to predict the result of a policy or action for which there are no existing data. Nonstatisticians will see a large volume of historical data and think it all contains information about behavior under the new policy.
A second challenge arises when using statistical methods to forecast the behavior of the enemy. Since surprise is an ancient principle of war, behaving in a manner that can be measured and predicted is unwise for anyone on the battlefield. An intelligent enemy has a significant incentive to be difficult to model.
Beyond this obvious challenge, if a statistician does find some predictable behavior by the enemy, military forces will take action against that element of the enemy force. This will hamper the model’s ability to predict behavior.
In instances where this issue comes up, I like to refer to Nassim Taleb’s story of the Thanksgiving turkey. If you ask a turkey what he thinks he will be doing on December 2, he will say, “Based on my analysis of the historical data, I will be sitting in my coop and eating corn.”
In situations such as these, the way we use statistics has to be adapted. When the underlying conditions influencing behavior have changed, statistical methods can detect and quantify changes resulting from the new conditions. That is, while statistics may not be able to predict the future, statistics can formally describe the future as it unfolds.
Since I can’t list every possible objection, I will close with a technique to unearth potential communication barriers. As a rule, the more senior an officer becomes, the less time he has and the more important concise communication becomes. Important work will be delegated to an “action officer,” who will likely be the statistician’s primary point of contact with the military.
Before briefing the senior military officer, arrange with the action officer to conduct a rehearsal with a junior military officer. The best candidate would be someone who is aware of your work, but not directly involved.
This will allow the statistician the opportunity to explore alternative explanations in a less time-constrained setting. Also, it will give the military people involved in the project time to absorb and translate what they have learned. Not only will this make the statistician more articulate, but it will enable the action officer to lay the groundwork with the senior leader.
LTC Charles Weko, Senior Training Analyst
Program Analysis and Evaluation Division
Training Branch, DAAR-RMP
Editor’s Note: The views expressed in this article are those of the author and do not reflect the official policy or position of the Department of the Army, Department of Defense, or U.S. Government.