Analytics and Modern Warfare

Dominance by the Numbers

M. Taillard

• Editorial: Palgrave Macmillan
• Año: 2014
• ISBN:
  9781137407870
• Categoría: Économie
• Categoría: Économétrie

This book details very simply and for even the most novice of potential analysts not only how to perform analytics which describe what is happening, predict what is going to happen, and optimize responses, but also places these analytics in the context of proactive strategy development.

PART I

Descriptive Analytics

Remaining strategically competitive can be difficult enough under the best of conditions, but unless one can accurately describe what exactly is happening with precision, it can be almost impossible. Most people think they understand what they see, but the limitations of our senses are severe and profound, making personal observation nearly useless. Confidence without proper analysis comes not from an understanding of what has happened, but from an inability to conceive of other possibilities. For example, during reconnaissance should a person observe several convoys leaving an opposition stronghold they may come to the conclusion that the location has reduced its operating strength as a result of having fewer people and fewer resources readily available. Should the observer be highly confident of this, it does not mean they are right, only that they have not considered the possibility that the movements witnessed may not significantly decrease functional strength at all, or that it is only a short exercise, or that it is only one element in a larger movement that includes incoming resources as well, or that the convoys were only moving a short distance to expand their regional operations, or any of a number of other possibilities. Casual observations about the movement of resources lead to common misperceptions during peacetime, as well, when the people of a nation believe that jobs or operations are being sent overseas when, in fact, national production continues to reach record highs and operations move from highly populated, high-cost areas to cheap, rural areas where fewer are there to witness the change. During peacetime, these types of sloppy conclusions drawn from observation are, at worst, an obnoxious political talking point, which begets little in the way of action, but during a conflict this can mean the difference between life and death for a number of people—the difference between winning and losing a battle. When there is no room for error, then one’s actions must be mathematically exact—the same unforgiving precision that is the bane of so many children during their math lessons must now become the bane of the opposition.

There are two broad categories of statistical analysis: descriptive and inferential. This can be a bit counterintuitive, because both are used descriptively, but descriptive statistics describes the thing being studied while inferential statistics uses descriptions of the data collected to infer descriptions of the thing being studied. Essentially, descriptive statistics provide information about the traits being measured while inferential statistics are used to determine whether the descriptions are valid, and the strength of the data in explaining or predicting some variable or relationship between variables. When testing data, the idea being tested is known as a hypothesis, which is a specific statement about the thing being studied that is being assessed for accuracy. If the results of the test are strong enough, they are said to be statistically significant, which means that the probability of the test results being caused by chance are so small that the very idea is rejected. This is known as rejecting the null hypothesis. The hypothesis for which you are testing is sometimes called the alternative hypothesis, while the null hypothesis states simply that the data being tested is not conclusive. It is a common mistake to say that the null hypothesis states that the alternative hypothesis is wrong but, in reality, you do not reject the alternative hypothesis, you are only incapable of rejecting the null hypothesis. When this happens, the possibility still exists that any significance in the data is merely the result of chance when doing sampling (discussed more in chapter 2). In order to reject the null hypothesis, the probability of chance being a factor must be very small, the most commonly chosen significance levels being 5 percent, 1 percent, 0.1 percent, and 0.00001 percent, depending on application; these are the milestones of probability, which must be surpassed—in other words, there must be less than 5 percent probability that the data collected were the result of chance rather than a systematic relationship.

When the results of a test are wrong, it falls into one of two types of errors. Type I errors are commonly known as false positives, which means the analysis shows something significant in the data when, in fact, there is nothing in reality. Formally, a type I error leads to the incorrect rejection of the null hypothesis. Type II errors, in contrast, are commonly known as false negatives, which means that the analysis shows that there is nothing of significance in the data when, in fact, there is something in reality. Formally, a type II error is the incorrect failure to reject the null hypothesis. It is in human nature that people are extremely prone to making type I errors. From the perspective of evolutionary psychology, the people who were prone to making type I errors would observe sounds and run away, thinking it was a predator rather than potential prey. They may lose a squirrel breakfast, but they will be alive to pass-on their genes. In contrast, those who are prone to making type II errors may observe sounds and think it is nothing significant when, in reality, they are about to be attacked by a predator. Over the millennia, this selective survival of specific psychological tendencies has created a human race extremely prone to finding patterns, which simply do not exist; they see faces in the moon or in a piece of toast, they form superstitions such as the belief that some action they took led to a particular outcome which then becomes a ritual such as before sporting events, they attribute particular outcomes or actions to the metaphysical despite a clearly explainable cause, and they make relationships between things when there simply isn’t one. Such errors are also possible when performing analysis, but the level of significance we place on the results of an analysis is critically important in evaluating the likelihood of such as error being the case. For example, the standard for announcing the discovery of a new particle in physics is 99.9999 percent certain, leaving only the very tiny probability of the data being the result of a type I error. In using this analytical approach, not only can we be extremely confident in what we are observing, but we can also be extremely precise in the degree to which we are uncertain, and assess the sources of the unexplained portion of the equation (discussed more greatly in chapter 2).

In the end, speculation, opinion, and observation—these are just different ways of stating the same thing: guessing; while measured and analyzed data provide facts with which truth can be derived, free of subjective interpretation, and error. Throughout part I readers will learn how to collect data and quantitatively describe it, how to compare the measurements and capabilities of two entities to determine whether there are any significant differences, as well as how to take advantage of those differences by developing models that both visually illustrates the nature of the problem at hand as well as creates standardized equations, which can be utilized repeatedly to find an accurate solution. Descriptive analytics are used extensively in the remainder of the book, as well, since the calculations and data included here are necessary to perform the predictive analytics discussed in part II, and the operational analytics discussed in part III.

CHAPTER 1

Descriptive Statistics

Most people can qualitatively describe what they see within certain limits—events that have occurred, the appearance of things, the direction in which things are moving, and so forth. Unfortunately, the usefulness of these basic observations is extremely limited, incapable of distinguishing between large volumes of things, identify traits that are not readily visible, and influenced by a high degree of subjectivity. Personal observation includes a large amount of estimation and the use of neurological responses that are frequently flawed. Illusions function by tricking the natural processes the brain uses to understand sensory information, and personal interpretations of information of events have a high rate of error resulting from insufficient information and the projection of one’s own beliefs and ideas on the external world. When even simple, qualitative analysis exists beyond the realm of our natural observational capabilities, it should then come as no surprise that our ability to accurately analyze anything with any amount of quantitative certainty is completely zero; our ability to measure anything in a manner that is useful in making precise decisions or predict future outcomes is limited to our access to measurement tools, which we do not possess naturally. Since such casual estimation lacks precision and functionality, this makes it little better than guessing. It is for this reason that descriptive statistics are necessary.

Descriptive statistics are some of the simplest yet most fundamental of analytics that one can perform; not only do they provide highly valuable insight into the world around us, but they are also the foundations upon which all other analytics are built. They allow us to understand the nature of things around us in a way that transcends our senses, or even what one might consider logical common sense. Measuring a single item can tell you a good deal about that item, and measuring several more of the same type of item will provide even greater information about the range of possible measurements you will encounter, what measurements are typical, and the kinds of trends you can expect to observe. For example, in measuring a single person, you can measure their physical endurance, their eyesight and how well they can aim, their response time, their knowledge of strategy and tactics, and much more that would tell you about how effective they will be on the battlefield. That one person, though, may or may not be typical, and even if they are, this says nothing of how probable it is to encounter the atypical.

These most basic of analytics describe reality, both the certain and uncertain, using quantitative measures and quantitative descriptions of qualitative variables, so that decisions can be made with certainty rather than by simply guessing. They allow us to take known facts, the data of exact measurements with little meaning independent of context, and use them to derive truths—answers when questions arise from imperfect access to information, or imperfect ability to evaluate and process information through casual observation. This chapter focuses on describing how to perform simple descriptive analytics so that one can do more than provide vague or broad descriptions, instead of allowing them to define with surgical precision the exact nature of the opposition, themselves, and the conflict at hand. This chapter also focuses heavily on introducing the basic mathematical concepts that will be utilized in later chapters to show how they can be used in decision-making, including defining the different types of variables, which can be used to describe, categorize, and measure various elements of the things around us.

In this world of data, more is better, because statistics are used not just to describe those things being measured, but also the data itself in order to derive as much information as possible. Each type of measurement taken is called a variable, which is any value that can vary. There are several types: nominal, ordinal, interval, and ratio. Nominal variables are those that categorize, but which have no definite order. It is extremely common to collect data on national citizenship, type of skill set, organizational affiliation, and other things which, by themselves, have no comparative values. Ordinal values are those which can be placed in some kind of order but which have no quantitative measurements, such as rank. Both nominal and ordinal variables are known at categorical, because they can take-on only a limited number of possible values which categorize data rather than measuring it. By contrast, interval and ratio variables are known as continuous variables, because they can take-on a theoretically limitless number of measured, quantitative values. Interval variables are similar to ordinal variables, except that they have meaningful quantitative values, such as time. 8.00 a.m. and 10.15 a.m. can be categorized quantitatively, and even incorporated into simple comparisons, such has having a 2.25 hour difference between them, but since it is in their nature that their values are derived comparatively as an interval, you could never multiply or divide between them; you could not divide 10.15 a.m. by 8.00 a.m. and get any sort of meaningful value. Ratio variables, however, have an absolute value using 0 as a reference point. Using time, again, as an example, 20 minutes and 5 minutes are each ratio variables; each have meaning derived from its own volume, 5 being ¼ or 25 percent of 20. All data belong to one type of variable, and each variable being measured has a single type of data, though the difference may be distinct. When studying camouflage, color is a nominal variable since there is no way to put into order such values as blue, green, gray, and so on, however, you can measure the amount of light reflected off each color as a ratio variable. In each case, you are measuring the same thing—color—but measuring different variables associated with color. Understanding the types of variables is extremely important not only because the types of statistical tests that are performed will depend on the types of variables being tested, but because understanding the different ways in which the world can be measured, evaluated, and compared, is a decisive factor in the success of an operation. The successes of Napoleon Bonaparte, in many ways, could be attributed to the importance he placed on mathematics and

science, and he often associated himself with mathematicians, physicists, and engineers, all of whom play a large role in the success of any military, as demonstrated by the US Army Corps of Engineers and the work they have accomplished, or Albert Einstein and his work in what was, at the time, theoretical physics, which led to the creation of the atomic bomb and shaped global military policy for nearly a century.

There is another way to categorize variables, which are dependent entirely on the relationship each variable has to other variables: by whether each is dependent or independent. A dependent variable is one whose value is shaped by the value of some other variable. How well a person can shoot, as measured by the rate of successes in hitting a target, is dependent on several other variables, such as the volume of time spent practicing, eyesight, and control over extraneous movements in the body and hands (keeping them steady). These variables that determine the value of the dependent variable are known as independent variables because their value is independent of any influence within this context. Each independent variable, though, is also a dependent variable in some other context; eyesight is dependent on genetic variables, environmental variables over the years which can damage the eyes, and corrective variables such as glasses or surgery. As all the variables around us determine outcomes and are, in turn, determined by other variables, by collecting sufficient volumes of the right types of data it becomes possible to “read” reality, understanding past, present, and even future.

Upon collecting data, the total number of measurements that have been collected of a single type is denoted by the letter n. In other words, if measuring 10 distances traveled and the amount of time it takes to make each trip, then 10 would be your n value, since there are 10 things being measured, which yield 10 of each type of variable. Simple counting such as this can provide extremely useful information, depending on the context. Heat maps, for exampl...

Índice

• Cover
• Title
• Introduction
• Part I Descriptive Analytics
• Part II Predictive Analytics
• Part III Operational Analytics
• Afterword
• Critique of Current Methods
• Index