Anomaly-Centric Analysis 092507

Executive Summary:

Anomaly-Centric Analysis (ACA) is an abbreviated from of the scientific method of hypothesis testing. ACA begins by the analyst developing a simple pseudo-logical model of the expected influential factors of analysis. Next the analyst tests the model and searches for anomalies. The analyst targets anomalies and analyzes them to produce an advanced understanding of the cause and effect relationship between the influential factors of analysis. ACA is designed to identify the nuances and indicators that most provide insight into the reality of the situation or entity being assessed.

The Model:

Analysts generally cue into certain factors they believe to be at the heart of cause and effect relationships and or correlative elements (indicators) to produce predictive analysis. This often runs the risk of “confirmation bias.” Confirmation bias is present when a premise precedes analysis. Often the analyst will ignore evidence to the contrary of the premise, seek confirming evidence, utilized cultural-specific maxims (mirror-imaging bias), etc. However, the fact is that analysts cannot afford to take a “Humeian” approach and question all that they know. We as analysts must develop certain procedures and cognitive schemas to develop timely and accurate analysis. ACA is a structured analysis approach. To be sure, the analyst develops a model, collects specifically indicated information and uses standardized techniques to develop analysis in all situations.

ACA begins by the analyst making logical conclusions about the influential elements of his intelligence requirement (the query). The analyst then develops a model that he/she believes will develop accurate analysis. This process not only allows the analyst to develop structure for analysis, but it identifies the analyst’s assumptions about the intelligence requirement (IR). For example, an analyst may conclude that the more small tribes an African country has, the higher the rate of tribal conflict present in that country. Therefore, the analyst’s model would consist of the number of tribes, and the sizes of tribes, correlating to the rate of tribal conflict in the country. The analyst may create an equation as such:


The analyst then begins to assess his model. Simply by producing a spreadsheet of the countries in the target region and their
respective “# of tribes” and the “avg. percent of population belonging to these tribes” the analyst can then create a third column (inserting the equation Column A• Column B, or cell A1• B1) calculating the conflict coefficient. The analyst can then order the sheet by the conflict coefficient column and begin to test his/her model. At this point the analyst should develop method of determining the rate of conflict in the country. Using open source methods should provide the analyst ample information to determine how many armed conflicts, demonstrations, etc. a country has experienced for a specified time period. The analyst may simply list the numbers of conflict incidents or on going struggles that a country is experiencing or assign a scaled number value (e.g. 1-10). The analyst may start by comparing the countries containing the highest and lowest conflict coefficient. If the analyst’s model is accurate, there should be a relatively proportionate correlative variance between the conflict coefficient and actual rate of conflict between the countries. The analyst should proceed by comparing the entire list. If the analyst’s conflict coefficient and actual conflict rate does not correlate then he/she must look to other potential influential elements.

Perhaps the analyst determines that geographic location respective to the tribes is an influential factor in tribal conflict. The analyst may then add another variable to his/her equation. Perhaps the analyst determines that the average number of tribes bordering a single tribe’s territory is an influential factor. Perhaps he/she determines that the number of disputed territory is an influential factor, or for simplicity’s sake the analyst may determine that population density will suffice to address the issue. Therefore the analyst would augment the model and perhaps develop an equation as such:


Again the analyst tests the conflict coefficient to see if it generally correlates to the actual rate of conflict in the target countries. The analyst should continue to develop the model adding columns to his/her spreadsheet, combining and add factors in various ways until he/she develops model that generally reflects the reality of the intelligence query. If you can divide the list of countries into 3 or 4 groups (depending on the size of the sample) and can say that groups generally fall into an accurate lower, middle, and upper 33, or 25 percentiles, respectively, then you have likely created a model that reflects the major influential factors of your query. Building an exact model is not likely possible. The analyst should not attempt to account for every nuance of the query. When the model generally reflects reality, the analyst is ready for anomaly centric analysis.

Before we continue, the question should be addressed whether an analyst is capable of producing such a model. Analysts produce assessments based on the information available. They use indicative information to determine unknown factors, such as analysis of future events (predictive analysis), analysis of enemy behavior and strategy (counter-intuitive analysis), trends, patterns, tactics, techniques, etc. The very production of analysis insinuates that the analyst is able to survey available information determines its meaning/relevance, determine its reliability, and determine what the information indicates. Failure to create a model that generally represents the influential factors of an intelligence query should bring the analyst pause. Of course individual situations require individual consideration; however, analyst who are using schemas and weighing information in a way that does not reflect the major influences relevant to their intelligence query should reassess their analysis. We began by developing a model based on the analysts assumptions about the intelligence requirement, if the analyst fails to produce a model that generally reflects reality then the analyst should examine his/her assumptions. In this way ACA helps to battle analytical biases and address the issues raise by modern research into the psychology of intelligence analysis.

Anomaly Centric Analysis:

Once a suitable model has been developed to represent the majority of influential concerns of a target query, the analyst should begin to target the anomalies. While the model is far from perfect, due to the fact that you will likely never develop a model that can respond to dynamism of the real world, the model does serve the purpose of identifying the entities that do not fit the model. This is where the analyst finds the nuances of individual situations. Any analyst will tell you that you must address each dimension of context (environmental, social, political, etc), specific to the intelligence requirement in order to create reliable analysis, anomaly centric analysis is where the analysts addresses these contextual elements. If a target-specific context exerts its influence it is likely to distort the target queries position in the model. For instance, perhaps a country does not fit the model. Country X has a few large tribes, which have historically secure borders. The majority of control in the government is determined by resources which reside well within the boundaries of tribal territory. The tribes have, for decades, welded relatively equivalent power and influence in the country. Conflicts have been restricted to a few border regions, and have not generally affected the government’s stability. However, over the past few years inter-tribal fighting has significantly increased and increasing deeper border incursion have been reported by one tribe. Assessments have increasingly indicated that stability in the region is deteriorating, and civil war seems all but a matter of time. You know that the country does not fit the model. You know that there is a factor beyond tribe number, size, and population density. You identify an anomaly and you target it for analysis. It does not matter what you find, perhaps an HIV epidemic has offset the balance of power between the tribes, perhaps there is a new Islamic or communist movement moving through the country. Your conclusions however, become an indicator. You may conclude that a regional rise in HIV is an indicator that tribal conflict will increase. The nuances identified in anomaly centric analysis can develop into a system of indicators and warnings, indicating trends and likely events. They may indicate the affects of an up coming election. An HIV epidemic may start in one country and grow. The identified nuance/indicator may grow to become a primary factor in the model. ACA can evolve over time to react to a dynamic environment. Any analyst knows that analysis is never done; it grows and changes with the times. Analysts can share, tweak, and discuss models and spreadsheets. Over time models can be perfected, and adapt to fit a changing world. ACA clearly presents the analyst’s assumptions for other analysts to evaluate, it is an interactive, competitive, and naturally select (based on success) process.


Although the above example is purely theoretical, it demonstrates the adaptive, analytical nature of ACA. ACA develops and tests the analyst’s assumptions, identifies the influential anatomy of an intelligence query, helps the analyst target nuances, and identifies indicators and warnings. ACA is a process built to evolve. Someday you may use the Blair Model of Tribal Conflict, and later the Blair-Smith Model, which competes with the Johnson-Rodriquez Model. The information age has brought analysts closer than ever, using ACA thousands of models may compete for dominance, judged by their accuracy and flexibility. Information systems coded with fuzzy logic, and genetic programming may someday produce automated intelligence, identifying anomalies and bringing them to the attention of the analyst. ACA’s structured approach provides some common ground from which assessments may be analyzed and compared to other assessments. ACA develops a platform from which analysis is converted from ambiguous processes to clearly understandable process, freeing up the analyst to concentrate on the nuanced nature of intelligence queries.