1 / 13

Conceptual Model: Assumption Testing

Conceptual Model: Assumption Testing. Robert Eberth Sanderling Research Corporation 25 March 2008 robert.eberth.src@cox.net 571.257.6997. Conceptual Model: Assumption Testing. Precepts: Overarching purpose is to test the ABS Val framework

neva
Télécharger la présentation

Conceptual Model: Assumption Testing

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Conceptual Model:Assumption Testing Robert Eberth Sanderling Research Corporation 25 March 2008 robert.eberth.src@cox.net 571.257.6997

  2. Conceptual Model: Assumption Testing • Precepts: • Overarching purpose is to test the ABS Val framework • ABS Val process based on modern scientific method and its falsifiability criterion • The null hypothesis is the research hypothesis that the model is valid (“sufficiently accurate”) for its specific intended application • Objective then is to falsify the null hypothesis • Failure to falsify the null does not prove the model valid for the specific intended purpose, but should increase confidence in its validity • Degree of confidence then depends on the rigor and power of the tests applied

  3. Conceptual Model: Assumption Testing (cont’d) • Plan: • I.D. the analytic questions at hand, their metrics, and degree that results are expected to shape decisions • Detailed review of all documentation • Interview Application Sponsor • With the App Sponsor, I.D. the referent; i.e., the proxy for the real world for accuracy comparisons • With the App Sponsor, I.D. the accreditation criteria • How “accurate” must the model be • How can/will accuracy be determined (quant/qual) • Criteria must establish lower bounds of acceptability

  4. Conceptual Model: Assumption Testing (cont’d) • Assess the validity of the referent • Confirm that no alternative – and preferable – referent is available or could be made available • Assumption test the referent (for other than empirical datasets): • I.D./derive inherent assumptions • Perform logical verification – adequacy and correctness vis-à-vis underlying theory and assumptions • I.D. the operational implications of the assumptions • Determine the bounds of validity imposed on the app’s problem space and on the model’s validity assessment by the referent’s assumptions • Determine whether the operational implications and bounds of validity are acceptable to the Application Sponsor • Independent SME reviews (ideally, contrarian reviews)

  5. Conceptual Model: Assumption Testing (cont’d) • Assess the validity of the conceptual model • Potentially three separate assessments • Theoretic sub-model • Mathematic sub-model (if it exists) • Algorithmic sub-model • Each in turn receives same assessment techniques • Logical verification – determining sub-model is an adequate and correct implementation of its predecessor. • Assumption testing • I.D./derive the assumptions that are inherent to/embedded in the sub-model • Determine the operational implications of the identified assumptions in the context of the particular application

  6. Conceptual Model: Assumption Testing (cont’d) • Determine the bounds of validity of the model that are the result of the identified assumptions • Determine whether the operational implications and bounds of validity are acceptable to the Application Sponsor for the intended application • For some models, it may prove necessary to reverse-engineer one or more sub-models from later models. It may even be necessary to reverse-engineer the conceptual model, or portions of it, from source code. • The referent serves as the predecessor for the theoretic sub-model

  7. Results • BLUF: Framework “worked,” (or more correctly, “is working”), but: • More work needed on how to select and assess the validity of the referent when empirical data are not available for use as the referent • Framework needs templates • Groundwork: • This portion of effort dealt with the theoretic model. Algorithmic model work yet to come • Began with study of Pythagoras Manual and related detailed discussions with Edd Bitinas. However, Pythagoras itself was not assessed (V&V assumed) • Interviews w/ LT Robin Sparling, USN, the COIN study’s Project Officer (in place of Steve Stephens), and study of several study-related documents she provided

  8. Results (cont’d) • Interview w/ Dr. Akst, the Application Sponsor • Purpose was to “make headway in developing a COIN model.” • Did not specify an ABS, let alone Pythagoras • Approved recommendation of using “sea versus land basing” as study’s analytic question, but did not specify it going in. • Approved stated Marine missions, and O.K. with implied mission • Insisted study must use real-world dataset

  9. Results (cont’d) • Findings: • Multiple objectives: • OAD was to “make headway” in developing a COIN model • NGMS was tasked to determine whether and how Pythagoras could be used to support IW analyses • Study at hand had the analytic objective of determining whether it was best to leave the MAGTF ashore or afloat in a Columbian HA/DR/Security scenario • USMC missions in Columbian scenario: • Refugee camp security • Humanitarian Assistance • Disaster Relief

  10. Results (cont’d) • But, study team found no way to directly evaluate the effectiveness of mission performance • Thus decided to use allegiance changes of population segments among several distinct affiliation possibilities – thus producing an “implied mission” of keeping the bad guys from gaining strength • Stated as “Do not allow illicit organizations to take advantage of situation” • But may also imply that ABSs in general and Pythagoras in particular cannot support traditional MOEs of mission performance • Alternative possibility: When the population is the Center of Gravity, may affiliation changes be the best MOE for mission performance (and not just for HA/DR)? • NB: The population is NOT always the COG of an insurgency – it depends on the goal of the insurgency

  11. Results (cont’d) • Several assumptions then had a large impact: • Modeling the transitions among affiliations as a Markov process (a “memoryless” process) • Constant transition probabilities across all time steps (except those during the Marines time in-country) • Constant transition probabilities across all time steps while the Marines were in-country (although different probabilities from the baseline) • Initial indications are that the above assumptions absolutely pre-determined the results and in a predictable way (i.e., the model became deterministic if allowed to run to steady-state)

  12. Results (cont’d) • Unfortunately, that may mean that OAD can’t make a solid determination on the usefulness of Pythagoras in the IW or COIN context from this particular application • It also may mean that the answer to the one analytic question (afloat or ashore) depends entirely on the methodology used to develop the transition probabilities – the “influence estimation” and “salience” parameters • And those are suspect because of potential bias in data collection/analysis methodology (semantic differential) and distinction between “data” and “context”

  13. Results (cont’d) • Initial indications wrt the study (again, only from assessment of the theoretical model, so subject to change): • Probably cannot yet give a defensible answer to the afloat/ashore analytic question • Implementation assumptions too limiting • Semantic differential data collection/analysis methodology far too suspect • BUT, study may represent a huge leap forward in IW analysis • Could/should cause a re-evaluation of COGs and MOEs for IW environments • Could/should lead to a series of studies on semantic differential and alternative methodologies for capturing the propensity of persons to change affiliations, particularly in response to actions/events rather than just presence

More Related