Better Models for Specialized Career Fields Published March 9, 2021 By Capt Brian Kruchkow Wild Blue Yonder -- It’s no secret the Air Force is currently unable to fill cockpits for its aircraft due to a pilot shortage.1 Even with COVID-19 increasing retention from 46 percent of pilots to 51 percent, the service is still 2,000 pilots short, with no end in sight.2 The Air Force is attempting to fill the shortage by increasing pilot production and simultaneously attempting to retain experienced pilots through different career incentives, especially the Aviation Incentive Bonus (AvB), which could be as high as $400,000.3 The root cause of this shortage is the Air Force does not understand how pilots make career decisions, and therefore the models the Air Force relies on for manning requirements hold incorrect assumptions. Before outlining the RAND Corporation’s Dynamic Retention Model (DRM), which projects the retention across the Air Force, a case should be made for why accurate manning models are necessary. First and most obviously, the size and competency of our military force is a national security issue with grave consequences. When the service relies on a model that projects a number of trained personnel choosing to voluntarily stay and the model is wrong, especially in cases where it overestimates retention, it is a costly mistake in terms of funding from Congress for readiness and capabilities. Second, the models portray the metrics leaders care about, and in turn, the standard to which they hold their subordinates accountable. Operation plans which require a certain number of highly skilled personnel rely on the Air Force to maintain those personnel. If the Air Force has a systemic problem with retention leaders, it should be put to task for their failure to meet objectives. Holding leaders accountable in this case is only possible with accurate models as it is all too easy to say the problem was insurmountable from the start. There was a recent success story in Air Education and Training Command (AETC) regarding modelling, and it proved how a good model will have positive organization wide effects. For years, the production of student pilots had been overestimated and every year the commanders would go back to their superiors and say the goal was impossible, and pilot training had underperformed its target again. On the one hand, the production target being given to the AETC commanders was unrealistic because headquarters did not account for how limited resources would be used to produce a pilot. But, on the other hand, the commanders themselves did not try to correct the target production rate, which led to an insidious cycle of setting and not meeting goals. A team of captains with the guidance of a group commander decided to model the production of a student pilot accounting for all resources which would limit production. The model that emerged was able to accurately forecast production, and it allowed headquarters to set a viable target and hold commanders accountable to that target. It helped the commanders as well as the line instructor pilots by lowering the student training requirement and giving them a realistic work schedule. It also gave senior leaders a realistic picture of the resources they would need to fight for if they wanted to increase production in the future. The Dynamic Retention Model is how the Air Force projects the number of officers and enlisted who will choose to stay in after their service commitment.4 For pilots specifically, the model looks at the end of their first Active-Duty Service Commitment (ADSC) and each successive point a pilot may separate thereafter. The formula determining whether a pilot will stay is compared to one determining whether they will leave and whichever is higher is calculated as either a retained or separated pilot respectively.5 In this calculation, there is a “taste” factor for military life based on assumptions and previous retention data, which assumes pilots choose to stay in the military based on specific and personal reasons known only to the pilot but which may be closely approximated.6 The problem with the DRM is it overestimates the decision to stay because it models a generic preference for military life, and does not include a preference for a specific job. When a person chooses to become a pilot in the military, they simultaneously display multiple preferences. First, they show some “taste” for the military, as they could have become a pilot in the civilian sector. There are economic barriers in the civilian sector to earning a pilot’s license and enough flight hours for an airline job, which may account for some of the taste for military, especially when the military will initially train you at no cost. In this case, the DRM would systemically overestimate the preference for military life as pilots who have gained the experience needed for airline jobs will no longer have the economic barriers which existed prior to flight training. Second, pilots display a “taste” for a specific career, as there are many career fields and jobs they could have requested within the military. This preference for a specific job is especially strong with officers who choose highly specialized career fields, i.e., aviation and medical, as each of these career fields require years of extensive specialized training followed by assignments focused on executing that knowledge and skill. This preference for a job is reflected in the many surveys the Air Force has conducted which overwhelmingly indicate that pilots prefer not to have extra duties, want opportunities to fly more, and do not want inflexible career paths which take them out of the cockpit. This should lead the Air Force to include a “taste” or preference for job in the DRM, possibly with a higher weight than the “taste” for military service. A “preference for job” factor could be modelled one of several ways. The easiest would be to take the number of people who express interest in being a pilot in the military but are not given the opportunity or do not complete training and determine the number of those people who do not stay in the military or get out as soon as they have the chance. This would not reflect the preference of those who complete pilot training though. Another option would be to take the difference in the Marines, Air Force, Army, and Navy pilot retention and do a factor analysis to isolate job preference over a military career. Interesting results would surface from the Army having warrant officers perform their flying, and the Navy and Marines mandating their pilots serve in dual hatted roles due to their force structure. Finally, the Air Force has decades of survey data which they could sift through and determine a factor for job preference amongst pilots. The point is that the available computing power, statistical techniques, and modelling capability allow the Air Force to conceivably construct this variable. The impacts of including a “preference for job” factor in the DRM are threefold. First, it would lead to a more accurate model of retention, which would give senior leaders a better tool for forecasting personnel and manning requirements. Second, the Air Force gathers metrics on what it cares about as a service. Currently, we are ignoring our people by not understanding their decision-making process and it leads to overestimation of the decision to stay in the service. Forcing accountability through metrics requires Air Force leaders to respond flexibly to their pilots’ desire to do their job over career advancement. Further, like the dynamic “taste” factor in the DRM currently, the “preference for job” factor reduces stochasticity by simply asking pilots what assignments they prefer and the degree to which they prefer them. With this data, the Air Force is appropriately armed to retain the needed number of pilots. Given this better retention model, how would the Air Force incorporate individual job preference to retain needed personnel without breaking the service? The Air Force, to their credit, has tried a guaranteed flying track (PHOENIX AVIATOR), and offered to exempt pilots from leadership roles to keep them in the cockpit longer. Neither solution worked due to the service’s need for bodies on staff or in the Air Operations Center. While there will always be a balance between staff and flying assignments, the underlying problem is not solved by low retention rates. Forcing a pilot who does not want a staff assignment to fill one results in the pilot either taking the assignment or separating. If they separate, then the Air Force must fill a flying billet and a staff billet with one less personnel. Here is where a better model helps; knowing the true number of pilots who would refuse out-of-cockpit duties can redirect assignments toward an optimal outcome, for both the pilots and the Air Force. Some staff jobs may have to go unfilled as well as some cockpits, but the Air Force is better served to understand the pilot has a say in this dynamic, and if the Air Force wants to maximize the number of billets filled it needs to understand how the pilot makes career decisions. Further recommendations are for the Air Force to allow more flexible career paths without negative impacts to members’ careers. Desires which do not fall into the traditional flying/leadership career track could be leveraged to increase retention when needed. For example, pilots who do not enjoy school but are excellent flyers could find exemption from forced PME schedules, pilots who want a certain school but may not be eligible could be granted waivers. The possibilities are only capped by commander’s willingness to listen and flexibility to respond so long as the pilot understands the tradeoffs they are accepting with their requests. This does not have to be widespread either; grant flexible assignments to commanders to retain pilots within a certain window, in specific career fields, which are hardest hit by low retention. In any case, competing with the private sector in terms of capital will always be a losing proposition for the Air Force. Having good models which reflect realistic decision making, flexibility, and listening to pilots may give the Air Force the advantage it needs to retain highly specialized Airmen. Appendix A In this appendix the stochastic and non-stochastic formulas for retention are laid out. I understand the average reader may not care about these formulas, but the desired take away is this: we can model retention to better reflect the way our pilots make decisions, and in doing so we will align our service with its core mission: to fly, fight, and win. As the model shows, a service member will stay (be retained) when the value of staying (VtS) is greater than the value of leaving (VtL).7 The military preference variable, γm, shows to what degree an individual prefers military life over civilian, and the model states it is hard to estimate from given data. What should be added for aviation (or any highly specialized career field) is a preference for job variable modelling the service members desire to do their job over performing other roles within the military. This will allow targeted retention in those career fields. The following is a look at the DRM to show where I believe the variable could go, mainly to replace γm:8 Captain Brian Kruchkow Captain Brian Kruchkow (BS, US Air Force Academy; MS, The Johns Hopkins University) is a First Assignment Instructor Pilot at Laughlin Air Force Base. He has served on staff implementing the Air Force’s push to innovate and worked on the team which modelled production across AETC. This paper was written as part of the SOS Air University Advanced Research (AUAR) elective, Artificial Intelligence section. The author would also like to thank Col Ron Garner, USMC; Dr Paul Hoffman, USAF retired; Capt Mark Verbruggen, USAF; Capt Stephen Conroy, USAF; and Capt Nicole De Luca, USA, for their significant contributions to this article. Notes 1 Air Force Magazine. “USAF Is Closing the Pilot Shortage, But Still Planning for Post-Pandemic Dip,” October 30, 2020. 2 Air Force Magazine. “USAF Is Closing the Pilot Shortage.” 3 US Air Force, “Air Force Announces Fiscal Year 2019 Aviation Bonuses,” Accessed December 15, 2020. 4 Michael G. Mattock and Jeremy Arkes, The Dynamic Retention Model for Air Force Officers: New Estimates and Policy Simulations of the Aviator Continuation Pay Program , Technical Report, TR-470-AF. Santa Monica, CA: RAND Project Air Force, 2007. 5 Mattock, The Dynamic Retention Model, 23-30. 6 Mattock, The Dynamic Retention Model, 23-30. 7 Mattock, The Dynamic Retention Model, 24. 8 Mattock, The Dynamic Retention Model, 23-25.