Military manpower lends itself, almost perfectly, to analysis and modeling.  We have large and detailed data bases and opportunities for both natural and controlled experiments.  And, overall, we have made considerable progress in modeling behavior and decision making.  But I think it safe to say that we have about reached the limits, perhaps exceeded the limits, of many of our traditional modeling approaches. What I want to do today is address three areas in which I think there is real progress to be made and offer my assessments of how progress might be realized in those areas.
         
First is the general area of inventory models, retention, and, in particular, the use of discount rates as a major factor in decision making.  ACOL models were an enormous leap forward, but remain of limited scope and questionable predictive power.

          While it is certainly true that money far in the future is valued less than money today, it need not be true that service members’ behavior is driven by, or even approximated by, traditional discount rate analyses.  Indeed, all the evidence available suggests that any attempt to use discount rates to explain service member decisions implies behavior somewhere between inconsistent and irrational.  Nor is there much support in the economic literature for this approach.  Going back to Irving Fisher, the time value of money was primarily a portfolio management concept for people and institutions with accumulated wealth.  It was not primarily a model for individuals’ preferences and behavior.   Among service members we routinely find behavior that is inconsistent with traditional net present value theory.

          We must, I think, conclude that while future income is certainly discounted in some way the net present value of expected future compensation is clearly not the primary, and perhaps not even a major, factor in service members’ decision making. 
          Discount rate analysis would be fine if it held predictive power, but it does not.  We are able to model behavior, after the fact, by plugging in the “correct” discount rate, but such exercises are, from a policy perspective, useless because the magic discount rate varies radically from situation to situation.  Plug the discount rate for VSI/SSB analyses into a retirement model and things look all wrong.  Our over focus on expected future earning has, I think, lead us in many unfortunate directions. 

          First, it has lead to a profound misunderstanding of the military retirement system.  Military retired pay is frequently characterized as expensive and back-loaded.  But, setting medical benefits aside, only about 16% of active duty compensation cost goes towards retirement pay, that’s less than the comparable percentage for civil servants.  And military retirement is far more front loaded.  A thirty-year-old active duty member is likely to start receiving benefits in about a decade,  while a 30 year old civil servant has about three decades to wait.  Overall, the military retirement system appears to be vastly more efficient than competing retirement systems.

          Second, it has created a mind set in which the way to solve inventory problems is to throw money at service members and would-be recruits.  Again, while it is true that increasing compensation will improve recruiting and retention, it may not be the most effective or efficient way, particularly once you get to a point where military compensation is competitive with, or better than, civilian opportunities.

          Third, by putting so much emphasis on compensation we have, to some degree, contributed to the increased costs of service members.  This, in turn, has lead to efforts to reduce the numbers of service members, which, in turn, exacerbates the need for quality and retention, driving up the average cost of personnel even more.

          I do not have an alternative model, but I do have some thoughts on how to develop one.  First, I think some combination of a rational expectations and permanent income hypothesis should replace expected net present value as the Holy Grail of retention.  The timing of promotions, for example, appears to drive retention decisions far more than its financial repercussions would suggest.  Expected future assignments, while difficult to measure, may also be quite significant.  Another important feature would be discontinuities, asymmetries, and other such properties on which standard econometric models tend to choke.

A second area in which I would encourage research is the topic of “requirements” at the service-wide, or macro, level.  The term “requirements” alone is sufficiently hideous to suggest vast room for improvement.  And while there has been some good work at the platform level, the current state of requirements analysis on a macro level is almost nonexistent.  Everyone more or less knows, or at least assumes, that the current inventory requirements are a function of the career profile which, in large part, is a function of the lack of lateral entry, the retirement system, and various limits on “control grades” and the like.  Trying to derive what the true “requirements,” or more properly the demand functions, might look like has been widely viewed as impossible.

          This has lead to an approach that is, I think, at least undesirable and, potentially, dangerous.  The alternative to deriving the underlying “requirements” has been to try to adopt more flexible and targeted compensation and personnel management in hopes that the underlying demand will emerge in a sort of Revealed Preferences scenario.  I see two problems with this approach.  First, the bulk of management and compensation remains under the traditional systems and any flexible features are simply add-ons.  Tools such as SRB and AIP have many desirable properties, but may well be simply added costs on top of a base compensation system that is adequate, or even above adequate.  Meanwhile, vastly expensive and inefficient benefits (such as Redux repeal, concurrent receipt, and TriCare for Life) drive costs up far more than any targeted reforms could hope to offset.  Second, flexible compensation tools often lead us to increase the notional requirements.  As an example, recruit standards may be needlessly high, in large part, because we are capable of recruiting that quality of individuals, but we do so only by offering signing bonuses, GI Bill, and entry at advanced rank.  The same may be true, in large part, with retention driven by SRBs.

          Difficult as it may be, a realistic analysis of requirements (and their associated costs) seems to be the only alternative.  This might start with a skeptical review of recruit standards, but would absolutely require a life-cycle costing approach to personnel.  This would mean, among other things, allocating retirement and dependent support costs more accurately across the personnel inventory.  Along these lines, I would suggest a construct in which recruiting and initial training costs are amortized over the first six years of service, retirement costs are amortized between six and twenty YOS, and the costs associated with dependents are amortized based on typical family size at careen point.

The third area I see ripe for improved analysis is the issue of outsourcing and mil/civ conversion.  The standard approach here has been to take as given the fact that military personnel are most costly, civil servants next and contract services, where practical,  the least.  We have, as a result, searched for conversion and outsourcing opportunities anywhere feasible.  But we are starting to see the downside risks to such an approach.  The new concept of “dwell” is an issue largely because the Army and, to a lesser degree, the Marines, don’t have enough total active duty to rotate through theatre at acceptable paces.  The number of combat units is not a binding constraint at the moment, it is the shortage of soldiers and marines to move through those units.   At least part of the reason for a shortage of personnel to rotate through the combat units is the reduction over the years of military personnel in non-military essential positions.

          I liken the approach so far to a just-in-time inventory model or to a  hub-spoke systems for an airline.  Those approaches probably do minimize cost when things are running properly, and they may well minimize long-run average costs.  Unfortunately, they also tend to lead to major breakdowns when the system is sufficiently disturbed.  In running a military, however,  we need to plan for operations during major disruptions and should built in a certain amount of redundancy and excess capacity, perhaps even deliberate inefficiency in peacetime.

          In the Navy we are finding that giving up so much of our in-house technical and engineering capabilities at the systems commands is contributing to acquisition problems, most notable in the LCS program, but elsewhere as well.  Additionally, the almost cult-like devotion to reducing manpower has lead to seemingly absurd decisions  in areas such as crew levels for surface combatants.  Unknown billions will have been spent on R&D, procurement, and shore-side maintenance and  support in order to save a relative handful of billets on each of the DDG-1000 class.  With only seven ships in the class, just the interest on those additional outlays (at real t-bill rates) would more than pay cost of the billets saved even priced at average cost.  But the billets saved are not of average cost, they are predominantly junior billets, which are not only cheaper, but may well be vital to creating experienced mid-grade sailors.

          Rather than looking for ways to reduce end-strength in hopes of saving some marginal (and perhaps nonexistent) cost difference between military personnel and civilians or contract support, it seems far more productive to try to understand why using military might be, or might appear to be, a more expensive option.  The military compensation system is not particularly efficient, but neither is the civil service system nor, presumably, contractors’ systems.  In any case, it seems unlikely that those differences in efficiency are large enough to drive a significant cost difference. Much of what we lose in efficiency on, say, military healthcare may be recouped with the military’s more efficient retirement.   But, on the other hand, if any added costs are largely a function of quality, flexibility, and productivity, then military personnel may well be the desirable, even cost effective, solution.  Just one such factor is the centralized nature of recruiting, detailing, training, and assignment of military personnel which creates easy to see costs in BuPers, but frees up the time of local supervisors and managers who would otherwise recruit and train civilians on an ad hoc basis or have to select and monitor contractors.  Finally, I would note that any end-strength conversion that causes a need for a reservist in the event of a major mobilization is almost certainly inefficient.

To summarize, I see three areas ripe for future manpower analysis.  These are:
-A better understanding of retention and reenlistment behavior, that deemphasizes discount rates.
-A better look at the underlying demand for military personnel at a service-wide level, primarily in terms of the experience and quality mix relative to true life-cycle costs.
-An understanding of the real trades-off and financial implications of conversion of active duty roles to civilian and contractors, particularly in terms of any flexibility lost with reduced active duty levels.