This chapter covers three presentations on workforce modeling in the federal government, the Federal Aviation Administration (FAA), and health care. Modeling methods and other factors important for the modeling process in each situation were discussed, as were the challenges involved in each case.
The session’s first presentation defined strategic workforce planning and described three historical examples of the application of such models in the federal government and the different challenges involved in each. For this session, the committee had asked the speaker to consider the following questions, which he touched on throughout the course of his presentation:
How have workforce planning and analysis efforts evolved over time in government agencies, in terms of application, tools, and techniques? How have specific legislations driven the importance and legitimacy of these efforts, and the application of a methodology or approach? And, how have these efforts grown in sophistication, moving from manpower analysis to strategic workforce planning?
Alex Manganaris (consultant, public-sector human capital) began by describing strategic workforce planning as a “supply and demand exercise” in which the supply component consists of the available workforce and the demand component consists of the workforce that is needed or will be needed in the future. The ultimate goal of workforce planning is matching supply with demand. Understanding workforce supply, he explained, involves knowledge about several characteristics of the workforce, including the mix of workers (employees or contractors, blue collar or white collar, etc.), whether each worker is mission critical or support, and a consideration of the way the workforce will change over time as a result of retirement and turnover.
On the demand side, Manganaris said, understanding the workforce needed involves several components, including a strategic business plan that takes into account any foreseeable changes in business processes and technology or legislative demands, as well as stated workload measures and known budgetary constraints. He stressed the importance of integrating workforce planning with the budgeting process because resourcing ultimately influences the work that can be accomplished and the staff that can be funded to perform that work. He noted that multiyear budgeting is helpful for the process of strategic workforce planning.
Manganaris then provided three examples of federal government projects in which strategic workforce planning was used. The first project was undertaken for the Air Force in the early 1990s, at a time of cuts to funding. In order to decrease the number of layoffs in two large, civilian-dominated commands, a spreadsheet model was created to determine the minimum funding necessary to avoid forced layoffs, using a 2-year, month-by-month pro rata attrition rate and accounting for upcoming functional transfers between commands. This model showed that the resourcing needed to prevent the layoffs was unaffordable, prompting the decision to move funding for civilian pay from other commands with less use of civilians to avoid the large-scale layoffs.
The second example Manganaris described was a project for the Internal Revenue Service (IRS), following the 1998 IRS Restructuring and Reform Act. This act required a transition from a regional structure to a business-line structure, which affected approximately 65,000 employees, or about two-thirds of the IRS workforce. Manganaris’s office designed a transition model that influenced both the new IRS structure and the personnel transition plan. The transition model used linear programming, which focused on matching the current workforce with the demand of the new structure, by grade, series, and location. The modeling process was iterative, with biweekly results briefed to the then-IRS commissioner, Charles Rossoti. On the basis of Rossoti’s feedback on the model’s personnel predictions, the design structure was altered to decrease the predicted number of unplaced revenue agents because those agents are a critical occupation for the IRS. Manganaris also noted that secondary methodologies were in place to calculate a method to transition out the remaining unassigned personnel.
For his third example, Manganaris briefly summarized work performed for the intelligence community. It was an attempt to impose a common structure on 17 decentralized intelligence organizations covering six different cabinet-level agencies. Manganaris noted the difficulty of this work due to the large variation in approaches and commitment demonstrated by the component organizations, and he explained that, because of this variation, negotiation was an especially important part of successful implementation.
To illustrate helpful ways to conceptualize workforce data, Manganaris presented several charts depicting various types of data and described the information that can be gleaned from each type of representation: see Figure 3-1. A years-of-service profile, plotting the number of employees against the number of years they have been with the organization, can both reflect events in the past, such as periods of decreased hiring, and predict future periods of attrition due to retirement eligibility. The figure also shows how bar graphs can be used to illustrate
the percentages of filled and unfilled positions in each role or to show the difference between present and future personnel requirements as workplace functions evolve.
In response to Manganaris’s presentation, Cheryl Paullin (committee member) asked whether, during his 30 years of experience, he had noticed any trends in the way organizations perform workforce planning and staffing analysis. Manganaris answered that in the military and government sectors, he has seen less in-house capability in these areas in recent years, but that it is possible that these functions are being contracted out. He observed that people who are often hired to do workforce planning are pulled from other peripherally related positions and may not have a quantitative background. Manganaris also observed that although there are much more data available these days, “there is a tendency to think more about graphic displays and less about the nuts and bolts which go into that, or at least the modeling assumptions.”
Robert Anselmi (committee member) commented that funding from the U.S. Department of Veterans Affairs (VA) is based on patient workload, and when there is an increase in a particular workload (such as mental health), Congress allocates additional funding, but none of that funding is specifically for facilities. Anselmi asked whether Manganaris has seen similar funding misalignments in his work. Manganaris replied that some organizations, including the IRS and the Department of Defense, use unit cost rates and modular costing methods to help prevent such misalignments. He noted: “So, yes, that is done a lot in the federal government, where you look at the total cost of the person instead of just, oh, we got more money to hire people. But somebody has to really make that argument. If it’s in the budget proposal, that certainly can be taken care of. . . . I would hope the VA has some way of costing that when they do their budget request.”
The discussion turned to qualitative aspects of decision making that can influence workforce planning. Manganaris used the example of the IRS, in which a 30 percent increase in staffing suggested by the staffing model was deemed “politically unsellable” by IRS decision makers, so the model was not implemented. Kim O’Keefe (committee member) asked whether the risk of understaffing was specifically addressed in such situations, and Manganaris replied that such a risk could be addressed through identifying and focusing on staffing the mission-critical occupations for the particular organization in question.
Much of the discussion focused on the details of the modeling process. Colin Drury (committee cochair) asked Manganaris about the validation process for the modeling examples he had offered. Manganaris replied that the process of validation is different when projects are not repeated, and instead, feedback on the model comes more in terms of “lessons learned.” “[I]n the sciences, validation is very important and much more emphasized,” but in the work he does “a lot of these things are one-off transitions. One and done. We didn’t know if it was repeatable because we didn’t intend to repeat the restructuring of the IRS anytime soon.”
In terms of staffing modeling, William Marras (committee member) asked Manganaris’s opinion on the most useful portions of models, in terms of which direction the committee should take when advising the Veterans Health Administration (VHA). Manganaris stressed the importance of accurate staffing requirements, and that the VHA should ask: “Do we have good models to reliably say, based on this plan to recapitalize our infrastructure, what critical elements of the workforce we need? This is really important, coming in.” Anselmi asked Manganaris whether the VHA could use a linear programming model for building maintenance staffing in hospital settings, given the strict requirements for preventive maintenance and testing mandated by the accreditation process and the low level of acceptable risk. Manganaris responded that since a linear programming model simplifies systems, one could potentially be used for individual aspects of a hospital setting, but he advised that the entire system could not be modeled linearly because “there are too many different things going on in too many different ways.”
To conclude, Manganaris stressed that predicting future workforce demand is a difficult challenge and that requirements for future staffing should be based on sound planning. Components of the plan should include workforce measures and forecasts of the changing needs of the organization. To deal with the challenge of predicting future workforce demand, Manganaris suggested the strategy of subdividing and simplifying the problem, focusing first on “mission critical” occupations, such as those that involve complex work that must be performed by trained individuals and thus cannot be easily accomplished through reassignment of personnel. He noted that less critical, support positions can be determined more easily, based on a ratio of overhead rates. Manganaris emphasized that strategic staffing planning must be linked to the resource allocation process in order to be effective and that cost models should be developed for that purpose.
The session’s second presentation covered a congressionally mandated staffing study for the FAA by a committee of the National Academies of Science, Engineering, and Medicine, which was completed in 2013.1
Bill Strickland, who was a member of the committee that carried out the study, explained that, at the time of the study, the FAA was staffing its facilities using an allocation model (a model aimed at distributing the available resources effectively, irrespective of their collective adequacy), rather than a sufficiency model (a model designed to predict the resources needed to sustain system performances at an acceptable level). During the study time frame, the total system specialist workforce was capped at 6,100 funded positions, so the allocation model just used certain parameters to distribute those out to districts. At the time, the FAA had access to a much better sufficiency staffing model, but it was not being used because the FAA was using the allocation model.
Strickland described key elements of the committee’s statement of task, shown in Box 3-1. To clarify the second bullet point, Strickland reiterated that the committee’s job was not to create a model, but to evaluate the FAA’s current models and advise the agency on the elements that should be included in the model the contractor was developing. In describing the committee’s final report, Strickland highlighted several key concepts and the committee’s recommendations.
A key concept reflected in the committee’s recommendations was that workforce planning models are not static. Strickland explained that this was particularly relevant for the FAA at the time because the agency was in the process of transitioning to a new system of air traffic control (NextGen) in which, within the next decade, satellite-based GPS [global positioning system] technology would replace ground-based radar as the method for tracking airplanes. Any staffing model developed for the FAA had to consider the staffing implications of such a system change. Strickland also noted that even appropriate models need continuous reviewing and updating in order to remain applicable and to accommodate technological improvements in data modeling. Thus, one of the committee’s recommendations to the FAA was that the effectiveness of the developed staffing model should be continually monitored and adjusted to enhance accuracy.
Another key concept addressed in the committee’s recommendations was that the FAA should plan for the effects of model development and implementation on employees. Stressing the importance of accurate data, Strickland described the committee’s recommendation that the FAA should, through direct observation, systematically validate historical data and data obtained from estimates by subject-matter experts. However, he acknowledged, data collection is time intensive and if the data collection needed to develop a model or keep it current has a negative impact on employees, then the data might not be collected appropriately, if at all. For this reason, the committee recommended to the FAA that ongoing data collection should not place an unacceptable burden on data providers. To build employee trust in the model, Strickland said that the committee recommended that the outputs of the model should be understandable by FAA’s internal users at all levels.
The committee also recommended that the FAA’s staffing model should be robust enough to account for all the varied aspects of the systems specialist job, in addition to time spent directly on maintenance tasks. Those additional considerations are numerous: they include training and certification; time dedicated to military reserve service or other leave; travel time to and from remote worksites and the environmental challenges posed by some of those worksites; fatigue mitigation plans; deficiencies in data reporting; aging workforce and succession planning; and nontechnical task demands, such as paperwork.
Through an evaluation of the FAA’s past models against the suggested model criteria, the committee was able to assess the shortcomings of the FAA’s then-current staffing models and make recommendations for the development of a new model. Strickland noted that some of the considerations for the recommended model included equipment inventory, failure rates, time to perform each task, and any valid allowances or accommodations. Furthermore, he said, the recommendation requested that the model developed be based on the different specialties of systems specialists, rather than providing only an overall staffing level at each facility. The committee also recommended to the FAA that the model structure account for both deterministic and unpredictable, stochastic events.
1 Available at https://www.nap.edu/catalog/18357/assessment-of-staffing-needs-of-systems-specialists-in-aviation, accessed March 2019.
The committee also recommended that output reports from the model should be able to predict consequences in terms of deferred maintenance and overtime. In Strickland’s words: “If the model says you need more people and the budget people say we cannot afford more people, then the model should say, ‘here is what the implications are going to be, and you are going to end up paying overtime, and you are going to end up breaking systems that you are going to have to buy later. . . .’” In this way, he said, the proper use of the model data could provide justification for requesting funding to hire additional personnel.
Strickland concluded with another key point, based on the committee’s recommendation of a timeline for model development and implementation: development of a useful model is a complex undertaking, involving a wide array of people and systems. He illustrated the complexity of the process with a diagram demonstrating that model development is only the first step, shown in Figure 3-2, and stressed that the FAA was informed that the process of developing and implementing a model takes time. “This will not be fast. And the FAA should prepare everyone for the fact that it is not going to be fast.”
Following Strickland’s presentation, Drury concluded the session by listing the obvious similarities and one major difference between the VHA’s staffing situation and the FAA’s systems specialist staffing situation. In terms of similarities, Drury noted that the issues faced by both organizations are partly stochastic. Also, in both settings, training is specialized and is measured in weeks, months, or years. Drury listed complexity as another similarity between the two situations, pointing out that both organizations exhibit variability between different worksites. Another important similarity is the fact that both organizations are developing staffing models in the face of overarching changes—NextGen for the FAA and the transition from inpatient to outpatient care for the VHA.
In terms of differences, Drury observed that the FAA tracks some major outcome measures that could ultimately be used to validate staffing standards, such as the time the national air space is up and running or the response time to an airspace outage. In contrast, the VHA does not currently track patient outcomes in the same way. Drury stated his hope that the comparison between the two situations would be helpful for the work of the current committee.
The session’s third presentation dealt with the application of workforce modeling in the medical industry, as well as methods that can be used for such modeling and other important factors that should be considered in the modeling process. A case study was provided to further illustrate many of these points. For this session, the committee had asked the speaker to address the following questions:
Are there any reasonable standard models, or model structures, that have seen broad validation and acceptance by customers? How do you know what level of aggregation is “best” for a model: individual departments, groups of departments, or whole organization? What factors do you consider in the selection of the correct technique(s) among the range of forecasting possibilities? And, how does the solution integrate with other corporate programs, such as budgeting and planning, to ensure that the forecast is consistent with corporate resources and plans?
Joe Crance (consultant, medical facilities staffing) first addressed the committee’s question regarding the existence of medical labor standards, stating that medical workforce requirements have been modeled since at least the 1980s by the Management Engineering Program of the U.S. Air Force.2 In terms of the best level of aggregation for a model, Crance explained that staffing standards should first be developed at the lowest level, focusing on individual functions containing homogeneous tasks. Once this approach is complete, related functions can be grouped to begin to create labor standards for an entire system.
In response to another one of the committee’s questions, Crance explained his view of how the requirements process should fit into the budgeting process. First, he said, an organization should clearly establish the levels of service that will be provided. Those levels of service should be used to create a lean organizational structure that will be able to deliver those levels of service. At that point in the process, Crance explained, staffing requirements to meet the established levels of service can be determined, and then the budget request for those requirements can be submitted. He stressed that the requirements must be based on workload content, illustrating this point with a circular diagram showing that levels of service should determine requirements, which should then in turn determine the budget received: see Figure 3-3. He acknowledged, however, that quite often the budget does not meet (i.e., fund) 100 percent of the requirements. In those cases, corporate decisions should be made to reduce levels of service and decrease workload in order to mitigate risk. Furthermore, Crance warned, unfunded requirements should be defined and kept in view because they are a key aspect of risk.
Crance next addressed the committee’s question about the selection of appropriate methods for building workforce models, explaining that the selection of method should be determined by the nature of the work being performed. A number of methods for determining cycle time are available: see Figure 3-4. He explained that in a
2 The Air Force Management Engineering Agency was the precursor to the current Air Force Manpower Analysis Agency (AFMAA), which was established in 1975; one of its units, the Air Force Medical Engineering Team (AFMEDMET), was designed to create manpower standards for medical activities. Although AFMEDMET no longer exists, the AFMAA continues to work with the medical community to build staffing tools for military medical activities.
health care setting, patient-centric functions, such as primary care and internal medicine, can often be modeled by correlation-regression analysis or ratio unit times. If the function is highly transactional, such as a pharmacy or customer call center, discrete event simulation or queuing models can be used. If diligence is required, for example, for emergency departments or ambulance crews that must be on standby, the minimum staffing model can be used. Management, such as a department head or hospital administrator, can be modeled using a simple staffing pattern. Crance noted that the selection of a method should also take into account the ease of collecting the data necessary for that method for the selected model, along with the budget and time constraints for model building.
Crance stressed that, regardless of the modeling method chosen, data accuracy is critical. He pointed out that counts of a work unit, such as the number of patients seen, are reliable sources of data, but that little credence is put into data obtained from worker-hour accounting systems, primarily because workers tend not to report their idle time. Even at best, Crance continued, this type of historical data only provides information about the amount of time it took to perform a task, not the amount of time it should have taken. To gather accurate worker-hour measurement data, Crance suggested spending the time to collect data properly at the process level, using traditional work measurement techniques.
Crance explained that those techniques can be divided into two groups: (1) “engineered” work measurement methods, including time study and predetermined time study systems and work sampling, and (2) “non-engineered” work measurement methods, including the use of a timing device to measure cycle time, the use of historical records, and estimates of technicians’ time, which involve interviews with subject-matter experts. Crance emphasized that the engineered techniques provide the most accurate data and recommended that organizations attempting to develop staffing standards should try to use engineered techniques as much as possible.
To illustrate several of the points made in his answers to the committee’s questions, Crance described a study performed in 1998, in which Wilford Hall Medical Center (WHMC) agreed to partner with the Air Force Management Engineering Agency to provide “proof of concept” that discrete event simulation modeling could be used to model its new TRICARE primary care clinic. For this project, Crance’s manpower study team first process-mapped primary care, which involved defining the processes, the employees needed to perform those processes, the inputs and outputs, and the business regulations governing those processes. Crance warned that process mapping is an involved technique and is not the same as simple flowcharting.
In terms of data collection, Crance’s team time-studied routine processes and used technicians’ estimates for those processes that rarely occurred. They also gathered a facility layout diagram to determine whether the layout and age of the facilities affected the staffing requirements. In analyzing the data, Crance explained, they determined the process mean, data dispersion, and shape of the distribution. Simulation models were built using the data to determine what the requirements should be. He noted that activity-based costing models were also part of the deliverable. He also noted that the time to complete the first pilot study at a single location was 125 days, which he described as “about average, probably, for a lot of single-point studies out there.”
Crance summarized the results of implementation of the study’s findings for the primary care clinic: (1) a savings of $100,000 annually through reduction in unnecessary nursing staff; (2) a reduction in congressional inquiries from patient complaints, from one per week to less than one per quarter; and (3) increased patient throughput by 30 percent, by using doctors more efficiently and decreasing their idle time. WHMC leadership was so impressed with the study that they asked Crance and his team to expand the study to the obstetrics-gynecology clinic, internal medicine, and the pharmacy. He noted that the obstetrics-gynecology and internal medicine studies took approximately 75 and 42 days to complete, respectively, and he did not know about the length of the pharmacy study.
Paullin asked Crance how modeling methods can be used to deal with similar functions across many different facilities. Crance explained that if a discrete event simulation model is used, data obtained from that model can be used to create a correlation-regression model that is run multiple times against different patient workloads to generate the necessary data points, allowing the model to be applied across the rest of the organization. He noted that individual locations will have variances that need to be considered, and he acknowledged that assessing these variances and making the appropriate adjustments to the model take time. Crance suggested starting with a simple, functional model, with the intent to get a more accurate model over time by adjusting it on the basis of workload differences at individual locations.
Steven Broskey (VHA) asked how work order backlogs in VHA engineering departments can be addressed using staffing models. Crance pointed out that for maintenance-based functions, it is desirable to have a certain amount of backlog because the nature of the work is such that it cannot always be performed on a set schedule, perhaps due to competing priorities for a given piece of equipment. He suggested that models be used to assess staffing needs only if the severity of the backlog becomes such that equipment breaks down because maintenance has been deferred. Broskey explained that sometimes backlogs occur when there is employee turnover, and Crance cautioned that he would not inflate the staffing requirement to account for turnover, but instead suggested that a contingency plan be developed, such as having an additional electrician from the private sector on retainer to fill in until a permanent replacement is found.
Crance summarized his presentation with four points: (1) Creating and implementing labor standards in a medical setting has been done before. (2) The desired levels of service should be established first and should then inform staffing requirements, which then are used for budget submissions. (3) There are many methods to forecast workload requirements; they all require accurate measurement of workload content. (4) Workload requirements modeling is an involved process, but it can reap huge benefits in mission accomplishment for the right cost.
This page intentionally left blank.