CIVFORS 101

1
CIVFORS 101

• Part 1 How CIVFORS Generates Forecasts

2
Basic Terms

• Goal vs. No Goal forecasts
• SAMAS (Structure and Manpower Allocation System)
• Steady state - special type of goal forecast
• History period, Projection period

3
CIVFORS No Goal Forecasts

• Build historical data
• Typically five years of history
• Focus on strength and personnel actions
• Compute rates at detailed levels
• Rates based on historical counts
• Rates computed for groups of people in workforce
• Five different methods to calculate rates
• Rates are projected into the future
• Generate forecast based on projected rates
• Start with strength at end of history
• Each quarter, age the workforce and apply rates
  to strength
• Strength is updated at each step

4
Cell Dimensions (example)
5
Strength and Personnel Actions
Cell 1426 Qtr 1 Str 46
Accessions 4 Resignations 3 Retirements
1 Promo In 2 Promo Out 4 NAG 3
Cell 1426 Qtr 2 Str 47
6
Rate Calculations
Accessions 4/46.087 (8.7) Resignations 3/46.065
(6.5) Retirements 1/46.021 (2.1) Promo
In 2/46.043 (4.3) Promo Out 4/46.087
(8.7) NAG 3/46.065 (6.5)

• Rates are calculated for each quarter and for
  each type of personnel action
• Weighting is applied (more recent time periods
  usually have more weight)
• Final rates for different personnel actions are
  applied to last quarter of history to calculate
  first quarter of the projection period

7
Projection Period - 1st Qtr
Cell 1426 Qtr 20 (last qtr of history) Str 53
Accessions 11.6 6.1 (6) Resignations 9.5 5.0
(-5) Retirements 3.1 1.6 (-2) Promo In 4.9 2.6
(3) Promo Out 7.2 3.8 (-4) NAG 6.1 3.2 (3)
Cell 1426 Qtr 21 (first projection qtr) Str 54

• Final rates are applied to the end strength of
  the last quarter of history to calculate the
  strength of the first quarter of the projection
  period...

8
Projection Period
Cell 1426 Qtr 21 Str 54
Cell 1426 Qtr 24 Str 57
Cell 1426 Qtr 23 Str 56
Cell 1426 Qtr 22 Str 55
Rates applied
Rates applied
Rates applied

• Final rates are applied to each successive cell
  to obtain the end strength for each quarter of
  the projection period.
• In a no-goal forecast, no constraints or external
  influences are applied. The end strength is
  strictly based on the application of the rates.

9
CIVFORS Goal Forecasts

• Almost same process as no goal
• Determine historical counts
• Calculate personnel action rates
• Project rates into the future
• Apply rates quarter by quarter
• Strength targets influence the forecast
• Manpower targets/goals used
• Optimizer determines hiring plan needed to meet
  targets
• Example 1000 nurses
• Other personnel actions can also be designated
  for optimization to meet targets instead of
  accessions such as promotions
• Forecast is based on optimal plan
• Start with strength at end of history
• Each quarter, age the workforce and apply rates
  to strength
• Historically-based rates can be over-ridden for
  any transaction

10
Projection Period - Goal Forecast
Cell 1426 Qtr 21 Str 54
Cell 1426 Qtr 24 Str 59
Cell 1426 Qtr 23 Str 57
Cell 1426 Qtr 22 Str 56
Rates applied
Rates applied
Rates applied
Accessions are added or subtracted above the
going rate

• In a goal forecast, rates can be over-ridden.
• Usually the primary optimization variable used is
  the number of accessions (reduced or increased in
  order to meet lower or higher manpower targets).
  Reason is that accessions are about the only
  controllable transaction type compared to others
  like resignations, retirements, reassignments
  etc

11
Life Cycle Modeling,Time Series Format

• Aging the workforce - age and years of service
  affected
• Gains
• Migrations - changes within the population
• Losses

12
Powerful Use of CIVFORS

• Forecasts many different levels simultaneously
• Allows optimization for goal runs
• Balances aggregate with detailed levels
• For Default models/forecasts selected the best
  predictors are automatically used to generating
  rates for DEFAULT published forecasts ONLY
• Forecasts different types of personnel actions
  depending on what the user needs to know
• Allows different data elements to be in models
  without reprogramming software
• Has been validated over time
• Aggregate 0.5-1 of actuals one year out
• Aggregate 3-5 of actuals 3 years out
• Detailed 1-3 of actuals one year out
  (typically)

13
CIVFORS 101

• Part 2 Basics of Building Populations, Models,
  and Forecasts

14
CIVFORS Terms Population, Model, Forecast

• Population
• What workforce segment do we wish to forecast?
• Model
• What levels of detail (data elements) do we need
  in the forecast?
• Which data elements should be predictive?
• What personnel actions (file types) do we wish to
  model?
• How many quarters of history and projection do we
  need?
• Forecast
• Is this a non-targeted or targeted forecast?
• When should the projection begin?
• How should personnel action rates be forecasted
  (rate options)?
• What strength targets should be used (if targeted
  forecast)?
• Should we make this forecast viewable to everyone
  (public)?

15
CIVFORS Terms Population
16
CIVFORS Terms Model
POPULATION D SSN_STATUS A OCC_SER 1515
17
CIVFORS Terms Forecast
POPULATION D SSN_STATUS A OCC_SER 1515
Model ORSA_NEW
Baseline No Goal Forecast
Baseline Goal Forecast
What-if Forecasts

• Same rate options as
• Baseline No Goal
• Targeted
• Choose targeted data
• elements
• Forecast based on
• meeting strength
• targets

• One or more of these
• Change rate options
• Override rates/targets
• using Rates Editor
• Change targeted data
• element selections
• Introduce constraints

• Default rate options
• Non-targeted
• Forecast based on
• history
• Runs faster than goal
• forecasts

18
Step 1 Edit a Current Population

• The easiest way to create new populations,
  models, and forecasts is to edit an existing
  population, model, and forecast, and save them as
  new ones.
• First, determine who you wish to forecast
  (population). Your population may be an
  occupation, installation, or command.
• 1. Click on Build Models - Edit -
  Population.
• 2. Select the population to be copied.
• 3. Change the name and description for your
  population.
• 4. Accept defaults on the Population Pre-Filter
  (USDH, Active)
• 5. Change the data element values to define your
  population.
• 6. Save the population

19
Rules of Thumb Population Building

• Dont change the population pre-filter defaults
• Include SSN_STATUSA (Active)
• Include POPULATIOND (Direct Hire)
• Population filter rules
• Folders are OR'd together.
• Values from the same data element within a folder
  are OR'd together.
• Values from different data elements within a
  folder are AND'd together (use CTRL key when
  selecting).
• The prefilter is AND'd to every folder.

20
Step 2 Edit a Current Model

• Choose a model built for a population that is the
  same type of population as yours. For example,
  if you are building a forecast for an occupation,
  select a model already built for an occupation.
• 1. Click on Build Models - Edit - Model.
• 2. Choose the model to be copied.
• 3. Change the name and description for your
  model.
• 4. Change the population selected to your
  population.
• 5. Click Next on the next three screens.
• 6. Click Save to save your new model.

21
Predictive and Proportionally Distributed Data
Elements

• Predictive data elements are used to apply
  historic rates to future projected strength.
• Proportionally Distributed data elements are
  applied proportionately in the forecast
• If 42 of the population is female, 42 of the
  forecasted information will be female.

22
File Types

• You must choose file types that account for all
  available gain and loss NOA transaction codes
  exactly once (and only once).
• Basic list
• ACCESSIONS NAG (non-acc gains)
• ACTGAIN NSL (non-sep losses)
• ACTLOSS RETIRES
• INVOLSEPS VOLSEPS

(Correct INVOL SEPS as of 8/2/2002)
23
File Type DefinitionsStrength, Targets, Gains

• Strength - The number of civilian personnel in
  the specified population at a given point in time
  
• Targets - The number of civilian personnel
  authorized (planned for) to be in a specific
  population at a given point in time
• Accessions (New-Hires) - New employees to the
  population from external sources as indicated by
  a personnel action record (NOA)
• NAG (Non-Accession Gains) - New employees to the
  population for which no personnel action record
  (NOA) was located (or not elsewhere counted)

24
File Type DefinitionsLosses

• Involuntary Separations - Losses to the
  population (employees leaving) where a personnel
  action record indicates that an agency-initiated
  action was the cause (specific series of NOAs
  that are applicable) a NOA record was located
  with a specific reason that meets the definition
• NSL (Non-Separation Losses) - Losses to the
  population (employees leaving) where NO personnel
  action record was located (or not elsewhere
  counted)
• Retirements (NOA 300-304) - Losses to the
  population (employees leaving) where a personnel
  action record was located that contained an NOA
  of 300-304
• Voluntary Separations - Losses to the population
  (employees leaving) where a personnel action
  record indicates that an employee-initiated
  action was the cause (specific series of NOAs
  that are applicable) a NOA record was located
  with a specific reason that meets the definition

25
File Type DefinitionsMigrations

• Change in a data element in an employees record
  from one period to another (fiscal year quarters)
• Employee is still a member of the population of
  interest but one or more of a specific set of
  data elements has changed
• Ex MACOM, Command, UIC, or CCPO_ID changes
  usually indicate that the employee has physically
  moved from one organization or geographic locale
  to another within the population of interest
• Changes in occupational series, career program,
  pay plan, pay grade, retirement eligibility, or
  year of service group may not be accompanied by a
  physical move but only a change in that
  particular data element
• Data elements tracked for the US Direct Hire
  Military Function forecasts population are
• CP_REASGN (Reassignment due to career program
  change)
• MC_REASGN (MACOM Transfers)
• PROMOTIONS (PAY_GRD Changes)
• YOS_REASGN (Reassignment due to change in YOS
  group)

26
Rules of Thumb Model Building

• Most models need 3-5 predictive data elements.
  Try to avoid predictive data elements with many
  different values.
• At least one of the predictive data elements
  should reflect aging (e.g., a grouping of YOS,
  AGE, or retirement eligibility).
• The other predictive data elements should be
  based on data mining analysis.
• Limit the number of proportionally distributed
  data elements to no more than 5 if possible.
• For each predictive data element, include its
  corresponding migration file type in the model.

27
Step 3 Build the Forecast

• Now that the model is built, we will build a
  non-targeted forecast using default settings.
• 1. From the same page where you saved the model,
  click on Run.
• 2. Give a name to your forecast such as
  OCSR1515_2003_NOGOAL. Also give a description
  so that you will know that this forecast was
  based on default settings.
• 3. Click Next twice. This will use the default
  rate options.
• 4. Change the projection start date to the start
  of the most recent fiscal year.
• 5. Click Next.
• 6. Click Save to save your forecast. To run
  the forecast at this point, we would click Run
  after clicking Save.

28
No Goal, Goal, Steady State

• No Goal Forecast reflects history repeating
  itself (no external influences)
• Goal Uses SAMAS manpower authorizations to
  influence forecast
• Steady State Uses strength on last day of
  history as target for forecasted quarters

29
Rate Options

• Small Cell, Medium Cell Tolerance threshold for
  application of small and medium cell size rules
• Outlier Threshold defines the threshold used
  when evaluating outliers in the rate processor.
• Extrapolation Method see next slide
• Yearly Weights can be used to suppress
  excessively turbulent periods of history, or to
  emphasize periods thought to be more
  representative of the future.

30
Extrapolation Methods

• Beginners, stick with the default (hybrid)
• Other options include
• Repeat last year
• Seasonal weighted average
• Weighted average
• Winters
• For more info, see the Help screens

31
Rate Features Help
32
Rate Features Help, cont
33
Target Options

• Targeted Data Elements define the level-of-detail
  for the targets. For example, if Gender and Grade
  are selected, targets (for each time period) will
  be generated for each combination of gender and
  grade.
• Aggregate Constraints (optional) apply to
  predictive data elements. Each data element
  selected will establish additional constraints
  rolled up to the level of detail represented by
  that data element. For example, if Gender and
  Grade are selected, separate aggregate targets by
  gender and by grade (for each time period) will
  be established.
• Total Constraints (optional) establishes an
  aggregate target across all data elements (for
  each time period) will be established.
• If Aggregate or Total Constraints are
  established, you will also need to establish an
  allowable slack percentage in the activated Edit
  boxes to the right of the data elements. Default
  values of 2 are provided, but you may choose any
  numeric value between 0 and 50.

34
Optimization Options

• The goal of the optimization is to have strength
  and targets as close together as possible. The
  first job in optimization is to examine projected
  strengths against targets. If there is a delta
  between projected strength and targets, Optimized
  File Types (Accessions, Promotions, etc.) are
  used to fill the gap. Since Accessions (unlike
  Losses) are routinely "managed", this file type
  is often selected as the Optimized File Type.
  Select the Optimized File Types by placing a
  check mark next to the appropriate file types.
• A Constraint is an equation that places limits on
  variable sets in the linear program formulation.
  Constraints are established in the Constraint
  Builder. Place a check mark next to each
  constraint that you wish to implement. If no
  constraints have been built for this model, then
  no constraints will be displayed.

35
Rules of Thumb Forecast Building

• Stick with default options until you become
  comfortable with the system.
• Non-targeted runs produce forecasts strictly
  based on history, while targeted runs produce
  forecasts based on meeting future strength
  targets.

36
Building Forecasts in CIVFORS Performance

• The amount of time required for the system to
  produce the forecast is dependent on
• How many other runs are running in the system.
• The size of the population.
• Whether the run is targeted or non-targeted.
• The number of predictive data elements in the
  model.
• The number of total data elements in the model.