Population Forecasting: What Method is Best?

1
Population Forecasting What Method is Best?

• Analysts that undertake population forecasting
  have a wide variety of method available to them,
  all with a mix of strengths and
  weaknesses. -Simple extrapolation -Complex
  Ratio -Complex extrapolation -Cohort
  Survival -Simple Ratio -Cohort Component
• The question then arisesHow should analysts go
  about choosing a method (or set of methods) to
  use to make population projections and a
  forecast? The answer?

• IT DEPENDS
• Pittenger and Smith et al make it clear that
  choice of projection method depends upon a number
  of factors. What factors do these authors
  identify as important when making the choice of
  forecasting method?

2
Factors Influencing the Choice of Forecasting
Method
Needs of the Users --Geographic
Detail --Demographic Detail --Temporal
Detail Are User Needs Satisfied?
Plausibility Do the Outputs Make Sense?
Face Validity --Availability of Data --Quality of
Data Are the Inputs Good?
Model Complexity --Ease of Application --Ease of
Explanation Can we do this? Can we
explain what we did?
Political Acceptability Are the Outputs
Acceptable?
Resources--Money--Personnel--TimeCan we
afford it?
Forecast Accuracy Is the Forecast Accurate?
3
Choosing a Forecasting MethodA Multi-Criteria
Decision Making Process

• What this decision boils down to is what is
  commonly called multi-criteria decision making
  (MCDM).
• MCDM attempts to answer the question What is the
  best method/answer for a given problem? given
  that there are numerous, conflicting criteria for
  making a choice.
• For example, What home should I buy? is a
  common problem faced by many individuals and
  families.
• This is a question that requires MCDM, although
  most dont recognize that they are using this
  method in making this decision.
• What factors go into the choice of a new home?
• --Location --Cost --Neighborhood --Size
  (SqFt) --Size (BRs) --Schools
  --Amenities --Commute Time --Etc.

4
The Simplified MCDM Method

• There are a number of MCDM methods for making
  decisions (some of which you will learn about in
  Policy Analysis (Methods IV)), but the overall
  approach to making a decision is based in a
  relatively simple procedure
• 1) Identify the criteria of interest
• 2) Rank the criteria in terms of their
  importance
• 3) Weight the criteria
• 4) Use these rankings/weightings to score the
  alternatives
• 5) Make a choice based upon this analysis
• What all of this points to is our original answer
  it depends
• 1) It depends upon what criteria are important
• 2) It depends upon the rankings and weightings
  of these criteria
• 12 3) It depends upon what mix of factors
  is deemed the most valuable to the analysts
  and users

5
Forecasting Accuracy

• In theory, the most important criteria for a
  forecast is its level of accuracy. We assume
  that a forecast that is off by only 2 is much
  better than one that is off by 20.
• However, the likelihood of forecast accuracy
  serving as the most important criteria in
  choosing a method depends on local conditions.
• For example, sometimes politics plays a very
  important role in the choice of a method and, by
  extension, the result generated by a forecast
  (Riverkeepers vs St. Joe in Franklin County).
• However, lets assume for a moment that forecast
  accuracy is indeed the most important criteria
  for choosing a method. What does the empirical
  evidence have to say about the different methods?

6
Forecasting Methods Evidence to Date

• Chapter 13 of the Smith et al book provides an
  excellent, detailed summary of the current state
  of knowledge concerning the various forecasting
  methods
• --Simple extrapolation (simple
  ratio) --Complex extrapolation (complex
  ratio) --Cohort component methods --Structural
  models (to be discussed)
• What we think we know
• Structural Models gt CC gt Complex Extrap gt Simple
  Extrap
• (OR More Complexity leads to Greater Accuracy)
• What are the major conclusions suggested by Smith
  et als review of the evidence? Does research
  suggest that this belief is correct?

7
Forecasting Methods Evidence to Date

• The evidence to date suggests that
• 1) More complex methods do not produce more
  accurate forecasts of total population.to date,
  neither the sophistication of structural models
  nor the complexity of cohort component models has
  led to greater accuracy for projections of total
  population than can be achieved by simple
  extrapolation techniques. (p. 312)
• 2) No single method is consistently more
  accurate than the other methods
• Why is this? Why doesnt complexity result in
  accuracy?
• --Uncertainty, uncertainty, uncertainty
  --The CC method still requires extrapolation
  extrapolation of birth, death, and migration
  rates --Structural models still require
  extrapolation from recent or historical data

8
Forecasting Methods Evidence to Date

• Research into forecast accuracy has yielded other
  conclusions
• Forecast accuracy generally increases with
  population size.
• Forecast accuracy generally increases for areas
  with slow, but steady positive growth rates. It
  decreases for areas with rapid population
  increases or population losses.
• Forecast accuracy generally declines as the
  projection horizon (distance from the launch
  year) increases.
• The rule of thumb on base period is generally
  found to true The length of the base period
  should generally correspond to the length of the
  projection horizon. --Short projection horizon
  (1-5 years), short base period--Long projection
  horizon (20 years), longer base period But
  evidence suggests that too much input data
  (greater than 10 years) may increase error

9
Responding to the Evidence

• Given these general conclusions, what can
  analysts do to improve their projections and
  ultimately their forecasts?
• 1) Combine forecasts Complete a number of
  projections and try to incorporate many of these
  in the final forecast. --It is assumed that
  every projection has error, but by completing and
  comparing different projections you can
  cancel out the errors across these
  projections and arrive at a more accurate
  forecast.--Methods for combining 1) Average
  different projection results 2) Weight and
  average different projection results 3)
  Composite method Find methods that work under
  certain circumstances and rely on these
• 2) Account for Uncertainty Use methods to
  incorporate the concept of uncertainty in our
  forecasts. 1) Complete a Range of Projections
  using different assumptions (High, Med, Low
  Series) (Easy) 2) Use Prediction Intervals (aka
  Confidence Intervals) to generate different
  forecasts (Difficult)