PI Aging Simulation Model

1
PI Aging Simulation Model
2
The Current ProblemSuccess to the Successful
More Funds to Older, Experienced PIs
Higher Success of Older, Experienced PIs
Allocation to Older, Experienced PIs Instead of
Younger, Inexperienced PIs
Lower Success of Younger, Inexperienced PIs
Less Funds to Younger, Inexperienced PIs
3
Basic Structure for Age Group
New PIs (i.e., first-time) that enter the NIH
pool in this age group.
Represents the number of PIs in the total pool
that are in this age group.
PIs in the system that have aged enough to move
to the next age group.
PIs in the system that have aged enough to move
into this age group.
PIs of this age group that leave the system.
4
Connecting Age Groups
5
Differences Between Models

• OB
• Spreadsheet methodology
• Statistical
• Focuses on data
• Static
• No feedback loops

• OER
• System Dynamics (SD) methodology
• Operational simulation
• Focuses on activities
• Dynamic
• Feedback loops

6
Limitations of Simulation Model

• Data begins in FY80, so momentum inherent in
  system prior to FY80 is not captured.
• Data available for approximately 65 of the R01
  equivalents only
• Age data invalid for roughly one-third of R01
  data set.
• Length of Service averages based on total years
  of service rather than continuous years of
  service.
• Currently, there are no feedback mechanisms
  incorporated into the model
• All trends are based on data and do not change
  dynamically or in relation with other variables.

7
Simulations

• Baseline (FY80-FY06)
• FY80-FY06 entrance rate data.
• FY80-FY86 duration averages, FY87-FY06 uses FY86
  duration averages.
• Scenario 1 (FY80-FY16)
• Same as Baseline except FY07-FY16 entrance rates
  use trends based on FY97-FY06 entrance rates.
• Scenario 3 (FY80-FY16)
• Same as Scenario 2 except FY07-FY16 entrance
  rates specified to try to keep the PI age
  distribution consistent with FY06.

8
Average Length of Service
9
Baseline Results
10
Total Number of PIs (FY80-FY06)
11
Baseline 1991
Actual
Simulation
Avg Age 45.6
Avg Age 42.7
12
Baseline 1996
Actual
Simulation
Avg Age 47.3
Avg Age 44.7
13
Baseline 2001
Actual
Simulation
Avg Age 49.0
Avg Age 46.3
14
Baseline 2006
Actual
Simulation
Avg Age 50.8
Avg Age 47.5
15
Scenario 1 Results
16
Total Number of PIs (FY80-FY16)
17
Scenario 1 1991
Avg Age 42.7
18
Scenario 1 1996
Avg Age 44.7
19
Scenario 1 2001
Avg Age 46.3
20
Scenario 1 2006
Avg Age 47.5
21
Scenario 1 2011
Avg Age 48.3
22
Scenario 1 2016
Avg Age 49.8
23
Scenario 2 Results
24
Scenario 2 Approach

• Objective is to keep average age and approximate
  age distribution consistent with 2006 values
• Average age 47.5
• Possible policy changes to test
• No new PIs older than 65 minimal impact
• Forced retirement at 70 minimal impact
• Forced distribution of 1500 new PIs
• No new PIs at all
• All new PIs
• All new PIs forced to fit a specific age
  distribution

25
Scenario 2, No New PIs 2006
Avg Age 47.5
26
Scenario 2, No New PIs 2011
Avg Age 50.5
27
Scenario 2, No New PIs 2016
Avg Age 54.3
28
Scenario 2, All New PIs Avg Age 47.5
29
Scenario 2, All New PIs Avg Age 44.0
30
Scenario 2, All New PIs Avg Age 41.3
31
What Does This Tell Us?

• We have a model that is capable of forecasting
  the age distributions of the PI pool given
  assumptions on influxes and tenures.
• Making dramatic changes can have dramatic impacts.

32
Scenario 2 New PI Distribution 1

• Constant rate of 1500 New PIs
• Age 25-35 25
• Age 36-40 20
• Age 41-45 20
• Age 46-50 15
• Age 51-55 10
• Age 56-60 10
• Age 61-80 0

33
Scenario 2, New PI Distribution 1 2006
Avg Age 47.5
34
Scenario 2, New PI Distribution 1 2011
Avg Age 47.6
35
Scenario 2, New PI Distribution 1 2016
Avg Age 48.2
36
What Does This Tell Us?

• The ideal age distribution for the PI pool is
  still an unknown target.
• With changes that occur due to feedback loops in
  the system, the established age distribution
  policy for new PIs for future years will likely
  change every few years.
• In other words, there is no constant age
  distribution policy for incoming new PIs that
  will provide the ideal PI pool age distribution
  over the long run.

37
Additional Test Scenarios for Final Workforce
Group Meeting

• November 14, 2007

38
Test Scenario Effect of the Number of New PIs on
the Average Age of the Total Pool
Age Distribution
24-35 25 36-40 20 41-45 20 46-50
15 51-55 10 56-60 10 61-90 0
39
Test Scenario Effect of the Number of New PIs on
the Average Age of the Total Pool
Age Distribution
24-35 25 36-40 20 41-45 20 46-50
15 51-55 10 56-60 10 61-90 0
40
Test Scenario Small Changes in the Age
Distribution of the New PI pool
Distribution 1
24-35 25 36-40 20 41-45 20 46-50
15 51-55 10 56-60 10 61-90 0
Distribution 2
24-35 25 36-40 40 41-45 15 46-50
10 51-55 5 56-60 5 61-90 0
Distribution 3
24-35 25 36-40 60 41-45 10 46-50 5 51-55
0 56-60 0 61-90 0
41
Test Scenario Small Changes in the Age
Distribution of the New PI pool
Distribution 1
Distribution 2
Distribution 3
1100 New PIs
1500 New PIs
42
Test Scenario Extreme Case Replacing the PI
Pool
43
Conclusions

• The model in its current state matches historical
  data qualitatively, but could use some
  improvement with quantitative accuracy.
• The current backbone aging model needs to be
  enhanced to increase the quantitative weaknesses.
• The simulation could be improved with the
  addition of recycling of PIs as well as
  feedback loops regarding how individuals and
  institutions act/react to changes in NIH
  policies.
• With improvements, the simulation model could be
  very useful in understanding the short-term and
  long-term consequences of NIH policies.
• The ideal age distribution for the PI pool is
  still undetermined.

44
Next Steps

• Based on feedback from the final workforce group
  meeting, develop a list of specific model
  enhancements to be incorporated in a follow-on
  effort.
• On this next effort, focus on increasing the
  quantitative accuracy of the model compared to
  historical data.
• Report back to workforce modeling group on
  results from enhanced model.