powered by NetLogo

WHAT IS IT?

This model shows how different modes of agency coordination--consensus mode vs. parallel mode--affect emergency response.

Consensus mode requires that ALL agencies agree. Each agency can issue an alert, but agency consensus is required before the system can successfully respond to an emergency, e.g. close schools.

In parallel mode, if ANY agency alerts an emergency, then the system can successfully respond to an emergency. Here agencies operate in parallel, without any need to reach consensus. This is a parallel, redundant system. Each agency is effectively a "back up" to the others, like each engine in a twin-engine airplane.

The model shows that these two modes have opposite effects on system success in emergency response. We think of system success first in terms of the true positive rates of the whole system (responding to an emergency when there really is an emergency) as a function of true positive rates (also called "sensitivity") of the agencies. No agency will have a perfect true positive rate. 100% sensitivity is extremely unlikely. Put another way, each agency is likely to make some "errors" of ommission, but having a backup agency reduces the system level errors of ommission.

To increase system success, emergency response systems ought to contain redundant, parallel systems so that failure of any one agency is effectively “backed-up” by another. Consistent with that hypothesis, this simulation model suggests that emergency response is:

* More likely to succeed when agencies can respond in parallel, without having to reach consensus. Errors of omission are lower.

* Less likely to succeed when agencies are forced to reach consensus before action can be taken. That is errors of omission are higher under consensus mode.

For example, if 2 agencies have to reach consensus then agency 1 and 2 both have to agree, or the local system will fail (e.g. not close schools). It is then less likely schools will close, error of ommission is higher with 2 agencies compared to 1.

Essentially:

IF agency 1 AND agency 2 succeeds in alerting an emergency THEN system responds successfully.

By contrast, if the 2 agencies operate in parallel so that if either one alerts an emergency then the system succeeds (and don't have to reach consensus), with 2 or more agencies, this redundancy reduces errors of ommission. Essentially we program that as If ANY agency

IF agency 1 OR agency 2 succeeds in alerting an emergency THEN system responds successfully.

Thus these two different modes of coordination have opposite effects on errors of ommission. The divergence is larger with more agencies (3 agencies vs. 2, higher agency redudancy).

Of course reducing agency error is always good, but having agency redudancy and parallelism can reduce errors of ommission.

So far we have focused on errors of ommission. We see the opposite effect for errors of commission, e.g. closing schools when there is not really going to be an outbreak. Errors of commission are lower under consensus mode because schools will not close unless all three agencies agree.

There is a unavoidable trafeoff between errors of ommission and errors of commission.
This model is a "closeup" of one aspect of my outbreak response model, which includes school closure by three agencies, also includes vaccination, education, spread of beliefs. There is so much in that model that it is difficult to see how agencies coordinate.


HOW TO USE IT

Try it with the default settings first. With two agencies, the system failure rate is a function of whether or not the two agencies have to reach consensus, or whether they operate in parallel.

Hit Setup. Hit Go.

Later try experimenting with the number of agencies involved. If it just one then the total system error rate is the same as for the single agency.

Try adjusting the sensitivity rate. If sensitivity is a perfect 100% (the error of ommission rate is zero) then it doesn't matter how the agencies interact or how many interact, and whether or not they must reach consensus.

Here the total denominator is seasons, or we could think of this as seasons per local system.


OTHER THINGS TO NOTICE

The total system error rate is erratic at first, but soon the average rate settles down to a predictable probability.

We have focused above on reducing errors of ommission. We reduce that type of error by letting each agency back each other up, essentially increasing the overall sensitivity rate of the system.

However, another type of error is error of commission, e.g. closing schools when you don't need to (or even vaccinating unecessarily). Error of commission at the agency level is alerting an emergency when there really is no emergency. This type of error is lower when "specificity" is high. Specificity is the probability of not alerting an emergency when there really is no emergency. You can adjust the specificity rate.

S ystems confront a paradox: Agencies operating in parallel, like the twin engines in an airplane, reduce the chance for system failure, but the redundancy entails the expense of infrastructure and the costs of coordination.


CREDITS AND REFERENCES

Chris Keane developed this model as an extension of his outbreak response model, with embedded spread of virus.

The nelogo program and model examples make the programming much easier. Netlogo Copyright 2004 Uri Wilensky. All rights reserved.
See http://ccl.northwestern.edu/netlogo/models/ProbGraphsBasic for terms of use.