This preview shows page 1. Sign up to view the full content.
Unformatted text preview: ases A and B), this
kind of model is commonly used to provide a cost
estimate for a new NASA system. The project manager
needs to understand what confidence he/she can have
in that estimate.
One set of uncertainties concerns whether the
input variables (for example, weight) are the proper
explanatory variables for cost, and whether a linear or
loglinear
form
is
more
appropriate.
Model
misspecification is by no means rare, even for strictly
engineering relationships.
Another set of model uncertainties that
contribute to the uncertainty in the cost estimate
concerns the model coefficients that have been
estimated from historical data. Even in a wellbehaved
statistical regression equation, the estimated coefficients
could have resulted from chance alone, and therefore
cost predictions made with the model have to be stated
in probabilistic terms. (Fortunately, the upper and lower
bounds on cost for any desired level of confidence can
be easily calculated. Presenting this information along
with the cost estimate is strongly recommended.)
The above uncertainties are present even if the
cost model inputs that describe a new system are
precisely known in Phase A. This is rarely true; more
often, model inputs are subject to considerable
guesswork early in the project life cycle. The uncertainty
in a model input can be expressed by attributing a
probability distribution to it. This applies whether the
input is a physical measure such as weight, or a
subjective measure such as a "complexity factor." Model
input uncertainty can extend even to a grassroots cost
model that might be used in Phases C and D. In that
case, the source of uncertainty is the failure to identify
and capture the "unknownunknowns." The model inputs  the costs estimated by each performing organization
 can then be thought of as variables having various
probability distributions.
5.4.2 Modeling Techniques for Handling
Uncertainty The effect of model uncertainties is to induce
uncertainty in the model's output. Quantifying these
uncertainties involves producing an over...
View
Full
Document
This document was uploaded on 02/26/2014 for the course E 515 at University of Louisiana at Lafayette.
 Spring '13
 Mr.Kau
 Systems Engineering, The American

Click to edit the document details