Question. VARS is a sensitivity and uncertainty analysis framework. Do two separate analyses need be run to obtain uncertainty and sensitivity? Often the range of an input variable, X, will often exceed the uncertainty in that variable. Would it be correct to say that for sensitivity we would want to provide VARS the full range of the input, but for uncertainty we would want to provide a smaller range? But with a smaller uncertainty range you wouldn't capture uncertainty conditional on X's full range of possible point estimates. How would you handle that issue?
Monte-Carlo simulation is a forward approach for propagating uncertainty from inputs to the output. SA may be viewed as a backward approach that tries to partition the uncertainty in the output and attribute its pieces to different inputs.
Let me explain it in a bit more simple way (with all due respect to Saman's explanations), in SA we keep all variable constant and change just one variable to see how that single variable could affect the outcomes, and we should repeat it for all input parameters and a weighting factor should be defined to quantify the effects of each variable on output value i.e. how much they are influential (I used this method in geotechnical engineering field, which is my field of expertise), but in Mote-Carlo we should establish a "Probabilistic Space, by having in hand PDF of all of our influential parameters" and sampling from each PDF is completely random based on skewness of the PDF, then the results will be reflected in PDF of our output parameter (i.e. a wide range of outputs will be gained based on probabilistic sampling and the relationship between those input parameters and the output value. Note: PDF stands for Probability Density Function (i.e. the bell shape curve, which could be skewed either positively or negatively). Hope it was useful. (Please correct me, Saman, if I am wrong). Regards, Mike
![ILLSLICK - KILLSHOT REMIX [FULL VERSION]](http://i.ytimg.com/vi/RIK8RrqIAzE/mqdefault.jpg)
Thanks Saman jan for the explanations, nice and visualized approach to present complex concepts.
Thank you, Saman. It was very resourceful and informative.
Merci Saman
Question. VARS is a sensitivity and uncertainty analysis framework. Do two separate analyses need be run to obtain uncertainty and sensitivity? Often the range of an input variable, X, will often exceed the uncertainty in that variable. Would it be correct to say that for sensitivity we would want to provide VARS the full range of the input, but for uncertainty we would want to provide a smaller range? But with a smaller uncertainty range you wouldn't capture uncertainty conditional on X's full range of possible point estimates. How would you handle that issue?
Thank you for the valuable information.
Could you please tell us what is the best software to perform sensitivity analysis?
What is the difference between SA and Monte Carlo in order to confront uncertainty? Thank you in advance.
Monte-Carlo simulation is a forward approach for propagating uncertainty from inputs to the output. SA may be viewed as a backward approach that tries to partition the uncertainty in the output and attribute its pieces to different inputs.
Let me explain it in a bit more simple way (with all due respect to Saman's explanations), in SA we keep all variable constant and change just one variable to see how that single variable could affect the outcomes, and we should repeat it for all input parameters and a weighting factor should be defined to quantify the effects of each variable on output value i.e. how much they are influential (I used this method in geotechnical engineering field, which is my field of expertise), but in Mote-Carlo we should establish a "Probabilistic Space, by having in hand PDF of all of our influential parameters" and sampling from each PDF is completely random based on skewness of the PDF, then the results will be reflected in PDF of our output parameter (i.e. a wide range of outputs will be gained based on probabilistic sampling and the relationship between those input parameters and the output value.
Note: PDF stands for Probability Density Function (i.e. the bell shape curve, which could be skewed either positively or negatively).
Hope it was useful.
(Please correct me, Saman, if I am wrong).
Regards,
Mike