# Need some guidance on sensitivity and uncertainty analysis

1. Apr 26, 2014

### knight92

Hi, I have a set of results for which I wish to carry out sensitivity analysis. I have already googled it and it always comes up with examples on Microsoft Excel but I dont want to use excel for the sensitivity analysis. So can someone explain to me what sensitivity and uncertainty analysis are? As I understand it, its introducing 'what-if' scenarios but I have already completed my experiment and have the results.

The experimental results consist of hot water going into a radiator where the inlet and outlet temperatures are recorded giving delta temperature. Air passes through the radiator dissipating heat and the air temperature before and after passing through the radiator are measured. So I will be calculating the heat transfers. In this case should I introduce my own values just to show how the calculation of the heat transfer will change in case of a "what-if" scenario ?

Thank you.

2. Apr 29, 2014

### Stephen Tashi

Why do you want to do uncertainty and sensitivity analysis?

Are you doing this project to learn what that type of analysis is? Did someone assign you to do such an analysis?

3. Apr 29, 2014

### knight92

Yes I want to learn what the analysis is, I have tried looking it up except I cant wrap my head around to what it actually is?

4. Apr 29, 2014

### MarneMath

Sensitivity analysis isn't just one thing. It requires many things and all require a good breadth of understanding of the subtle assumptions built into the test. If you wish to calculate a what-if scenario you may run a simulation and perform many of such runs and see how it handles in different environment. You may wish to assume (or see) if there exist a linear relationship and build a linear regression and then test the so called beta coefficients. You may wish to augment this by using a partial F-test by changing one variable at a time, this method being fairly common.

Uncertainty analysis is even boarder. Some of it may just be using the appropriate statistical estimation, at other time it's about accounting for bias, and even so it may even reference back to the experiment design, which is another thing all together.

So the first thing to do in this is to try to determine what kind of model do you wish to build.