Recognitions:
Gold Member

## Comparing Results from an Experiment: What Statistical Measure is Important?

Hi folks

I have an experiment in which I take an image of a flame. I then run a software routine that tells me what the concentrations of OH (hydroxyl) is at different heights above the flame. I first have to give it a calibrated image of a flame with known data and it then is able to give me data about an unknown flame as stated above. Here is what I have done:

1. Give calibrated the software with a known flame image with known data

2. Imaged 2 different flames that are unknown. the software then returns the OH concentrations of these tow flames.

I am curious to know what statistical measures are important? I suppose I should compare the 2 data sets from the unknown flames so I can show that the data sets are statistically similar and not outside of the bounds of experimental error.

 PhysOrg.com science news on PhysOrg.com >> 'Whodunnit' of Irish potato famine solved>> The mammoth's lament: Study shows how cosmic impact sparked devastating climate change>> Curiosity Mars rover drills second rock target

Recognitions:
 Quote by Saladsamurai I am curious to know what statistical measures are important?
That is a question for combustion engineering, not for for statistics

My observation of human nature (not yours particularly, but from my sampling of forum postings) is that when people are faced with complicated problems involving probability, they avoid the necessary step of figuring out exactly what they are trying to accomplish. Instead of answering that question, they ask what "statistics" says they should do.

Unfortunately, mathematics and statistics don't tell you anything until you know what you are trying to accomplish. To give some completely hypothetical examples, suppose the flame is a flame in the burner of a furnace. Suppose one consideration for such flames is whether they get hot enough to damage the combustion chamber. Then you would be interested whether hot areas of the flame are near the surface of the chamber, how hot they are, how long this lasts. Perhaps physics tells you that a very tiny hot spot that only lasts for a short time won't burn a pinhole in the chamber, so you would want to know the probability that there would be hot spots that would be big enough and last long enough to be dangerous.

You have to figure out what properties of the flame are important from the viewpoint of combustion engineering.

Recognitions:
Gold Member
 Quote by Stephen Tashi That is a question for combustion engineering, not for for statistics My observation of human nature (not yours particularly, but from my sampling of forum postings) is that when people are faced with complicated problems involving probability, they avoid the necessary step of figuring out exactly what they are trying to accomplish. Instead of answering that question, they ask what "statistics" says they should do. Unfortunately, mathematics and statistics don't tell you anything until you know what you are trying to accomplish. To give some completely hypothetical examples, suppose the flame is a flame in the burner of a furnace. Suppose one consideration for such flames is whether they get hot enough to damage the combustion chamber. Then you would be interested whether hot areas of the flame are near the surface of the chamber, how hot they are, how long this lasts. Perhaps physics tells you that a very tiny hot spot that only lasts for a short time won't burn a pinhole in the chamber, so you would want to know the probability that there would be hot spots that would be big enough and last long enough to be dangerous. You have to figure out what properties of the flame are important from the viewpoint of combustion engineering.
Hi Stephen

Thanks for replying. I may have convoluted my intent with my wordiness (as usual). Again, I am specifically looking to:
 Quote by Me ... compare the 2 data sets from the unknown flames so I can show that the data sets are statistically similar and not outside of the bounds of experimental error.
I took a stats course awhile back and I could swear there was a parameter that could help me determine whether the 2 data sets obtained via an experiment under the same conditions are statistically similar, or whether experimental error could be hiding something.

Any thoughts on this?