Anyone here able to answer a stats question?Follow

I have 2 data points I want to use to form a ratio. Each data point has some uncertainty associated with it. What's the best way to express the uncertainty associated with the ratio?

For example let's say data point one is: 48.513, 50.058, 29.513

and data point 2 is 259.519, 296.952, 317.482

I'll calculate the mean and std dev for each point and get a ratio, but I'm not sure how to get an uncertainty number for the error associated with the ratio. That and our stats guy is gone for the weekend.

Anyhoo, I'm google-ing away but feeling tired, just thought I'd see if I could get you people to do my homework for me.

The odds are low.

That's fine, at least we have some pie now. Would you like some?

Actually the odds are really high. What is low are the odds that someone will actually bother to answer one.

Um... Calculate the highest and lowest possible ratios and go from there? So lowest of one set to highest of the other, then the other way around. That gives you a single range of ratios. This assumes that the results aren't linked together directly (so one set is the variable output of one thing, the other the output of something else). If the sets are linked (each pair is the result of a interdependent test of some kind), then you'd want to treat them as three separate ratios and then go from there.


How you analyze it has to take into account how the data was generated. Otherwise, it's just garbage you used math(tm) on.

You're looking at 6 independent measurements. In 3 of them the protein is at one concentration and in 3 it is at another level. The sampling technique is analogous to estimating the number of, say, red balls in the bin by blindly reaching in and grabbing a few. Those numbers above being the number of 'red balls' you grabbed (the decimal places come from normalization based on the variable number of successful sampling attempts).

Along the lines you were thinking though, I was considering making 6 ratios out of the points by taking the three points and comparing them to each of the other three, and using those values to address the question. The problem is that ratios don't necessarily behave the same way when you do something like average them. So that's where I got stuck.

Looks like I may be grabbing my old stats book from college and having some fun reading over it this weekend...

Rate-ups for the brainstorming helps.

Edited, Feb 10th 2012 6:11pm by someproteinguy

Mashed potatoes.

That's the badger.

You're probably right on the students knowing more, though. Especially if you're doing labs week in week out. Tends to stick in the mind a bit. As you said, you do the experiment and you have a guy who specifically does the stats portion. It looks like you're on the right track, though!

That's the part I have done. Now let's say I form want to express the difference between the two data points as a ratio (i.e. 5.2:1). How does the uncertainty from the two measurements factor into the uncertainty of the resulting ratio?

Something about how it all equals 42, but they are taking the long way to get there.