Which statistical test to use, and when?
June 15, 2009 2:16 PM Subscribe
Does anyone know of a good source that can help me figure out which statistical test to use, and under what circumstances?
My current problem is trying to figure out what test to use to compare two signal-to-noise ratios (behavioral measurements, not electrical engineering.) I am not even sure where to start with comparing these two ratios to see if there is a statistically significant difference between them.
Not homework, by the way. I'm just an independent researcher trying to muddle my way through some data, and it's been a while since I took a statistics course.
My current problem is trying to figure out what test to use to compare two signal-to-noise ratios (behavioral measurements, not electrical engineering.) I am not even sure where to start with comparing these two ratios to see if there is a statistically significant difference between them.
Not homework, by the way. I'm just an independent researcher trying to muddle my way through some data, and it's been a while since I took a statistics course.
When I took stats many years ago, the book was IPS. This text tells you, depending on things like whether or not your two groups are Gaussian, whether or not they have equal variances, and whether or not they're paired, which test you should be using.
That said, if you're an independent researcher, are you still affiliated with any academic institution? Most will generally have a department of statistics, or a biostatistics / computational biology 'core' which provides consulting and services to researchers, sometimes even on a walk-in, 'clinic' type basis.
posted by NucleophilicAttack at 3:27 PM on June 15, 2009
That said, if you're an independent researcher, are you still affiliated with any academic institution? Most will generally have a department of statistics, or a biostatistics / computational biology 'core' which provides consulting and services to researchers, sometimes even on a walk-in, 'clinic' type basis.
posted by NucleophilicAttack at 3:27 PM on June 15, 2009
The textbooks that I used in my undergrad years were Introduction to the Practice of Statistics mentioned by NucleophilicAttack above, and Statistical Methods for Psychology. Obviously, the latter book has an emphasis on behavioural methods, which might be helpful for you.
You might also find this statistics test flow chart to be useful. These are not uncommon, so you should be able to find more like this if you google around enough. I have to advise you, though, to actually learn what these tests are for and how they work. Too often, people blindly and haphazardly apply stat tests everywhere without understanding what they do.
posted by tickingclock at 9:49 PM on June 15, 2009 [1 favorite]
You might also find this statistics test flow chart to be useful. These are not uncommon, so you should be able to find more like this if you google around enough. I have to advise you, though, to actually learn what these tests are for and how they work. Too often, people blindly and haphazardly apply stat tests everywhere without understanding what they do.
posted by tickingclock at 9:49 PM on June 15, 2009 [1 favorite]
« Older How to tell the bull market from the bear market. | Need timer softwarer to control a PC Newer »
This thread is closed to new comments.
For common questions, sure. Look at the top subject area journal for your field, and you are probably coming to the same questions about data sources as those researchers on a regular basis. The tests and techniques which they use to address a similar question are a good starting point. You can almost always google the name of a test or method (or look at their citations) and get results. There are looks of "statistic test guide" or "statistic guide" "statistic choose" pages, but without knowing what data you're using, they may or may not be relevant. There's a bias towards epidemiology and experimental design type tests.
Signal to noise is probably the mean / SD? Are the data longitudinal? If they're identical draws for a population, the mean over the square root of the variance would be a t-distribution scaled by n^-1/2. You could construct a confidence interval from that. The inverse, SD/mean has a name " coefficient of variation" and googling "coefficient of variation test" yields ideas on how to test it.
posted by a robot made out of meat at 3:07 PM on June 15, 2009