Dismiss Notice
Join Physics Forums Today!
The friendliest, high quality science and math community on the planet! Everyone who loves science is here!

Statistical test

  1. Mar 19, 2008 #1
    Hi all,

    I am hoping someone can recommend a useful statistical test. I have a set of data on an x-y plot which varies about the y=0 line in a seemingly random way. Each data point has a y error bar, which appears to be, in general, smaller than the standard deviation of the data.

    I would like to apply a rigorous statistical test to calculate the probability that for the given individual one-sigma errors on each data point, the observed scatter about the y=0 line is plausible, or whether there is evidence for some kind of oscillation pattern?

    A null hypothesis test was my first thought but this does not take into account the individual errors bars on the data points. Also I have noticed that many mathematicians don't hold the the null-hypothesis in much respect, nor does it offer a probability of the data points being randomly spread.

    Thanks for any advise.

  2. jcsd
  3. Mar 19, 2008 #2
    It sounds like you're looking for a p-value, which is associated with null hypothesis testing. I would first use a computer and calculate the line of best fit through the data points. Let's call this line y = f(x). Then, your null hypothesis can be: "The equation which best models this phenomenon (or whatever it is) is f(x)." Your alternative hypothesis can be: "No. It's not."

    Then for all the data points (X1,Y1).... (Xn,Yn), calculate ((Yi - f(Xi))^2)/(f(Xi)) for all i = 1, 2, 3....n. Add the numbers from these calculations up and that would give you a chi-square statistic. Using a table or calculator, you can find the p-value corresponding to this score. The p-value, or the probability of observing this data given the equation y = f(x) is the true model for this phenomenon.

    Hope this helps.
Share this great discussion with others via Reddit, Google+, Twitter, or Facebook