1. Here's a description of how I calculated COR:

    • Formula $COR=√(h/H)$
    • I have 7 intervals where I measured $h$ and $H$
    • I did 5 trials for measuring $h$ and $H$ for each interval (in total 35 measurements for each of them)
    • The absolute uncertainty is the same for both $h$ and $H$ and is 0.002 m
  2. My question is:

    • How do I propagate the uncertainty for the COR?
  • $\begingroup$ What do you already know about your question? $\endgroup$ Feb 17, 2017 at 22:48
  • $\begingroup$ @NowIGetToLearnWhatAHeadIs I have all the values of h and H as well as their absolute uncertainty (0.002 as stated above) and also the values of COR. one way I thought of to calculate the COR uncertainty is: 0.5*((SD/Mean of h)+(SD/Mean of H)) but I am not sure if it is correct this way $\endgroup$
    – Bayan
    Feb 17, 2017 at 22:51

2 Answers 2


You have values for 7 different heights. For each height you can compute the error using standard methods. Combining the 7 values you want to take the weighted mean of the 7 CORs. Do you know how to compute the error for a weighted mean?

  • $\begingroup$ no, I don't :) can you explain briefly? $\endgroup$
    – Bayan
    Feb 17, 2017 at 23:37
  • $\begingroup$ See slide 3 of colorado.edu/physics/phys2150/phys2150_sp14/phys2150_lec4.pdf $\endgroup$
    – Floris
    Feb 17, 2017 at 23:42
  • $\begingroup$ Perhaps you should explain how to calculate error for the weighted mean in your answer, instead of providing a link. $\endgroup$ Feb 18, 2017 at 17:27
  • $\begingroup$ @sammygerbil that would be better but I am traveling and it's hard to do this on my phone. Thought I would help OP get started. "Teach a man to fish..." $\endgroup$
    – Floris
    Feb 18, 2017 at 17:30
  • $\begingroup$ Excuses excuses ... :) $\endgroup$ Feb 18, 2017 at 18:04

When doing a statistical analysis you can ignore the absolute uncertainty in the measurements, particularly when it is the same for each.

I would plot a graph of $h$ vs $H$, either using all 35 data points or using the 7 average values of $h$ if the values of $H$ were all the same. This will show you if the COR is constant over the range of your experiment. If it is not changing, you can fit a trend line of the form $y=bx$; the slope of this line is $b=e^2$ where $e$ is COR. You can use the squared deviations of data points from this line to calculate the Standard Error in the slope. If the slope is changing you can fit a quadratic $y=bx+cx^2$ then again use the value of $b$, which is the slope at small values of $x$.

You can do all this in a spreadsheet. Excel has built-in statistical tools which can help you. See http://www.chem.mtu.edu/~fmorriso/cm3215/UncertaintySlopeInterceptOfLeastSquaresFit.pdf. Or you could use data analysis software such as Origin or CurveFit to get the Standard Error.

When you have the Standard Error $S_b$ in slope $b$ then the COR is given by
$e^2=b\pm S_b=b(1\pm\frac{S_b}{b})$
$e=\sqrt{b}(1\pm \frac{S_b}{b})^{\frac12}\approx\sqrt{b}\pm\frac{S_b}{2\sqrt{b}}$.

If the above is too much effort, calculate the COR for each of the 35 readings, then use the data analysis tools in a spreadsheet to calculate the mean $m$ and standard deviation $\sigma$. Even basic models have AVERAGE and STDEV. The Standard Error in the mean is $S_m=\frac{\sigma}{\sqrt{N}}$. Your result is $m\pm S_m$.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.