The title may seem a bit off topic. I will explain my doubts with an example. Let there be a situation where we are measuring gravity using the formula enter image description here

Now, if the least count or error in the measurement of $l$ and $T$ is given, I can easily find the net error or relative error by taking a natural log on both the sides of the equation and differentiation.

But how to include the number of trials? For example, I learned somewhere that if the least count of a stop watch is 1 second, and the number of trials, say to measure the oscillation period of a pendulum is 20, then should the error in measuring the time period be cut down to 1/20 seconds?

  • 5
    $\begingroup$ Possible duplicate of How to combine measurement error with statistic error $\endgroup$ Commented Apr 16, 2017 at 11:05
  • 1
    $\begingroup$ By my reading, none of the proposed duplicates answer the question. They come close, but don't quite do it. If the OP is not satisfied with the answers he should say so in a comment. For now, I vote to reopen. $\endgroup$
    – garyp
    Commented Apr 16, 2017 at 13:13
  • $\begingroup$ I did check the link. As @garyp said, yes I did not find the answer there. Maybe the answer there is too technical for me to understand or it is not answering my question at all. $\endgroup$ Commented Apr 16, 2017 at 14:07
  • $\begingroup$ There are two things to deal with. First: Either a.) your reaction time estimate is wrong, or b.) you have not captured all sources of error. In either case, you have to deal with finding the best estimate of the time, and the best estimate of its error given a set of measured times. The best estimate of the time is the mean. The best estimate of the uncertainty in that value is the standard error of the mean (or just standard error). If the question reopens, I'll elaborate. In the meantime, maybe searching on "standard error" will help. $\endgroup$
    – garyp
    Commented Apr 16, 2017 at 21:16

1 Answer 1


There's two possible things going on in such a measurement:

a) measure the time taken for $n$ oscillations and then your systematic error will indeed be reduced, e.g. minimum stopwatch interval $ / n$; and

b) do the $n$-oscillation measurement $N$ times to estimate the statistical uncertainty.

As described by the answers at How to combine measurement error with statistic error (thanks to Emilio for the link), these error sources should be added in quadrature. The statistical error will converge to zero as $N \to \infty$, but the systematic limitation on $T$ remains fixed... unless you make $n$ bigger, assuming that $T$ remains constant through a long "run".

  • $\begingroup$ Thanks for confirming. Well, where can I study more about error analysis for JEE Advanced 2017 (if you are familiar with the exam). I really cannot find extensive details about the topic in any of my textbooks, however questions do come which require deeper understanding. $\endgroup$ Commented Apr 16, 2017 at 14:09
  • $\begingroup$ The issue in this question is not systematic error. $\endgroup$
    – garyp
    Commented Apr 16, 2017 at 21:11
  • $\begingroup$ @garyp I am really confused now as we weren't taught the classification of errors :( $\endgroup$ Commented Apr 17, 2017 at 2:50
  • $\begingroup$ It depends on the nature of the timer limitations. If there's the possibility of a constant, unknown offset -- e.g. your reaction time always makes things later, never earlier -- then there's an uncertainty related to the unknown nature of that systematic offset. You could attempt to estimate (i.e. calibrate) or reduce that effect. On the other hand, if the inaccuracy of the timing is random, then it will reduce as $\sigma_\mathrm{timer} / \sqrt{N}$, along with other purely statistical effects. $\endgroup$ Commented Apr 17, 2017 at 10:24

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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