0
$\begingroup$

I have the following:

  • x is mean of X={4,6} and equals to 5 and standard deviation of X equals to 1.41
  • y is mean of Y={3,9} and equals to 6 and standard deviation of Y equals to 4.24

How can I calculate the standard deviation of z which is equal to:

$$ z=\frac{y-x}{x}100 = \frac{6-5}{5}100=20$$

$\endgroup$
  • $\begingroup$ $z$ is a number: it has no standard deviation. Perhaps you are asking about the standard deviation of the random variable $100(Y-X)/X$? If so, you haven't provided enough information, because it's impossible to determine the degree of association of $X$ and $Y$. $\endgroup$ – whuber Aug 9 '17 at 20:27
1
$\begingroup$

Suppose you have quantities $a,b$ and their standard deviations are $\Delta a, \Delta b$. Then if $s=a+b$ or $s=a-b$ then $\Delta s = \sqrt{(\Delta a)^2+(\Delta b)^2}$.

And if $q = a/b$ then $\dfrac{\Delta q}{q} = \sqrt{\left(\dfrac{\Delta a}{a}\right)^2+\left(\dfrac{\Delta b}{b}\right)^2}$

Your problem involves only subtraction and division, so you should be able to apply these easily. Look up "propagation of uncertainty" for more information.

$\endgroup$
  • $\begingroup$ This solution implicitly makes many strong assumptions, of which the foremost is that $a$ and $b$ are not correlated. At the least you should identify the important assumptions you are making. $\endgroup$ – whuber Aug 9 '17 at 20:28
1
$\begingroup$

Your $20$ is based on means, but since your function is non-linear it is unlikely to be the mean of $Z$

You need more information, as shown in these examples

  • Consider $X$ taking the values $4$ and $6$ each with probability $\frac12$ and independently $Y$ taking the values $4.5$ and $7.5$ each with probability $\frac12$. Then for $Z=\frac{Y-X}{X} \times 100$, the expectation of $Z$ is $25$ and the standard deviation is about $40.5$

  • Consider $X$ taking the values $0$ with probability $\frac1{26}$ and $5.2$ with probability $\frac{25}{26}$ and independently $Y$ taking the values $4.5$ and $7.5$ each with probability $\frac12$. Then for $Z=\frac{Y-X}{10} \times 100$, $Z$ does not have an expectation or standard deviation. Similar examples can produce very large expectations and standard deviations

If $X$ and $Y$ are normally distributed and independent with the parameters you state, then simulation may suggest something similar to my first example. This would be misleading: in fact the distribution of $Z$ has heavy tails and no moments, similar to my second example

$\endgroup$
  • $\begingroup$ Hey Henry, thank you very much for your answer. After reading your answer it made me realise that I was not precise enough and that I made a mistake in asking my question. Now I edit it and fixed it. Please check now. $\endgroup$ – yuk Aug 9 '17 at 18:45
  • $\begingroup$ It is still unclear (for example I do not see how you calculated your new standard deviations), but if $X$ and $Y$ can each take two possible values independently, then $Z$ can take four possible values and you can find the mean and standard deviation of $Z$ $\endgroup$ – Henry Aug 9 '17 at 18:59
  • $\begingroup$ I calculated the standard deviations using the formula for Sample Standard Deviation. Just to explain what is what here: X elements are measurements of weight of the same object at time t0 and Y elements are also measurements of weight of the same object at time t0. z is the increase/decrease of weight. Hope this makes it a bit clearer. $\endgroup$ – yuk Aug 9 '17 at 19:41
  • $\begingroup$ (+1) These are good points. In the first bullet, (1) you mean $Z=(Y-X)/X\times 100$ and (2) the standard deviation is not "about 40.5": it is exactly $\sqrt{4375/2}\approx 46.77$. $\endgroup$ – whuber Aug 9 '17 at 20:35
  • $\begingroup$ @whuber: You are correct about $Z$. Perhaps my standard deviation calculations are hopeless, but on the numbers in my example in the first bullet point, I get $\sqrt{\frac{13125}{8}}$. Since then the question has been edited and I would now get $\sqrt{\frac{9375}{2}} \approx 68.465$ as the standard deviation of $Z$ $\endgroup$ – Henry Aug 9 '17 at 21:39

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.