Sample deviation is calculated as population standard deviation divided by square root of sample size $σ_x=\left(\frac{σ}{\sqrt x}\right)$. Then why in an example like this below, is the answer $5.086$ ? I understand that it comes from $4.2+1.96 \times \left(\frac{1.5}{\sqrt{11}}\right)$, but why are we dividing with $\sqrt {11}$ when going from sample to population, when according to the formula we should multiply with it?

The total amount of vegetation held by the earth’s forests is called the “biomass”. A random sample of $11$ one-square-metre plots was taken in Canada’s northern forests; the sample mean biomass was $4.2$ kilograms per square metre, and the sample standard deviation was $1.5$ kilograms per square metre. What is the upper limit of a $95%$ confidence interval for the mean biomass (kilograms per square metre) in all of Canada’s northern forests?

  • $\begingroup$ Welcome to math.Se! Please, use mathjax. $\endgroup$ – user228113 Dec 14 '15 at 1:16

Your goal is to get a 95% confidence interval for the mean $\mu$ of a normal population. Let's begin by pretending we know the population variance.

A point estimate of $\mu$ is the sample mean $\bar X$ of the data. In this situation, can show that $E(\bar X) = \mu$ and that $Var(\bar X) = \sigma^2/n.$ Also, that the distribution of $\bar X$ is given by $\bar X \sim Norm(\mu, \sigma/\sqrt{n}),$ where the second argument is the standard deviation.

Standardizing, we see that $Z = \frac{\bar X - \mu}{\sigma/\sqrt{n}}$ is standard normal. Because 95% of the area under the standard normal curve lies between -1.96 and 1.96, we have $$P(-1.96 < Z < 1.96) = P\left(-1.96 < \frac{\bar X - \mu}{\sigma/\sqrt{n}} < 1.96 \right) = 0.95.$$

By manipulating the inequality on the right, we can see that $$P(\bar X - 1.96\sigma/\sqrt{n} < \mu < \bar X - 1.96\sigma/\sqrt{n}) = 0.95.$$

If $\sigma$ is known, this relationship is the basis to say that a 95% confidence interval for $\mu$ has endpoints $\bar X \pm 1.96\sigma/\sqrt{n}.$

In this estimation context, $\sigma/\sqrt{n}$ is called the 'standard error' of the mean $\bar X.$ This is a way of saying that $SD(\bar X) = \sigma/\sqrt{n}.$ (The terminology 'standard error' is usually used only with estimators.) The denominator $\sqrt{n}$ indicates that the variability of the mean of a sample of size $n$ gets smaller as the sample gets larger. This is a different thing than the population standard deviation $\sigma$. For a single randomly chosen observation $X_i$, one could write $SD(X_i) = \sigma.$

In your specific problem, if $\sigma = 1.5,$ then the upper confidence limit is $$\bar X + 1.96\sigma/\sqrt{n} = 4.2 + 1.96(1.5)/\sqrt{11},$$ based on the information provided.

However, in case $\sigma$ is not known, it is usually estimated by the sample standard deviation $S = \sqrt{\frac{\sum_{i=1}^n(X_i - \bar X)^2}{n-1}}.$ In that case, $T = \frac{\bar X - \mu}{S/\sqrt{n}}$ has a Student t distribution with $n-1$ degrees of freedom, and a number from printed tables or software must be substituted for 1.96 above. For example, with $n = 11,$ your 95% CI would have the upper limit $\bar X + 2.228 S/\sqrt{n}.$ The R code below shows the computation of 2.228.

 qt(.975, 10)
 ## 2.228139

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.