Confidence Intervals Explained

A confidence interval gives a range of plausible values for something you can’t measure directly, such as a population mean. It tells you both your best estimate and how precise it is.

A worked example

You weigh 40 apples from an orchard. The mean is 72.5 g and the standard deviation is 8.4 g. What is the mean weight of all the apples in the orchard?

  1. Standard error: SE = s ÷ √n = 8.4 ÷ √40 = 1.33 g.
  2. Critical value: for 95% confidence with df = 39, t* = 2.02. You can read this from the t table.
  3. Margin of error: 2.02 × 1.33 = 2.69 g.
  4. Interval: 72.5 ± 2.69, which is 69.81 g to 75.19 g.

Run the apple example in the confidence interval calculator

What “95% confident” means

Imagine repeating the whole study many times: a new 40 apples each time, a new interval each time. About 95% of those intervals would contain the true mean. The other 5% would miss it.

For any one interval, the true mean is either inside it or not. You don’t know which. The 95% describes how often the method works, not the chance for this one interval. In practice, “we are 95% confident the true mean is between 69.8 and 75.2 g” is a fine way to say it.

Two readings to avoid: it is not the range where 95% of apples fall (that range is far wider, roughly 56 to 89 g), and it is not the range where 95% of future sample means would land.

What makes an interval wider or narrower

ChangeEffect on the apple interval
Raise confidence from 95% to 99%Wider: 68.90 to 76.10 g (± 3.60)
Weigh 160 apples instead of 40Narrower: 71.19 to 73.81 g (± 1.31)
Apples that vary more (larger SD)Wider, in proportion to the SD

Notice that four times the data only halves the width. Precision grows with the square root of the sample size, which is why the last bit of precision is always the most expensive.

Confidence intervals and hypothesis tests

A 95% interval and a two-tailed test at α = 0.05 always agree. If a value is outside the interval, a test against that value would be significant. If the orchard claims a mean of 78 g, the interval 69.8 to 75.2 already tells you that claim doesn’t fit the data.

The interval gives you more than the test, though. A test just says “different from 78”. The interval says by how much.

Intervals for other things

Check yourself

A poll reports 54% support with a 95% interval of 51% to 57%. Can you say a majority supports the policy?

The whole interval is above 50%, so the data are consistent with majority support, at 95% confidence. If the interval had been 48% to 60%, you couldn’t say that.

You want to halve your margin of error. How much more data do you need?

Four times as much, because the margin of error shrinks with √n.