Skip to content
Tutorials Center
My learning

Topics

Lesson 8 of 45 ยท Measurement, Error and Statistics

Statistics of replicates: mean, standard deviation and confidence intervals

Summarize replicate measurements with the mean, sample standard deviation, RSD and a t-based confidence interval, and see how many replicates you need.

Suppose you titrate five portions of the same solution and read 24.82, 24.78, 24.85, 24.80 and 24.79 mL. What is "the" volume? And how much would you expect a sixth measurement to differ? Statistics turns a handful of numbers into an honest statement: a best estimate and an interval in which the true value probably lies.

A bell curve centered on the mean, with shaded regions for plus or minus 1, 2 and 3 standard deviations containing 68.3, 95.4 and 99.7 percent.
Caption: random error often follows a bell curve, and the standard deviation sets its width.

What you will learn

  • Calculate the mean, sample standard deviation, variance and relative standard deviation.
  • Explain the normal distribution and the 68-95-99.7 rule.
  • Build a confidence interval for the mean using Student's t.
  • Estimate how many replicates are needed for a target precision.

Key terms

Mean (xฬ„): the sum of the results divided by their number n. Sample standard deviation (s): the typical spread of single results about the mean. Variance: sยฒ, the square of the standard deviation. RSD (relative standard deviation, coefficient of variation): 100 ร— s / xฬ„, in percent. Standard error of the mean: s / โˆšn, the uncertainty of the mean.

The mean and the standard deviation

For n replicates x_1 โ€ฆ x_n:

xฬ„ = ฮฃ x_i / n and s = โˆš( ฮฃ (x_i โˆ’ xฬ„)ยฒ / (n โˆ’ 1) )

The divisor n โˆ’ 1 (the degrees of freedom) is used because the mean was estimated from the same data, which makes the deviations a little too small. For our titration: the mean is 24.808 mL, the squared deviations sum to 0.00308, so s = โˆš(0.00308/4) โ‰ˆ 0.028 mL and RSD โ‰ˆ 0.11 %.

If the whole population of possible results is known, the symbols are ฮผ (true mean) and ฯƒ (population standard deviation). In practice you only have a sample, so you estimate ฮผ with xฬ„ and ฯƒ with s.

The normal distribution

Random error is often described by the bell curve with density f(x) = 1/(ฯƒโˆš(2ฯ€)) ร— exp(โˆ’(x โˆ’ ฮผ)ยฒ/(2ฯƒยฒ)). About 68.3 % of results fall within ยฑ1ฯƒ of ฮผ, 95.4 % within ยฑ2ฯƒ and 99.7 % within ยฑ3ฯƒ. That is why a result more than three standard deviations from the mean is suspicious.

The standard error and confidence intervals

Averaging n values shrinks the uncertainty of the mean to s/โˆšn. To turn that into a statement with a stated probability, use Student's t, which accounts for the fact that s itself is only an estimate:

ฮผ = xฬ„ ยฑ t ร— s / โˆšn

Typical values of t for 95 % confidence: 4.303 (ฮฝ = 2), 3.182 (3), 2.776 (4), 2.571 (5), 2.262 (9), approaching 1.960 for very large n. Here ฮฝ = n โˆ’ 1.

For our titration, n = 5, so t = 2.776 and the interval is 24.808 ยฑ 2.776 ร— 0.0277/โˆš5 = 24.808 ยฑ 0.034 mL. Read this as: "if the method were repeated many times, 95 % of such intervals would contain the true mean." It is not the statement that 95 % of individual results lie in this interval.

Because the width falls as 1/โˆšn, quadrupling the replicates only halves the interval. That is why a handful of replicates (three to six) is the usual compromise.

How many replicates?

If you need the mean within ยฑE and you know s approximately, then n โ‰ˆ (t s / E)ยฒ. Since t depends on n, start with t = 1.96 and iterate. With s = 0.028 mL and E = 0.02 mL: n โ‰ˆ (1.96 ร— 0.028/0.02)ยฒ = 7.5. But with only 8 replicates t is about 2.4, not 1.96, so you iterate: the answer settles at about 10 replicates.

Common misconceptions

"The confidence interval contains 95 % of the data." It describes the mean, not individual values.

"A small s proves the result is accurate." s measures random scatter only; a bias is invisible in it.

"Use n instead of n โˆ’ 1 for a small sample." That underestimates s; use n โˆ’ 1.

Try it

Python Mean, s, RSD and the 95 % confidence interval

Press Run (or Ctrl + Enter) to see the result here.

Program input ยท what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

JS How many replicates do I need?

Press Run (or Ctrl + Enter) to see the console output here.

No output.

Check your understanding

Quick check ยท 5 questions

of 5 answered You scored / 5 ยท perfect!

  1. 1. Results: 10.1, 10.3, 9.9, 10.2, 10.0. The mean is

    The sum is 50.5; divided by 5 it is 10.1.
  2. 2. Why is the standard deviation of a sample divided by n โˆ’ 1?

    Deviations from the sample mean are smaller than deviations from the unknown true mean; dividing by n โˆ’ 1 corrects that bias.
  3. 3. About what fraction of normally distributed results lies within ยฑ2ฯƒ of the mean?

    ยฑ2ฯƒ contains 95.4 %.
  4. 4. n = 4, xฬ„ = 50.00, s = 0.20, t(95 %, 3 d.f.) = 3.182. The 95 % confidence interval is

    t s / โˆšn = 3.182 ร— 0.20 / 2 = 0.318 โ‰ˆ 0.32.
  5. 5. To halve the width of a confidence interval (same s) you need roughly

    The width scales with 1/โˆšn, so halving it needs about four times the data.

Summary

  • Report the mean, s (with n โˆ’ 1) and RSD for replicates.
  • Random error follows the normal distribution: 68 %, 95 % and 99.7 % within 1, 2 and 3 ฯƒ.
  • The mean's uncertainty is s/โˆšn; the t-based interval xฬ„ ยฑ t s/โˆšn has a stated confidence.
  • Precision improves slowly with n: choose n from the target interval.

Next: Significance tests