EART60061 Chap.4 Confidence Limits, Standard Error and t Values
Confidence Limits, Standard Error and t Values
Confidence Limits is the take-home message of making measurement meaningful: two numbers can be compared only through the confidence we have in each. A confidence limit is the half-width of the range within which the true mean lies at a chosen confidence. Two corrections turn 1.96 standard deviations into the right limit.
Means vary less than single measurements, so the standard error, which divides the standard deviation by root n, replaces the standard deviation.
Small samples deserve less confidence, so a t value for n minus 1 degrees of freedom replaces 1.96. The chapter applies the method to biomonitoring copper with seaweed around a mine-affected coast, where sampling controls variation and replicates measure what remains, and it shows how the results figure, error bars and comparison lines earn marks in the short paper.
The chapter also covers precision in reporting, since the course warns that a limit given to the wrong number of decimal places can be marked wrong, and the follow-up question on analytical error from standards and reference material.
What this chapter covers
- 01
Variation of means versus measurements
- 02
Standard error
- 03
Degrees of freedom and t values
- 04
TINV and two-decimal precision
- 05
Comparing sites with overlapping error bars
- 06
Controlling variation in seaweed sampling
- 07
Instrumental and procedural analytical error
Worked example · free
Test a lead mean against a guideline with a t-based confidence limit
- 2Standard error = 2.6 / square root of 6 = 2.6 / 2.449 = 1.061 mg/kg.
- 2Confidence limit = 2.571 times 1.061 = 2.73 mg/kg, so the interval is 28.67 to 34.13 mg/kg.
- 1The upper end, 34.13, lies below 35, so the mean is below the guideline at 95% confidence.
Key terms
- Biomonitoring
- Measuring contaminants in living organisms, such as seaweed, which integrate exposure over time and concentrate metals.
- Correction factor
- A ratio between two species' concentrations at a shared site, used to make data from a substitute species comparable.
- Procedural blank
- A sample containing none of the analyte that is processed alongside real samples to reveal contamination or bias introduced by the procedure.
- t distribution
- The distribution that replaces the normal for means of small samples, wider for few degrees of freedom and approaching it as they grow.
Confidence Limits, Standard Error and t Values FAQ
When should I use the standard error instead of the standard deviation?
Use the standard error whenever the error bar belongs to a mean of several measurements. Means vary less than single values; the module shows ten means of 50 heights with a standard deviation near 0.8 against 5.81 for the individual heights. The standard error divides by the square root of n to capture this.
Why is t larger than 1.96 for small samples?
A mean from three replicates is less certain than one from thirty, and the t distribution builds that into the multiplier through degrees of freedom. With two degrees of freedom t is 4.303; above about 30 it is close to 1.96, and the two match only for an infinite sample.
What must the legend of a results figure say?
The model answer to the coastal paper expects a legend beneath the figure stating that error bars are drawn at 95% confidence. Without that sentence a reader cannot interpret whether overlapping or separate bars mean anything, so the figure loses much of its value as evidence.
How does TINV work for a 95% confidence limit?
TINV takes the probability outside the interval and the degrees of freedom. For 95% confidence the first argument is 0.05, and for a mean the second is the number of replicates minus one. Three replicates give 4.303 and ten give 2.262, the values used in the course exercises.
Why does a correction factor appear in the seaweed method?
One species was absent at a site, so a different species was analysed there and at a site where both grew. The ratio of the two means at the shared site converts the substitute data, and the method must state openly that this correction was made.
Exam move
Practise the four-line routine until it is automatic: mean and standard deviation, standard error, t for n minus 1, limit and comparison. Then rework one dataset three ways, with 1.96 times the standard deviation, 1.96 times the standard error and t times the standard error, and describe how the conclusion changes. Finally, draw one results figure by hand with error bars, reference lines and a complete legend.
Finish by explaining, in two sentences, why the two corrections to 1.96 times the standard deviation push the error bar in opposite directions.