Confidence intervals
Generally, for a given or measured result, a tolerance interval or a confidence interval may be assigned within which the assigned or measured value is expected to fall.
This interval may be considered a Type A measurement uncertainty and depends in particular on the sample size.
The confidence interval is conventionally expressed as 100(1 − α)%, where α is a small number less than 0.5, representing the risk. If α = 0.05, then the confidence interval is 95%.
Case of a simple mean:
A confidence interval may be assigned to the mean x̄ in the following form:

With s: experimental standard deviation, calculated with ν = n − 1 degrees
of freedom, and t
Let us consider an example: a laboratory performs n = 60 boron analyses on
a batch of steel parts and determines a mean of x̄ = 23 ppm. The calculated
standard deviation is S = 16 ppm, and for a 95% confidence interval, we have
seen that α = 0.05, so 1 − α/2 = 0.975 and therefore t

If we had considered the distribution to be normal, we would have multiplied (S=σ)/√n par 2 (n→∞)
Difference between two simple means:

The difficulty lies in calculating a standard deviation that is representative of the difference between the values of x̄ and ȳ̄:


with f=fx+fy (degrees of freedom), fx=nx-1 (degrees of freedom on x), fy=ny-1 (degrees of freedom on y)
Let us consider an example: a laboratory performs nx = 60 boron analyses on a batch of steel parts and determines a mean value of x̄ = 23 ppm. The calculated standard deviation is Sx = 16 ppm. The laboratory then performs ny = 45 boron analyses on a second batch of steel parts and determines a mean value of ȳ = 28 ppm. The calculated standard deviation is Sy = 18 ppm.


with f=59+44=103 ( total degrees of freedom), fx=60-1=59 (degrees of freedom on x), fy=45-1=44 (degrees of freedom on y)
![]()
If the laboratory performs the same number of analyses with different standard deviations, the formulas are simplified:



with f=2(n-1)
Confidence interval on a linear curve:
We consider a curve with the equation Y = aX + b, obtained by titration using the values from the data sheet: Slope uncertainty
| Data |
Y |
X |
| Unity | mV | ml |
| U | ± 10 | ± 0,2 |
| 1 |
75 | 1,0 |
| 2 |
109 | 1,5 |
| 3 | 150 | 2,0 |
| 4 | 175 | 2,5 |
| 5 | 210 | 3,0 |
| 6 | 243 | 3,5 |
| 7 | 279 | 4,0 |
a is obtained by linear regression:

We can then easily determine b by:
![]()

with n=7
We will calculate the standard deviation of the curve using the following formula:

is obtained using the equation Y=67,1X+9,4
![]()
The confidence interval for Y can be expressed as follows:

t(1-α/2)= 2,5706 (f=n-2=7-2=5)

with a=67,1 et b=9,4
Note: The values obtained cannot be compared with the calculations in the worksheet: Slope uncertainty because we had defined the measurement uncertainties for each of the input quantities, which is not the case here; however, it is easier to compare the values obtained with the technical data sheet: Numerical Example Based on the New VIM Definition of Calibration because we used the Monte Carlo method, which only uses the data provided and the titrant uncertainty (±2%), the uncertainty value for Y at 150 mV was 4.6 mV. The confidence interval is 4.3 mV on average here, and 3.5 mV for a reading of 150 mV.

Percentiles of Student’s t-distribution
