01 · Recover totals
Five observations have mean 3 and descriptive SD 2. Find T and U.
Hint
Use U = n(SD² + mean²).
Worked solution
T = 15; U = 5(4 + 9) = 65.
Understand · explore · practise
Recover sums from means and standard deviations, correct errors, add or remove observations and combine disjoint datasets without averaging their standard deviations.
Before you startMean, variance, standard deviation and combined means.
01 / Keep a three-part record
Write n, T = Σx and U = Σx².
Then mean = T/n and descriptive variance = U/n − (T/n)².
Work with these totals when data are corrected or combined. A mean alone loses the count; a standard deviation alone loses both the count and the location.
Start with values 2, 4, 6, 8: count 4, sum 20, squared sum 120. Replace the recorded value 8 by c.
Sum = 20 − 8 + 8 = 20.
Squared sum = 120 − 8² + 8² = 120.
Mean = 5; variance = 5; SD ≈ 2.236068.
A replacement keeps n = 4. Adding or removing a whole observation would also change the count.
02 / Recover the underlying totals
T = n×mean = 6×5 = 30
Recover the ordinary sum.
Variance = 2² = 4
Do not insert SD directly into the variance formula.
U = n(variance + mean²) = 6(4 + 25) = 174
Recover the squared sum.
Five observations have mean 3 and descriptive SD 2. Find T and U.
Use U = n(SD² + mean²).
T = 15; U = 5(4 + 9) = 65.
Ten observations have mean 4 and descriptive variance 9. Find U.
Nine is already a variance.
U = 10(9 + 16) = 250. Do not square nine again.
03 / Replace one recorded value
Tnew = Told − a + b; Unew = Uold − a² + b².
Here a was recorded and b is the corrected replacement. The count stays the same.
Pause, replay or seek freely. The notes explain the same idea and stay in view.
Four recorded values have T = 30 and U = 380. One value was written as 18 instead of 8. Find the corrected mean and descriptive variance.
Replace 18 by 8 in both totals.
Tnew = 30 − 18 + 8 = 20. Unew = 380 − 324 + 64 = 120. Mean = 5; variance = 120/4 − 25 = 5.
For n = 5, T = 25, U = 165, a recorded 9 should be 7. Find corrected mean and variance.
The squared-sum correction is −81 + 49.
Tnew = 23; Unew = 133. Mean = 23/5 = 4.6; variance = 133/5 − (23/5)² = 136/25 = 5.44.
04 / Adding and removing change n
For the data 2, 4, 6, 8, remove 8. Find the new n, T, U and variance.
Subtract one from the count as well.
n = 3; T = 12; U = 56. Mean = 4; variance = 56/3 − 16 = 8/3.
Add 10 to the original data 2, 4, 6, 8. Find the new mean and variance.
n = 5; add 10 to T and 100 to U.
T = 30; U = 220. Mean = 6; variance = 220/5 − 36 = 8.
Why is subtracting (18 − 8)² = 100 from U wrong when correcting 18 to 8?
The required change is a difference of squares, not a squared difference.
The correct reduction is 18² − 8² = 324 − 64 = 260. In general a² − b² is not (a − b)².
05 / Add totals for disjoint groups
TA = 12; UA = 3(8/3 + 16) = 56
Recover group A’s sums.
TB = 20; UB = 2(4 + 100) = 208
Recover group B’s sums.
Combined n = 5, T = 32, U = 264
The groups are disjoint and measured in the same units.
Mean = 6.4; variance = 264/5 − (32/5)² = 296/25 = 11.84
The combined SD is √296/5 ≈ 3.441.
A is 0, 0 and B is 10, 10. Each has SD zero. Find their combined SD.
The combined mean is five.
Squared deviations are all 25, so descriptive variance = 25 and SD = 5. Averaging the group SDs would incorrectly give zero.
Why must you not add the sums for a whole class and for the girls within that same class as if they were separate groups?
One group is contained in the other.
That would count the girls twice. Use disjoint groups or remove the overlap using enough information.
06 / Separate within-group and between-group variation
Combined variance = Σ nᵢ[vᵢ + (mᵢ − m)²] / Σnᵢ
m is the combined mean; mᵢ and vᵢ are each group’s mean and descriptive variance.
The first term measures spread within groups. The second measures the separation of their means from the overall mean. This explains why simply averaging group variances usually misses part of the variation.
For groups 0, 0 and 10, 10, which part of the formula supplies the combined variance?
Both within-group variances are zero.
All variation is between groups: their means are each five from the combined mean. Weighted squared distances therefore give variance 25.
If disjoint groups have the same mean, does their combined descriptive variance equal the count-weighted average of their variances?
The between-group distances are then zero.
Yes, for these descriptive variances and compatible units. Their SDs still should not generally be averaged; take the square root after combining variances.
07 / Check the denominator convention
If supplied s² = Sxx/(n − 1), recover Sxx = (n − 1)s² and then U = Sxx + n mean². Do not multiply s² by n. Once you have n, T and U, combine them and use the denominator required by the final question.
A group has n = 3, mean 4 and sample variance s² = 4 using n − 1. Recover T and U.
Sxx = 2×4.
T = 12; Sxx = 8; U = 8 + 3×16 = 56.
The combined dataset in the worked example has n = 5, T = 32 and U = 264. Find its sample variance using n − 1.
Sxx = U − T²/n.
Sxx = 264 − 1024/5 = 296/5. Divide by four: sample variance = 74/5 = 14.8.
08 / Respect rounded summaries
If means or SDs were rounded, the recovered T and U are estimates. Small differences in rounded numbers can affect a variance calculation. Preserve precision where possible and check that the final variance is non-negative.
A source reports a mean and SD to one decimal place. Can you claim the recovered squared sum is necessarily exact?
The rounded summaries do not uniquely determine the unrounded values.
No. Treat it as an estimate based on the reported summaries and state the precision limitation.
09 / Update the records before the summaries
Record n, T and U; make the appropriate changes to all three; then recompute the mean and requested variance or SD. Check disjoint groups, common units, denominator conventions and rounding.
Section 1 of 9 · Keep a three-part record