Method
Why a published table rarely adds up exactly
Rounded components and a rounded total are computed separately, so the columns often fail to sum. A worked demonstration of when a table's arithmetic should balance, when it cannot, and how to tell a rounding artefact from a real error.
By Journal Info Editorial Team · Published 2026-11-20
A table of percentages that does not sum to 100 is something a careful reader will often notice, and it is often not a mistake. Figures are usually rounded for display after each one is computed, so the displayed parts and the displayed total need not add up. Understanding when a table should balance and when it cannot is the difference between catching a real error and chasing a rounding artefact for an hour.
This article demonstrates the mechanism with a small table, states the rule for when summation is a valid check, and gives a procedure for verifying a table from its underlying counts.
The mechanism
Suppose a league has three teams and you want each team’s share of total goals. The exact shares are 33.33%, 33.33% and 33.33%, computed from 100, 100 and 100 goals out of 300. Displayed to one decimal place, the three figures are 33.3, 33.3 and 33.3, which sum to 99.9. Nothing is wrong: each figure was rounded independently and the rounding errors do not cancel.
Shift the counts and the gap moves. With 1,000 goals split 333, 333 and 334, the exact shares are 33.3%, 33.3% and 33.4%, which sum to exactly 100. With 149, 149 and 149 goals the shares are 33.33% each and the display to one place gives 99.9 again. The shortfall is a property of the rounding, not of the data.
A six-row example shows the same thing with whole numbers. These goal totals are invented: six teams with 29, 27, 26, 24, 23 and 21 goals, 150 in all.
| Team | Goals | Exact share | Displayed (whole number) |
|---|---|---|---|
| Team A | 29 | 19.33% | 19% |
| Team B | 27 | 18.00% | 18% |
| Team C | 26 | 17.33% | 17% |
| Team D | 24 | 16.00% | 16% |
| Team E | 23 | 15.33% | 15% |
| Team F | 21 | 14.00% | 14% |
| Total | 150 | 100.00% | 99% (displayed parts sum to 99) |
Three of the six shares round down by a third of a point, and nothing rounds up, so the displayed column sums to 99 while the exact shares sum to 100. Counts, shares and total are all correct.
Two different questions
The confusion comes from mixing two questions that look identical:
- Does the displayed table sum to the displayed total? This is a property of the rendering, and the honest answer is often no.
- Do the underlying counts add up? This is a property of the data, and it must hold.
Only the second question is a real check. If the raw counts do not add to the raw total, something is wrong with the data. If the raw counts are consistent and the display does not balance, the table is fine and the reader has simply confronted the display.
This is why the procedure for verifying any table is to go back to counts. A percentage, a rate or an average is a derived quantity, computed from counts and then rounded.
A worked check
Here is a three-row table with counts and derived rates. The counts are invented, and the win rate is wins divided by games (the same hits-over-attempts logic as a batting average, which is hits divided by at-bats according to MLB Glossary — Batting Average).
| Team | Wins | Losses | Games | Win rate |
|---|---|---|---|---|
| A | 30 | 20 | 50 | 60.0% |
| B | 25 | 25 | 50 | 50.0% |
| C | 20 | 30 | 50 | 40.0% |
| Total | 75 | 75 | 150 | 50.0% |
Check the counts first: wins sum to 75, losses sum to 75, games sum to 150, and 75 + 75 = 150. The counts are consistent. Now check the rates. The displayed rates sum to 150.0%, which is meaningless, because averaging rates is not the same as pooling the counts. The correct total rate is 75 ÷ 150 = 50.0%, which is what the table shows. The table balances on counts and correctly refuses to balance on rates.
That last point generalises: a rate of a combined group is not the average of the group’s rates unless the groups have equal denominators. Adding percentages across unequal groups is the single most common error in reader-side table checking, and it produces confident but meaningless totals.
When summation is a valid check
Summation is a valid test on counts, on totals explicitly defined as sums, and on any column labelled as a partition of something. It is invalid on rates, averages, indices and any figure described as “per” something, unless the denominators happen to be equal.
A quick rule: look at the column heading. If it is a count, the column should sum. If it is a rate or an average, the column should not be expected to sum, and the total row should be computed from the pooled counts.
Verifying a table against its source
When a table’s figures matter, the check has four steps:
- Find the counts. A table without counts is a table you cannot verify, only trust.
- Check that counts close. Sums must match totals, and totals must match the stated denominator.
- Recompute each derived figure from the counts and compare to the displayed value. Differences of one unit in the last displayed place are rounding; larger differences are errors.
- Check the direction of a discrepancy. If every derived figure is off by the same small amount, the cause is usually a different rounding rule. If one figure is off and the rest are exact, the cause is usually a transcription error in that one cell.
Step 4 is the diagnostic that saves the most time. Random, isolated discrepancies indicate a data problem; systematic, small discrepancies indicate a convention.
What this means for a data notebook
In our worked examples we show the counts alongside the derived figures for exactly this reason: a reader should be able to see the inputs, repeat the division, and decide whether the result holds. A table whose parts do not sum should say so, and say why.
Method
All counts and tables here are invented for illustration. The rounded shares, the sums and the exact percentages were computed by script (round half up). The batting-average definition used in one example was read on 2026-10-31. The rounding behaviour described is arithmetic and needs no outside source.
Sources
- MLB Glossary — Batting Average (AVG): hits divided by at-bats, as an example of a rate derived from counts.
AI assistance recorded as: AI draft, not edited