The short answer
Keep the displayed percentages and the observed vote total as separate inputs. Rounded percentages need not sum to 100%, and they may not determine exact votes even when the total is known. Use the original worked examples to understand the ambiguity; their assumed half-up rounding rule is not a verified LinkedIn rule.
What you’ll take away
- Preserve the observed display instead of forcing it to sum to 100%.
- Several integer vote allocations can fit the same rounded percentages.
- Poll responses describe the observed responding group, not a representative population.
First record what the poll actually shows
Save the question, exact option labels, displayed percentages, visible total and observation time together. Mark a live poll as live. If screenshots come from different moments, a changed total can create a mismatch before rounding enters the problem. Keep a missing vote total unavailable rather than estimating it from impressions.
LinkedIn’s poll FAQ describes who can see how someone voted. That permission is not a reason to publish identities in a recap. The numerical exercise needs no names, account export or attempt to identify individual voters.
Why a rounded total can be 99% or 101%
In the constructed 40-vote example, counts of 13, 12 and 15 correspond to 32.5%, 30% and 37.5%. Under the explicitly chosen rule that an exact half rounds upward, the display is 33%, 30%, 38%: a total of 101%. The unrounded fractions still sum to 100%.
With six votes split 2, 2, 2, each share is one third. Rounded whole percentages are 33%, 33%, 33%, totaling 99%. Neither example measures LinkedIn behavior. They demonstrate why normalizing an observed display could erase information or manufacture precision.
A known total can still leave ten answers
Take the constructed display 33%, 33%, 34% with 400 total votes. Under the same assumed rule, the first two options can each have 130 through 133 votes; the third can have 134 through 137. The counts must also add to 400, leaving ten possible triples.
Two complete witnesses are 130/133/137 and 133/133/134. Both sum to 400 and reproduce the display. Multiplying each displayed percentage by 400 would produce one convenient triple, 132/132/136, without establishing it as the actual allocation. Report a range or leave exact counts unconfirmed.
Reproduce the finite example locally
Download and unzip the complete packet. Run node reproduce-examples.mjs with an already installed Node.js 22 or newer. The poll fixture records the assumed rule and every expected result. The checker enumerates the first two integer counts, computes the third from the total, and retains only matching displays.
The exercise rejects a zero total and bounds enumeration to 1,000 votes. This keeps a small teaching calculation predictable; it is not an estimator for arbitrary polls. The integer formula avoids a floating-point tie changing the intended half-up example. Read the assumption before applying its output.
When no allocation fits, stop the inference
With ten votes, the same 33/33/34 display has no solution under the exercise’s rule. That establishes an incompatibility among these particular inputs and this rule. It does not prove manipulation: the total, moment, display precision, option set or assumed rounding may differ.
Do not search for a plausible person-level story to repair the numbers. Reopen the observed source, check whether the poll changed, and record any unknown policy. Where exact counts are available through an authorized interface, preserve those observations separately from a reconstructed guess.
Finish with a claim the record supports
A safe constructed summary is: “The display was 33%, 33%, 34% at a shown total of 400 votes. Exact option counts are not established; ten allocations fit our illustrative rounding assumption.” That is more informative than silently inventing exact counts or branding a 101% total an error.
Complete the blank poll record with the source state, total, percentages, assumption and bounded conclusion. The local checker confirms arithmetic only. Participation, repeated collection, option wording and who encountered the poll still limit any broader conclusion about your audience.
Sources & product references
Product details reflect the publication date. External features and interfaces may change.
