# Two-Way Tables in Business Reports: Name the Denominator Behind Each Percentage

<a id="main-content"></a>

Operations &amp; Systems

October 28, 2026  ·  8 min read  ·  Victor Valentine Romo

On this page

1. [Define the Observation Before the Categories](<#define-the-observation-before-the-categories>)
2. [Reconcile the Counts at Both Margins](<#reconcile-the-counts-at-both-margins>)
3. [Calculate Three Shares From the Same Cell](<#calculate-three-shares-from-the-same-cell>)
4. [Read the Condition Out Loud](<#read-the-condition-out-loud>)
5. [Keep Unknown Outcomes Visible](<#keep-unknown-outcomes-visible>)
6. [Distinguish Zero From an Undefined Share](<#distinguish-zero-from-an-undefined-share>)
7. [Copy the Table-Review Worksheet](<#copy-the-table-review-worksheet>)
8. [Hand Off the Question With the Number](<#hand-off-the-question-with-the-number>)
9. [Sources](<#sources>)

The report says exports account for forty percent of completions. Someone reads that as a forty percent completion rate for exports.

The cell count stays the same. The question changes.

Put the denominator beside the percentage before the report leaves the worksheet.

[Penn State's lesson on two-way tables](<https://online.stat.psu.edu/stat504/Lesson03>) describes counts cross-classified by two categorical variables. It distinguishes joint proportions from conditional proportions. The denominator determines which distribution a share describes.

The service-request report below is fictional. Its counts, category rules, worksheet, and review sequence illustrate a reporting method. They describe no customer, organization, or measured result.

<a id="define-the-observation-before-the-categories"></a>

## Define the Observation Before the Categories

One observation represents one eligible service request. It does not represent one email, one status change, or one person. Several messages about the same request still produce one observation.

The fictional cohort contains requests received in one reporting week. Every included request has reached its five-business-day assessment deadline before the extract closes. Completion means [a recorded first completion](</articles/throughput-busy-time-service-team-reports/>) within that window.

A request completed after its deadline belongs in the not-complete-by-deadline category. That category describes the deadline result. It does not claim the request remains unfinished forever.

The row variable is request type: account update or report export. Each request receives exactly one type under the intake dictionary. [A request requiring both activities](</articles/decision-table-hit-policy-business-rules/>) follows its designated primary type. If that designation is missing, hold it outside this table and disclose the exclusion.

The column variable is deadline outcome: completed, not complete, or unknown. Unknown means the extract lacks enough outcome evidence to assign either known result. It does not mean an eligible request missed its deadline.

[Record those definitions before grouping](</articles/business-metric-dictionary-template/>). Otherwise, one reviewer may count status events while another counts unique requests. Both can produce a tidy grid from different observations.

<a id="reconcile-the-counts-at-both-margins"></a>

## Reconcile the Counts at Both Margins

The synthetic extract begins with 107 rows. Four are repeated rows for already represented requests. Remove those repetitions under the stated one-request rule. Three remaining requests have no usable type. Hold those three for classification review.

That leaves 100 eligible, uniquely represented, typed requests. Fifteen have unknown deadline outcomes. Keep them inside the table under the unknown column.

The table has two row categories and three column categories. The total row and total column summarize those categories. They are not additional request types or outcomes.

| Request type | Completed within five business days | Not complete by deadline | Unknown outcome | Row total |
| --- | --- | --- | --- | --- |
| Account update | 36 | 9 | 5 | 50 |
| Report export | 24 | 16 | 10 | 50 |
| Column total | 60 | 25 | 15 | 100 |

Read the first row as 36 + 9 + 5 = 50. Read the second as 24 + 16 + 10 = 50. The row totals reconcile to 100.

Then read down. Completed requests total 36 + 24 = 60. Known deadline misses total 9 + 16 = 25. Unknown outcomes total 5 + 10 = 15. The column totals also reconcile to 100.

[Penn State calls these sums marginal totals](<https://online.stat.psu.edu/stat504/Lesson03>). A row margin sums across columns. A column margin sums across rows. The grand total includes every observation in the table once.

Keep the extract reconciliation beside the grid: 107 starting rows, four duplicate rows removed, three requests held for missing type, 100 represented requests. A matching pair of margins [cannot reveal excluded records](</articles/data-quality-dimensions-business-reports/>) by itself.

<a id="calculate-three-shares-from-the-same-cell"></a>

## Calculate Three Shares From the Same Cell

Use the report-export completion cell. It contains 24 requests. That count supports three different questions.

| Question | Calculation | Result | Denominator meaning |
| --- | --- | --- | --- |
| What share of represented requests are completed report exports? | 24 / 100 | 24% | All requests in the table |
| Among report exports, what share are recorded complete by deadline? | 24 / 50 | 48% | All report-export requests |
| Among completed requests, what share are report exports? | 24 / 60 | 40% | All completed requests |

The joint share is 24%. It describes the intersection of a request type and an outcome within the whole table. Both labels belong in its statement.

The row share is 48%. It conditions on report-export membership. Start with those fifty requests, then ask how many have the recorded completion outcome.

The column share is 40%. It conditions on completion. Start with the sixty completed requests, then ask how many are report exports.

The arithmetic uses one numerator and three denominators. The percentages differ because the populations differ. None is a substitute for either of the others.

This follows the distinction between joint and conditional sample proportions in [Penn State's notation and structure section](<https://online.stat.psu.edu/stat504/Lesson03>). The business labels and counts here are original synthetic examples.

For the account-update completion cell, the same review gives 36 / 100 = 36% jointly. Its row share is 36 / 50 = 72%. Its column share is 36 / 60 = 60%.

The two completion-column shares add to 100%: forty percent exports and sixty percent account updates. That check describes [the composition of completions](</articles/simpsons-paradox-business-group-mix/>). It does not calculate either type's completion rate.

<a id="read-the-condition-out-loud"></a>

## Read the Condition Out Loud

A sentence beginning “among report exports” names the report-export denominator. A sentence beginning “among completed requests” names the completion denominator.

Write that opening before choosing the percentage. Then underline the phrase after “among.” Confirm that the table margin contains exactly those observations.

For this example, a defensible statement reads: “Among fifty report exports, twenty-four have recorded completion within five business days, a forty-eight percent share.”

The reversed condition reads: “Among sixty requests completed within five business days, twenty-four are report exports, a forty percent share.”

Those statements can appear together. Keep their denominator labels attached when shortening them for a slide. A chart legend saying only “exports, forty percent” leaves the reader to supply the population.

Use separate [headings](</articles/accessible-headings-b2b-pages/>) when displaying several percentage grids. Label one “percentage within each request type” and another “percentage within each outcome.” Avoid a single unlabeled percent column containing different denominator rules.

Moving the type variable from rows to columns changes the layout. It does not change the underlying question. Recalculate the display labels after transposing the table. A row percentage in the first layout becomes a column percentage in the transposed layout.

<a id="keep-unknown-outcomes-visible"></a>

## Keep Unknown Outcomes Visible

The report-export row has ten unknown outcomes. Its forty-eight percent recorded completion share retains all fifty exports in the denominator. That choice preserves the stated cohort.

A reviewer may instead ask about requests with known outcomes. That is a different conditional question. For exports, the known-outcome denominator becomes 24 + 16 = 40. The corresponding completion share is 24 / 40 = 60%.

Label it “completion share among report exports with known outcomes.” State that ten exports are excluded from that calculation because their deadline outcomes are unknown.

For account updates, the equivalent denominator is 36 + 9 = 45. The known-outcome completion share is 36 / 45 = 80%. Five account updates remain outside that conditional calculation.

Across both types, sixty completions among eighty-five known outcomes give approximately 70.6%. Across all one hundred represented requests, recorded completions account for 60%. Keep both populations identifiable.

Neither calculation resolves the unknown records. An unknown outcome could later become a recorded completion or a recorded deadline miss. Do not assign either status to make a report appear complete.

If the next extract fills those fields, [preserve the earlier extract date](</articles/late-arriving-data-report-revisions/>) and explain the revision. Changing outcome evidence changes the table. It does not automatically establish a change in the underlying service process.

<a id="distinguish-zero-from-an-undefined-share"></a>

## Distinguish Zero From an Undefined Share

Suppose a separate synthetic week contains no eligible report exports. The export row total is zero. Its completion cell is also zero.

The row completion share would require 0 / 0. Leave it undefined. Display “not applicable: no eligible exports” rather than zero percent.

Now suppose ten eligible exports appear and none completes by deadline. The completion share is 0 / 10 = 0%. That statement has a defined denominator and an observed zero numerator.

The same rule applies down a column. When a week has no completed requests, the share of completions belonging to exports is undefined. An empty completion column cannot describe the composition of a completed population.

Keep [a numeric zero, a missing outcome](</articles/null-blank-zero-business-data/>), and an undefined ratio distinct in exports as well as on the page. A spreadsheet cell that replaces all three with a dash hides the reporting condition.

If a category cannot occur under the service definition, describe that restriction separately. An impossible combination and an observed empty cell require different explanations. This worksheet does not supply a statistical test for either case.

<a id="copy-the-table-review-worksheet"></a>

## Copy the Table-Review Worksheet

Use one worksheet row per claim. The following version is a proposed editorial tool, not a Penn State reporting requirement.

| Field | Filled synthetic example |
| --- | --- |
| Question | Among report exports, what share has recorded completion by deadline? |
| Observation | One eligible unique request received in the reporting week |
| Row variable | Primary request type: account update or report export |
| Column variable | Five-business-day outcome: completed, not complete, or unknown |
| Cell count | 24 report exports completed within the window |
| Chosen denominator | 50 report exports, including ten unknown outcomes |
| Exclusions | Four repeated extract rows removed; three requests held for missing type |
| Statement | Recorded completions account for 48% of fifty represented report exports |

Add the extract [timestamp](</articles/event-time-processing-time-business-reports/>), classification dictionary version, and reviewer to the working document. Those fields let a later reader reconstruct the population. Keep private request records in their controlled source.

Check each statement against the worksheet before comparing categories. If the question changes, create another row with its own denominator. Reusing the first percentage under a new heading silently changes its meaning.

Use the [weighted-average review](</articles/weighted-averages-business-reports/>) when combining rates across categories. Use the [percentage-point guide](</articles/percentage-points-business-reporting/>) when describing a difference between shares. Those tasks follow denominator selection.

For the published grid, apply the [comparison-table guidance](</articles/accessible-b2b-comparison-tables/>) to [captions](</articles/b2b-video-caption-quality-review/>), headers, and narrow screens. Read a cell with its row and column labels to check whether it makes sense outside its position.

<a id="hand-off-the-question-with-the-number"></a>

## Hand Off the Question With the Number

Before sending the report, ask a second reader to reconstruct one percentage from counts alone. Give them the question and the grid. Ask which observations belong below the fraction line.

If they choose a different margin, revise the wording before discussing the result. A clearer label is part of the finding.

Keep significance testing, [forecasting](</articles/sales-forecasting-b2b/>), and causal explanations outside this table review. The immediate deliverable is a reconciled count layout and a sentence whose denominator the reader can identify.

Name the population. Keep the count beside the share.

<a id="sources"></a>

## Sources

- [Penn State STAT 504, Lesson 3: Two-Way Tables](<https://online.stat.psu.edu/stat504/Lesson03>). Cross-classification, marginal totals, and joint and conditional sample proportions. Checked 2026.10.28. The service-request counts, missing-record rules, worksheet, and review sequence are synthetic editorial examples. No statistical testing or population inference is performed.
