Translate | Percentages, Percentage Points and Basis Points — Preserve the Baseline and the Size of the Change

Translating percentages accurately means preserving the comparison behind the number. A rate that rises from 20% to 25% has increased by five percentage points, but it has increased by 25% relative to its starting value. Those two descriptions can both be correct. They are not interchangeable translations, because they answer different questions about the same change. Before translating a percentage, identify what is being counted, what it is being divided by, and which starting value the sentence uses.

This guide explains how to translate percentage change, percentage points and basis points in reports, dashboards, research summaries, product information and everyday educational texts. Its specific job is to help you preserve numerical claims when a source moves between proportions, rate differences and relative comparisons. It is not a replacement for a complete financial-report translation workflow or a general guide to statistics. The examples are invented teaching cases, not claims about actual organisations, investments or research findings.

The practical method is to separate the source statement into a small mathematical record before rebuilding the sentence in another language. Record the original value, final value, population, period, unit and comparison operator. Then decide which words can change naturally and which relationships must remain fixed. This lets you translate fluent sentences without accidentally changing “increased to” into “increased by,” “percentage points” into “percent,” or a change in the sample into a change in the people being measured.

A fifty-second orientation

A percentage expresses a quantity per hundred. A percentage-point change subtracts one percentage value from another. A relative percentage change divides the change by a stated baseline. A basis point is one hundredth of a percentage point. Each has a different role, even when all four appear in the same report. The Office for National Statistics guidance on percentages distinguishes percentage changes from percentage-point differences; European Central Bank reporting guidance also illustrates the relationship between basis points, percentages and decimal representation.

For a first review, ask three questions: “Of what?”, “Compared with what?” and “Over which period?” If one answer is missing, do not use a smoother sentence to hide the gap. Keep the original form where possible and raise a precise query. A translator is allowed to make the language clearer; changing the statistical claim requires a separate, documented decision.

1. Build the comparison before choosing the wording

Consider the sentence: “The completion rate increased from 40% to 50%.” There are two endpoints and a direction. The source does not merely contain the numbers 40 and 50. It says that a defined measure was at one level before and another level afterward. The noun “completion” identifies the event; “rate” signals a denominator; and “from” and “to” assign roles to the endpoints.

A useful working note is: earlier rate 40%; later rate 50%; change +10 percentage points; relative increase 25% of the earlier rate. You do not have to add all four statements to the translation. The note is a check on whichever statement you do publish. If your target draft says “completion increased by 50%,” the note immediately exposes the error: 50% is the destination rate, not the relative increase.

This habit is especially helpful when target-language grammar places modifiers differently. The final value may move closer to the verb during rewriting. Keep a label beside it in your working notes so proximity does not change its role. Think of the note as a small diagram in words: baseline, change, endpoint. The translated sentence must preserve those connections even when its word order changes completely.

2. A percentage always has a denominator, even when the sentence hides it

“Thirty percent completed the course” is incomplete without a population. It might mean 30% of enrolled learners, 30% of those who started, or 30% of those who submitted the final assignment. These are different claims. A translator who replaces “participants who began” with the more convenient “participants” may enlarge the denominator without changing a single visible number.

Suppose 200 people enrolled, 160 began, and 120 completed. Completion is 60% of enrolments and 75% of starters. Neither percentage is inherently more truthful. The question is which population the source names. In a faithful translation, the denominator travels with the result, whether it appears in the same sentence, a table heading or a footnote.

Before translating a table, look beyond the individual cell. A column headed “% of valid responses” does not include unanswered questions. The next column might use “% of all respondents.” Keep these headings distinct. A number can survive character for character while its meaning changes because a qualifier was removed from the heading. In data-rich translation, headings and footnotes are part of the evidence, not optional decoration around it.

3. Percentage points describe a gap on the percentage scale

To calculate a percentage-point difference, subtract the earlier percentage from the later percentage. A change from 72% to 78% is six percentage points. A change from 78% to 72% is a fall of six percentage points. No division by the starting value is involved in this particular calculation.

The advantage of this wording is that it preserves an absolute gap between rates. It is often the clearest way to compare shares, completion percentages or the proportion of items passing a check. But clarity does not authorise substitution. If the source explicitly reports relative growth, translate relative growth rather than silently replacing it with a percentage-point description.

When target-language space is limited, resist shortening the unit to a percent sign alone. “+6%” and “+6 percentage points” are different statements. An abbreviation can be used when the publication defines it, but its meaning must be visible to the intended reader. For an educational page or a general report, spelling out the unit at first mention is often the least ambiguous editorial choice. Thereafter, follow the agreed house style consistently rather than switching between several shortened forms.

4. Relative change needs a starting value

Relative percentage change asks how large the change is compared with the baseline. For a positive nonzero starting value, calculate the difference, divide by that starting value, and multiply by 100. In the invented example of 80 completed tasks becoming 100, the increase is 20 tasks. Twenty divided by 80 is 0.25, so the relative increase is 25%.

Notice what happens when the direction reverses. A fall from 100 tasks to 80 is a decrease of 20%, not 25%. The change in task count has the same magnitude, but the baseline is different. Translation must therefore preserve both chronology and the comparison direction. “Twenty-five percent more than” cannot automatically become “twenty-five percent less than” when the subjects exchange positions.

A useful diagnostic question is: “If I start with the source baseline and apply the target sentence’s claimed change, do I reach the source endpoint?” Starting at 80 and adding 25% gives 100. Starting at 80 and adding 20% gives 96. This reverse check tests the actual claim rather than the visual similarity of the numbers. It is particularly useful when the source contains several percentages close together.

5. Basis points are small rate differences, not another word for percent

One basis point equals 0.01 percentage point. Therefore, a change of 25 basis points is a change of 0.25 percentage point. In decimal proportion notation, that same change is 0.0025. These are three representations of one rate difference, not three different economic events. Keep the distinction between a percentage value and its decimal storage representation explicit.

Take a fictional published rate moving from 4.00% to 4.25%. It has increased by 25 basis points. The increase relative to the original rate is 6.25%, because 0.25 divided by 4.00 is 0.0625. Translating the sentence as “the rate increased by 25%” would create a very different endpoint: 5.00% rather than 4.25%.

Basis-point abbreviations also need context. Preserve the source’s defined abbreviation when readers are expected to recognise it, or spell out the term on first use. Do not attach a percent sign to a basis-point count as a stylistic embellishment. A clean review checks the label, the numerical scale and the direction separately. The arithmetic here is illustrative; it is not advice on which rate or financial product anyone should choose.

6. “To,” “by” and “of” carry the mathematics

“The error rate fell to 3%” gives an endpoint. “The error rate fell by 3%” gives a relative change if that is the source convention. “Three percent of submissions contained an error” gives a share of a population. The same digits appear, but the prepositions connect them to different mathematical objects.

When translating a compressed headline, expand it into a complete sentence in your working notes. “Errors down 3%” may need the report body to determine whether it means three percentage points, a 3% relative reduction, or a rounded count comparison. Do not invent a baseline simply because a familiar target phrase expects one. The correct next action may be to inspect the accompanying table or query the author.

The same care applies to “up to,” which describes a ceiling rather than a result necessarily achieved. “Up to 30% fewer errors” must not become “30% fewer errors” without authority. “More than” excludes equality; “at least” includes it. These small expressions deserve a deliberate comparison pass after the main language edit. Their grammatical size is small, but they often control the largest practical difference between two translations.

7. Change in a count is not automatically change in a rate

Imagine that one workshop had 20 successful submissions out of 40 attempts and another had 36 out of 60. The number of successes rose from 20 to 36, an 80% increase. The success rate rose from 50% to 60%, a ten-percentage-point increase or a 20% relative increase in the rate. Three changes describe the same pair of workshops.

A translator must preserve which change the sentence owns. “Successful submissions increased by 80%” should not become “the success rate increased by 80%.” The first describes a count; the second describes a proportion. The population grew too, so the two cannot be exchanged.

A practical working record puts the numerator and denominator beside each rate. This may feel excessive for a short paragraph, but it removes uncertainty from later editing. When a noun is shortened from “number of successful submissions” to “success,” you can check whether the shortened wording remains clear. Natural style is valuable only while the measured object remains identifiable. Where a shorter expression could be read as either count or rate, keep the fuller noun phrase.

8. Percent of a subgroup is not percent of the whole

Suppose 40% of an invented learner group selected an optional module. Among those learners, 25% chose the advanced version. The advanced version therefore accounts for 10% of the entire group, because 0.40 multiplied by 0.25 is 0.10. However, the source sentence “25% of module participants chose the advanced version” should not be translated as “25% of learners chose it.”

Subgroup relationships are easily lost when a target language uses fewer repeated nouns. Replacing a repeated phrase with a pronoun can be perfectly natural, but the pronoun must point to the subgroup, not the larger population. Re-read the paragraph with the group names written out to check the reference.

Do not add the derived 10% unless the assignment permits explanatory additions. The calculation belongs in your verification notes when the public task is translation alone. This distinction protects source fidelity: knowing more about the arithmetic helps you avoid errors, but it does not give you unrestricted authority to rewrite the author’s selection of facts. A translator’s private verification can be more extensive than the final wording.

9. Averaging percentages can hide unequal group sizes

An invented small class has nine successful learners out of ten, giving 90%. A larger class has 45 out of 90, giving 50%. The simple average of those two percentages is 70%. The combined success rate, however, is 54 successful learners out of 100, or 54%. Both calculations can be defined, but they answer different questions.

If the source says “the average class success rate,” it may mean an average that treats every class equally. If it says “the percentage of all learners who succeeded,” the combined numerator and denominator are the relevant objects. A translator should not decide between them by choosing the result that sounds more plausible. Preserve the stated measure and raise a query if the source wording conflicts with the table.

This example also shows why a number check cannot stop at copying digits. The phrase “overall rate” carries a claim about aggregation. During review, ask whether the source and target both weight the same units: people, classes, schools, products or reporting periods. A change in weighting can be a larger distortion than a small rounding difference, even though every percentage in the table remains unchanged.

10. Successive changes compound rather than cancel automatically

A quantity of 100 increases by 20% to 120. A later decrease of 20% takes it to 96, not back to 100. The second percentage is applied to a different baseline. Consequently, the sentence “a rise of 20% followed by a fall of 20%” does not mean “no overall change.”

Keep the sequence intact in translation. Words such as “then,” “subsequently,” “compared with the previous period” and “relative to the original total” identify which base each operation uses. In a compressed narrative, they may be the only clues that the baseline changes between steps.

If a source table reports separate year-on-year changes, do not add them to create a cumulative change unless the intended calculation supports it. A 10% rise followed by another 10% rise gives a 21% cumulative increase from the original baseline. An author may have intentionally reported individual changes rather than the cumulative result. Translate the reported series, and use the combined calculation only as a private check or a clearly approved explanatory addition.

11. Zero and negative baselines require a stop rule

A change from zero to five cannot be expressed with the ordinary relative-change formula because that formula would divide by zero. “Five more cases” is meaningful. “A 500% increase from zero” is not supported by that calculation. Do not introduce such a percentage while making a source paragraph more concise.

Negative baselines need equal care. Moving from a loss of ten units to a gain of five is an improvement of fifteen units, but a mechanically calculated percentage can be confusing unless the reporting convention is explicitly defined. This is a good place to preserve the actual endpoints instead of inventing a headline percentage.

The translation workflow should have a clear query threshold: when the baseline prevents ordinary interpretation, flag the ambiguity. A focused query might ask, “Does the reported change compare the signed values, absolute magnitudes, or a separately defined benchmark?” That is much more useful than a general request to check the numbers. The question names the missing decision and lets the source owner resolve it without guessing what the translator found difficult.

12. Percentage change and percentage-point change can coexist in one sentence

A source might say: “The rate increased by four percentage points to 24%, representing 20% growth from the previous level.” Work backward to check the relationship. If the endpoint is 24% after an increase of four percentage points, the earlier rate was 20%. Four divided by 20 is 20%, so the statement is internally coherent.

A translator may rearrange the sentence for readability, but the target should still distinguish the difference, endpoint and relative growth. Repetition of the word “percentage” is not necessarily bad writing when it protects different concepts. A shorter target sentence is not better if readers must infer which number is which.

One useful alternative, when the editorial brief allows sentence splitting, is to state the endpoints first and give the relative interpretation in the next sentence. Splitting changes presentation, not the relationship, provided nothing is omitted or added. Keep the source attribution attached to both sentences when the numbers are someone else’s claim. Otherwise, restructuring can accidentally turn a reported calculation into an assertion made directly by the translator or publisher.

13. Rounding must not manufacture a disagreement

Imagine underlying rates of 19.6% and 20.4%. Rounded to whole percentages, both may display as 20%, even though the unrounded difference is 0.8 percentage point. A source that provides a separately calculated change may therefore appear inconsistent with rounded endpoints without actually being wrong.

The right response is not to force every displayed figure to agree through silent alteration. Check whether the source notes its rounding convention or supplies more precise values elsewhere. Preserve the original precision and identify any uncertainty. If the translation is for a data interface, distinguish the stored value from the displayed value before proposing changes.

Avoid increasing apparent precision during conversion. A source estimate of “about 20%” should not become “20.000%” because a spreadsheet or calculator displays trailing zeros. Nor should a carefully measured percentage be casually rounded to an integer just because the target sentence is shorter that way. Precision, approximation and presentation are separate decisions. Record which one you are changing, and make sure the assignment actually permits that change.

14. A whole worked passage: repair the relationships, not just the words

Use this invented source passage: “The pilot enrolled 250 learners. Completion rose from 64% in the earlier cohort to 72% in the pilot. The report describes an eight-percentage-point improvement, equivalent to a 12.5% relative increase in completion rate. Thirty percent of pilot completers selected an additional exercise.”

Start with the endpoints: 72 minus 64 gives eight percentage points. Eight divided by 64 gives 0.125, confirming the 12.5% relative increase. The pilot completion count is 180 learners. Thirty percent of those completers is 54 learners. The earlier cohort’s total size is not supplied, so the passage does not establish the change in the number of completers between cohorts.

Now consider a flawed draft: “The pilot’s 250 learners completed 72% more work, an improvement of 8%, and 30% of all learners selected an additional exercise.” This draft invents an amount of work, changes percentage points into a percentage, and expands the final denominator. None of those changes is justified by the source.

A repaired target-language version should retain the source’s defined objects: enrolled learners, completion rate, earlier cohort, pilot cohort and pilot completers. The arithmetic notes help you see why each term matters. They are not a licence to add the derived 54 learners when the source only gives a subgroup percentage. For a faithful translation, preserve the original information selection; for an approved explanatory adaptation, label any added calculation clearly.

15. A second worked passage: basis points and reporting formats

Here is another fictional source: “The published reference rate moved from 3.50% to 3.75%, a rise of 25 basis points. The export file stores the new rate as 0.0375. The explanatory note rounds the rate to one decimal place for its summary chart.”

The first check is mathematical: 3.75 minus 3.50 equals 0.25 percentage point, which is 25 basis points. The second check concerns representation: the proportion 0.0375 corresponds to 3.75%. The third concerns display. Depending on the stated rounding rule, the summary chart may show a rounded value while the export preserves the more precise number. These are different layers, not necessarily contradictory data.

A translator should not replace 0.0375 in the export with 3.75 merely because the user-facing label says percent. That would change the stored quantity if the application expects a proportion. Conversely, displaying 0.0375% would understate the rate by a factor of one hundred. Keep field semantics visible in the translation brief.

This guide does not prescribe an interest-rate reporting system. It demonstrates how to translate a specified one. When the receiving system is unknown, ask whether its field expects a percentage value, a decimal proportion or a basis-point count. That narrow question is more useful than asking simply which number format the client prefers.

16. Query the source with an answerable question

A source query should identify the exact conflict and the smallest decision needed to resolve it. Instead of “Please check this percentage,” write: “The table changes from 12% to 15%, but the paragraph says an increase of 3%. Should that read three percentage points, or should the endpoints be different?” The question presents the evidence without silently deciding which source element is wrong.

A second useful pattern is a denominator query: “Does 40% refer to everyone invited or only to respondents?” A third is a baseline query: “Is the 10% improvement measured against the original result or the immediately preceding period?” Each can be answered directly by the person who owns the data.

Keep the unresolved source wording and your proposed alternatives separate. Do not place a speculative correction inside a polished target paragraph where a later reviewer might mistake it for approved text. Once the answer arrives, update the affected sentence and every linked caption or summary. The decision should travel with the claim so another editor does not later restore the earlier ambiguity in the name of consistency.

17. Review terminology, arithmetic and attribution in separate passes

For the terminology pass, check words such as rate, share, proportion, count, percentage point, relative change and basis point. Ask whether the target uses the same type of quantity. A word can be a natural dictionary equivalent but still be wrong for this particular mathematical role.

For the arithmetic pass, ignore elegance. Reconstruct the baseline and endpoint, recompute any stated changes, and check signs. A decrease should not become an increase after sentence restructuring. A range should keep both endpoints. A percentage-point number should not acquire a percent sign through an automated formatting rule.

For the attribution pass, check who owns the claim. “The report estimates,” “the team observed” and “the model projects” express different evidence status. A fluent translation must not turn an estimate into a measured result. This separation makes review manageable: one pass tests vocabulary, one tests relationships, and one tests the authority and uncertainty attached to the numbers. All three are needed before the prose can be treated as reliable.

18. Practice clinic with explained answers

Practice one: endpoints and difference. A fictional quality rate rises from 82% to 88%. The percentage-point increase is six. The relative increase is 6 divided by 82, approximately 7.32%. Preserve whichever form the source reports. Do not translate a six-point improvement as a six-percent relative increase, because applying 6% to the original rate gives 86.92%, not 88%.

Practice two: a falling rate. A fictional error rate declines from 8% to 6%. It falls by two percentage points and by 25% relative to the original rate. The direction is a decline in errors, which may be described as an improvement only when the source makes that evaluation. Numerical direction and evaluative language are different layers.

Practice three: basis points. A made-up rate rises from 2.10% to 2.35%. The difference is 0.25 percentage point, or 25 basis points. The target must not say 25 percentage points. A reviewer can catch the error by adding the claimed difference back to 2.10% and checking whether it reaches 2.35%.

Practice four: subgroup scope. Half of 80 participants choose a project, and 20% of those project participants choose an optional extension. Eight people select the extension. Translate “20% of project participants” with its subgroup intact. Writing “20% of participants” would imply sixteen people and would double the derived count.

Practice five: equal percentage changes. A fictional stock of 200 items rises by 10%, then falls by 10%. The endpoints are 220 and 198. The overall change is a decrease of 1% from the original stock. The translation must not describe the two changes as cancelling unless the source is reporting something other than successive relative changes.

Practice six: missing denominator. A table says “Satisfied: 70%” but does not say whether unanswered surveys were excluded. The correct translation action is to preserve the stated number and query the denominator when it matters. Choosing an interpretation because it is common in another project would import an unsupported methodology.

Practice seven: zero baseline. A source says applications increased from zero to twelve. Preserve those endpoints. The ordinary relative-growth formula is undefined at a zero baseline, so do not invent a percentage increase. An explanatory adaptation may say twelve additional applications, but a new statistical measure needs a defined basis.

Practice eight: unequal groups. One group records 8 successes out of 10 and another 18 out of 30. The combined rate is 26 out of 40, or 65%. The simple average of the two group rates, 80% and 60%, is 70%. Check whether the source reports an average of group rates or a rate across all individuals before choosing the target wording.

19. Frequently asked translation questions

Must every percentage be converted into words? No. Preserve the publication’s expected representation where it is clear. The important question is whether the reader can identify the quantity and comparison. Spelling out “percentage points” may remove ambiguity, but spelling out every digit is not necessary.

Can I add a relative change that the source does not report? Only when the assignment permits explanatory additions, and then the added calculation should be identifiable as such. For a translation-only task, calculate privately to verify the claim and preserve the source’s selection of information publicly.

What should I do when a headline conflicts with the table? Record the conflict and ask a precise source query. Do not assume that the table is always right or the headline is always wrong. Either may contain a transcription error, a different denominator or a different reporting period.

Can one house-style rule settle percent versus percentage points? No. House style can govern spelling, spacing and abbreviations, but the mathematics determines which quantity the source means. A style preference cannot turn one measure into the other.

What is the most useful final test? Reconstruct the source endpoints from the target claim. If the target says a value rose by a certain relative percentage, apply that change to the original baseline. If it says a percentage-point change, add or subtract that difference on the percentage scale. Failure to recover the endpoint is a clear reason to review the wording.

20. “Twice as much” and “200% more” need different endpoints

An invented baseline of 30 units provides a useful language test. Twice that amount is 60 units. An amount that is 200% of the baseline is also 60 units. An amount that is 200% more than the baseline is 90 units: the original 30 plus an increase of 60. The words “of” and “more than” are therefore not interchangeable ways of making the sentence sound natural. They define different operations.

A translation should preserve the source operation even when the target language expresses multiplication differently. Write a concrete endpoint in your working notes and check whether a target reader could reconstruct it. When a source uses an expression such as “two times higher,” whose intended interpretation is unclear in context, do not settle it merely by choosing your preferred reading. Inspect the reported values or ask whether the intended endpoint is twice the baseline or the baseline plus twice the baseline.

This problem becomes more difficult when a marketing sentence combines emphasis with measurement. “Up to three times as many completed tasks” contains a ceiling, a multiplier and a counted object. Preserve all three. It does not say every user achieved that result, and it does not automatically mean a 300% increase. A precise translation can still be lively; it simply must not add a stronger numerical promise than the source supports.

21. Complementary rates do not have identical relative changes

Suppose an invented pass rate rises from 90% to 95%. If every item is classified either pass or fail, the corresponding failure rate falls from 10% to 5%. The pass rate rises by five percentage points and the failure rate falls by five percentage points. However, the relative rise in the pass rate is about 5.56%, while the relative fall in the failure rate is 50%. Complementary endpoints do not produce identical relative changes.

The assumption about complete classification matters. If some items are pending, withdrawn or unassessed, pass and fail may not add to 100%. A translator should not manufacture a failure rate by subtracting the pass rate from 100 unless the source categories justify that operation. Read the legend before treating the two labels as exact complements. This is another example of arithmetic depending on definitions rather than the other way around.

A source may deliberately frame the same improvement as fewer failures rather than more passes. Preserve that framing when it is part of the author’s meaning, while retaining the correct numerical scale. Do not change “failures halved” to “passes doubled” for stylistic variety. In this example, the former is true and the latter is false. A final review should ask not only whether the direction sounds positive, but whether the translated sentence names the same measured event.

22. Continue through the existing translation architecture

The broader factual method belongs to Translate | Names, Numbers, Dates and Units. The complete meaning-transfer architecture remains Master Art of Translation. Use the Vocabulary Learning Hub to develop distinctions between related words, and How English Works for the grammar of modifiers, comparisons and reference. This specialist guide adds a worked mathematical check; it does not replace those broader owners.

Before release, confirm that every percentage still names the same population, every change uses the same baseline, every percentage-point or basis-point label preserves its scale, and every estimate keeps its evidence status. Then read the target text naturally. The goal is neither stiff word-for-word copying nor attractive statistical rewriting. It is clear language that leaves the size and meaning of the original claim unchanged.