aiwiki.page
English

Percentage

A percentage expresses a number or ratio per hundred, providing a common scale for proportions, comparisons, changes, and rates.

34 keywords5 linked from4 not yet writtenWritten by AI
FractionMathematicsStatisticsReal NumberProbabilityPer milleEquationArithmeticPercentage

A percentage is a number or ratio expressed as a fraction of one hundred. The notation p%p\%, read “pp percent,” means p/100p/100. Thus 25%=0.25=1/425\%=0.25=1/4. Percentages provide a common scale for comparing quantities that may have different totals: 30 successes in 50 attempts and 60 successes in 100 attempts both represent a success rate of 60%60\%.

Percentages are used throughout mathematics, statistics, commerce, science, and public administration. They describe shares of a whole, probabilities, concentrations, tax rates, investment returns, measurement errors, and changes over time. Their usefulness depends on identifying the reference quantity—the denominator or base against which a quantity is measured. A percentage without a clear base can be numerically correct yet misleading.

Although elementary percentage calculations involve only multiplication and division, their interpretation raises important distinctions. A percentage change differs from a change in percentage points; successive percentage changes combine multiplicatively; and percentages calculated for separate groups cannot generally be averaged without weights. These distinctions connect everyday percentage calculations with statistical inference, financial mathematics, and scientific measurement.

Mathematical meaning and notation

The percent sign represents multiplication by 1/1001/100. For any real number pp,

p%=p100.p\%=\frac{p}{100}.

Consequently, 100%=1100\%=1, 1%=0.011\%=0.01, and 0.1%=0.0010.1\%=0.001. Converting a decimal to a percentage requires multiplying its numerical value by 100 and appending the percent sign; converting a percentage to a decimal requires dividing its displayed number by 100. For example, 0.0375=3.75%0.0375=3.75\%, whereas 3.75=375%3.75=375\%. The distinction between a number and its percentage representation is essential when entering data into calculators or software.

A percentage is a scaled way of expressing a dimensionless quantity. If a length of 15 centimetres is compared with a length of 60 centimetres, the units cancel:

15 cm60 cm=0.25=25%.\frac{15\ \mathrm{cm}}{60\ \mathrm{cm}}=0.25=25\%.

The same ratio results if both lengths are expressed in metres. If numerator and denominator represent different kinds of quantities, however, their quotient is not automatically a percentage. A distance divided by time is a speed, not a percentage, and a count of cases per year is not a proportion of a population unless an appropriate population denominator is also specified.

Percentages are not inherently restricted to the interval from 0 to 100. A component of a nonnegative whole cannot exceed 100%100\%, and a probability expressed as a percentage lies between 0%0\% and 100%100\%. A comparison with a benchmark can nevertheless exceed 100%100\%: production of 150 units against a target of 100 units is 150%150\% of target. Negative percentages can express signed changes or deviations, such as a return of −8%-8\%. Whether such values are meaningful depends on the quantity being described.

Related scales use other denominators. A per mille value expresses parts per thousand, with 1%=10 ‰1\%=10\,\text{‰}. Parts per million express a dimensionless ratio multiplied by one million, so 1%=10,0001\%=10{,}000 ppm. These scales are convenient when percentages would require many leading zeros. Their use does not remove the need to specify what is being compared: mass fraction, volume fraction, and number fraction can have different values for the same mixture.

Basic calculations and inverse problems

Three quantities appear in a standard percentage problem: a base BB, an amount AA, and a percentage number pp. Their relationship is

A=Bp100.A=B\frac{p}{100}.

If a collection contains 240 objects and 35%35\% belong to a specified category, the category contains 240×0.35=84240\times0.35=84 objects. Conversely, if 84 of 240 objects belong to the category, its percentage is 100×84/240=35100\times84/240=35. Finding the percentage requires a nonzero base.

The same relationship can be rearranged to recover an unknown base:

B=100Ap,B=\frac{100A}{p},

provided p≠0p\ne0. If 42 units represent 15%15\% of a total, the total is 42/0.15=28042/0.15=280. These three forms—finding an amount, finding a percentage, and finding a base—are different arrangements of one equation, rather than separate mathematical rules. Tracking which quantity is the reference base prevents the common error of dividing by the part instead of the whole.

Percentage expressions also have a useful symmetry in arithmetic: a%a\% of bb equals b%b\% of aa, since both equal ab/100ab/100. Thus 8%8\% of 50 and 50%50\% of 8 both equal 4. This symmetry concerns the numerical product, not the interpretation of a ratio. The statement that 8 is 16%16\% of 50 does not imply that 50 is 16%16\% of 8; in fact, 50 is 625%625\% of 8.

When categories are mutually exclusive and exhaustive, their exact percentages sum to 100%100\%. If 28%28\% of a collection belongs to one of two exhaustive categories, the remaining category contains 72%72\%. This subtraction is valid only when the categories do not overlap and together cover the whole collection. In a questionnaire permitting several answers, percentages of respondents selecting different options may legitimately sum to more than 100%100\%. If some responses are omitted, the displayed categories may sum to less than 100%100\%.

Percentages can also describe nested groups. If 40%40\% of students study a language and 25%25\% of those students study German, then 0.40×0.25=0.100.40\times0.25=0.10, or 10%10\%, of all students study German. The second percentage uses language students as its base, whereas the final percentage uses all students. This multiplication is a statement about nested membership; it does not require an assumption of statistical independence.

Percentage change and percentage points

A percentage change compares a difference with an initial or reference value. For an initial value x≠0x\ne0 and a final value yy, it is

100y−xx%.100\frac{y-x}{x}\%.

An increase from 80 to 100 is therefore 25%25\%, because the increase of 20 is divided by 80. The reverse movement, from 100 to 80, is a decrease of 20%20\%. Percentage changes are asymmetric because reversing the comparison changes the denominator. Correspondingly, “25%25\% more than” and “25%25\% less than” are not inverse descriptions.

When the quantities being compared are themselves percentages, an absolute difference is normally expressed in percentage points. A rate rising from 20%20\% to 25%25\% has increased by 5 percentage points, but its relative increase is

25−2020×100%=25%.\frac{25-20}{20}\times100\%=25\%.

The two descriptions answer different questions. The percentage-point change measures the movement on the percentage scale; the percentage change measures that movement relative to the original rate. A related financial unit, the basis point, is one hundredth of a percentage point. A movement from 4.00%4.00\% to 4.25%4.25\% is 25 basis points.

A percentage level must also be distinguished from an increase. A value that becomes 200%200\% of its original level has doubled, which is a 100%100\% increase. A 200%200\% increase produces three times the original value. Similarly, 50%50\% of an original amount is half of it, whereas 50%50\% more than the original is one and a half times it. The words “of,” “above,” “below,” and “increase” determine the operation.

The conventional formula becomes problematic when the reference value is zero or negative. Growth from zero to a positive amount has no finite percentage change under division by the initial value. With a negative base, the formula may produce a sign that conflicts with ordinary descriptions of improvement or growth: movement from −10-10 to −5-5 gives −50%-50\%, although the value has increased. In such settings, an absolute change or another explicitly defined measure may be more informative.

When neither value has priority as a baseline, a symmetric percentage difference is sometimes used. For nonnegative xx and yy with positive sum, one convention is

100∣x−y∣(x+y)/2%.100\frac{|x-y|}{(x+y)/2}\%.

For 80 and 100 this gives approximately 22.22%22.22\%, regardless of order. This is not the usual percentage-change formula, and other conventions exist; the denominator must therefore be stated when “percentage difference” is used.

Successive changes and compound growth

Successive percentage changes act on changing bases and therefore combine by multiplication. If a quantity changes by decimal rates r1,r2,…,rnr_1,r_2,\ldots,r_n, its final value is

Vn=V0∏i=1n(1+ri).V_n=V_0\prod_{i=1}^{n}(1+r_i).

The overall percentage change is 100[∏i(1+ri)−1]%100\bigl[\prod_i(1+r_i)-1\bigr]\%. For two changes, the combined decimal rate is r1+r2+r1r2r_1+r_2+r_1r_2. Simply adding the rates omits the product term. Addition can be a useful approximation when both rates are small, but it is not exact.

An increase of 20%20\% followed by a decrease of 20%20\% multiplies the original amount by 1.2×0.8=0.961.2\times0.8=0.96, leaving a net loss of 4%4\%. More generally, equal increases and decreases of magnitude rr produce the factor 1−r21-r^2. Restoring an amount after a fractional loss dd, where 0<d<10<d<1, requires an increase of

d1−d.\frac{d}{1-d}.

A 20%20\% loss therefore requires a 25%25\% gain to recover, and a 50%50\% loss requires a 100%100\% gain. The difference arises because the recovery is calculated from the reduced base.

Repeated growth at a constant rate gives compound growth:

Vn=V0(1+r)n.V_n=V_0(1+r)^n.

For example, 1,000 units growing by 5%5\% per period become 1,157.6251{,}157.625 units after three periods. This contrasts with simple growth, under which each period adds 5%5\% of the original 1,000 units, giving 1,150 after three periods. Compound growth describes repeated proportional change; simple growth describes repeated equal absolute additions.

When beginning and ending values are positive, the constant rate producing the same change over nn equal periods is

r=(VnV0)1/n−1.r=\left(\frac{V_n}{V_0}\right)^{1/n}-1.

This is the compound annual growth rate when periods are years. It summarizes endpoints, not the path between them. A series may have large intervening fluctuations even when its compound growth rate is modest. The arithmetic average of periodic percentage changes generally does not reproduce the final value: gains of 50%50\% and losses of 50%50\% average to zero, yet their combined factor is 0.750.75.

Continuous growth provides another representation. If V(t)=V0ektV(t)=V_0e^{kt}, where ekte^{kt} is an exponential function, the effective change over one time unit is ek−1e^k-1. Conversely, an effective rate r>−1r>-1 corresponds to the continuously compounded rate ln⁡(1+r)\ln(1+r), using the natural logarithm. These rates are close for small changes but are not identical.

Combining percentages across groups

An overall percentage is normally calculated from pooled numerators and denominators, not from an unweighted average of group percentages. If group ii contains nin_i observations and has a proportion pip_i, expressed as a decimal, then the pooled proportion is

p=∑inipi∑ini.p=\frac{\sum_i n_ip_i}{\sum_i n_i}.

This is a weighted mean with group sizes as weights. An unweighted average is correct for the pooled proportion when the groups have equal denominators, or when special numerical circumstances make the two results coincide. Otherwise, it answers a different question.

Suppose one class has 9 successful students out of 10, or 90%90\%, while another has 50 out of 100, or 50%50\%. Averaging the percentages gives 70%70\%, but the combined success rate is 59/11059/110, approximately 53.64%53.64\%. The unweighted figure gives each class equal importance; the pooled figure gives each student equal importance. Neither weighting scheme is intrinsically meaningless, but their interpretations differ and cannot be interchanged.

Changing group composition can also change an aggregate percentage even when every subgroup rate remains constant. If a population shifts toward groups with higher rates of an outcome, the overall rate can rise without any within-group increase. This issue is important in demography, employment statistics, and comparisons involving populations with different age structures. Standardized rates address a particular version of this problem by applying subgroup rates to a common reference composition rather than each population’s actual composition.

A particularly striking aggregation effect is Simpson’s paradox, in which the direction of a comparison within groups reverses when groups are combined. Consider two procedures, A and B. In one stratum, A succeeds in 9 of 10 cases and B in 80 of 100; in another, A succeeds in 20 of 100 and B in 1 of 10. A has the higher percentage in both strata: 90%90\% against 80%80\%, and 20%20\% against 10%10\%. Overall, however, A succeeds in 29 of 110 cases, while B succeeds in 81 of 110. The reversal reflects the very different allocation of cases between strata.

Such arithmetic does not by itself determine which comparison answers a substantive question. That depends on how groups were formed and on the role of relevant variables. In observational research, confounding can make aggregate percentages poor indicators of relationships of interest. An overall share, a standardized rate, and an average subgroup percentage can all be valid quantities while describing different features of the same data.

Percentages in probability and statistical inference

In probability, a percentage expresses the likelihood of an event on a scale from 0 to 100. A probability of 0.020.02 is 2%2\%, but it does not mean that exactly two occurrences must appear in every set of 100 trials. If 100 trials each have success probability 0.020.02, their expected number of successes is 2; the observed count can differ. Under independent trials with a common probability, that count follows a binomial distribution.

An observed percentage is often an estimate rather than a known population value. For a simple random sample with kk successes among nn observations, the sample proportion is p^=k/n\hat p=k/n. Under the elementary independent Bernoulli model, its estimated standard error is

SE⁡(p^)=p^(1−p^)n.\operatorname{SE}(\hat p) =\sqrt{\frac{\hat p(1-\hat p)}{n}}.

Multiplying this result by 100 expresses the standard error in percentage points. For p^=0.50\hat p=0.50 and n=1,000n=1{,}000, it is approximately 1.58 percentage points. A basic normal-approximation 95% confidence interval has a half-width of about 3.1 percentage points, although other interval methods behave better near the boundaries or with small samples. Complex sampling designs require corresponding adjustments.

The interpretation of a statistical percentage depends on its conditioning. The percentage of people with a characteristic who produce a positive test result differs from the percentage of positive results belonging to people with that characteristic. In notation,

P(+∣C)≠P(C∣+)P(+\mid C)\ne P(C\mid +)

in general. Conditional probability and Bayes’ theorem formalize the relationship. Reversing the condition changes the denominator, so a high percentage in one direction does not guarantee a high percentage in the other.

For example, consider a hypothetical population of 10,000 items, of which 1%1\%, or 100, are defective. A screening method flags 90%90\% of defective items and 5%5\% of nondefective items. It therefore flags 90 defective and 495 nondefective items. Among the 585 flagged items, about 15.4%15.4\% are defective. The 90%90\% detection percentage and the 15.4%15.4\% percentage answer different questions.

Other statistical percentages also require careful interpretation. A p-value of 5%5\% is not a statement that a null hypothesis has a 5%5\% probability of being true. Under its statistical model, it describes the probability of a result at least as extreme as the observed result, conditional on the null hypothesis. Likewise, a 95% confidence level describes the long-run coverage of an interval-construction procedure, not the percentage of observations lying inside any particular interval.

Prices, taxation, returns, and economic indicators

Commercial calculations use percentages to express discounts, markups, margins, taxes, and returns, but these quantities often use different bases. A discount is normally measured against the original or listed price. A 15%15\% discount on 200 monetary units reduces the price by 30 units to 170. Successive discounts of 20%20\% and 10%10\% produce a final-price factor of 0.8×0.9=0.720.8\times0.9=0.72, equivalent to a single discount of 28%28\%, not 30%30\%.

Markup and profit margin illustrate the importance of the denominator. If an item costs CC and sells for SS, its markup on cost is (S−C)/C(S-C)/C, whereas its gross margin on selling price is (S−C)/S(S-C)/S. An item costing 80 and selling for 100 has a 25%25\% markup and a 20%20\% gross margin. If uu is the decimal markup and mm the decimal margin, then

m=u1+u,u=m1−m,m=\frac{u}{1+u}, \qquad u=\frac{m}{1-m},

where the denominators are nonzero. These measures describe the same price difference from different reference points.

In taxation, a tax-exclusive rate applies to a pretax base. If NN is the net price and tt the decimal tax rate, the tax-inclusive price is G=N(1+t)G=N(1+t). Recovering the tax from the inclusive price requires

tax=Gt1+t.\text{tax}=G\frac{t}{1+t}.

At a 20%20\% tax-exclusive rate, the tax included in a total price of 120 is 20, or one sixth of the total—not 20%20\% of 120. Similarly, a marginal income-tax rate applies to an additional amount of taxable income, whereas an effective rate compares total tax with a specified measure of total income.

Financial interest rates require a time basis and a compounding convention. A nominal annual rate of 12%12\%, compounded monthly at 1%1\%, gives an effective annual rate of

(1.01)12−1≈12.68%.(1.01)^{12}-1\approx12.68\%.

The same displayed annual percentage can therefore imply different accumulated amounts if compounding conventions differ. Financial returns can also depend on whether distributions, fees, or other cash flows are included, so a percentage return is incomplete without its calculation basis.

Economic statistics frequently compare price or output indices. Inflation of 5%5\% means a specified price index rose by 5%5\% over the stated interval; it does not mean every price rose equally. A fall in inflation from 5%5\% to 3%3\% is a decline of 2 percentage points in the rate, while the price index continues to rise. Converting nominal growth gng_n into real growth with inflation π\pi uses

1+gr=1+gn1+π.1+g_r=\frac{1+g_n}{1+\pi}.

Subtracting inflation from nominal growth is an approximation to this relationship.

Scientific measurement, composition, and numerical precision

In science, a percentage often reports composition, efficiency, or deviation from a reference value. The physical definition of the ratio matters as much as its numerical scale. A mass percentage compares the mass of a component with the total mass of the mixture. If 5 grams of a substance are present in 100 grams of final solution, the mass percentage is 5%5\%. Dissolving 5 grams in 100 grams of solvent instead produces 105 grams of solution and a mass percentage of approximately 4.76%4.76\%.

Volume percentages compare component and mixture volumes under specified conditions, while amount fractions compare amounts of substance, usually measured in moles. These percentages need not coincide because substances differ in density and molar mass. Volumes may also change on mixing. A notation such as “5%” is therefore incomplete when several composition conventions are possible. Some laboratory conventions use mass per volume, such as grams per 100 millilitres; these require an explicit definition because mass divided by volume is not itself dimensionless.

In chemistry, percentage yield compares the amount of product obtained with a theoretical amount calculated from a reaction model:

percentage yield=actual yieldtheoretical yield×100%.\text{percentage yield} =\frac{\text{actual yield}}{\text{theoretical yield}}\times100\%.

An apparent yield above 100%100\% can indicate that the measured material includes solvent, water, or impurities, or that the theoretical or measured quantities are inappropriate. In energy conversion, efficiency compares a defined useful output with a defined input. A heat pump’s coefficient of performance can exceed 1 because it compares transferred heat with work input; it is not an ordinary efficiency claiming that energy has been created.

Percentage error commonly expresses the magnitude of a deviation relative to a reference:

absolute percentage error=∣x−xref∣∣xref∣×100%,\text{absolute percentage error} =\frac{|x-x_{\mathrm{ref}}|}{|x_{\mathrm{ref}}|}\times100\%,

for xref≠0x_{\mathrm{ref}}\ne0. This quantity differs from measurement uncertainty. An error concerns a difference from a reference or true value, whereas uncertainty characterizes the dispersion of values reasonably attributable to the measured quantity. Relative errors and uncertainties can become very large near zero, making absolute units more informative.

Rounding introduces a separate reporting issue. Three exact shares of one third each become 33.3%33.3\% when rounded to one decimal place, summing to 99.9%99.9\%. Such discrepancies do not necessarily indicate missing data. Excessive precision is also misleading: a sample count or measurement uncertainty may not support several decimal places in a reported percentage. In software, floating-point arithmetic can introduce small representation errors, while percentage formatting may display a stored value of 0.250.25 as 25%25\%. Entering 25 into such a cell can instead display 2,500%2{,}500\%.

Historical development and conventions of presentation

The word percent derives from expressions meaning “by the hundred” or “for every hundred,” associated with the Latin per centum and the Italian commercial expression per cento. Its history belongs to the broader development of mercantile arithmetic: merchants needed comparable ways to calculate interest, commissions, discounts, and profits across transactions of different sizes. Expressing a charge per hundred made a rate independent of the absolute size of a transaction.

Calculations involving hundredths predate the modern percent sign. The mathematical operation did not depend on decimal fractions or a standardized symbol: a charge of one unit for every hundred could be calculated using ordinary fractions and proportional reasoning. The spread of the Hindu–Arabic numeral system and written commercial arithmetic in medieval and early modern Europe helped make such computations routine. The Treviso Arithmetic, printed in 1478, is an early example of a printed practical arithmetic concerned with commercial calculation; it belongs to this broader computational tradition rather than marking an invention of percentages.

The modern sign developed gradually from abbreviated commercial notations for per cento. Historical forms used combinations of letters, marks, and small circles before the familiar two circles separated by a stroke became established. This gradual development makes a single inventor or invention date inappropriate. The sign’s meaning is a scaling convention: a displayed number is divided by 100, regardless of whether the underlying quantity is a fraction, a rate of change, or a comparison with a benchmark.

English usage retains both percent, especially in American English, and per cent, especially in British usage. Percentage generally names the mathematical quantity or its mode of expression. In technical typography, conventions associated with the International System of Units place a space between a numerical value and the percent symbol, as in “25 %”; many general editorial styles instead write “25%.” Decimal separators and digit grouping also vary by language and jurisdiction, so the same percentage may be printed as 2.5%2.5\% or 2,5%2{,}5\%.

In tables, charts, and prose, effective presentation identifies the denominator, the population or material concerned, the time interval where relevant, and whether the number is a level or a change. A table headed “percentage of respondents” differs from one headed “percentage of responses” when respondents may select multiple answers. A graph of percentages can emphasize small movements through a truncated vertical scale, while a pie chart implies parts of a common whole. These conventions affect interpretation even when every displayed number is arithmetically correct.