Odds, in probability theory and statistics, represent the ratio of the probability that an event will occur to the probability that it will not occur.[1] This measure is expressed as a pair of numbers, such as a:b, where a denotes the favorable outcomes (or "successes") and b the unfavorable ones (or "failures"), and can be converted to a probability p using the formula p = a / (a + b).[2] Unlike probability, which ranges from 0 to 1, odds can exceed 1 and are particularly useful in contexts where multiplicative effects or ratios are analyzed, such as in logistic regression models.[3]Odds are commonly categorized as odds in favor (favorable to unfavorable) or odds against (unfavorable to favorable), with even odds (1:1) indicating a 50% probability of the event occurring.[4] In practice, odds are presented in various formats to suit different regions and applications: fractional odds (e.g., 5/1, common in the UK for betting), decimal odds (e.g., 6.00, multiplying the stake to yield total payout), and moneyline odds (e.g., +500 or -200, prevalent in the US to indicate profit or stake required).[5]A key application of odds is in gambling and betting, where they quantify the implied probability of an outcome and determine potential payouts, allowing bookmakers to balance risk while incorporating a house edge.[5] For instance, in betting, fractional odds of 3/1 imply a 25% probability (1/(3+1)) and a payout of four times the stake if successful. In statistics and epidemiology, the odds ratio (OR) extends this concept to measure the association between an exposure and an outcome, calculated as the ratio of odds in one group to odds in another; an OR greater than 1 indicates increased odds of the outcome with the exposure.[6] This tool is fundamental in case-control studies and meta-analyses, providing interpretable effect sizes for binary data.[7]
Fundamentals
Definition and Interpretation
In probability theory, odds represent the ratio of the probability that an event will occur to the probability that it will not occur. In discrete cases with a finite number of outcomes, odds are the ratio of the number of favorable outcomes to the number of unfavorable outcomes.[8][9] This is formally expressed as the odds in favor of the event equaling p / (1 - p), where p is the probability of the event occurring.Odds of 1:1, also known as even odds, correspond to a probability of 0.5, signifying an equal likelihood of the event happening or not, as in the outcome of a fair coin flip.[10] Odds greater than 1:1 indicate that the event is more likely than not to occur, while odds less than 1:1 suggest it is less likely.[11]For example, consider a biased coin where the probability of heads is p = 0.75; the odds in favor of heads are then $0.75 / 0.25 = 3, or 3:1, meaning three chances for heads to one against. Conversely, the odds against heads would be 1:3. This distinction highlights "odds in favor" as the ratio favoring success and "odds against" as the inverse ratio.[12]The term "odds" dates from the early 16th century, originally meaning "inequality" or "difference in amount," with the sense related to wagering first appearing around 1597.[13]
Mathematical Relations
The odds o of an event are derived from its probability p as the ratio of the probability of the event occurring to the probability of it not occurring, given by the formulao = \frac{p}{1 - p}.This expression arises directly from the definition of odds as a measure contrasting success and failure outcomes in probability theory.[14] The inverse relation converts odds back to probability:p = \frac{o}{1 + o}.This bidirectional transformation ensures that odds and probabilities are equivalent representations, with odds emphasizing the relative scale between favorable and unfavorable outcomes.[14]A key property of odds emerges in Bayesian updating, where they exhibit a multiplicative structure under independent evidence. In the odds form of Bayes' theorem, the posterior odds of a hypothesis given evidence E equal the prior odds multiplied by the likelihood ratio \Lambda = \frac{P(E \mid H)}{P(E \mid \neg H)}:O(H \mid E) = O(H) \cdot \Lambda.For multiple independent pieces of evidence, the likelihood ratios multiply, allowing the posterior odds to incorporate each factor sequentially by multiplication. This property facilitates the combination of independent observations without requiring direct probability computations.[15][16]Logarithmic odds, or logit, defined as \log o = \log \left( \frac{p}{1 - p} \right), possess additive properties in statistical models such as logistic regression. Here, the log-odds are expressed as a linear combination of predictors:\log o = \beta_0 + \beta_1 x_1 + \cdots + \beta_k x_k,enabling the effects of independent variables to accumulate additively on the log-odds scale, which simplifies parameter estimation and interpretation in generalized linear models.[17]Odds also relate directly to likelihood as the ratio comparing the likelihood of the event to its complement under a uniform prior, where o = \frac{L(H \mid data)}{L(\neg H \mid data)} when priors are equal. This framing positions odds as a likelihood-based measure, integral to inference where evidence updates beliefs multiplicatively via Bayes' rule.[15]
Historical Development
Origins in Early Probability
The concept of odds emerged in the context of 16th- and 17th-century European gambling practices, where games of chance like dice throwing and lotteries prompted early attempts to quantify unequal chances. Dice games, prevalent across Europe, involved calculating the likelihood of specific outcomes, such as the probability of rolling a particular sum or number. For instance, Gerolamo Cardano, in his manuscript Liber de ludo aleae (written around 1525 and published posthumously in 1663), systematically analyzed dice throws, determining that the chance of rolling a double six with two dice is 1 in 36, based on the 36 possible outcomes, and for three dice, the probability of at least one six is 91 in 216.[18] Cardano's work, though not framed in modern probabilistic terms, introduced combinatorial methods to assess these ratios, reflecting practical concerns in gambling where players sought to evaluate risks without a formal theory of probability. Lotteries, introduced in Europe in the 16th century—such as the Genoa lotto in 1530—further embedded these ideas, as they required dividing prizes based on drawn lots, often using rudimentary equiprobable assumptions.[19]The foundational developments in early probability theory, which implicitly incorporated odds-like ratios, arose from the 1654 correspondence between Blaise Pascal and Pierre de Fermat on the "problem of points." This problem addressed fair division of stakes in an interrupted game of chance, such as one where players alternate winning rounds until one reaches a set number of points. Fermat proposed enumerating remaining possible outcomes to compute shares, for example, assigning a player with two points needed out of four a stake ratio of 11:5 based on winning combinations in the remaining throws.[20] Pascal refined this by considering expected values across all scenarios, dividing 64 pistoles as 48 to the leader and 16 to the opponent when two points were needed, using proportional allocations derived from the number of favorable outcomes.[20] Their exchange, prompted by the gambler Chevalier de Méré, laid the groundwork for expectation in unequal chances, treating probabilities as ratios without yet using the term "odds."Christiaan Huygens formalized these ideas in his 1657 treatise De ratiociniis in ludo aleae, the first published book on probability, where he extended the problem of points to various games like dice and tennis. Starting from an axiom that a fair game has equal value to both players, Huygens derived expectations using ratios of favorable to unfavorable cases for interrupted plays. For example, in a dice game interrupted after some throws, he calculated the equitable division by the ratio of ways each player could still win, such as 32:1 for a scenario with 32 favorable outcomes versus one unfavorable.[21] This approach directly built on Pascal and Fermat, applying ratio-based divisions to ensure fairness in gambling disputes.[22]The explicit terminology of "odds" in English betting contexts appeared in the late 16th to early 18th centuries, evolving from its earlier sense of inequality. By the 1590s, "odds" denoted an advantage given to balance unequal chances in wagers, as seen in Shakespeare's Henry IV, Part 2 (1597), where it refers to disproportionate conditions in disputes.[23] In gambling, this shifted to payoff ratios by the 1700s, with phrases like "ten to one" common in English betting on horse races and games, marking the term's integration into probabilistic language for expressing chances.[23]
Evolution in Modern Contexts
The formalization of odds within probability theory began with Jacob Bernoulli's posthumously published Ars Conjectandi in 1713, where he expressed odds as ratios tied to expected values in games of chance, laying groundwork for later refinements in statistical inference.[24] Building on early probability concepts, 19th-century mathematicians like Pierre-Simon Laplace advanced the use of probability ratios—precursors to modern odds—in his Théorie Analytique des Probabilités (1812), applying them to inverse problems and error analysis in astronomical data.[25] By the 1830s, these ideas permeated actuarial science, with practitioners adopting odds-based calculations for life insurance premiums; for instance, the British actuary Charles Babbage and others used probability ratios derived from mortality tables to assess risk odds, enabling the growth of equitable insurance societies.[26]In the 20th century, Ronald A. Fisher elevated odds ratios as a key tool in genetics during the 1920s, employing them in analyses of contingency tables to quantify associations in inheritance patterns, as seen in his foundational work on population genetics. Post-World War II, the application of odds expanded significantly in epidemiology, where odds ratios became central to case-control studies for measuring exposure-disease links, exemplified by Jerome Cornfield's 1959 advocacy for their use in smoking-lung cancer research;[27] concurrently, econometrics integrated odds into discrete choice models, such as logit regressions, to model binary outcomes in economic behavior. These developments underscored odds' versatility beyond gambling, embedding them in empirical sciences for robust inference under uncertainty.Legal frameworks further institutionalized odds in regulated gambling during this period. The UK's Gaming Act 1968 marked a pivotal reform by legalizing off-course betting shops in an expanding commercial sector. In the digital era post-2000, online platforms revolutionized odds through algorithmic computation, enabling dynamic adjustments based on real-time data like betting volume and market events; this shift, powered by machine learning models, has transformed global sports betting into a data-driven industry valued at billions, with firms like Betfair pioneering exchange-based dynamic odds since 2000.
Statistical Applications
Odds Ratios
The odds ratio (OR) is a statistical measure used to quantify the strength of association between an exposure and an outcome by comparing the odds of the outcome occurring in one group versus another.[28] It is particularly common in case-control studies and logistic regression analyses where direct probabilities may not be available.[6]Formally, the odds ratio is defined as the ratio of the odds in the first group to the odds in the second group, where odds are the ratio of the probability of success to the probability of failure.[28] In a 2×2 contingency table with cell counts a (exposed with outcome), b (exposed without outcome), c (unexposed with outcome), and d (unexposed without outcome), the odds ratio is calculated as:\text{OR} = \frac{a/b}{c/d} = \frac{ad}{bc}[28]Equivalently, in terms of outcome probabilities p_1 (for group 1) and p_2 (for group 2), the odds ratio simplifies to:\text{OR} = \frac{p_1 / (1 - p_1)}{p_2 / (1 - p_2)} = \frac{p_1 (1 - p_2)}{p_2 (1 - p_1)}[29]An OR of 1 indicates no association between the exposure and outcome, as the odds are equal across groups.[6] An OR greater than 1 suggests a positive association, where the odds of the outcome are higher in the first group (e.g., exposed), while an OR less than 1 indicates a negative association or protective effect.[28] Confidence intervals for the OR are typically constructed on the logarithmic scale to ensure symmetry, with the standard error approximated as \sqrt{1/a + 1/b + 1/c + 1/d}; for stratified data, the Mantel-Haenszel method provides a pooled estimate and corresponding interval to account for confounding factors.[28][30]For illustration, consider a case-control study examining the association between an environmental exposure and disease occurrence, structured in the following 2×2 contingency table:
The odds of disease among the exposed are 25/10 = 2.5, and among the unexposed are 10/10 = 1; thus, OR = 2.5 / 1 = 2.5 (or equivalently, (25 × 10) / (10 × 10) = 2.5).[28] This indicates that the exposed group has 2.5 times the odds of developing the disease compared to the unexposed group.[6]
Uses in Statistics and Science
In epidemiology, odds ratios serve as a key measure in case-control studies to assess the association between risk factors and disease outcomes. For instance, the seminal 1950 study by Doll and Hill examined the link between smoking and lung cancer by comparing smoking histories among 709 lung cancer patients and 709 matched controls, finding that the odds of smoking were substantially higher among cases, yielding an odds ratio of approximately 14 for heavy smokers, which highlighted smoking as a major risk factor.[31] This approach allows retrospective analysis of rare diseases where cohort studies may be infeasible, enabling estimation of relative risks under certain conditions.[32]In genetics, log-odds ratios are central to genome-wide association studies (GWAS), where they quantify the association between single nucleotide polymorphisms (SNPs) and disease susceptibility by modeling allele frequencies in cases versus controls via logistic regression. This logit transformation of odds facilitates the identification of genetic variants with small effect sizes, as seen in large-scale GWAS for traits like type 2 diabetes, where log-odds ratios around 0.1 correspond to modest per-allele risk increases.[33] The use of log-odds also aids in meta-analyses, combining evidence across studies while accounting for population stratification.[34]Machine learning employs odds from logistic regression for binary classification tasks, where the model outputs log-odds to predict the probability of class membership, interpretable as the change in odds for a unit increase in predictors. For multi-class problems, this extends to softmax regression, which normalizes log-odds across categories to produce a probability distribution summing to one, commonly used in applications like image recognition.[35] The resulting odds ratios provide intuitive effect sizes, such as in credit risk models where a feature's coefficient indicates how it alters default odds.In actuarial science, odds ratios inform insurance premium calculations through logistic models of claim probabilities, approximating mortality or loss ratios for low-probability events like policyholder death. For example, in life insurance experience studies, odds ratios from categorical predictors like age or health status adjust base rates to set equitable premiums, ensuring solvency while reflecting risk heterogeneity.[36]In physics, particularly for rare event modeling in particle detection, odds ratios appear in Bayesian frameworks to evaluate evidence for signals amid background noise, such as in searches for new particles at colliders. The Bayes factor, akin to an odds ratio between hypotheses, quantifies the likelihood of data under a signal-plus-background model versus background alone, crucial for assessing detections like the Higgs boson where false positives must be minimized.[37]A key limitation of odds ratios is their approximation to relative risks, which holds reliably only for rare events (typically incidence below 10%), as higher baseline risks inflate the odds ratio beyond the true risk increase; for common outcomes, direct risk estimation via cohort designs is preferable.[38]
Gambling and Betting
Fractional Odds
Fractional odds, prevalent in the United Kingdom and Ireland, express the potential profit as a ratio of two positive integers separated by a slash, denoted as a/b or a:b, where a is the numerator representing the profit units and b the denominator representing the stake units.[39] This format indicates that for every unit staked, the bettor wins a units of profit if successful, excluding the return of the original stake.[40] The profit is calculated as \text{profit} = \text{stake} \times \frac{a}{b}, while the total return, including the stake, is \text{total return} = \text{stake} \times \left( \frac{a}{b} + 1 \right).[41]In this system, odds greater than 1/1 signify underdogs, offering higher potential returns to reflect lower perceived likelihood of winning; for instance, 3/1 odds on a £1 stake yield £3 profit plus the £1 stake returned, for a total payout of £4.[42]Even money, or 1/1 odds, provides a straightforward double of the stake—£1 profit on a £1 bet, totaling £2—balancing risk and reward without favoritism.[43] A practical example in horse racing involves 5/2 odds: a £2 stake results in £5 profit ($2 \times \frac{5}{2}) plus the £2 stake, yielding £7 totalreturn upon victory.[40]This format originated in the 18th century amid the rise of organized horse racing in Britain, where bookmakers used simple fractional ratios to quote prices, aligning with the era's currency system of pounds, shillings, and pence.[44] It remains a cornerstone of UK and Irish betting culture, particularly in traditional markets like thoroughbred racing, due to its intuitive representation of profit multiples.[45]
Decimal Odds
Decimal odds, also known as European odds, represent the total payout per unit stake, including the original stake, and are expressed as a single decimal number greater than or equal to 1.00.[39] This format is the standard in continental Europe, Australia, New Zealand, and Canada, where it simplifies calculations for bettors compared to other systems.[39][46]The total return on a winning bet is calculated as the stake multiplied by the decimal odds value, while the profit is the stake multiplied by (decimal odds minus 1). For example, with decimal odds of 2.50 on a €1 stake, the total return is €2.50, yielding a €1.50 profit.[47][39] For favorites, such as odds of 1.50 on a €1 stake, the total return is €1.50, resulting in a €0.50 profit.[48] Lower decimal values (closer to 1.00) indicate favorites with smaller payouts, while higher values signify underdogs with larger potential returns.[49]One key advantage of decimal odds is their ease of use in calculating payouts for accumulator bets, also known as parlays, where the overall odds are obtained by simply multiplying the individual decimal odds together.[50] This multiplicative property makes it straightforward to determine combined returns without additional conversions, enhancing efficiency for multi-leg wagers.[47]Decimal odds are the predominant format on betting exchanges like Betfair, particularly for soccer markets, where they facilitate quick back-and-lay trading among users.[51][52]Betfair's Exchange exclusively uses decimal odds to streamline global participation and real-time betting adjustments.[51]
Moneyline Odds
Moneyline odds, commonly referred to as American odds, represent a betting format predominantly utilized in the United States for sports wagering, where the odds directly indicate the potential profit relative to a standardized $100 unit. Positive odds denote underdogs and specify the profit earned on a $100 stake, while negative odds signify favorites and indicate the amount required to wager to secure $100 in profit. This system simplifies quick assessments of risk and reward in straight-win bets, without incorporating point spreads or other adjustments.[39]The payout structure for moneyline odds follows specific formulas based on the sign of the odds. For positive odds (e.g., +200), the profit is calculated as stake multiplied by (odds divided by 100), with the total payout being the stake plus the profit. For negative odds (e.g., -150), the profit is stake multiplied by (100 divided by the absolute value of the odds), again adding the stake to obtain the total payout. These calculations ensure bettors can determine exact returns upfront, accounting for the bookmaker's vigorish embedded in the lines.[53][54]Practical examples illustrate the application: A +300 moneyline on an underdog team means a $50 stake yields a profit of $50 × (300 / 100) = $150, for a total payout of $200 if the team wins. In contrast, a -110 line, frequently seen in near-even matchups, requires a $110 stake to win $100 profit, totaling $210 upon success. Such formats are staples in betting markets for professional sports like the NFL and NBA, where they facilitate straightforward wagers on outright victors.[55][39]A rough conversion to implied probability from moneyline odds distinguishes between signs: for positive odds, 100 / (odds + 100); for negative odds, |odds| / (|odds| + 100). This provides an estimate of the event's likelihood as perceived by the odds without delving into bookmaker adjustments.[56]
Other Formats
Wholesale odds represent the fair or "true" probabilities of an event occurring without incorporating the bookmaker's profit margin, often referred to as a 100% book where the sum of probabilities across all outcomes equals exactly 100%. These internal odds are used by bookmakers to set their public offerings, providing a baseline for pricing bets on high-volume events.[57]Hong Kong odds, a regional format popular in Asian markets, express the net profit per unit staked, equivalent to decimal odds minus one, making them simpler for quick calculations of winnings. For example, Hong Kong odds of 5.0 correspond to decimal odds of 6.0, meaning a $1 bet returns $5 in profit (total $6 including stake). This format omits the stake from the quoted figure, focusing solely on the payout multiplier for the underdog or even-money scenarios, and is favored in high-stakes Asian betting environments for its brevity.[58]Indonesian odds adapt the American moneyline style but scale to unit stakes rather than $100, using negative values for favorites (indicating the stake needed to win one unit) and positive values for underdogs (indicating the profit from one unit staked). A favorite at -1.67 Indonesian odds requires staking 1.67 units to win 1 unit, while an underdog at +2.50 yields 2.50 units profit per unit staked; this format is prevalent in Southeast Asian sportsbooks, emphasizing concise risk assessment similar to moneyline but adjusted for local betting volumes.[59]Probability odds, expressed directly as the ratio of success probability to failure probability (p : 1-p), find niche application in academic betting models for their mathematical convenience in statistical analysis, such as logistic regression and Bayesian inference, allowing precise modeling of event likelihoods without normalization constraints. These odds facilitate advanced simulations in research on market efficiency and risk, prioritizing conceptual probability over commercial payouts.[3]
Relation to Probabilities
Converting Odds to Probabilities
Converting odds to probabilities involves straightforward mathematical formulas that express the likelihood of an event occurring as a proportion between 0 and 1. In probability theory, odds in favor expressed as a ratio o (decimal form, where o = a/b for integer a:b) yield probability p = \frac{o}{1 + o}.[60] This assumes odds represent the ratio of the probability of success to failure.In betting contexts, odds formats imply probabilities differently, often as odds against the event. For fractional odds \frac{a}{b} (common in the UK, where a/b is profit per unit stake, equivalent to odds against a:b), the implied probability is p = \frac{b}{a + b}.[5] For decimal odds o (total payout per unit stake, common in Europe), p = \frac{1}{o}. For moneyline odds, positive +x (underdog) gives p = \frac{100}{100 + x}, and negative -x (favorite) gives p = \frac{|x|}{|x| + 100}.[61]To convert between odds formats, one can first transform to a common intermediary like decimal odds. Decimal odds to fractional odds are obtained by subtracting 1 from the decimal value and expressing the result as a fraction in lowest terms.[5] For moneyline odds, positive values +x convert to decimal odds via \frac{x}{100} + 1, while negative values -x use \frac{100}{|x|} + 1.[61] These conversions facilitate comparisons across formats without altering the underlying probability. Note that in betting, these implied probabilities include the bookmaker's margin and sum to more than 1 across outcomes.The following table illustrates conversions for common betting examples (implied probabilities without vig adjustment):
These values are derived using the formulas above, with probabilities rounded to three decimal places for clarity.[5][61]The inverse process converts a probability back to odds. In general probability terms, for p, the odds in favor are o = \frac{p}{1 - p}, and fractional form p : (1 - p) simplified. For fair betting odds (no vig), decimal odds D = \frac{1}{p}, moneyline is + \frac{1-p}{p} \times 100 if underdog or - \frac{p}{1-p} \times 100 if favorite. For even odds (p = 0.5), this yields general odds of 1:1, betting decimal of 2.00, fractional 1/1, and moneyline +100.[60]
Implied Probabilities and Adjustments
In betting markets, odds provided by bookmakers imply probabilities for each outcome, but these are inflated beyond true probabilities to incorporate the bookmaker's profit margin, known as the vigorish (vig) or overround. The implied probability for a single outcome is derived from the odds—specifically, for decimal odds o, it is $1 / o—representing the bookmaker's assessment of the event's likelihood adjusted for their advantage. When these implied probabilities are summed across all mutually exclusive outcomes in a market, the total exceeds 100%, ensuring the bookmaker's long-term profitability regardless of the result. This overround varies by sport, market type, and competition level, typically ranging from 2% to 10% in efficient markets.[62]For a simple two-outcome market, such as a tennis match or a two-way bet without a draw, the overround can be explicitly calculated. The formula for the overround r is:r = \frac{1}{o_1} + \frac{1}{o_2} - 1where o_1 and o_2 are the decimal odds for each outcome. For example, if both outcomes are priced at 1.95 decimal odds, the implied probabilities are each approximately 51.28% ($1 / 1.95), summing to 102.56% and yielding an overround of 2.56%. In American moneyline format, a common balanced line of -110 on both sides converts to decimal odds of about 1.91 ($1 + 100/110), with implied probabilities of 52.38% each, totaling 104.76% and a vig of 4.76%. This structure guarantees the bookmaker collects more in total stakes than it pays out over many bets.[63][64]Bookmakers adjust fair odds—those reflecting true probabilities without margin—by shortening them to build in the vig. The relationship is expressed as bookmaker odds o_b = o_f / (1 + m), where o_f is the fair decimal odds and m is the margin (overround as a decimal). Equivalently, fair probabilities are obtained by normalizing the implied probabilities to sum to 100%: fair probability for outcome i = (implied probability _i ) / (total implied probabilities). This adjustment reveals the "no-vig" or true odds, helping bettors identify value. If the summed implied probabilities from odds across multiple bookmakers fall below 100%, an arbitrage opportunity arises, enabling guaranteed profit by wagering proportionally on all outcomes, though bookmakers actively monitor and limit such inefficiencies.[5][65][66]In multi-outcome markets like soccer's three-way (home win, draw, away win), the overround is the excess over 100% in the sum of implied probabilities. For a match with decimal odds of 2.00 (home win), 4.00 (draw), and 3.00 (away win), the implied probabilities are 50%, 25%, and 33.33%, respectively, summing to 108.33% and indicating an 8.33% margin. This example illustrates how bookmakers distribute the vig across outcomes, often higher in less liquid markets to account for risk.[67]