A ratio distribution (also known as a quotient distribution) is a probability distribution constructed as the distribution of the ratio of random variables having two other known distributions.
Given two (usually independent) random variables X and Y, the distribution of the random variable Z that is formed as the ratio Z = X/Y is a ratio distribution.
An example is the Cauchy distribution (also called the normal ratio distribution), which comes about as the ratio of two normally distributed variables with zero mean.
Two other distributions often used in test-statistics are also ratio distributions:
the t-distribution arises from a Gaussian random variable divided by an independent chi-distributed random variable,
while the F-distribution originates from the ratio of two independent chi-squared distributed random variables.
More general ratio distributions have been considered in the literature.[1][2][3][4][5][6][7][8][9]
Often the ratio distributions are heavy-tailed, and it may be difficult to work with such distributions and develop an associated statistical test.
A method based on the median has been suggested as a "work-around".[10]
The ratio is one type of algebra for random variables:
Related to the ratio distribution are the product distribution, sum distribution and difference distribution. More generally, one may talk of combinations of sums, differences, products and ratios.
Many of these distributions are described in Melvin D. Springer's book from 1979 The Algebra of Random Variables.[8]
The algebraic rules known with ordinary numbers do not apply for the algebra of random variables.
For example, if a product is C = AB and a ratio is D=C/A it does not necessarily mean that the distributions of D and B are the same.
Indeed, a peculiar effect is seen for the Cauchy distribution: The product and the ratio of two independent Cauchy distributions (with the same scale parameter and the location parameter set to zero) will give the same distribution.[8]
This becomes evident when regarding the Cauchy distribution as itself a ratio distribution of two Gaussian distributions of zero means: Consider two Cauchy random variables, and each constructed from two Gaussian distributions and then
where . The first term is the ratio of two Cauchy distributions while the last term is the product of two such distributions.
A way of deriving the ratio distribution of from the joint distribution of the two other random variables X , Y , with joint pdf , is by integration of the following form[3]
If the two variables are independent then and this becomes
This may not be straightforward. By way of example take the classical problem of the ratio of two standard Gaussian samples. The joint pdf is
Defining we have
Using the known definite integral we get
which is the Cauchy distribution, or Student's t distribution with n = 1
The Mellin transform has also been suggested for derivation of ratio distributions.[8]
In the case of positive independent variables, proceed as follows. The diagram shows a separable bivariate distribution which has support in the positive quadrant and we wish to find the pdf of the ratio . The hatched volume above the line represents the cumulative distribution of the function multiplied with the logical function . The density is first integrated in horizontal strips; the horizontal strip at height y extends from x = 0 to x = Ry and has incremental probability .
Secondly, integrating the horizontal strips upward over all y yields the volume of probability above the line
Finally, differentiate with respect to to get the pdf .
From Mellin transform theory, for distributions existing only on the positive half-line , we have the product identity provided are independent. For the case of a ratio of samples like , in order to make use of this identity it is necessary to use moments of the inverse distribution. Set such that .
Thus, if the moments of and can be determined separately, then the moments of can be found. The moments of are determined from the inverse pdf of , often a tractable exercise. At simplest, .
In the Product distribution section, and derived from Mellin transform theory (see section above), it is found that the mean of a product of independent variables is equal to the product of their means. In the case of ratios, we have
which, in terms of probability distributions, is equivalent to
Note that i.e.,
The variance of a ratio of independent variables is
When X and Y are independent and have a Gaussian distribution with zero mean, the form of their ratio distribution is a Cauchy distribution.
This can be derived by setting then showing that has circular symmetry. For a bivariate uncorrelated Gaussian distribution we have
If is a function only of r then is uniformly distributed on with density so the problem reduces to finding the probability distribution of Z under the mapping
We have, by conservation of probability
and since
and setting we get
There is a spurious factor of 2 here. Actually, two values of spaced by map onto the same value of z, the density is doubled, and the final result is
When either of the two Normal distributions is non-central then the result for the distribution of the ratio is much more complicated and is given below in the succinct form presented by David Hinkley.[6] The trigonometric method for a ratio does however extend to radial distributions like bivariate normals or a bivariate Student t in which the density depends only on radius . It does not extend to the ratio of two independent Student t distributions which give the Cauchy ratio shown in a section below for one degree of freedom.
In the absence of correlation , the probability density function of the two normal variables X = N(μX, σX2) and Y = N(μY, σY2) ratio Z = X/Y is given exactly by the following expression, derived in several sources:[6]
Under several assumptions (usually fulfilled in practical applications), it is possible to derive a highly accurate solid approximation to the PDF. Main benefits are reduced formulae complexity, closed-form CDF, simple defined median, well defined error management, etc... For the sake of simplicity let's introduce parameters: , and . Then so called solid approximation to the uncorrelated noncentral normal ratio PDF is expressed by equation [11]
Under certain conditions, a normal approximation is possible, with variance:[12]
A transformation to the log domain was suggested by Katz(1978) (see binomial section below). Let the ratio be
.
Take logs to get
Since then asymptotically
Alternatively, Geary (1930) suggested that
has approximately a standard Gaussian distribution:[1]
This transformation has been called the Geary–Hinkley transformation;[7] the approximation is good if Y is unlikely to assume negative values, basically .
This is developed by Dale (Springer 1979 problem 4.28) and Hinkley 1969. Geary showed how the correlated ratio could be transformed into a near-Gaussian form and developed an approximation for dependent on the probability of negative denominator values being vanishingly small. Fieller's later correlated ratio analysis is exact but care is needed when combining modern math packages with verbal conditions in the older literature. Pham-Ghia has exhaustively discussed these methods. Hinkley's correlated results are exact but it is shown below that the correlated ratio condition can also be transformed into an uncorrelated one so only the simplified Hinkley equations above are required, not the full correlated ratio version.
Let the ratio be:
in which are zero-mean correlated normal variables with variances and have means
Write such that become uncorrelated and has standard deviation
The ratio:
is invariant under this transformation and retains the same pdf.
The term in the numerator appears to be made separable by expanding:
to get
in which and z has now become a ratio of uncorrelated non-central normal samples with an invariant z-offset (this is not formally proven, though appears to have been used by Geary),
Finally, to be explicit, the pdf of the ratio for correlated variables is found by inputting the modified parameters and into the Hinkley equation above which returns the pdf for the correlated ratio with a constant offset on .
Contours of the correlated bivariate Gaussian distribution (not to scale) giving ratio x/y
pdf of the Gaussian ratio z and a simulation (points) for
The figures above show an example of a positively correlated ratio with in which the shaded wedges represent the increment of area selected by given ratio which accumulates probability where they overlap the distribution. The theoretical distribution, derived from the equations under discussion combined with Hinkley's equations, is highly consistent with a simulation result using 5,000 samples. In the top figure it is clear that for a ratio the wedge has almost bypassed the main distribution mass altogether and this explains the local minimum in the theoretical pdf . Conversely as moves either toward or away from one the wedge spans more of the central mass, accumulating a higher probability.
The ratio of correlated zero-mean circularly symmetric complex normal distributed variables was determined by Baxley et al.[13] and has since been extended to the nonzero-mean and nonsymmetric case.[14] In the correlated zero-mean case, the joint distribution of x, y is
where
is an Hermitian transpose and
The PDF of is found to be
In the usual event that we get
Further closed-form results for the CDF are also given.
The graph shows the pdf of the ratio of two complex normal variables with a correlation coefficient of . The pdf peak occurs at roughly the complex conjugate of a scaled down .
The ratio of independent or correlated log-normals is log-normal. This follows, because if and are log-normally distributed, then and are normally distributed. If they are independent or their logarithms follow a bivarate normal distribution, then the logarithm of their ratio is the difference of independent or correlated normally distributed random variables, which is normally distributed.[note 1]
This is important for many applications requiring the ratio of random variables that must be positive, where joint distribution of and is adequately approximated by a log-normal. This is a common result of the multiplicative central limit theorem, also known as Gibrat's law, when is the result of an accumulation of many small percentage changes and must be positive and approximately log-normally distributed.[15]
If two independent random variables, X and Y each follow a Cauchy distribution with median equal to zero and shape factor
then the ratio distribution for the random variable is[16]
This distribution does not depend on and the result stated by Springer[8] (p158 Question 4.6) is not correct.
The ratio distribution is similar to but not the same as the product distribution of the random variable :
More generally, if two independent random variables X and Y each follow a Cauchy distribution with median equal to zero and shape factor and respectively, then:
The ratio distribution for the random variable is[16]
If X has a standard normal distribution and Y has a standard uniform distribution, then Z = X / Y has a distribution known as the slash distribution, with probability density function
where φ(z) is the probability density function of the standard normal distribution.[17]
defines , Fisher's F density distribution, the PDF of the ratio of two Chi-squares with m, n degrees of freedom.
The CDF of the Fisher density, found in F-tables is defined in the beta prime distribution article.
If we enter an F-test table with m = 3, n = 4 and 5% probability in the right tail, the critical value is found to be 6.59. This coincides with the integral
For gamma distributionsU and V with arbitrary shape parametersα1 and α2 and their scale parameters both set to unity, that is, , where , then
If , then . Note that here θ is a scale parameter, rather than a rate parameter.
If , then by rescaling the parameter to unity we have
which includes the regular gamma, chi, chi-squared, exponential, Rayleigh, Nakagami and Weibull distributions involving fractional powers. Note that here a is a scale parameter, rather than a rate parameter; d is a shape parameter.
In the ratios above, Gamma samples, U, V may have differing sample sizes but must be drawn from the same distribution with equal scaling .
In situations where U and V are differently scaled, a variables transformation allows the modified random ratio pdf to be determined. Let where arbitrary and, from above, .
Rescale V arbitrarily, defining
We have and substitution into Y gives
Transforming X to Y gives
Noting we finally have
Thus, if and then is distributed as with
The distribution of Y is limited here to the interval [0,1]. It can be generalized by scaling such that if then
Then is approximately normally distributed with mean and variance .
The binomial ratio distribution is of significance in clinical trials: if the distribution of T is known as above, the probability of a given ratio arising purely by chance can be estimated, i.e. a false positive trial. A number of papers compare the robustness of different approximations for the binomial ratio.[citation needed]
In the ratio of Poisson variables R = X/Y there is a problem that Y is zero with finite probability so R is undefined. To counter this, consider the truncated, or censored, ratio R' = X/Y' where zero sample of Y are discounted. Moreover, in many medical-type surveys, there are systematic problems with the reliability of the zero samples of both X and Y and it may be good practice to ignore the zero samples anyway.
The probability of a null Poisson sample being , the generic pdf of a left truncated Poisson distribution is
which sums to unity. Following Cohen,[21] for n independent trials, the multidimensional truncated pdf is
and the log likelihood becomes
On differentiation we get
and setting to zero gives the maximum likelihood estimate
Note that as then so the truncated maximum likelihood estimate, though correct for both truncated and untruncated distributions, gives a truncated mean value which is highly biassed relative to the untruncated one. Nevertheless it appears that is a sufficient statistic for since depends on the data only through the sample mean in the previous equation which is consistent with the methodology of the conventional Poisson distribution.
Absent any closed form solutions, the following approximate reversion for truncated is valid over the whole range .
which compares with the non-truncated version which is simply . Taking the ratio is a valid operation even though may use a non-truncated model while has a left-truncated one.
Then substituting from the equation above, we get Cohen's variance estimate
The variance of the point estimate of the mean , on the basis of n trials, decreases asymptotically to zero as n increases to infinity. For small it diverges from the truncated pdf variance in Springael[22] for example, who quotes a variance of
for n samples in the left-truncated pdf shown at the top of this section. Cohen showed that the variance of the estimate relative to the variance of the pdf, , ranges from 1 for large (100% efficient) up to 2 as approaches zero (50% efficient).
These mean and variance parameter estimates, together with parallel estimates for X, can be applied to Normal or Binomial approximations for the Poisson ratio. Samples from trials may not be a good fit for the Poisson process; a further discussion of Poisson truncation is by Dietz and Bohning[23] and there is a Zero-truncated Poisson distribution Wikipedia entry.
This distribution is the ratio of two Laplace distributions.[24] Let X and Y be standard Laplace identically distributed random variables and let z = X / Y. Then the probability distribution of z is
Let the mean of the X and Y be a. Then the standard double Lomax distribution is symmetric around a.
This distribution has an infinite mean and variance.
If Z has a standard double Lomax distribution, then 1/Z also has a standard double Lomax distribution.
The standard Lomax distribution is unimodal and has heavier tails than the Laplace distribution.
is proportional to the product of independent F random variables. In the case where X and Y are from independent standardized Wishart distributions then the ratio
In relation to Wishart matrix distributions if is a sample Wishart matrix and vector is arbitrary, but statistically independent, corollary 3.2.9 of Muirhead[26] states
The discrepancy of one in the sample numbers arises from estimation of the sample mean when forming the sample covariance, a consequence of Cochran's theorem. Similarly
^Note, however, that and can be individually log-normally distributed without having a bivariate log-normal distribution. As of 2022-06-08 the Wikipedia article on "Copula (probability theory)" includes a density and contour plot of two Normal marginals joint with a Gumbel copula, where the joint distribution is not bivariate normal.
^Fieller, E. C. (November 1932). "The Distribution of the Index in a Normal Bivariate Population". Biometrika. 24 (3/4): 428–440. doi:10.2307/2331976. JSTOR2331976.
^ abPham-Gia, T.; Turkkan, N.; Marchand, E. (2006). "Density of the Ratio of Two Normal Random Variables and Applications". Communications in Statistics – Theory and Methods. 35 (9). Taylor & Francis: 1569–1591. doi:10.1080/03610920600683689. S2CID120891296.
^Díaz-Francés, Eloísa; Rubio, Francisco J. (2012-01-24). "On the existence of a normal approximation to the distribution of the ratio of two independent normal random variables". Statistical Papers. 54 (2). Springer Science and Business Media LLC: 309–323. doi:10.1007/s00362-012-0429-2. ISSN0932-5026. S2CID122038290.
^Of course, any invocation of a central limit theorem assumes suitable, commonly met regularity conditions, e.g., finite variance.
^ abcKermond, John (2010). "An Introduction to the Algebra of Random Variables". Mathematical Association of Victoria 47th Annual Conference Proceedings – New Curriculum. New Opportunities. The Mathematical Association of Victoria: 1–16. ISBN978-1-876949-50-1.
^"SLAPPF". Statistical Engineering Division, National Institute of Science and Technology. Retrieved 2009-07-02.
^Hamedani, G. G. (Oct 2013). "Characterizations of Distribution of Ratio of Rayleigh Random Variables". Pakistan Journal of Statistics. 29 (4): 369–376.
^Katz D. et al.(1978) Obtaining confidence intervals for the risk ratio in cohort studies. Biometrics 34:469–474
^Cohen, A Clifford (June 1960). "Estimating the Parameter in a Conditional Poisson Distribution". Biometrics. 60 (2): 203–211. doi:10.2307/2527552. JSTOR2527552.