MvtInvBeta: Multivariate Inverted Beta Distribution

Description

Calculation of density function, cumulative distribution function, equicoordinate quantile function and survival function, and random numbers generation for multivariate inverted beta distribution with a scalar parameter parm1 and a vector of parameters parm2.

Usage

dmvinvbeta(x, parm1 = 1, parm2 = rep(1, k), log = FALSE)
pmvinvbeta(
  q,
  parm1 = 1,
  parm2 = rep(1, k),
  algorithm = c("numerical", "MC"),
  nsim = 1e+07
)
qmvinvbeta(
  p,
  parm1 = 1,
  parm2 = rep(1, k),
  interval = c(1e-08, 1e+08),
  algorithm = c("numerical", "MC"),
  nsim = 1e+06
)
rmvinvbeta(n, parm1 = 1, parm2 = rep(1, k))
smvinvbeta(
  q,
  parm1 = 1,
  parm2 = rep(1, k),
  algorithm = c("numerical", "MC"),
  nsim = 1e+07
)

Arguments

vector or matrix of quantiles. If $x$ is a matrix, each row vector constitutes a vector of quantiles for which the density $f(x)$ is calculated (for $i$-th row $x_i$, $f(x_i)$ is reported).

parm1

a scalar parameter, see parameter $a$ in Details.

parm2

a vector of parameters, see parameter $l_i$ in Details.

log

logical; if TRUE, probability densities $f$ are given as $log(f)$.

a vector of quantiles.

algorithm

method to be used for calculating cumulative probability. Two options are provided as (i) numerical using adaptive multivariate integral and (ii) MC using Monte Carlo method. Recommend algorithm numerical for $(k <= 4)$ dimension and MC for $(k > 4)$ dimension based on running time consumption. Default option is set as numerical.

nsim

number of simulations used in algorithm MC.

a scalar value corresponding to probability.

interval

a vector containing the end-points of the interval to be searched. Default value is set as c(1e-8, 1e8).

number of observations.

dimension of data or number of variates.

Value

dmvinvbeta gives the numerical values of the probability density.

pmvinvbeta gives a list of two items:

$\quad$ value cumulative probability

$\quad$ error the estimated relative error by algorithm = "numerical" or the estimated standard error by algorithm = "MC"

qmvinvbeta gives the equicoordinate quantile. NaN is returned for no solution found in the given interval. The result is seed dependent if Monte Carlo algorithm is chosen (algorithm = "MC").

rmvinvbeta generates random numbers.

smvinvbeta gives a list of two items:

$\quad$ value the value of survial function

$\quad$ error the estimated relative error by algorithm = "numerical" or the estimated standard error by algorithm = "MC"

Details

Multivariate inverted beta distribution is an alternative expression of multivariate F distribution and is a special case of multivariate Lomax distribution (Balakrishnan and Lai, 2009). Its probability density is given as $$f(x_1, \cdots, x_p) = \frac{\Gamma(\sum_{i=1}^{p} l_i + a) \prod_{i=1}^{p} x_i^{l_i-1}}{\Gamma(a) [\prod_{i=1}^{p} \Gamma(l_i)] (1+\sum_{i=1}^{p} x_i)^{\sum_{i=1}^{p} l_i + a}},$$ where $x_i>0, a>0, l_i>0, i=1,\cdots, p$.

Cumulative distribution function $F(x_1, \dots, x_k)$ is obtained by multiple integral $$F(x_1, \dots, x_k) = \int_{0}^{x_1} \cdots \int_{0}^{x_k} f(y_1, \cdots, y_k) dy_k \cdots dy_1.$$ This multiple integral is calculated by either adaptive multivariate integration using hcubature in package cubature (Narasimhan et al., 2018) or via Monte Carlo method.

Equicoordinate quantile is obtained by solving the following equation for $q$ through the built-in one dimension root finding function uniroot: $$\int_{0}^{q} \cdots \int_{0}^{q} f(x_1, \cdots, x_k) dx_k \cdots dx_1 = p,$$ where $p$ is the given cumulative probability.

The survival function $\bar{F}(x_1, \cdots, x_k)$ is obtained either by the following formula related to cumulative distribution function $F(x_1, \dots, x_k)$ (Joe, 1997) $$\bar{F}(x_1, \cdots, x_k) = 1 + \sum_{S \in \mathcal{S}} (-1)^{|S|} F_S(x_j, j \in S),$$ or via Monte Carlo method.

Random numbers $X_1, \cdots, X_k$ from multivariate inverted beta distribution can be generated through parameter substitutions in simulation of generalized multivariate Lomax distribution by letting $\theta_i = 1, i = 1, \cdots, k$.

References

Balakrishnan, N. and Lai, C. (2009). Continuous Bivariate Distributions. 2nd Edition. New York: Springer.

Joe, H. (1997). Multivariate Models and Dependence Concepts. London: Chapman & Hall.

Narasimhan, B., Koller, M., Johnson, S. G., Hahn, T., Bouvier, A., Ki<U+00EA>u, K. and Gaure, S. (2018). cubature: Adaptive Multivariate Integration over Hypercubes. R package version 2.0.3.

Nayak, T. K. (1987). Multivariate Lomax Distribution: Properties and Usefulness in Reliability Theory. Journal of Applied Probability, Vol. 24, No. 1, 170-177.

Examples

Run this code

# NOT RUN {
# Calculations for the multivariate inverted beta with parameters:
# a = 7, l1 = 1, l2 = 3
# Vector of quantiles: c(2, 4)

dmvinvbeta(x = c(2, 4), parm1 = 7, parm2 = c(1, 3)) # Density

# Cumulative Probability using adaptive multivariate integral
pmvinvbeta(q = c(2, 4), parm1 = 7, parm2 = c(1, 3))

# }
# NOT RUN {
# Cumulative Probability using Monte Carlo method
pmvinvbeta(q = c(2, 4), parm1 = 7, parm2 = c(1, 3), algorithm = "MC")
# }
# NOT RUN {
# }
# NOT RUN {
# Equicoordinate quantile of cumulative probability 0.5
qmvinvbeta(p = 0.5, parm1 = 7, parm2 = c(1, 3))
# }
# NOT RUN {
# Random numbers generation with sample size 100
rmvinvbeta(n = 100, parm1 = 7, parm2 = c(1, 3))

smvinvbeta(q = c(2, 4), parm1 = 7, parm2 = c(1, 3)) # Survival function

# }

Run the code above in your browser using DataLab