| Title: | Monotonic Association on Zero-Inflated Data |
| Version: | 0.0.2 |
| Author: | Alice Albasi [aut, cre] |
| Maintainer: | Alice Albasi <albasialice@gmail.com> |
| Description: | Methods for calculating and testing the significance of pairwise monotonic association from and based on the work of Pimentel (2009) <doi:10.4135/9781412985291.n2>. Computation of association of vectors from one or multiple sets can be performed in parallel thanks to the packages 'foreach' and 'doMC'. |
| Depends: | R (≥ 3.3.0) |
| Imports: | foreach |
| License: | GPL-3 |
| Encoding: | UTF-8 |
| RoxygenNote: | 6.0.1 |
| Suggests: | doMC, gamlss.dist, knitr, testthat, R.rsp, rmarkdown |
| VignetteBuilder: | R.rsp, knitr |
| NeedsCompilation: | no |
| Packaged: | 2022-05-08 20:22:25 UTC; alicealbasi |
| Repository: | CRAN |
| Date/Publication: | 2022-05-09 07:10:07 UTC |
Associate pairwise vectors form one or two sets
Description
Given two matrices m_1 and m_2, computes all pairwise correlations of each
vector in m_1 with each vector in m_2. Thanks to the package foreach,
computation can be done in parallel using the desired number of cores.
Usage
associate(m1, m2, parallel = FALSE, n_cor = 1, estimator = "values", d1,
d2, p11 = 0, p01 = 0, p10 = 0)
Arguments
m1, m2 |
matrices whose columns are to be correlated. If no estimation calculations are needed, default is NA. |
parallel |
should the computations for associating the matrices be done in parallel? Default is FALSE |
n_cor |
number of cores to be used if the computation is run in parallel. Default is 1 |
estimator |
string indicating how the parameters |
d1, d2 |
sets of vectors used to estimate |
p11 |
probability that a bivariate observation is of the type (m,n), where m,n>0. |
p01 |
probability that a bivariate observation is of the type (0,n), where n>0. |
p10 |
probability that a bivariate observation is of the type (n,0), where n>0. |
Details
To find pairwise monotonic associations of vectors within one set m, run
associate(m,m). Note that the values on the diagonal will not be necessarely
1 if the vectors contain 0's, as it can be seen by the formula p_{11}^2 t_{11} + 2 * (p_{00} p_{11} - p_{01} p_{10})
Value
matrix of correlation values.
Examples
v1=c(0,0,10,0,0,12,2,1,0,0,0,0,0,1)
v2=c(0,1,1,0,0,0,1,1,64,3,4,2,32,0)
associate(v1,v2)
m1=matrix(c(0,0,10,0,0,12,2,1,0,0,0,0,0,1,1,64,3,4,2,32,0,0,43,54,3,0,0,3,20,1),6)
associate(m1,m1)
m2=matrix(c(0,1,1,0,0,0,1,1,64,3,4,2,32,0,0,43,54,3,0,0,3,20,10,0,0,12,2,1,0,0),6)
associate(m1,m2)
p-value for Pimentel's tau_b
Description
Computes an estimated p-value for Kendall's Tau_b for zero inflated continuous data as in Pimentel(2009).
Usage
calc_p(x, y, estimator = "values", p11 = 0, p01 = 0, p10 = 0)
Arguments
x, y |
vectors to be correlated. Must be numeric. |
estimator |
string indicating how the parameters $p_11$, $p_01$, $p_10$, $p_00$ are to be estimated. The default is 'values', which indicates that they are estimated based on the entries of x and y. If estimates=='mean', each $p_ji$ is estimated as the mean of all pairs of column vectors in m1, and of m2 if needed. If estimates=='own', the $p_ji$'s must be given as arguments. |
p11 |
probability that a bivariate observation is of the type (m,n), where m,n>0 |
p01 |
probability that a bivariate observation is of the type (0,n), where n>0 |
p10 |
probability that a bivariate observation is of the type (n,0), where n>0 |
Value
p-value of correlation.
combine
Description
Designed to combine the matrix of correlation values with the matrix of p-values so that in the cases when the null hypothesis cannot be rejected with a level of confidence indicated by the significance, the correlation is set to zero. Thanks to the package foreach, computation can be done in parallel using the desired number of cores.
Usage
combine(m1, m2, sl = 0.05, parallel = FALSE, n_cor = 1,
estimator = "values", d1, d2, p11 = 0, p01 = 0, p10 = 0)
Arguments
m1, m2 |
matrices whose columns are to be correlated. If no estimation calculations are needed, default is NA. |
sl |
level of significance for testing the null hypothesis. Default is 0.05. |
parallel |
should the computations for associating the matrices be done in parallel? Default is FALSE |
n_cor |
number of cores to be used if the computation is run in parallel. Default is 1 |
estimator |
string indicating how the parameters |
d1, d2 |
sets of vectors used to estimate |
p11 |
probability that a bivariate observation is of the type (m,n), where m,n>0. |
p01 |
probability that a bivariate observation is of the type (0,n), where n>0. |
p10 |
probability that a bivariate observation is of the type (n,0), where n>0. |
Details
To test pairwise monotonic associations of vectors within one set m, run
combine(m,m). Note that the values on the diagonal will not be necessarily
significant if the vectors contain 0's, as it can be seen by the formula
p_{11}^2 t_{11} + 2 * (p_{00} p_{11} - p_{01} p_{10}). The formula for the
variance of the estimator proposed by Pimentel(2009) does not apply in case
p_{11}, p_{01},p_{10}, p_{00} attain the values 0 or 1. In these cases the R
function cor.test is used. Note that while independence implies that the
estimator is 0, if the estimator is 0, it does not imply that the vectors are
independent.
Value
matrix of combined association values and p-values.
no_null
Description
computes association if significance level excludes the null hypothesis
Usage
no_null(x, y, sl, estimator = "values", p11 = 0, p01 = 0, p10 = 0)
Arguments
x, y |
vectors to be correlated. Must be numeric. |
sl |
level of significance for testing the null hypothesis. Default is 0.05. |
estimator |
string indicating how the parameters $p_11$, $p_01$, $p_10$, $p_00$ are to be estimated. The default is 'values', which indicates that they are estimated based on the entries of x and y. If estimates=='mean', each $p_ji$ is estimated as the mean of all pairs of column vectors in m1, and of m2 if needed. If estimates=='own', the $p_ji$'s must be given as arguments. |
p11 |
probability that a bivariate observation is of the type (m,n), where m,n>0. |
p01 |
probability that a bivariate observation is of the type (0,n), where n>0. |
p10 |
probability that a bivariate observation is of the type (n,0), where n>0. |
Value
correlation value if significantly different from 0 or 0 otherwise.
p_01 estimator
Description
computes estimate of parameter p_01 based on sample proportions.
Usage
prop_01(x, y)
Arguments
x, y |
vectors to be correlated. Must be numeric. |
Value
p_01 estimator
p_10 estimator
Description
computes estimate of parameter p_01 based on sample proportions.
Usage
prop_10(x, y)
Arguments
x, y |
vectors to be correlated. Must be numeric. |
Value
p_10 estimator
p_11 estimator
Description
computes estimate of parameter p_11 based on sample proportions.
Usage
prop_11(x, y)
Arguments
x, y |
vectors to be correlated. Must be numeric. |
Value
p_11 estimator
Pimentel's tau_b
Description
Computes the estimator for Kendall's tau_b for zero inflated continuous data proposed by Pimentel(2009).
Usage
tau_p(x, y, estimator = "values", p11 = 0, p01 = 0, p10 = 0)
Arguments
x, y |
vectors to be correlated. Must be numeric and have the same length. |
estimator |
string indicating how the parameters $p_11$, $p_01$, $p_10$ are to be estimated. The default is 'values', which indicates that they are estimated based on the entries of x and y. If estimates=='own', the $p_ji$'s must be given as arguments. |
p11 |
probability that a bivariate observation is of the type (m,n), where m,n>0. Default is 0. |
p01 |
probability that a bivariate observation is of the type (0,n), where n>0.Default is 0. |
p10 |
probability that a bivariate observation is of the type (n,0), where n>0.Default is 0. |
Value
correlation values
test_associations
Description
To test pairwise monotonic associations of vectors within one set m, run
test_associations(m,m). Note that the values on the diagonal will not be
necessarily significant if the vectors contain 0's, as it can be seen by the
formula p_{11}^2 t_{11} + 2 * (p_{00} p_{11} - p_{01} p_{10}). The formula for the
variance of the estimator proposed by Pimentel(2009) does not apply in case
p_{11}, p_{00}, p_{01}, p_{10} attain the values 0 or 1. In these cases the R
function cor.test is used. Note that while independence implies that the
estimator is 0, the estimator being 0 does not imply that the vectors are
independent.
Usage
test_associations(m1, m2, parallel = FALSE, n_cor = 1,
estimator = "values", d1, d2, p11 = 0, p01 = 0, p10 = 0)
Arguments
m1, m2 |
matrices whose columns are used to estimate the |
parallel |
should the computations for combiing the matrices be done in parallel? Default is FALSE. |
n_cor |
number of cores to be used if the computation is run in parallel. Default is 1. |
estimator |
string indicating how the parameters |
d1, d2 |
sets of vectors used to estimate |
p11 |
probability that a bivariate observation is of the type (m,n), where m,n>0 |
p01 |
probability that a bivariate observation is of the type (0,n), where n>0. |
p10 |
probability that a bivariate observation is of the type (n,0), where n>0. |
Details
Given two matrices m_1 and m_2, computes all pairwise correlations of each
vector in m_1 with each vector in m_2. Thanks to the package foreach,
computation can be done in parallel using the desired number of cores.
Value
matrix of p-values of association.
Examples
v1=c(0,0,10,0,0,12,2,1,0,0,0,0,0,1)
v2=c(0,1,1,0,0,0,1,1,64,3,4,2,32,0)
test_associations(v1,v2)
m1=matrix(c(0,0,10,0,0,12,2,1,0,0,0,0,0,1,1,64,3,4,2,32,0,0,43,54,3,0,0,3,20,1),6)
test_associations(m1,m1)
m2=matrix(c(0,1,1,0,0,0,1,1,64,3,4,2,32,0,0,43,54,3,0,0,3,20,10,0,0,12,2,1,0,0),6)
test_associations(m1,m2)
m3= matrix(abs(rnorm(36)),6)
m4= matrix(abs(rnorm(36)),6)
test_associations(m3,m4)