Type: Package
Title: Record Linkage for Empirically Motivated Priors
Version: 1.1.0
Depends: R (≥ 3.0.2), stringdist, plyr
Imports: stats, utils
Suggests: knitr, rmarkdown
Encoding: UTF-8
VignetteBuilder: knitr
Description: An implementation of the model in Steorts (2015) <doi:10.1214/15-BA965SI>, which performs Bayesian entity resolution for categorical and text data, for any distance function defined by the user. In addition, the precision and recall are in the package to allow one to compare to any other comparable method such as logistic regression, Bayesian additive regression trees (BART), or random forests. The experiments are reproducible and illustrated using a simple vignette. LICENSE: GPL-3 + file license.
License: GPL-3
LazyData: TRUE
RoxygenNote: 7.1.1.9000
NeedsCompilation: no
Packaged: 2020-09-30 21:00:40 UTC; rebeccasteorts
Author: Rebecca Steorts [aut, cre]
Maintainer: Rebecca Steorts <beka@stat.duke.edu>
Repository: CRAN
Date/Publication: 2020-10-06 09:50:02 UTC

RLdata500

Description

Data on synthetic generation of German names with 500 total records and 10 precent duplication.

Usage

RLdata500

Format

A data frame with five variables: fname_c1,lname_c1, by, codebm, bd.

Check whether 2 records which are estimated to be linked have the same IDs

Description

Check whether 2 records which are estimated to be linked have the same IDs

Usage

check_IDs(recpair, identity_vector)

Arguments

recpair

A record pair

identity_vector

A vector of the unique ids

Value

Whether or not two records which are estimated to be linked have the same unique ids

Examples

id <- c(1,2,3,4,5,1,7,8,9,10,11,12,13,14,15,16,17,18,19,20)
rec1 <- 6
rec2 <- 1
check_IDs(recpair=c(rec1,rec2),identity_vector=id)

identity.RLdata500

Description

Unique identifier for RLdata500 the corresponds to the record number format A vector that contains the codeid

Usage

identity.RLdata500

Format

An object of class numeric of length 500.

Description

Function that returns the shared MPMMS (except with an easier condition to code than JASA paper). Function to make a list of vectors of estimated links by "P(MPMMS)>0.5" method Note: The default settings return only MPMMSs with multiple members.

Usage

links(lam.gs = lam.gs, include.singles = FALSE, show.as.multiple = FALSE)

Arguments

lam.gs

The estimated linkage structure with a default of 10 iterations

include.singles

Do not include the singleton records

show.as.multiple

Always return MPMMSs that have more than one member

Value

Returns the shared MPMMS

Examples

lam.gs <- matrix(c(1,1,2,2,3,3,5,6,4,3,4,5,3,2,4,1,2,3,4,2),ncol=20,  nrow=4)
links(lam.gs)

This function takes a set of pairwise links and identifies correct, incorrect, and missing links (correct = estimated and true, incorrect = estimated but not true, missing = true but not estimated)

Description

This function takes a set of pairwise links and identifies correct, incorrect, and missing links (correct = estimated and true, incorrect = estimated but not true, missing = true but not estimated)

Usage

links.compare(est.links.pair, true.links.pair, counts.only = TRUE)

Arguments

est.links.pair

The number of estimated links

true.links.pair

The number of true links

counts.only

State whether or not counts only is true or false

Value

Gives a vector of the estimated and true links, estimated but not true links, and the true but not estimated links

Examples

id <- c(1,2,3,4,5,1,7,8,9,10,11,12,13,14,15,16,17,18,19,20)
lam.gs <- matrix(c(1,1,2,2,3,3,5,6,4,3,4,5,3,2,4,1,2,3,4,2),ncol=20, nrow=4)
est.links <- links(lam.gs)
true.links <- links(matrix(id,nrow=1))
est.links.pair <- pairwise(est.links)
links.compare(est.links.pair, true.links=id)

Function to compute a record's Maximal Matching Set (MMS) based on a single linkage structure

Description

Function to compute a record's Maximal Matching Set (MMS) based on a single linkage structure

Usage

mms(lambda, record)

Arguments

lambda

The linkage structure

record

A vector of records

Value

Computes a records MMS

Examples

lambda <- matrix(c(1,1,2,2,3,3),ncol=3)
record <- c(1,10,3,5,20,2)
mms(lambda=lambda, record=record)

Function to compute a record's MPMMS based on a Gibbs sampler. Note: It returns a list of the MPMMS ($mpmms) and its probability ($prob)

Description

Function to compute a record's MPMMS based on a Gibbs sampler. Note: It returns a list of the MPMMS ($mpmms) and its probability ($prob)

Usage

mpmms(lam.gs, record)

Arguments

lam.gs

The gibbs sampler

record

A specific record

Value

Returns a list of the MPMSS and the associated probabilities.

Examples

lam.gs <- matrix(c(1,1,2,2,3,3,5,6,4,3,4,5,3,2,4,1,2,3,4,2),ncol=20,  nrow=4)
record <- c(1,3,1,3,1,3,1,3,1,3,1,3,1,3,1,3,1,3,1,3)
mpmms(lam.gs=lam.gs, record=record)

Function to take links list that may contain 3-way, 4-way, etc. and reduce it to pairwise only (e.g., a 3-way link 12-45-78 is changed to 2-way links: 12-45, 12-78, 45-78

Description

Function to take links list that may contain 3-way, 4-way, etc. and reduce it to pairwise only (e.g., a 3-way link 12-45-78 is changed to 2-way links: 12-45, 12-78, 45-78

Usage

pairwise(.links)

Arguments

.links

A vector of records that are linked to one another

Value

Returns two ways links of records

Examples

id <- c(1,2,3,4,5,1,7,8,9,10,11,12,13,14,15,16,17,18,19,20)
lam.gs <- matrix(c(1,1,2,2,3,3,5,6,4,3,4,5,3,2,4,1,2,3,4,2),ncol=20, nrow=4)
est.links <- links(lam.gs)
est.links.pair <- pairwise(est.links)

Gibbs sampler for empirically motivated Bayesian record linkage

Description

Gibbs sampler for empirically motivated Bayesian record linkage

Usage

rl.gibbs(
  file.num = file.num,
  X.s = X.s,
  X.c = X.c,
  num.gs = num.gs,
  a = a,
  b = b,
  c = c,
  d = d,
  M = M
)

Arguments

file.num

The number of the file

X.s

A vector of string variables

X.c

A vector of categorical variables

num.gs

Total number of gibb iterations

a

Shape parameter of Beta prior

b

Scale parameter of Beta prior

c

Positive constant

d

Any distance metric measuring the latent and observed string

M

The true value of the population size

Value

lambda.out The estimated linkage structure via Gibbs sampling

Examples

data(RLdata500)
X.c <- as.matrix(RLdata500[c("by","bm","bd")])[1:3,]
p.c <- ncol(X.c)
X.s <- as.matrix(RLdata500[c(1,3)])[1:3,]
p.s <- ncol(X.s)
file.num <- rep(c(1,1,1),c(1,1,1))
d <- function(string1,string2){adist(string1,string2)}
lam.gs <- rl.gibbs(file.num,X.s,X.c,num.gs=2,a=.01,b=100,c=1,d, M=3)