# 1.4.3 Added functions for assessing uncertainty before performing matching: # rho.bounds: uses available data (A=(Y,X) and B=(Z,X)) to get bounds for Pearson's correlation between Y and Z using the available Xs; # rho.bounds.pred: as rho.bounds but intead of using the X directly, uses them as predictors for respectively Y and Z to get the corresponding # predictions and use these latter ones to assess uncertainty (usefule when there are many Xs including categorical ones); # Fbounds.pred: assesses uncertainty characterizing the contigency table crossing Y and Z, with two samples and using Xs to predict # respectively Y and Z # # pw.assoc: correction of a bug related to the call of the weighting variable # # mahalanobis.dist: now allows estimation of the covariance matrix in a robust manner # # ################################################################################################################### # 1.4.2 Added function for using output of hotdeck functions for imputing missing values in a dataset # # create.imputed is a NEW function that uses the output of hotdeck functions to impute missing values # # #################################################################################################################### # # 1.4.1 Added function for comparing distributions, improved graphical comparisons # # comp.cont is a NEW function for empirical comparison of the marginal distributions of the same numerical variable(s) # but estimated from two different data sources # # plotCont plots and compares also the empirical cumulative distribution function estimated # from two different data sources # # # ###################################################################################################################### # # 1.4.0 Addedd functions for plotting results, changes to some code for better management of the NAs # # NND.hotdeck and RANDwNND.hotdeck NO longer trasform the categorical matching variables in dummies # when the chosen distance function is defined only for numerical variables; in practice, mixed-type matching variables # can only be used with the Gower's distance # # fact2dummy: when a NA is observed for a categorical variable then the function puts NAs in all the dummy # variables generated from it # # pw.assoc discards NAs before calculation of the associaione or PRE measures; removal follows the pairwise # deletion rule (units where one of both the values are missing are discarded) # # plotTab is a NEW function for comparing the marginal distributions of the same categorical variable(s) but estimated # from two different data sources # # plotCont is a NEW function for comparing the marginal distributions of the same numerical variable but # estimated from two different data sources # # plotBounds is a NEW function providing a graphical summary of the width of the Frechet Bounds estimated with # the Frechet.bounds.cat function # # # ######################################################################################################################### # # 1.3.0 changes in the functions related to uncertainty investigation when dealing with categorical variables # # Frechet.bounds.cat now permits to align marginal distributions of X variables via IPF algorithm # (previously harmonization had to be done befor calling it by using harmonize.x function) # # Fbwidths.by.x provides penalty measures because of the increase of cells to estimate when increasing the number of Xs. # Sparsness of tables is explicitly considered. # # New function selMtc.by.unc() permits to identify best subset of matching variables which minimize a penalized # uncertainty estimate, as in D'Orazio, Di Zio, Scanu 2017 paper (see ref in help pages) # # Updates in pw.assoc() to allow computation of bias corrected Cramer's V, mutual information (also # normalized), AIC and BIC. Results can be organized in a data.frame. Changes in the documentation layout # to achieve coherence with documentation of other functions in the package # # Please note that Vignette is frozen to StatMatch 1.2.5, therefore it will not provide new feauter related to investigation # of uncertainty and more in general selecting of matching variables. # New vignette related to uncertainty topic is expected to be realesed in future. # # ##################################################################################################################### # # 1.2.5 gower.dist is faster and more efficient due improvements of Jan van der Laan (also thanks to Ton de Waal ) # # NND.hotdeck allows performing constrained search of donors, allowing donor to be selected not more than k times (k>=1). # argument k is set by the user # # fixed a minor bug in RANDwNND.hotdeck (not affecting results) # # richer output in Frechet.bounds.cat and Fb.widths.byx # # 1.2.4 added the new function pBayes for applying pseudo-Bayes estimator to sparse contingency tables # # modified comb.samples to handle a continuous target variable (Y or Z) # # Faster versions of Frechet.bound.cat and Fbwidths.by.x. # # Fbwidths.by.x now provides a richer output. # # 1.2.3 corrected a bug in RANDwNND.hotdeck. Thanks to Kirill Muller # # 1.2.2 added 3 data sets used in the function's help pages and in the vignette # # modified the RANDwNND.hotdeck function to identify the subset of the donors by # simple comparing the values of a single matching variable # # Minor modification of the hotdeck functions to handle and monitor the processing # when dealing with donation classes # # 1.2.1 now Frechet.bounds.cat() can be called just to compute the uncertainty bounds # when no X variables are available. # # RANDwNND.hotdeck can search for the closest k nearest neighbours by using the # function nn2() in the package RANN (wrap of the Artificial Neural Network # implemented in the package ANN). It is very fast and efficient when dealing # with large data sources. # # Fix of a minor bug in mixed.mtc() # # 1.2.0 new function comp.prop() for computing similarities/dissimilarities # between marginal/joint distributions of one or more categorical variables # # new function pw.assoc() to compute pairwise association measures among # categorical response variable and a series of categorical predictors # # rankNND.hotdeck() can perform constrained matching too # # rankNND.hotdeck(), NND.hotdeck() and mixed.mtc() solve constrained problems # more efficiently and faster by using solve_LSAP() in package "clue" # or (slower) by means of functions in the package "lpSolve". # It is no more possible to solve constrained problems by means # of functions in package "optmatch" # # NDD.hotdeck(), RDDwNND.hotdeck() and rankNND.hotdeck() are more # efficient in handling donation classes (thanks to Alexis Eidelman # for suggestion). # # fixed a bug in mahalanobis.dist (thanks to Bruno C. Vidigal) # # 1.1.0 The function comb.samples() now allows to derive predictions # at micro level for the target variables Y and Z # # 1.0.5 fixed some minor bugs # # 1.0.4 fixed some minor bugs # # 1.0.3 now mixed.mtc() can handle also categorical common variables # # fixed a bug in comb.samples() when handling factor levels # # new error messages in RANDwNND.hotdeck() when computing ditances # between units with missing values # # 1.0.2 new function mahalanobis.dist() to compute the mahalanobis distance # # fixed a bug in mixed.mtc() when computing the range of admissible values # for rho_yz # # fixed a bug in NND.hotdeck() and RANDwNND.hotdeck() when # managing the row.names # # 1.0.1 new functions harmonize.x() and comb.samples() to perform statistical # matching when dealing with complex sample survey data via # weight calibration. # # new function Frechet.bounds.cat() to explore uncertainty when dealing with # categorical variables. The function Fbwidths.by.x() permits to # identify the subset of the common variables that performs better in reducing # uncertainty # # New function rankNND.hotdeck() to perform rank hot deck distance # # Update of RANDwNND.hotdeck() to use donor weight in selecting a donor # # new function maximum.dist() that computes distances according to the # L^Inf norm. A rank transformation of the variables can be used. # # 0.8 fixed some bugs in NND.hotdeck() and RANDwNND.hotdeck()