Distributional comparison of synthesised data set with the original (observed) data set using propensity scores.
utility.synds(object, data, method = "cart", cp = 0.0001, maxorder = 1,
deviance = FALSE, null.utility = FALSE, syn.only = FALSE, all.comb = FALSE, ...)# S3 method for utility.synds
print(x, …)
an object of class synds, which stands for 'synthesised
data set'. It is typically created by function syn() and it includes
object$m synthesised data set(s).
the original (observed) data set.
a single string or a vector of strings specifying the
method(s) for modeling the propensity scores. Methods can be selected
from "cart", "logit" and "poly".
See details for more.
complexity parameter for classification with method "cart".
maximum order of interactions to be considered in
"logit" method. For model without interactions 0 should be
provided.
a logical value indicating whether deviance statistic for propensity score model should be calculated. This statistic gives similar inference as the default utility value based on the variance.
a logical value indicating whether the mean squared error (MSE) of the propensity score value for the methods chosen according to the null distribution should be estimated. See details for more.
a logical value indicating whether any non-synthesized variables present in the released data should be included in the estimation of the propensity scores. Default is FALSE which means that non-synthesized variables are included.
alternative method for estimating propensity scores where all
synthetic data from each m imputation is combined with the original
data. Only one utility value is returned rather than one for each synthetic
data set.
an object of class utility.synds.
An object of class utility.synds which is a list include the mean and
sd of the propensity score utility values (if m > 1) and also the raw
utility values for each synthetic set.
The list also contains the chosen method for modeling the propensity scores. If multiple methods were provided, the list will contain a matrix of utility value results for each of the different methods.
If null.utility is set to TRUE, two more elements are given in
the output list. nullstats.summary provides both the differences and
ratios between observed and null mean utility statistics (MSEs) for each
method. nullstats.raw gives utility statistics from each pair of
synthetic data sets compared.
If "deviance" is set to TRUE, the function also returns the
deviance statistic. Calculated as -2[log likelihood (estimated model)] / N,
where the estimated model uses the predicted propensity scores and the null
model uses the true proportion of records in the synthetic data (i.e. no
distinguishability between original and synthetic data).
This function follows the method for evaluating the utility of masked data as given in Snoke et al. (forthcoming) and originally proposed by Woo et al. (2009). Propensity scores, as detailed in Rosenbaum and Rubin (1983), are estimated for probability of membership in the synthetic data set. The value returned then is the mean squared difference between these probabilities and the probability of indistinguishability (perctentage of combined records in the synthetic data set).
Propensity scores can be modeled in a variey of ways. Commonly a simple
logistic regression is used with all variables in the data set as predictors,
implemented here as method "logit". Alternative modeling options
available here are classification and regression trees as method "cart"
(default) and multivariate adaptive regression splines using the
"polspline" packing under function "polyclass"
as method "poly" (in testing).
If missing values exist, indicator varibales are added and included in the
model as recommended by Rosenbaum and Rubin (1984). The missing cell is then
either imputed as zero or the mean. For cateogrical variables, NA is
treated as a new category. Thanks to https://github.com/markmfredrickson/optmatch/blob/master/R/fill.NAs.R
for useful code chunks on flagging and filling.
Null propensity score MSEs can be estimated and returned for each selected method. These values are estimated as proposed in Snoke et. al. (forthcoming) and give an estimate of the expected values under a correct synthesis model. These are used to produce a difference or ratio between the observed and null MSEs to improve accuracy of the utility estimate.
Woo, M-J., Reiter, J.P., Oganian, A., and Karr, A.F. (2009). Global Measures of Data Utility for Microdata Masked for Disclosure Limitation. Journal of Privacy and Confidentiality, 1(1), 111-124.
Rosenbaum, P.R. and Rubin, D.B. (1984). Reducing Bias in Observational Studies Using Subclassification on the Propensity Score. Journal of the American Statistical Association, 79(387), 516-524.
Snoke, J., Raab, G.M., Nowok, B., Dibben, C. and Slavkovic, A. (forthcoming). General and specific utility measures for synthetic data.
# NOT RUN {
ods <- SD2011[1:1000, c("age", "bmi", "depress", "alcabuse", "englang")]
s1 <- syn(ods, m = 5)
utility.synds(s1, ods)
# }
Run the code above in your browser using DataLab