algorithmUDP: Differential privacy algorithm
In schwenzfeier/udp: Statistically Useful Differential Privacy

Description Usage Arguments Value Examples

Calculates a quantity of interest in a differentially-private way. Note that many returned items are not differentially-private and are simply used for debugging and illustrative purposes.

1
2
3

algorithmUDP(data, statistic, B, n, P, lambda, lambda_var, delta,
  epsilon = 0.1, epsilon_alpha = 0.1, censoring_cutoff = 0.9,
  bias_cutoff = 0.1, parallelize = F, ...)

`data`	Input data
`statistic`	Function that calculates quantity of interest
`B`	Number of bootstraps to run via BLB algorithm
`n`	Split size
`P`	Number of partitions
`lambda`	Bounding parameter for the QOI
`lambda_var`	Bounding parameter for the variance
`delta`	Privacy parameter
`epsilon`	Privacy budget for the QOI
`epsilon_alpha`	Privacy budget for estimating alpha^dp
`censoring_cutoff`	Maximum amount of censoring to allow
`bias_cutoff`	Maximum amount of censoring to allow withou doing bias correction
`parallelize`	Whether to parallelize the BLB calculations
`...`	Parameters necessary for `statistic`

`theta_tilde`	Differentially private estimate of quantity of interest
`theta_hat`	Differentially private estimate of QOI, unadjusted for bias introduced by censoring
`var_est`	Estimate of variance of theta_tilde
`a_1`	Estimate of left-sided censoring
`a_2`	Estimate of right-sided censoring
`alpha_noise`	SD of differentially private noise added to alpha estimate
`theta_noise`	SD of differentially private noise added to theta_hat estimate
`blb_thetas`	Vector of QOIs calculated in each partition during bag of little bootstraps procedure
`sigma_hat`	Estimated SD of true QOI
`alpha_too_high_halt`	Indicator for whether alpha was greater than the censoring cutoff
`bias_adj_no_converge`	Indicator when bias adjustment procedure has failed
`theta_tilde_var_sims`	Simulated theta_tilde draws from the variance simulation
`mvn_draws`	Matrix of draws from the multivariate normal from the variance simulation
`var_sigma_mat_not_pos_def`	Indicator for whether the variance simulation covariance matrix needed to be adjusted
`sigma_marix`	Covariance matrix used in variance simulation
`var_theta_hat_dp_nonoise`	Variance of theta_hat_dp before accounting for variance introduced by dp noise
`orig_sigma_mat`	Covariance matrix used in variance simulation before any adjustment to make it positive definite
`fix_indicator`	Indicator for whether theta_hat_dp was outside the range of the bounding parameter and was brough back in
`two_sided_ba_ind`	Indicator for whether the two sided bias adjustment procedure was used

1
2
3

## Not run: algorithmUDP(dat, statistic = coefFn, B = 100, n = 100, P = 1000, lambda = 3.1,
lambda_var = 0.025, form = as.formula(Y1 ~ X), coef = 'X')
## End(Not run)