find.q {elec}

find.q

Description

Find q, the minimum number of precints with w\_p's greater than given t.stat that can hold an entire election shift in them.

Usage

find.q(
  V,
  t.stat,
  bound.col,
  M,
  threshold = 1,
  w_p = weight.function("no.weight"),
  drop = NULL
)

Arguments

V

The data.frame of votes–the subwing of a elec.data object, usually.

t.stat

The worst error found in the audit (weighted, etc.)

bound.col

The name of the column in V to be used for the passed size (max number of votes, total votes, incl undervotes, etc.) to the error function.

M

The margin to close. Usually 1 for proportional. Can be less if error from other sources is assumed.

threshold

The total amount of error to pack in the set of tainted precincts

w_p

The weight function for errors.

drop

Drop precincts with this column having a "true" value–they are previously audited or otherwise known, and thus can't hold error. Can also pass a logical T/F vector of the length of nrow(V)

Details

This number is behind the SRS methods such as CAST. If we know how many precincts, at minimum, would have to hold substantial error in order to have the reported outcome be wrong, we can compute the chance of finding at least one such precinct given a SRS draw of size n.

Find the number of precints that need to have "large taint" in order to flip the election. This is, essentially, finding a collection of precints such that the max error (e.max) plus the background error (the w\_p-inverse of the t.stat) for the rest of the precints is greater than the margin (or 1 if done by proportions).

Value

integer, number of badly tainted precints needed to hold 'threshold' error

Author(s)

Luke W. Miratrix