Package 'ddi'

Data Defect Correlation

Description

The Data Defect Correlation (ddc) is the correlation between response and group membership. It quantifies the correlation between the outcome of interest and the selection into the sample; when the sample selection is independent across members of the population, the ddc is zero. Currently both variables are binary. The data defect index (ddi) is the square of ddc. Squaring the d.d.c. is more useful for characterizing the asymptotics of ' MSE.

Usage

ddc(mu, muhat, N, n, cv = NULL)

Arguments

mu	Vector of population quantity of interest
muhat	Vector for sample estimate
N	Vector of population size
n	Vector of sample size
cv	Coefficient of variation of the weights, if survey weights exist and muhat is the weighted proportion. The coefficient of variation is a summary statistic computed by sd(weights) / mean(weights).

Value

A vector of d.d.c. of the same length of the input, or a scalar if all input variables are scalars.

References

Meng, Xiao-Li (2018) <doi:10.1214/18-AOAS1161SF>, "Statistical Paradises and Paradoxes in Big Data (I): Law of Large Populations, Big Data Paradox, and the 2016 US Presidential Election." Annals of Applied Statistics 12:2, 685–726.

Examples

library(tibble)
library(dplyr)

data(g2016)

# 1. scalar input
select(g2016, cces_pct_djt_vv, cces_n_vv, tot_votes, votes_djt) %>%
  summarize_all(sum)

## plug those numbers in
ddc(mu = 62984824/136639786, muhat = 12284/35829, N = 136639786, n = 35829)

# 2. vector input using "with"
with(g2016, ddc(mu = pct_djt_voters, muhat = cces_pct_djt_vv, N = tot_votes, n = cces_n_vv))

# 3. vector input in tidy tibble
transmute(g2016, st,
 ddc = ddc(mu = pct_djt_voters, muhat = cces_pct_djt_vv, N = tot_votes, n = cces_n_vv))

---

2016 General Election Results and Survey Estimates

Description

Donald Trump's voteshare in each U.S. state, with survey estimates from the Cooperative Congressional Election Study (pre-election wave). See Meng (2018) referenced below for more details. We focus on unweighted estimates to capture the response patterns, before correcting for any imbalances through weights.

Usage

g2016

Format

A data frame with 51 rows (all U.S. states and D.C.)

state

state (full name)

st

state (abbreviation).

pct_djt_voters

Donald J. Trump's voteshare, the estimand.

cces_pct_djt_vv

CCES unweighted proportion of Trump support, one estimate.

cces_pct_djtrund_vv

CCES unweighted proportion counting Republican undecideds as Trump voters.

votes_djt

Total number of votes by Trump.

tot_votes

Turnout in Presidential as total number of votes cast.

cces_totdjt_vv

Validated voters intending to vote for Trump. Used as the numerator for the above CCES estimates.

cces_n_vv

Validated voters in survey sample. Used as the denominator for the above CCES estimates.

vap

Voting Age Population in the state.

vep

Voting Eligible Population in the state (estimate from the US Election Project).

Source

Cooperative Congressional Election Study (CCES) https://cces.gov.harvard.edu/ and the United States Election Project http://www.electproject.org/2016g. Created under https://github.com/kuriwaki/poll_error.

References

For an explanation in the context of d.d.i., see Meng (2018) <doi:10.1214/18-AOAS1161SF>

Examples

library(dplyr)
data(g2016)

transmute(g2016,
          st,
          ddc = ddc(mu = pct_djt_voters,
                    muhat = cces_pct_djt_vv,
                    N = tot_votes,
                    n = cces_n_vv))

---