Português

Poll Simulator

= ·

Tap a term to see what it means, or .

Simulate

Example: in an electorate of 100,000 people, 38.6% vote for candidate A. Each poll interviews 1,000 voters drawn without replacement and publishes a 95% confidence interval. Draw polls and see how many intervals contain the true value. The electorate is split into neighbourhoods of about 500 people; switch the design to draw whole neighbourhoods (clusters) or to split the sample across regions (strata), and see what happens to the coverage when the poll publishes the margin of a simple random sample. The intraclass correlation says how alike neighbours vote.

Step by step

    What each term means

    Exact Coverage (C)
    The share of all possible samples whose confidence interval contains the true value P. It is what "95% confidence" promises, and it is almost never exactly 95%: with 1,000 out of 100,000 voters and P = 38.6%, it is 94.98%. The coverage observed in the simulator fluctuates around it.
    Sum over the Hits
    Runs only over the counts k whose interval, computed with p̂ = k/n and the standard error estimated by the poll itself, contains P. Since k moves in steps of 1, the range of hits is a set of integers, and the ends lost to rounding make the coverage drift from the nominal level.
    Ways to Choose k Voters for A
    How many groups of k people can be formed from the K = P·N voters for A. In the simulator, K is the true proportion times the population, rounded.
    Ways to Complete the Sample
    How many groups of n − k people can be formed from the other N − K voters. Every choice of voters for A pairs with any of these, so the two counts multiply.
    All Possible Samples
    The number of samples of size n that can be drawn from N people without replacement, all equally likely. With n = 1,000 and N = 100,000, it exceeds 102,430. The whole fraction is the hypergeometric distribution, the exact probability that the poll finds k voters for A.