November 1963 · Harris/Newsweek Survey

Nonresponse Stress Test

Stress test on the published headlines

This tool does not estimate nonresponse bias. It asks how far the released headline would move under assumptions you choose about a hypothetical group the survey may have missed.

Runs on all three poll outcomes — Kennedy's job approval, JFK-vs-Goldwater vote intention, and support for the proposed tax cut. This is a sensitivity check on the released score, separate from the explanatory model.

Based on the archived November 1963 Harris–Newsweek survey. The public archive contains 1,283 respondents; the valid-case base can differ by outcome.

How to Use This Tool

Pick an outcome, set the sliders, and click Calculate Scenario. Expand for details.

1. Select an Outcome

Choose which released score to stress-test: Presidential Approval, Vote Intention (JFK vs. Goldwater), or Tax Cut Support. The percentage shown is the baseline used by this demonstration.

2. Set the Sliders

Two sliders define your sensitivity scenario. How much of the sample should the scenario replace? sets the replacement share. What would they have said? sets that group's positive-outcome rate; its sub-heading names the outcome you selected. Neither slider estimates actual nonresponse.

3. Calculate the Scenario

Click Calculate Scenario. The tool applies the replacement assumptions directly and reports the resulting headline and percentage-point shift. The calculation is deterministic: it does not use Monte Carlo simulation or resampling.

4. Read the Verdict

The results section asks: how consequential is this assumption for the headline? It compares the scenario shift with a declared precision benchmark. In this historical demonstration the benchmark is calculated from the outcome's Kish effective sample size; for presidential approval it is about ±2.9 points. It is not a survey-reported margin of error and does not estimate nonresponse bias. The page also flags a majority flip when the headline crosses 50 percent.

More options

Save and compare

Each run is saved automatically. Expand the Pinned Scenarios drawer at the bottom to compare previous runs side by side.

Reset and iterate

Click Reset to restore defaults. Try fixing one slider and sweeping the other to map where the headline would change.

Base

—% (benchmark ± —)

Baseline used by this demonstration

Scenario

—% (exact under assumptions)

Exact result under the stated replacement assumptions

Scenario replaces — records (—% of sample)

Could non-response have changed the headline?

How big is the shift?

Scenario shift —× the precision benchmark

Direction of change —

Magnitude of change — points

Verdict

Benchmark result —

The headline is — under this scenario.

The difference between the headline number and your scenario —.

Detailed Analysis

Baseline used here: the outcome at —%. Illustrative precision benchmark: ±— points.

Scenario-implied headline: —%. This value follows directly from the stated replacement assumptions; no resampling or Monte Carlo simulation is used.

The scenario replaces —% of the analytic base with a hypothetical group at —% on the outcome (gap: — pts).

Why a precision benchmark? The archived 1963 materials do not provide the information needed to reconstruct a fully design-based margin of error. This demonstration therefore uses an illustrative benchmark derived from the outcome's Kish effective sample size. It is a comparison yardstick, not an estimate of nonresponse bias. A live release should use its reported margin of error or another declared precision threshold.

Baseline vs. scenario

Exact scenario calculation

The replacement calculation is deterministic: p_scenario = p₀ + f × (r − p₀). The only uncertainty yardstick shown on this page is the separately declared precision benchmark used to judge the size of the shift.

Tipping frontier

Every combination of replacement share and replacement-group rate that would move the headline past — points — the precision benchmark the assessment above already uses. Inside the shaded band the headline holds; outside it, the headline tips. Your current scenario is the indigo dot.

The curves are exact, not simulated: the scenario-implied rate is p₀ + f × (r − p₀), so the tipping condition solves to r = p₀ ± T/f. This reports sensitivity to stated assumptions about people the survey may have missed; it does not estimate actual non-response bias.