JFK Approval Simulator
Published model · Part 2 of 2: Fine-Tuning Scenarios · ← Part 1: All-or-Nothing Simulator
Change the distribution of how respondents rated JFK on key issues and see what the fitted model implies for overall approval.
Shift the mix of responses gradually — moving some respondents from "Poor" to "Only fair," say, instead of pinning everyone at once — and see what the fitted model implies for the headline.
Curious what's happening inside the model when you apply a scenario? See the companion walkthrough Inside the Model — it traces one respondent's answers through the equation and shows how the headline emerges from all respondents' answers.
Think a different specification explains this better? Challenge this model — opens the Model Builder with these predictors loaded, so you can add, drop, or replace them and refit.
How to Use This Tool
Shift response distributions gradually and inspect the fitted model's implied approval
How to Use This Tool
Shift response distributions gradually and inspect the fitted model's implied approval
1. Pick the outcome (optional)
Three of the Harris/Newsweek survey’s questions serve as outcomes here: Presidential Approval, 1964 Vote Intention, and Tax Cut Support. Two of them can be explored more than one way. Approval has the binary approve/disapprove cut, the full ordered 4-point scale (an ordered-logit model on the Excellent-to-Poor ratings that shows how the whole approval distribution shifts, not just the top box), and a five-category version that keeps “Not sure” as its own answer instead of dropping it (an unordered model across all five responses, with the Excellent-or-Pretty-good share reported for comparison with the binary cut). Vote Intention has the binary Kennedy-vs-Goldwater cut and the full three-way leaned vote (a multinomial model across Kennedy / Goldwater / still-undecided that shows how the whole three-way split shifts, with P(Kennedy) as the headline). You're on the published Approval model — the reference specification used for the primary case study. Use the outcome picker at the top to explore the others as exploratory auto-built models — same data, same sliders, different predictors chosen by Auto-Build.
2. Filter to a subgroup (optional)
Leave the filter at All respondents to use the published full-sample model, or filter to a subgroup like Republicans or college-educated voters. When you pick a subgroup, the simulator refits the published specification on that subgroup's respondents — the same six predictors with coefficients re-estimated there, the refit the paper's Appendix D reports. A Build a custom model button lets you ask Auto-Build for a subgroup-specific specification instead; its variables may differ from the headline model, and the page labels it as an alternative rather than the published account.
3. Set the distribution
Use the preset buttons for a quick start, or drag the sliders to change the mix of responses for any variable. Unlike pinning everyone to one answer, sliders let you shift just some respondents — for example, moving 10% of Poor ratings to Only fair. Respondents who move keep all their other answers, so the result stays an all-else-equal what-if that meets the All-or-Nothing simulator exactly at 100%. The colored bar under each header shows the variable's model-implied range across its displayed response levels under the fitted model. Each variable's percentages must sum to 100%.
4. Run the scenario
Click Run scenario to send your slider settings through the active model. The result shows the model-implied approval rate under that scenario and a 95% confidence interval. This is a conditional implication of the fitted model, not an estimate of what causing those conditions would do. The How certain is this result? chart shows 10,000 simulated outcomes so you can see how much the baseline and your scenario overlap.
5. Explore each factor
Click Explore Each Factor after a run to see a full per-variable sensitivity breakdown. Each card shows the model-implied approval if that one variable's distribution were set to 100% at each level, holding your other sliders fixed. Combined high/low scenario cards show the model-implied result of setting every variable to the displayed level associated with the highest or lowest fitted outcome.
6. Switch to All-or-Nothing
Fine-Tuning shifts distributions, approximating the way public opinion usually moves. If you'd rather pin every respondent to one specific answer per question — useful for testing extreme scenarios that make the model's assumptions visible — use the All-or-Nothing Simulator link above the scenario panel. Your subgroup filter carries over.
7. Save, share, and reset
Every run is saved to the Saved Scenarios drawer at the bottom of the page — click Load on any card to restore its slider settings, or Remove to drop it. Use Share scenario to copy a link with your settings encoded, or Reset to baseline to return sliders to the starting distributions. The baseline shown is the model's estimate for whichever filter is active.
Loading simulator…
Couldn't load the simulator
Want to set every respondent to one specific value instead? All-or-Nothing Simulator →
Auto-Build Actionable found no movable predictors for this outcome, so this simulator uses the best Standard model instead — its predictors may not be classified as directly movable.
Set the distribution
The number on each card is the variable's incremental contribution — the gain in Tjur R² it adds once the other variables are included, the same quantity reported elsewhere on this site. Bars and ordering use it too, largest first. Hover a number to see how much of the variable's one-at-a-time separation that represents. It is a model-specific diagnostic, not a measure of causal leverage or actionability.
Note: this page doesn't pin everyone to one specific value. It moves a share of respondents from the answers you shrink to the answers you grow, and everyone who moves keeps all their other answers.
Predicted outcome
How did you change the survey responses?
Each pair of bars compares the actual distribution (green) to the distribution under your scenario (red where you changed it). This is what you changed — not the predicted impact, just the input.
How certain is this result?
Every prediction has wiggle room — these histograms show how much. The green bars are the plausible answers for the baseline; the red bars are the plausible answers for your scenario. Where the colors overlap, the two answers are close enough that the model can't cleanly tell them apart.
Support and calibration
This panel answers two questions. First, is there enough survey data behind your scenario? It shows how many people your change moves to a different answer, and warns you if any answer they're moved into was given by too few people to trust. Second, how well does the model match the survey?
Explore each factor
For each variable, see how the predicted outcome would change if you switched that one distribution to 100% at each level — holding all your other selections fixed. This shows which individual levels produce the largest model-implied changes in the current scenario.