Read a simulated vote without inventing a consumer study
Does a high instant-coffee percentage prove that consumers prefer it?
No. Every dot is a browser-generated Bernoulli vote. The sensory profiles, utility weights, 77% baseline and branding adjustment are chosen inputs; there are no participants, measured coffee chemistry or imported study responses. The statistics summarize one synthetic draw.
This is an independent teaching simulation. The Drexel news notice confirms that Food Lab authors compared instant and drip coffee in a 2025 column, but does not supply this simulator’s data, weights, 77% calibration or branding effect. No replication or institutional endorsement is established.
Try a worked example
In the baseline preset, matching acidity values force p=0.77 even when body or temperature changes. Revealing labels subtracts 0.41, giving p=0.36: a 41 percentage-point change, not a measured effect. At N=100 the expected instant count is 77, but each run can differ. If a run produces 77 votes, its Wilson interval is approximately 67.8% to 84.2%; the normal-approximation z statistic against 50% is 5.4. These values describe the generated votes, not real consumers.
Separate evidence from a chosen probability
The university notice identifies the authors and a drip-versus-instant column dated October 6, 2025. It supplies no participant-level dataset or methodological detail. Use it to locate the original discussion; this page cannot establish its study design or numerical results. Drexel: coffee comparison notice
Outside the baseline override, the page weights low acidity by 0.35, body by 0.30, sweetness by 0.25 and aroma by 0.10. Bitterness appears on the radar but has no utility weight. Temperature below 60 degrees reduces drip utility by 0.05 per degree. These are authored rules, not calibrated sensory findings.
Interpret uncertainty within the simulation
The displayed interval uses the Wilson score formula with z=1.96. NIST describes this proportion interval and distinguishes it from other methods. Its coverage concerns repeated samples under a binomial model; it is not a 95% probability that this particular interval contains a real consumer preference. NIST: proportion confidence intervals
The p-value uses a two-sided normal approximation against p=0.5, without a continuity correction. It is not an exact binomial test. NIST states sample-size conditions for the approximation. The control permits N=10, so small-sample values need extra care; this tool does not switch to an exact method. NIST: testing a proportion
Make the controls and export useful
Every slider change, preset selection and label toggle draws fresh votes. This confounds a parameter change with random variation; it does not compare the same participants before and after. Run Trial also assigns new cup codes; those codes do not influence the vote probability. Reset restores all baseline profiles and draws a fresh sample rather than replaying an earlier run.
The JSON includes the synthetic vote vector and chosen profiles, allowing counts and displayed statistics to be checked. It does not record a random seed or real identities, so it is a snapshot rather than a reproducible study. In a real comparison, record the recruitment, coffee preparation, allocation, blinding, temperature, outcome question and analysis plan before interpreting a preference.
Sources and further reading