Questionnaire methodology and privacy

Methodology: how the comparison works

Rank, answer, review, compare. No answer is preselected. The first issue receives N points for N included issues; the last receives one. Excluding an issue recomputes the weights. Every predefined subquestion gets an equal share of its parent issue's weight. An unscored half stays unscored; its points do not move to the other half.

Only directly stated positions or votes on the specific proposition can qualify. Both candidates must have documented positions on the same proposition and a meaningful difference. A vote on a large package does not establish a position on every provision. Silence is not opposition. Historical, inferred, mixed and unresolved positions are excluded.

Agreement answers use +1 for strongly agree, +0.5 for somewhat agree, −0.5 for somewhat disagree and −1 for strongly disagree. Candidate codes are +1 for support and −1 for opposition; they describe policy direction, not intensity. Mixed / depends, Unsure and Skip are unscored. Free text never creates a match.

agreement = 1 − |visitor position − candidate position| / 2. Same-side Strongly earns 100%; same-side Somewhat 75%; opposite-side Somewhat 25%; opposite-side Strongly 0%. Weighted candidate points divided by the shared scored weight give each candidate's independent agreement percentage. Scores need not sum to 100%.

Categorical answers use explicit, sourced option mappings, with no invented distance between categories. Ambiguous mappings are excluded. A candidate may match multiple options only when the evidence supports it.

Coverage is scored weight divided by all included weight, including skipped answers and unresolved evidence. Below 60% the result says Limited comparison and gives no overall alignment headline. This is an editorial display rule, not a statistical confidence interval. No scorable answers means no percentage or leader. Ties use an epsilon of 0.000000001; small gaps receive restrained wording.

The sensitivity check repeats the calculation with equal weight per included issue and the same answers and evidence decisions. It reports whether the apparent lead changes. Source publication dates and dates checked are separate. A recently opened older statement is still an older statement.

Answers remain in memory, with no server submission or browser saving. Candidate positions are hidden in the normal flow, not cryptographically secret. Inspecting the public page data can reveal them. The report is available only through your explicit print or download action.

Questions about the questionnaire