Next Impact Hub · poll analysis

A read of the “Where should the next Impact Hub be?” poll (28 responses), comparing how different groups and time horizons rank the candidate cities across a dozen measures.

Everyone = 27 unique respondents (deduplicated) · Cohort = 10 · Non-cohort = 17 · Seriously considering = 18 (long-term raters). Long-term uses a 5-point excitement scale; short-term a 4-point likelihood scale. Blanks are excluded from counts and means. Use the tabs above; click any column header to sort.

Long-term — Everyone

23 respondents · excitement 1–5 · colour-coded per column (green = better → red = worse) · hover a heading for its meaning · click to sort

All respondents — cohort and non-cohort pooled — with the one duplicate submission deduplicated.

Short-term — Everyone

26 respondents · likelihood 1–4 · colour-coded per column (green = better → red = worse) · hover a heading for its meaning · click to sort

All respondents — cohort and non-cohort pooled — with the one duplicate submission deduplicated.

Condorcet winner: ★ Chiang Mai, Thailand

Long-term — Cohort

10 respondents · excitement 1–5 · colour-coded per column (green = better → red = worse) · hover a heading for its meaning · click to sort

Short-term — Cohort

10 respondents · likelihood 1–4 · colour-coded per column (green = better → red = worse) · hover a heading for its meaning · click to sort

Condorcet winner: none — preferences form a cycle (no city beats all others head-to-head).

Long-term — Non-cohort

13 respondents · excitement 1–5 · colour-coded per column (green = better → red = worse) · hover a heading for its meaning · click to sort

Short-term — Non-cohort

16 respondents · likelihood 1–4 · colour-coded per column (green = better → red = worse) · hover a heading for its meaning · click to sort

Condorcet winner: none — preferences form a cycle (no city beats all others head-to-head).

Long-term — Seriously considering a 12+ month base

Interest threshold — toggle to widen or tighten:

Methodology

Scales. Long-term answers are excitement, encoded Not at all = 1, Slightly = 2, Moderately = 3, Very = 4, Extremely = 5. Short-term answers are likelihood, encoded Unlikely <20% = 1, Mod. Unlikely 20–50% = 2, Mod. Likely 50–80% = 3, Likely 80%+ = 4. A blank is excluded from every measure except Expected attendees, where it contributes 0. n denotes the number of respondents who rated a given city.

Display. Each column is colour-scaled independently — green = better, red = worse (for Detractors, Meh and the regrets, lower is better, so green marks low values; Net is centred at 0). Hover a heading for its definition, click to sort; ★ marks a Condorcet winner.

Weighting (long-term only). Each LT panel has an Unweighted / Weighted by interest toggle. In the weighted view each respondent contributes with a weight based on their answer to the 12+ month base interest question — Very = 3, Moderately = 2, Slightly = 1, no answer = 0 — so people seriously considering a long-term base count more, and respondents who didn't engage with the interest question drop out entirely. Every measure (Total, Likert, Mean, Superfans, Keen, Detractors, Meh, Net, E[att], regrets, Borda, Condorcet) is recomputed under those weights. Short-term tables stay unweighted (there is no per-respondent overall ST interest question).

Total

Sum of the ratings from the n respondents who scored the city. Confounds turnout with intensity, and is reduced by non-responses (blanks are not counted). Higher is better.

Likert sum

Like Total but with the lowest rating zeroed: Σ (rating − 1) over respondents who rated the city. Long-term scores 0/1/2/3/4 for Not at all / Slightly / Moderately / Very / Extremely; short-term 0/1/2/3 for Unlikely <20% / Mod. Unlikely / Mod. Likely / Likely 80%+. Algebraically equals Total − n, so it ranks identically to Total within a fixed rater set and diverges only through differing blanks — useful when a non-enthusiastic vote should contribute 0 rather than mild positive support. Higher is better.

Mean

Total / n, over the respondents who rated the city. Higher is better.

Median

Median of the city's ratings. Higher is better; integer/half-integer valued, so ties are common.

Superfans

Number of ratings at the top level: = 5 long-term (Extremely), = 4 short-term (Likely 80%+). Higher is better.

Keen

Number of ratings in the top two levels: ≥ 4 long-term, ≥ 3 short-term. Equivalent to approval voting with the threshold set at the top two levels. Higher is better.

Detractors

Number of ratings at the bottom level, = 1 (Not at all / Unlikely <20%). Lower is better.

Meh

Number of ratings in the bottom two levels, ≤ 2. Lower is better.

Net

Keen − Detractors. Centred at 0; positive means top-two ratings outnumber bottom ratings.

Expected attendees

Σ p over all respondents, where each response maps to a probability p and a blank contributes 0. Short-term uses the bucket midpoints: <20% → 0.10, 20–50% → 0.35, 50–80% → 0.65, 80%+ → 0.90. Long-term has no probability in the data, so excitement is mapped to an assumed probability of commitment — an explicit modelling assumption, not a measurement. The main table uses the Linear map 1→0.10, 2→0.30, 3→0.50, 4→0.70, 5→0.90; the sensitivity table also reports Conservative 0 / 0.05 / 0.20 / 0.45 / 0.75 and Optimistic 0.15 / 0.40 / 0.65 / 0.85 / 0.97, so the ranking's dependence on the assumption is visible. Higher is better.

Total regret

For each respondent i who rated the city, let m_i be the highest rating i gave any city; the city's regret from i is m_i − rating_i. Total regret = Σ_i (m_i − rating_i) over those respondents. Lower is better. When all cities share the same raters this equals (Σ m_i) − Total, i.e. it ranks identically to Total within a group and diverges only through differing blanks.

Max regret

max_i (m_i − rating_i) — the largest single-respondent regret (a minimax / worst-off criterion). Lower is better; integer-valued and bounded by the scale, so ties are frequent.

Borda

Per respondent, restricted to the cities they actually rated, the city earns 1 point for each rated city it outranks and ½ for each tie; summed across respondents. Uses rank order only, not magnitudes, so it can disagree with Total and Mean. Higher is better.

Condorcet winner

The city that wins every pairwise majority: for each other city d, more respondents who rated both rank it above d than below. Reported only if such a city exists; if the majority relation cycles (e.g. A>B, B>C, C>A) there is none. ★ marks it in the tables.