Standards & Methodology
Every figure we publish is built to a stated standard — a known sample, a known method, and a margin of error we print ourselves. Here is exactly how it works, so any result can be judged on the evidence rather than taken on trust.
For example: if a poll at our political minimum reported 52%, its margin of ±3.5% or better at 95% confidence would place the true figure very likely between 48.5% and 55.5%. The numbers here are for illustration only.
Sampling standards
Political and public-opinion polls are held to a fixed minimum, because they’re published and contested. Commercial research is sized to each client’s question — with the precision of every choice shown plainly.
Held high on purpose. Public polls get broken down by group — by sex, age, and island — and a larger sample keeps those breakdowns statistically honest, not just the headline.
Test packaging with 200 or run a major brand study at 1,000 — the right sample depends on the decision. Whatever you choose, we show the margin of error before you launch, so precision is a choice you make with open eyes.
Margin of error at 95% confidence, calculated at the most conservative split (50/50). Smaller samples widen the band; larger ones narrow it. Demographic subgroups always carry a larger margin than the full sample.
Who answers
Online panels live or die on who’s really behind the answers. Ours are checked at signup and watched continuously — the two attacks on web polling, fake respondents and people answering from abroad, are closed off by design.
Every respondent confirms a working Bahamian mobile number with a one-time code at signup. No valid number, no entry.
We confirm respondents are in The Bahamas using network and device location signals, screening out answers from outside the country.
Every respondent confirms they are 18 or older at signup — the voting-age population, and the only audience whose views belong in a national poll.
Each respondent carries a quality score that updates with their behaviour. Low-quality and suspicious accounts are flagged and removed before they ever reach a published result.
Individual answers that fail attention or qualification checks are excluded — quality is enforced at the level of each response, not just each person.
Representativeness
We tell you exactly how a result was produced, and we never dress a raw sample up as something it isn’t. Every poll is labelled for what it is.
Results are reported in one of two ways, and we say which on every poll. A descriptive result reports what our respondents said, plainly — “of [n] verified respondents, X% said…”. A representative result is weighted to the Bahamian population by age, sex, and island, drawing on national census figures, so it can speak for the country as a whole. We use the representative label only when the sample can genuinely support it.
Independence
Signal Insights is independent and non-partisan. We are not affiliated with, funded by, or aligned to any political party, candidate, or campaign. We publish our findings as the data shows them — regardless of who they favour.
Reading a result
Margin of error tells you how close a poll is likely to be to the figure you’d get if every adult in the country were asked. We print it on every poll, because a result without it is only half a fact.
“95% confidence” is the standard the industry holds to: run the same poll 100 times and about 95 of them would land within that band. It is the honest acknowledgement that a poll is a precise estimate, not a census.
On every poll
Each result we put into the public record carries the same disclosure — sample, dates, method, and margin of error. Nothing to look up, nothing to ask for.
Fielded by Signal Insights · [field dates]
n = [sample size] verified Bahamas-based adults (18+)
Method: [descriptive / weighted to population by age, sex, island]
Margin of error ±[X]% at 95% confidence — larger for subgroups
Respondents phone-verified and location-checkedTrack record
As we publish, each poll appears here in full — topic, field dates, sample, method, and margin of error — so anyone can check the work behind a headline. The archive grows with every poll we put into the public record.