← Back to sample work

Research sample Illustrative data

Research Brief

Turnout & Persuasion Analysis

City Council At-Large Race — General Municipal Election

Prepared by Eno Analytics · Sample deliverable · Confidential draft format

About this sample. This brief demonstrates the format and rigor of an Eno Analytics research engagement. The candidate, jurisdiction, and all figures are fictional and for illustration only. No real voter data was used. It is not a forecast of any actual election.

Executive summary

This analysis sizes the electorate for a competitive at-large council race, segments voters by likelihood of voting and current support, and tests three campaign messages through a randomized survey experiment. Three findings drive the recommendation:

  • 1
    Turnout will be low and uneven. Our model projects roughly 20,300 voters (42% of registrants), concentrated heavily among a high-propensity tier that is already largely decided.
  • 2
    The race runs through ~5,500 persuadable voters. The candidate's base (≈7,700) is short of a majority; winning requires converting roughly 44% of the persuadable pool.
  • 3
    One message clearly outperforms. A schools-and-parks message lifted support +8.0 points (95% CI: +2.4 to +13.6) over control — the only arm with a large, statistically clear effect.

Recommendation: concentrate persuasion spending on the schools-and-parks message, targeted to the ~3,900 persuadable voters who are also likely to turn out, while running a parallel turnout program to defend the base. Details and caveats follow.


Data & method

Two components underlie the brief. Both are documented so a third party could reproduce them.

ComponentSource (illustrative)Method
Turnout & support modelRegistered-voter file with prior municipal vote historyVote-propensity tiers from prior participation; support scores calibrated to the survey below
Message testLive-caller + text-to-web survey of likely votersRandomized experiment, 4 arms, pre-registered outcome

Survey: n = 2,400 likely voters; 600 per arm. Full-sample margin of error ±2.0 points at 95% confidence; per-arm margin ±4.0 points. Fielded over one weekend; weighted to the modeled electorate by age, region, and vote history.


Finding 1 — The electorate is small and top-heavy

We divide registrants into three vote-propensity tiers based on how many of the last four municipal elections they voted in, then apply modeled turnout rates to each. The high-propensity tier is barely a third of registrants but more than half of actual voters.

Propensity tierRegisteredModeled turnoutProjected votersShare of voters
High (voted 3–4 of last 4)14,50078%11,31055.6%
Mid (voted 1–2 of last 4)18,70041%7,66737.7%
Low (voted 0 of last 4)15,0009%1,3506.6%
Total48,20042.2%20,327100%
High 11,310 Mid 7,667 Low 1,350 0 6,000 12,000
Figure 1. Projected voters by propensity tier. High-propensity voters cast an estimated 56% of ballots despite being 30% of registrants.

So what. Persuasion aimed at low-propensity registrants is largely wasted — they mostly won't vote. The efficient targets are people who will vote but haven't decided.


Finding 2 — The race is decided by the persuadable middle

Applying calibrated support scores to the projected 20,327 voters yields the split below. A simple majority requires 10,164 votes. The base alone falls about 2,440 votes short.

Support segmentShareProjected votersImplication
Base (strong / lean support)38%7,724Protect via turnout program
Persuadable (undecided / soft)27%5,488Primary persuasion target
Opposition35%7,114Do not invest
Total100%20,327

To reach 10,164 votes, the campaign must win 2,440 of the 5,488 persuadables — about 44%. That is achievable, but only with a message that measurably moves this group.


Finding 3 — Only one message moves the needle

We randomly assigned 2,400 likely voters to a control arm or one of three message arms, then measured stated support. The schools-and-parks message produced the largest and most statistically reliable lift; the public-safety message could not be distinguished from no message at all.

ArmnSupportLift vs. control95% CIVerdict
Control (no message)60041%
A — Property tax / economy60047%+6.0+0.4 to +11.6Significant (marginal)
B — Public safety60044%+3.0−2.6 to +8.6Not significant
C — Schools & parks60049%+8.0+2.4 to +13.6Significant
0 +5 +10 +15 Support lift over control (percentage points) A · Economy B · Public safety C · Schools & parks
Figure 2. Estimated support lift over control with 95% confidence intervals. An interval that crosses zero (Arm B) means the effect is statistically indistinguishable from no effect. Arm C is both the largest effect and the most reliable.

Reading the intervals. Message B's point estimate is positive (+3.0) but its interval spans zero, so we cannot rule out that it does nothing. We would not spend on it. Message C's entire interval sits above +2.4 points — a real effect.


Recommendation

  1. Lead persuasion with Message C (schools & parks). It is the only message with a large, statistically clear effect on support.
  2. Target the persuadable-and-likely voters. Of the 5,488 persuadables, roughly 3,900 are also mid-to-high propensity. Concentrate paid persuasion there; suppress spend on low-propensity and opposition voters.
  3. Run turnout in parallel for the base. The 7,724-vote base is necessary but not sufficient; a defensive GOTV program protects it while persuasion closes the ~2,440-vote gap.
  4. Retire Message B. Public safety did not move this electorate in testing. Re-test only if the strategic environment changes.

Limitations & assumptions

  • Turnout rates are modeled from prior participation and will shift with the real information environment; treat projected voter counts as central estimates, not certainties.
  • Stated support in a survey overstates persuasion durability; field effects are typically smaller than survey effects.
  • The message test measures short-term movement in a single contact. It does not capture repetition, backlash, or opponent response.
  • All figures are illustrative. A live engagement would report weighting diagnostics, response rates, and full crosstabs in a technical appendix.

Prepared by Eno Analytics as a demonstration of deliverable quality. For a scoped engagement, see Research & Analytics or get in touch.

Want work like this for your question?