The forecast uses the State Board of Elections’ mail-in figures through
October 30, 2026, its unofficial early-voting counts through October 29,
2026 and its same-day registration counts through October 29, 2026.
Every number on this page is read from the model’s committed outputs in
data/turnout_model/. Ballots cast counts every ballot:
election day, early voting, mail-in and provisional. A turnout rate is
always ballots cast divided by a named denominator: active registration,
eligible registration (active plus inactive registrants) or citizen
voting-age population (CVAP).
The central forecast is 2,330,865 ballots cast, 54.2% of active registration. The range to quote runs from 2,060,722 to 2,553,256.
| Scenario | Ballots cast | 90% interval | Rate on active | Rate on eligible | Rate on CVAP |
|---|---|---|---|---|---|
| Low (a 2022-like midterm) | 2,102,288 | 2,060,722 to 2,143,646 (within scenario) | 48.9% | 45.3% | 47.6% |
| Central (the level rule’s blend) | 2,330,865 | 2,289,592 to 2,371,828 (within scenario) | 54.2% | 50.2% | 52.8% |
| High (a 2018-like midterm) | 2,513,478 | 2,473,047 to 2,553,256 (within scenario) | 58.4% | 54.1% | 56.9% |
| Headline range (low’s 5th to high’s 95th percentile) | central: 2,330,865 | 2,060,722 to 2,553,256 (across scenarios) | 47.9% to 59.4% | 44.4% to 55.0% | 46.6% to 57.8% |
| 2022 general, actual | 2,031,635 | 49.3% | 45.7% | not computed | |
| 2018 general, actual | 2,335,128 | 59.1% | not available | not computed |
Low assumes each party votes at its 2022 statewide level, a quiet midterm. High assumes each party votes at its 2018 level, an energized midterm. Central blends the two, party by party, with weights set by the written rule below. The intervals within each scenario are narrow by construction, because each draw is calibrated to a past election. They carry only cell-level noise and the shift in the 2026 electorate’s mix. The real uncertainty is the spread between low and high. That is why the headline range runs from low’s 5th percentile to high’s 95th. The 2022 rate on eligible divides by that election’s active plus inactive registrants, from the board’s inactive counts. No inactive counts exist for 2018, so its rate on eligible is not available. CVAP comes from a single ACS vintage matched to 2026, so it is not used for the past elections.
The 2026 forecasts include 7,142 ballots from same-day registration beyond 2022’s level, the same in every scenario; see Ballots already cast.
The 2026 rate on eligible is projected. There is no 2026 inactive count yet, so eligible is projected active registration times one plus the average of the 2022 and 2024 inactive-to-active ratios. The board’s own published 2026 percentage will use its actual eligible count and will differ.
The model treats each county and party group as a cell, 72 in all. In each cell, ballots cast are a binomial share of active registration. The model has three layers. Structure is estimated: how each cell’s turnout sits above or below its party’s statewide level in a midterm. Composition is computed: how the cell’s 2026 electorate differs in age and newness from 2022’s. Level is set by a written rule and bracketed between the 2022 and 2018 midterms.
The structure layer is a hierarchical binomial regression, fitted
with lme4::glmer on 360 county x party rows from six
elections (2018 and 2022 generals, 2022 and 2026 primaries, 2024 primary
and general). It has a free statewide level for each election and party,
four county census covariates, and random effects for county x party,
county x election type and each observation. The formula, from
fit_summary.json:
cbind(ballots_cast, active - ballots_cast) ~ 0 + ep + z_share_ba_plus + z_median_age + z_share_moved_past_year + z_share_owner_occupied + (1 | county:party_group) + (1 | county:election_type) + (1 | obs)
| Random effect | Variance (logit scale) | Standard deviation |
|---|---|---|
| Observation (cell noise) | 0.0144 | 0.120 |
| County x election type | 0.0048 | 0.070 |
| County x party | 0.0290 | 0.170 |
The full formula and the one without (1 | county:party_group:election_type) are singular: the county x party x type term is estimated at 0 (it is nearly the same as (1 | obs), with at most three rows per cell and election type), and (1 | county) is estimated at 0 once the covariates are in: the covariates and (1 | county:party_group) take up all county-level variation. Dropping (1 | county) as well leaves the estimates unchanged. zero_sd_terms in each attempt lists the terms at 0.
The rule puts Democrats halfway between 2022 and 2018 (w = 0.5), Republicans halfway between 2022 and 2018 (w = 0.5) and Other (unaffiliated and minor parties) closer to 2018 (w = 0.7). The weight w is the share of the central level taken from 2018. The votes that moved it:
The rule, verbatim from level_rule_text.csv:
Three scenarios per party p, on the logit scale, per draw k:
low: alpha = alpha*_{GG22,p,k}
high: alpha = alpha*_{GG18,p,k}
central: alpha = (1 - w_p) * alpha*_{GG22,p,k} + w_p * alpha*_{GG18,p,k}
w_p starts at 0.5 and moves by 0.1 per index vote, bounded to [0.2, 0.8].
Votes, per party p (DEM, REP, OTH), each in {-1, 0, +1} where +1 means 2018-like and -1 means 2022-like:
1. Registration momentum: position = (m26 - m22) / (m18 - m22) using indices.csv reg_momentum, reg_momentum_2022, reg_momentum_2018 (statewide row for p). Vote +1 if position >= 2/3, -1 if <= 1/3, else 0. If |m18 - m22| < 0.002 the analogues cannot discriminate: vote 0.
Amendment: a position outside [-0.5, 1.5] extrapolates beyond the two analogues and its sign is not interpretable: vote 0 when position < -0.5 or > 1.5; otherwise the thresholds above apply.
2. Mail requests, per registrant: project GG26 final sent for p as a linear trend over the available daily snapshots extended to 2026-10-27 (the request deadline; an assumption: no ballots are sent after it). Index = (projected GG26 final sent_p / 2026 projected active_p from denominators_gg26.csv) / (GG22 final sent_p (2022-11-28 file) / GG22 active_p from training.csv), so registration growth is not read as engagement. Vote +1 if >= 1.05, -1 if <= 0.95, else 0. (No 2018 analogue exists: 2018 predates the permanent mail-in list, so this index can only say "more or less engaged than 2022".)
3. Primary participation, per registrant: index = (GP26 ballots_p / GP26 active_p) / (GP22 ballots_p / GP22 active_p), from training.csv (DEM and REP; GP26 split estimated; OTH gets vote 0 because primaries are closed). Vote +1 if >= 1.05, -1 if <= 0.95, else 0.
Caveat: GP26 had an incumbent governor and GP22 an open seat, which depresses this index for reasons unrelated to November.
4. Ballots banked before election day, per registrant: banked = early-voting ballots (GG26: the board's unofficial eight-day county-by-party file at its latest as_of; GG22: the certified early-voting count in training.csv) plus mail ballots received back by the same number of days before election day (sent-returned county files; GG22 at the latest daily file on or before its election day minus that many days). Index = (GG26 banked_p / 2026 projected active_p) / (GG22 banked_p / GG22 active_p). Vote +1 if >= 1.05, -1 if <= 0.95, else 0.
Caveat: voters have been moving from election day to early voting and mail, so part of a high index is a change of habit rather than more voters; the vote is one step of 0.1, like the others. (No 2018 analogue: 2018 has no daily mail-return files and predates the permanent mail-in list.)
level_rule.csv also records the raw-count ratios (mail_count_ratio, primary_count_ratio, banked_count_ratio) beside the per-registrant indices that vote.
| Quantity | DEM | REP | OTH |
|---|---|---|---|
| Registration momentum, Sept 2025 to Sept 2026 | 1.0019 | 0.9869 | 1.0302 |
| Registration momentum, Sept 2021 to Sept 2022 | 0.9999 | 1.0057 | 1.0459 |
| Registration momentum, Sept 2017 to Sept 2018 | 1.0173 | 1.0029 | 1.0298 |
| Position between 2022 (0) and 2018 (1) | 0.12 | 6.66 | 0.97 |
| Vote: registration momentum | -1 | 0 | +1 |
| Mail ballots sent, latest | 435,023 | 101,260 | 118,593 |
| Mail ballots sent, used (latest; sending is over) | 435,023 | 101,260 | 118,593 |
| Mail ballots sent, 2022 final | 438,093 | 102,183 | 99,172 |
| Active registration, 2026 (projected) | 2,218,637 | 1,007,172 | 1,075,725 |
| Active registration, 2022 | 2,235,977 | 992,188 | 895,991 |
| Mail sent, raw count ratio 2026/2022 | 0.993 | 0.991 | 1.196 |
| Mail sent per registrant, index 2026/2022 | 1.001 | 0.976 | 0.996 |
| Vote: mail requests | 0 | 0 | 0 |
| Primary ballots, 2026 | 654,148 | 219,736 | 28,934 |
| Primary ballots, 2022 | 687,517 | 301,958 | 44,428 |
| Primary ballots, raw count ratio 2026/2022 | 0.951 | 0.728 | not used |
| Primary ballots per registrant, index 2026/2022 | 0.956 | 0.705 | not used |
| Vote: primary participation | 0 | -1 | 0 |
| Early votes, 2026 (4 days out) | 365,824 | 148,200 | 90,372 |
| Mail ballots received, 2026 (4 days out) | 261,782 | 57,635 | 57,550 |
| Early votes, 2022 | 219,829 | 111,858 | 49,187 |
| Mail ballots received, 2022 (4 days out) | 252,067 | 61,065 | 47,222 |
| Banked ballots, raw count ratio 2026/2022 | 1.330 | 1.190 | 1.534 |
| Banked ballots per registrant, index 2026/2022 | 1.340 | 1.173 | 1.278 |
| Vote: ballots banked before election day | +1 | +1 | +1 |
| Sum of votes | 0 | 0 | +2 |
| Weight on 2018 (w) | 0.5 | 0.5 | 0.7 |
The mail-request deadline has passed. The board’s October 30, 2026 snapshot is dated after the October 27 deadline, so sending is over and every projection uses the latest count of ballots sent. The sensitivities below equal the rule.
| Projection | Party | Projected sent | Mail index | Mail vote | w |
|---|---|---|---|---|---|
| Rule: line through all snapshots | DEM | 435,023 | 1.001 | 0 | 0.5 |
| Rule: line through all snapshots | REP | 101,260 | 0.976 | 0 | 0.5 |
| Rule: line through all snapshots | OTH | 118,593 | 0.996 | 0 | 0.7 |
| Latest count held | DEM | 435,023 | 1.001 | 0 | 0.5 |
| Latest count held | REP | 101,260 | 0.976 | 0 | 0.5 |
| Latest count held | OTH | 118,593 | 0.996 | 0 | 0.7 |
| Line through last three snapshots | DEM | 435,023 | 1.001 | 0 | 0.5 |
| Line through last three snapshots | REP | 101,260 | 0.976 | 0 | 0.5 |
| Line through last three snapshots | OTH | 118,593 | 0.996 | 0 | 0.7 |
| Scenario | Ballots cast | Difference from central | Rate on active | Rate on eligible (projected) |
|---|---|---|---|---|
| Central (rule) | 2,330,865 | 54.2% | 50.2% | |
| Central, latest mail count held | 2,330,865 | +0 | 54.2% | 50.2% |
| Central, last three snapshots | 2,330,865 | +0 | 54.2% | 50.2% |
No 2018 analogue exists: 2018 predates the permanent mail-in list, so the mail index can only say more or less engaged than 2022. The rule’s GG26 projection is an OLS line of sent on calendar day over every daily snapshot, extended to 2026-10-27, the mail-in request deadline (assumed to be when sending stops), floored per county x party at the latest observed count. The mail vote depends on the projection choice: the sent series is step-like, so the all-snapshot slope is much steeper than the last few days’, and holding the latest count or extending the last three snapshots gives lower indices (see the _latest and _recent columns of level_rule.csv and the central_mail_latest and central_mail_recent scenarios). No GG22 trajectory of daily sent counts exists to anchor the choice. Once a snapshot dated on or after the deadline is in, every method uses the latest count, since sending is over, and the sensitivities equal the rule.
Caveats on the rule:
Early voting ended on October 29. Voters cast 604,396 ballots in the eight days, by the board’s unofficial count, against 380,874 in 2022’s certified count, a rise of 59% on registration that grew 4%. By October 30, 2026, 4 days before the election, 376,967 mail ballots had come back, against 360,354 at the same point in 2022. The model uses these numbers in three places:
| Scenario | Ballots cast | Early | Mail-in | Banked share | Election day and provisional | Registrants not yet voted | Rate among them |
|---|---|---|---|---|---|---|---|
| 2026 low | 2,102,288 | 604,396 | 560,543 | 55.4% | 937,349 | 3,136,595 | 29.9% |
| 2026 central | 2,330,865 | 604,396 | 560,543 | 50.0% | 1,165,926 | 3,136,595 | 37.2% |
| 2026 high | 2,513,478 | 604,396 | 560,543 | 46.3% | 1,348,539 | 3,136,595 | 43.0% |
| 2018 actual | 2,335,128 | 663,188 | 120,517 | 33.6% | 1,551,423 | 3,170,322 | 48.9% |
| 2022 actual | 2,031,635 | 380,874 | 541,990 | 45.4% | 1,108,771 | 3,201,292 | 34.6% |
People who have already voted are the likeliest voters. With a larger share of registrants banked than in 2022, the ones left are, all else equal, less likely to vote than 2022’s were. That makes this check one-sided: a scenario that needs a higher rate than 2022’s from them is asking for more enthusiasm than in 2022, not just a change of habit. Read against that, the low scenario asks for less on election day than 2022 delivered (29.9% against 34.6%), from a pool that has already given up its likeliest voters to early voting and mail.
Same-day registration. During early voting, 3,802 people registered and voted at once, against 2,442 in 2022, and 3,562 registered voters updated their address and voted (1,730 in 2022). New registrants are not in the September-based 2026 registration, so they add ballots the cells cannot see. 2022’s same-day registrants are already in the reference levels, because certified ballots include them, so the model adds only 2026’s excess: election-day same-day registration is projected at 2022’s county counts (11,456 statewide) times the growth in early-voting same-day registration (1.56x), for 21,638 new registrants in all, against 14,496 for 2022 scaled to 2026 registration. The difference, 7,142 ballots, is added to every scenario. Address changes add nothing: those voters are already on the rolls.
Early voting is read from the board’s unofficial eight-day files (early-voting, county-party), which the State Board prints while the campaign runs; the certified early-voting count will differ slightly. Mail ballots received back are the sent-returned county files’ received measure (envelopes back, before the canvass accepts or rejects them). The 2022 comparison point is the GG22 daily file the same number of days before its election day (November 8, 2022), from the backfilled daily reports. The banked index treats early and mail ballots alike, so a voter who moved from election day to early voting since 2022 raises it without adding a ballot. Same-day registration adds voters who are not in the September-based 2026 active registration. The reference levels already carry the same-day registrants of their own year, because certified ballots cast include them, so the forecast adds only the excess of 2026’s same-day new registrants over GG22’s, scaled to 2026 registration: extra_c = sdr26_c - sdr22_c * active26_c / active22_c, where sdr22_c is GG22’s new registrants during early voting and on election day (same-day county file) and sdr26_c is GG26’s early-voting new registrants (latest same-day county file) plus GG22’s election-day new registrants in the county times the statewide growth of early-voting same-day registration from 2022 to 2026. Each county’s extra is split by party in proportion to GG26’s early-voting same-day new registrants (county-party file; the statewide party mix where a county has none) and added to every scenario as ballots with no noise: every same-day registrant votes. Address changes at the polls are voters already on the rolls and add nothing. GG22 serves as the same-day reference for all scenarios: 2018 allowed same-day registration only during early voting and its same-day files are not in the data. Rates on active keep the September-based denominator.
Central turnout rates on active registration run from 43.3% in Baltimore City to 65.4% in Talbot. In 23 of 24 counties the central rate lies between the county’s 2022 and 2018 rates. Relative to its forecast, the 90% interval is wider than the statewide one (3.5% of the forecast) in 24 of 24 counties, because cell noise averages out across the state.
| County | Ballots cast | 90% interval (within-scenario) | Rate on active | Rate on eligible (projected) | Share of state ballots | 2022 rate on active | 2022 rate on eligible | 2018 rate on active |
|---|---|---|---|---|---|---|---|---|
| Talbot | 18,916 | 17,667 to 20,184 | 65.4% | 61.6% | 0.8% | 62.7% | 59.7% | 68.4% |
| Kent | 9,264 | 8,638 to 9,891 | 64.3% | 60.5% | 0.4% | 62.0% | 58.4% | 67.0% |
| Howard | 148,756 | 138,708 to 158,399 | 61.8% | 58.1% | 6.4% | 57.4% | 54.0% | 67.6% |
| Queen Anne’s | 25,758 | 24,122 to 27,430 | 61.7% | 58.7% | 1.1% | 59.0% | 56.1% | 64.9% |
| Carroll | 78,232 | 72,740 to 83,745 | 60.1% | 56.8% | 3.4% | 57.1% | 54.1% | 62.6% |
| Worcester | 26,811 | 24,870 to 28,788 | 60.1% | 55.9% | 1.2% | 57.6% | 53.7% | 62.4% |
| Calvert | 41,539 | 38,374 to 44,686 | 59.2% | 54.5% | 1.8% | 55.9% | 51.5% | 61.1% |
| Harford | 116,626 | 109,077 to 123,853 | 58.8% | 54.8% | 5.0% | 55.2% | 51.6% | 61.9% |
| Frederick | 122,194 | 114,124 to 130,128 | 58.8% | 54.8% | 5.2% | 55.0% | 51.6% | 62.9% |
| Garrett | 12,195 | 11,084 to 13,151 | 57.7% | 54.1% | 0.5% | 56.3% | 53.1% | 57.0% |
| Montgomery | 400,011 | 371,083 to 428,691 | 57.0% | 51.3% | 17.2% | 51.5% | 46.3% | 63.0% |
| Anne Arundel | 238,116 | 221,694 to 253,589 | 56.7% | 52.1% | 10.2% | 53.7% | 49.6% | 60.1% |
| St. Mary’s | 43,273 | 40,170 to 46,520 | 54.9% | 49.7% | 1.9% | 50.5% | 45.9% | 58.0% |
| Dorchester | 13,302 | 12,257 to 14,343 | 54.8% | 52.3% | 0.6% | 51.5% | 49.1% | 58.9% |
| Baltimore County | 317,742 | 296,070 to 339,075 | 54.2% | 50.9% | 13.6% | 49.3% | 46.1% | 59.4% |
| Somerset | 7,973 | 7,308 to 8,627 | 54.1% | 52.0% | 0.3% | 49.3% | 47.5% | 60.2% |
| Caroline | 12,347 | 11,297 to 13,307 | 53.9% | 50.5% | 0.5% | 50.4% | 47.5% | 57.5% |
| Allegany | 24,098 | 22,099 to 26,118 | 53.8% | 49.9% | 1.0% | 50.5% | 47.2% | 55.8% |
| Cecil | 38,814 | 35,747 to 41,772 | 52.5% | 48.6% | 1.7% | 47.5% | 43.8% | 54.5% |
| Charles | 68,420 | 61,705 to 74,636 | 52.4% | 48.3% | 2.9% | 45.5% | 42.0% | 58.5% |
| Washington | 52,972 | 48,887 to 57,181 | 52.3% | 46.9% | 2.3% | 49.9% | 45.0% | 54.2% |
| Wicomico | 34,537 | 31,336 to 37,476 | 51.3% | 47.7% | 1.5% | 47.5% | 44.4% | 55.5% |
| Prince George’s | 300,118 | 273,969 to 326,570 | 48.2% | 45.0% | 12.9% | 40.6% | 38.1% | 55.8% |
| Baltimore City | 178,851 | 158,717 to 200,040 | 43.3% | 40.5% | 7.7% | 37.3% | 34.9% | 48.1% |
The forecast’s electorate is less Democratic than UMBC’s likely voters, with a larger share of other voters. In the central scenario, Democrats cast 53.9% of ballots, against 60% in UMBC’s sample. Other voters cast 19.9% of ballots in the central scenario, up from 16.0% in 2022. Their active registration has grown 20% since then, which drives the change.
| Electorate | Democrats | Republicans | Other (unaffiliated and minor) |
|---|---|---|---|
| 2026 low | 53.9% | 27.6% | 18.5% |
| 2026 central | 53.9% | 26.2% | 19.9% |
| 2026 high | 54.7% | 25.5% | 19.8% |
| 2022 actual | 55.9% | 28.1% | 16.0% |
| 2018 actual | 57.3% | 27.4% | 15.3% |
| UMBC Institute of Politics likely voters, September 22 to September 27, 2026 (n = 776, unweighted) | 60% | 25% | 15% |
The forecast’s party groups are registered party. A poll’s party shares come from its sample, which may record party as respondents describe it. Many unaffiliated registrants call themselves Democrats or Republicans when asked. Likely-voter screens also keep the respondents most certain to vote, and unaffiliated voters are the least likely to say so: UMBC’s own enthusiasm figures are lower for them. UMBC’s shares are unweighted by party, so they describe who answered, not the weighted estimate. The poll is a cross-check here and does not move the forecast.
Composition favours turnout in every cell, but the forecast uses it only to shift turnout between counties. Older electorates and fewer brand-new registrants raise each party’s expected 2026 rate. The statewide level of that boost cannot be trusted, so it is removed.
| Party | Age factor | New-registrant factor | Composition (comp) | County range of comp |
|---|---|---|---|---|
| DEM | +0.035 | +0.016 | +0.051 | +0.036 (Montgomery) to +0.078 (Talbot) |
| REP | +0.049 | +0.031 | +0.080 | +0.060 (Prince George’s) to +0.099 (Queen Anne’s) |
| OTH | +0.023 | +0.011 | +0.034 | +0.021 (Montgomery) to +0.058 (Carroll) |
| County | Party | 2026 registrants (voter file) | comp | comp, centred within party |
|---|---|---|---|---|
| Talbot | DEM | 11,108 | +0.078 | +0.026 |
| Somerset | DEM | 5,502 | +0.077 | +0.026 |
| Wicomico | DEM | 27,611 | +0.074 | +0.023 |
| Harford | DEM | 67,703 | +0.074 | +0.023 |
| Carroll | OTH | 36,019 | +0.058 | +0.022 |
Every scenario uses comp_centred_cp = comp_cp minus the weighted mean over counties of comp for party p (weights: 2026 projected active registration, denominators_gg26.csv). composition_factors.csv’s comp is positive in every cell (statewide DEM +0.051, REP +0.080, OTH +0.034), but its statewide level is not identified: the 2022 mix comes from today’s voter file, which lacks the ~13% of the 2022 electorate who have since died or moved, and that bias is as large as the factor. Centred, composition only redistributes turnout across counties within a party and the level comes from the calibrated scenario. central_comp_level uses the uncentred comp with the central level; its statewide difference from central is the measure ballots_minus_central in forecast_statewide.csv. The midterm structure effect s_cp is treated as describing the 2022 composition, and comp is a change from it.
The factors are centred within party for one reason. The 2022 age mix comes from today’s voter file, which lacks the roughly 13% of the 2022 electorate who have since died or moved. That gap is as large as the factor itself, so its statewide level is not identified. Using the uncentred factor as a sensitivity adds +55,641 ballots to the central forecast, +1.3 points on the rate on active registration (55.5% against 54.2%).
Leaving out one midterm and predicting it from the other elections, the model beats a naive copy of the other midterm on 10 of 12 measures. Its county turnout rates on active registration miss by 2.2 points on average for 2022 and 2.2 for 2018. Both held-out tests calibrate statewide party totals to the actual result, so they test the county pattern, not the level.
| Held-out election | Grain | Measure | Model | Naive | Better |
|---|---|---|---|---|---|
| 2022 general | County | Mean abs. error, rate on active (points) | 2.25 | 2.54 | model |
| 2022 general | County | Largest abs. error, rate on active (points) | 4.16 | 6.99 | model |
| 2022 general | County | Mean abs. error, share of state ballots (points) | 0.17 | 0.17 | naive |
| 2022 general | County x party | Mean abs. error, rate on active (points) | 2.69 | 3.02 | model |
| 2022 general | County x party | Largest abs. error, rate on active (points) | 8.68 | 12.13 | model |
| 2022 general | County x party | Mean abs. error, share of state ballots (points) | 0.06 | 0.07 | model |
| 2018 general | County | Mean abs. error, rate on active (points) | 2.17 | 2.52 | model |
| 2018 general | County | Largest abs. error, rate on active (points) | 8.08 | 7.08 | naive |
| 2018 general | County | Mean abs. error, share of state ballots (points) | 0.12 | 0.14 | model |
| 2018 general | County x party | Mean abs. error, rate on active (points) | 2.63 | 3.04 | model |
| 2018 general | County x party | Largest abs. error, rate on active (points) | 11.34 | 11.36 | model |
| 2018 general | County x party | Mean abs. error, share of state ballots (points) | 0.05 | 0.06 | model |
| Held-out election | County | Actual rate on active | Model error (points) | Naive error (points) |
|---|---|---|---|---|
| 2018 general | Garrett | 57.0% | +8.1 | +7.1 |
| 2018 general | Washington | 54.2% | +4.5 | +4.8 |
| 2018 general | Allegany | 55.8% | +4.1 | +3.5 |
| 2022 general | Garrett | 56.3% | -4.2 | -7.0 |
| 2022 general | Kent | 62.0% | -4.1 | -4.0 |
| 2022 general | Prince George’s | 40.6% | +4.0 | +4.2 |
The model loses to naive in 2 places:
Mail ballots make up a smaller share of the forecast vote than in 2022, 24.0% against 26.7%. The largest mode is election day, at 46.9%.
| Mode | Ballots cast, 2026 central | Share, 2026 central | Share, 2022 actual |
|---|---|---|---|
| Mail-in | 560,543 | 24.0% | 26.7% |
| Early voting | 604,396 | 25.9% | 18.7% |
| Election day | 1,093,167 | 46.9% | 51.5% |
| Provisional | 72,759 | 3.1% | 3.1% |
| Total | 2,330,865 | 100.0% | 100.0% |
Early voting is the board’s unofficial count, 604,396 statewide; early voting is over. Mail ballots are the 376,967 received by October 30, 2026 times 2022’s late-arrival multiplier in each county (certified vote-by-mail over the ballots back the same number of days before that election), capped at the ballots sent. That is 560,543 mail ballots, 85.6% of the 654,876 sent. Provisional ballots keep their 2022 share of each county’s vote, and election day is the rest. No county hit the cap of 90% of its ballots by mail.
| County | Ballots cast | Mail-in | Early voting | Election day | Provisional | Mail share | Mail share, 2022 | Early share | Early share, 2022 | Return rate |
|---|---|---|---|---|---|---|---|---|---|---|
| Allegany | 24,098 | 3,581 | 3,598 | 16,333 | 586 | 14.9% | 17.7% | 14.9% | 10.5% | 74.4% |
| Anne Arundel | 238,116 | 67,547 | 63,281 | 100,838 | 6,450 | 28.4% | 27.5% | 26.6% | 17.8% | 97.6% |
| Baltimore City | 178,851 | 36,615 | 33,389 | 98,405 | 10,442 | 20.5% | 31.9% | 18.7% | 15.2% | 66.1% |
| Baltimore County | 317,742 | 70,736 | 78,631 | 160,042 | 8,333 | 22.3% | 28.6% | 24.7% | 20.1% | 74.5% |
| Calvert | 41,539 | 8,827 | 13,478 | 18,494 | 740 | 21.2% | 23.7% | 32.4% | 23.2% | 85.4% |
| Caroline | 12,347 | 1,311 | 3,098 | 7,793 | 145 | 10.6% | 13.1% | 25.1% | 18.8% | 73.2% |
| Carroll | 78,232 | 13,688 | 18,498 | 44,828 | 1,218 | 17.5% | 20.0% | 23.6% | 19.5% | 77.8% |
| Cecil | 38,814 | 5,877 | 9,564 | 22,632 | 741 | 15.1% | 17.3% | 24.6% | 18.7% | 75.9% |
| Charles | 68,420 | 11,946 | 22,258 | 32,426 | 1,790 | 17.5% | 21.6% | 32.5% | 21.6% | 75.4% |
| Dorchester | 13,302 | 2,009 | 3,335 | 7,771 | 187 | 15.1% | 19.1% | 25.1% | 17.5% | 79.5% |
| Frederick | 122,194 | 26,770 | 28,642 | 63,847 | 2,935 | 21.9% | 23.3% | 23.4% | 14.8% | 85.0% |
| Garrett | 12,195 | 1,649 | 2,774 | 7,604 | 168 | 13.5% | 15.0% | 22.7% | 20.8% | 81.1% |
| Harford | 116,626 | 17,844 | 33,337 | 63,450 | 1,995 | 15.3% | 18.1% | 28.6% | 21.5% | 77.6% |
| Howard | 148,756 | 33,052 | 38,861 | 73,391 | 3,452 | 22.2% | 25.9% | 26.1% | 20.1% | 79.7% |
| Kent | 9,264 | 1,657 | 2,554 | 4,870 | 183 | 17.9% | 21.0% | 27.6% | 22.4% | 83.3% |
| Montgomery | 400,011 | 141,085 | 93,190 | 149,726 | 16,010 | 35.3% | 34.2% | 23.3% | 16.1% | 100.0% |
| Prince George’s | 300,118 | 79,554 | 96,772 | 110,534 | 13,258 | 26.5% | 28.4% | 32.2% | 21.4% | 89.9% |
| Queen Anne’s | 25,758 | 3,818 | 8,059 | 13,394 | 487 | 14.8% | 16.4% | 31.3% | 27.1% | 81.0% |
| Somerset | 7,973 | 998 | 1,983 | 4,826 | 166 | 12.5% | 16.0% | 24.9% | 20.1% | 73.8% |
| St. Mary’s | 43,273 | 7,570 | 14,325 | 20,473 | 905 | 17.5% | 19.3% | 33.1% | 23.8% | 82.9% |
| Talbot | 18,916 | 4,061 | 6,920 | 7,680 | 255 | 21.5% | 23.7% | 36.6% | 29.3% | 89.3% |
| Washington | 52,972 | 9,662 | 13,112 | 29,219 | 979 | 18.2% | 20.3% | 24.8% | 14.9% | 78.9% |
| Wicomico | 34,537 | 6,111 | 9,635 | 18,001 | 790 | 17.7% | 19.6% | 27.9% | 19.8% | 83.8% |
| Worcester | 26,811 | 4,575 | 5,102 | 16,590 | 544 | 17.1% | 20.2% | 19.0% | 17.2% | 78.5% |
Run the whole model and this report from the repo root, after
make build-db:
make turnout-model
The scripts use the renv project in
scripts/turnout_model/ (renv.lock); see
scripts/turnout_model/README.md. The forecast reads the
board’s mail-in files through October 30, 2026 and its early-voting and
same-day files through October 29, 2026. Every input and output has a
provenance file:
| File | Provenance file | Made by | Run |
|---|---|---|---|
| data/turnout_model/inputs/covariates.csv | data/turnout_model/inputs/covariates_provenance.json | 01_build_inputs.R | 2026-11-01 |
| data/turnout_model/inputs/denominators_gg26.csv | data/turnout_model/inputs/denominators_gg26_provenance.json | 01_build_inputs.R | 2026-11-01 |
| data/turnout_model/inputs/inactive_ratio.csv | data/turnout_model/inputs/inactive_ratio_provenance.json | 01_build_inputs.R | 2026-11-01 |
| data/turnout_model/inputs/indices.csv | data/turnout_model/inputs/indices_provenance.json | 01_build_inputs.R | 2026-11-01 |
| data/turnout_model/inputs/polls_2026.csv | data/turnout_model/inputs/polls_2026_provenance.json | 01_build_inputs.R | 2026-10-17 |
| data/turnout_model/inputs/training.csv | data/turnout_model/inputs/training_provenance.json | 01_build_inputs.R | 2026-11-01 |
| data/turnout_model/gg26/backtest_cells.csv | data/turnout_model/gg26/backtest_cells_provenance.json | 02_fit.R | 2026-10-17 |
| data/turnout_model/gg26/backtest.csv | data/turnout_model/gg26/backtest_provenance.json | 02_fit.R | 2026-10-17 |
| data/turnout_model/gg26/composition_factors.csv | data/turnout_model/gg26/composition_factors_provenance.json | 03_compose.R | 2026-10-17 |
| data/turnout_model/gg26/early_vote_check.csv | data/turnout_model/gg26/early_vote_check_provenance.json | 04_forecast.R | 2026-11-01 |
| data/turnout_model/gg26/election_level_draws.csv.gz | data/turnout_model/gg26/election_level_draws_provenance.json | 02_fit.R | 2026-10-17 |
| data/turnout_model/gg26/election_levels.csv | data/turnout_model/gg26/election_levels_provenance.json | 02_fit.R | 2026-10-17 |
| data/turnout_model/gg26/forecast_county_party.csv | data/turnout_model/gg26/forecast_county_party_provenance.json | 04_forecast.R | 2026-11-01 |
| data/turnout_model/gg26/forecast_county.csv | data/turnout_model/gg26/forecast_county_provenance.json | 04_forecast.R | 2026-11-01 |
| data/turnout_model/gg26/forecast_statewide.csv | data/turnout_model/gg26/forecast_statewide_provenance.json | 04_forecast.R | 2026-11-01 |
| data/turnout_model/gg26/level_rule.csv | data/turnout_model/gg26/level_rule_provenance.json | 04_forecast.R | 2026-11-01 |
| data/turnout_model/gg26/level_rule_text.csv | data/turnout_model/gg26/level_rule_text_provenance.json | 04_forecast.R | 2026-11-01 |
| data/turnout_model/gg26/mode_split.csv | data/turnout_model/gg26/mode_split_provenance.json | 04_forecast.R | 2026-11-01 |
| data/turnout_model/gg26/reference_levels.csv | data/turnout_model/gg26/reference_levels_provenance.json | 04_forecast.R | 2026-11-01 |
| data/turnout_model/gg26/same_day_registration.csv | data/turnout_model/gg26/same_day_registration_provenance.json | 04_forecast.R | 2026-11-01 |
| data/turnout_model/gg26/structure_draws.csv.gz | data/turnout_model/gg26/structure_draws_provenance.json | 02_fit.R | 2026-10-17 |
| data/turnout_model/gg26/structure_effects.csv | data/turnout_model/gg26/structure_effects_provenance.json | 02_fit.R | 2026-10-17 |
| data/turnout_model/gg26/fit_summary.json | (self-describing) | 02_fit.R | 2026-10-17 |