Five Democratic groupings, 1994–2022
Every Democratic general-election candidate of the last eight cycles, grouped by the issue positions their roll calls, sponsorships, questionnaires and endorsements actually record — then measured against the fixed WAR reference model.
Five position profiles, formed without looking at any election result
Clustering uses the eight ontology issue families only. Residual WAR, wins and incumbency are attached afterwards, so no electoral outcome can place a member in a group. Unscored candidate-cycles are seats with no contested general election; they are not zeros.
Where the groups actually differ
Each mark is a group's mean on one issue family. Every row puts the pole this project codes as liberal on the left and the other pole on the right, so left always means the same thing. The count beside each row is how many people have evidence on that family; thin rows should be read as thin.
Each mark is one person with evidence on both families, shaped and colored by group. Choose a group to outline it, or drag a box over marks to list those people in the members table.
Overperformance against the reference model
WAR is the candidate-oriented race residual in two-party margin points: the actual legislative-minus-ticket gap minus the fitted structural expectation of the fixed 2018–24 reference model. Pre-2016 cycles are scored against modern partisan expectations, so early-era Democrats show large positive WAR partly by construction.
Which groups the Democratic field contained, by cycle
Counts are candidate-cycles whose person was grouped. A cycle’s bar excludes candidates without enough issue evidence, so bars are not chamber totals.
Every grouped person, with the evidence behind them
Select a row for that person’s issue profile. “Families observed” is how many of the eight issue families their record covers; three is the minimum to be grouped.
| Person | Cycles | Group | Families | Career WAR | Scored |
|---|
Who is in the analysis, and who is not
Nobody is dropped silently. People without enough issue evidence are counted and reported with the reason.
Cluster sensitivity, stated plainly
Each bar is an adjusted Rand index between the published grouping and the grouping produced by an alternative reasonable choice: 1.0 is identical, 0 is no better than chance. These are descriptive tendencies, not formal caucus membership.
What these groups are not
They are not the Alabama Legislative Black Caucus or any other organised caucus, which have real membership rolls this analysis does not use. They are not a pre-election snapshot: a member’s later roll calls help define the group used to read their earlier race. They are not causal — a group’s mean WAR describes the races its members happened to run, including opponent quality and local conditions this measure cannot separate.
How this page was produced
Grouping
WAR definition
The residual framing follows Split Ticket’s WAR methodology; this Alabama implementation and its estimates are independent. Race WAR = actual legislative-minus-ticket gap minus the fitted structural expected gap. Democratic WAR is the race residual and Republican WAR is its exact negative. One fixed reference model, fitted on 2018–24 Southern races, scores every cycle; no era-specific refit is applied. Career cumulative WAR is the sum of a person’s scored cycle WAR. No pooled individual effect, fundraising term, or ideology term enters WAR.
Evidence
Source credit
Roll-call and bill records from LegiScan (CC BY 4.0) and the Alabama Legislature; candidate questionnaires and interest-group ratings summarised from Vote Smart; election identities and totals from the Alabama Secretary of State. Only derived summaries are published here.