# Sources — Bipartisan Voting Chart
## "Congress Has Stopped Deliberating"
### Essay: It Begins With "Why?" (GitHub Issue #107)
### Last updated: August 2026 (see Corrections below)

---

## What This Chart Shows

**The share of House roll-call votes on which the two parties did *not* split** — where a majority of Democrats and a majority of Republicans ended up on the same side — from 1969 to 2025. It falls from 73% in 1970 to 15% in 2025.

### Polarity note (2026-08)

This chart plots the **complement** of the CQ Roll Call party-unity measure, i.e. `100 − party-unity-vote %`. The source series counts votes that *split* the parties; this chart counts the ones that didn't, so the line falls as the capacity for cross-party agreement falls. No new data is involved — every plotted value is one subtraction away from the published figure, and both columns are in `data.csv` (`house_bipartisan_votes_pct` and the original `house_party_unity_votes_pct`).

Where the draft quotes CQ Roll Call's own language — "2025 was the highest level of party-unity voting in the history of the study" — quote their 85.3%, not this chart's 14.7%. Same fact, two directions; mixing them inside a paragraph will lose the reader.

### Two metrics, both from the same CQ Roll Call study

1. **House bipartisan votes as a percentage of all House votes** (the plotted line) — the share of roll calls where the two party majorities agreed. Derived as 100 minus the party-unity-vote share.

2. **Average member defection rate** (callout stats 3 and 4) — the percentage of party-unity votes on which the average member voted *against* the majority of their own party. Derived as 100 minus the published party unity score. Not plotted; included in `data.csv` as `house_dem_defection_rate` and `house_gop_defection_rate`.

The chart displays metric 1 only as the primary visual argument, shown as the jagged annual series (navy) with an 11-year (~10-year) centered moving-average trend (oxblood) — the year-to-year series is genuinely volatile (swings of 15–35 points between adjacent years are normal throughout the range, e.g. 2010's 60.0% to 2011's 24.2%), and the trend line is there to make the underlying direction legible without smoothing away that real volatility.

### What "bipartisan" does and does not mean here

This is the residual of a specific technical definition, not a measure of deliberation, and the chart says so in its caveat box. A vote counts as bipartisan whenever the two party majorities land on the same side — which includes unanimous votes, near-unanimous votes, and routine procedural business. A high bipartisan share in 1970 is therefore not proof that the House was debating well.

The claim the collapse *does* support is narrower and harder to dismiss: the set of questions on which the two parties can end up on the same side at all has shrunk from roughly three votes in four to roughly one in seven. That is a statement about the range of available agreement, not about the quality of argument, and the draft should not stretch it past that.

---

## Primary Sources

### Brookings Institution — Vital Statistics on Congress, Chapter 8
- **URL:** https://www.brookings.edu/wp-content/uploads/2024/11/Chpt-8.pdf
- **Tables used:** Table 8-3 (Party Unity Votes in Congress, 1953–2022) and Table 8-4 (Party Unity Scores in Congressional Voting, 1954–2022)
- **Coverage:** 1969–2022
- **Notes:** Brookings republishes the CQ Roll Call party unity series. Their notes state that recent updates use tabulations based on Voteview roll-call data. The original series was compiled by Congressional Quarterly beginning after World War II. Vital Statistics on Congress was first published in 1980 as a joint effort by Thomas E. Mann (Brookings) and Norman J. Ornstein (AEI), in collaboration with Michael Malbin (Campaign Finance Institute).
- **Definition used:** A party unity vote is "a recorded vote in the Senate or House that splits the parties — that is, a majority of voting Democrats opposed a majority of voting Republicans. Party unity scores represent the percentage of such votes on which a member voted 'yea' or 'nay' in agreement with a majority of his or her party. Percentages are normalized to eliminate the effects of absences."

### CQ Roll Call Annual Vote Studies (2023–2025)
- **URL (2024 study):** https://rollcall.com/2025/02/18/congress-party-unity-vote-studies/
- **URL (2023 study):** https://rollcall.com/2024/02/08/house-gop-had-lowest-win-rate-on-party-unity-votes-since-1982/
- **Coverage:** 2023–2025
- **Notes:** CQ Roll Call has published annual party unity studies since the end of World War II. The Brookings series draws from this same source. For years beyond the Brookings 2022 endpoint, CQ Roll Call's published summaries are used directly. The 2025 figure of 85.3% is cited as the highest level in the history of CQ's study. The 2024 House figure (65.3%) is drawn from this same 2024 study and is now plotted on the chart (see Corrections).

### Voteview (supplementary cross-check)
- **URL:** https://legacy.voteview.com/Party_Unity.htm
- **Coverage:** 35th–113th Congresses
- **Notes:** Used for cross-checking Brookings figures. Voteview provides member-level party unity scores by Congress in downloadable Excel format.

---

## Methodology Notes

- **Party-line vote definition is consistent** across the full dataset: majority of one party opposed majority of the other. This is the standard CQ Roll Call definition used since the series began. The bipartisan share plotted here is its exact arithmetic complement, so the definition governs both.
- **Defection rates shown** in the callout statistics are House averages across both parties, derived as 100 minus the average unity score. They are computed from the *rounded* published unity figures (1970: 72% unity → 28% defection; 2025: 94% → 6%) so that the two charts' numbers sum to 100 for a reader comparing them. Computing from the unrounded party scores would give 28.5% for 1970. Individual party scores and derived defection rates are both in data.csv.
- **The 2025 data point** is sourced from CQ Roll Call's report covering that year, which noted it as the highest level in the history of their study. **Caveat carried over from the prior version and worth resolving:** this figure covers both chambers combined, while every other point in the series is House-only. It is also the chart's headline number. Worth confirming against a House-only 2025 figure before the chapter goes to press.
- **Trend line:** the chart's oxblood line is an 11-year centered moving average of the House bipartisan-votes series (`house_bipartisan_trend_10yr_centered` in data.csv; the un-inverted original is retained as `house_trend_10yr_centered`). Because inversion is a linear transform, the inverted moving average equals 100 minus the original — no re-smoothing was done and the inversion introduces no smoothing artifact — i.e., the true annual-data equivalent of the "~10-year" window the companion agenda-resolution chart (`book/obs01-unresolved-agenda`) uses via its 5-Congress centered average, so the two charts describe their trend windows the same way. A shortened window is used at the series' endpoints (1969, 2025). It is computed from, but not a substitute for, the raw annual series, which remains the primary line on the chart. The moving average bridges the 2023 data gap using the surrounding years. (This window was widened from an initial 5-year average — see Corrections.)
- **Genuine data gap — 2023:** No source (Brookings, CQ Roll Call, or Voteview) publishes a House party-unity-votes percentage for 2023; CQ Roll Call's 2023 study year did not break out this particular figure. This is a real gap, not an oversight — data.csv leaves the 2023 house_party_unity_votes_pct cell blank, and the chart shows a visible break in the raw (navy) line at 2023 rather than silently connecting 2022 to 2024 as if no year were missing. (The Senate party-unity-votes figure for 2023, 81.0%, is available and included in data.csv, since it comes from a different part of the same underlying CQ study.)
- **The 1974 reference line has been removed from this chart** (as of July 2026 — see Corrections). The marker was built for the earlier essay (#107, "It Begins With Why?"), which had already named the FECA/transparency-reform argument; this chart now supports Chapter 2 (observation), which hasn't made that argument yet, so the marker was visually importing a case this chapter doesn't assert. This is a chapter-wide fix, not specific to this chart — the same marker was stripped from every other Ch. 2 observation chart that had one (Gridlock, Legislation Decline, Moderate Decline, Congressional Approval). The Convergence synthesis chart intentionally keeps its 1974 indexing, since that chart is where the reform argument is actually made. (Secondary, chart-specific note: on this series 1974 also sits in a local dip — 29.4%, lower than both 1971 and 1973 — so the line wouldn't have lined up with anything visible here even if kept. That's supporting color, not the main reason.)

---

## Known Limitations and Disputes

- **"Bipartisan" here is a residual category, not a quality judgment.** See "What 'bipartisan' does and does not mean here" above. This is the chart's main interpretive risk: the inverted framing invites a reader to hear "bipartisan" as "deliberative," which the measure does not support.
- Party unity scores measure voting alignment, not causal relationship. The chart does not itself prove that transparency reforms caused increased party-line voting — that argument is made in the essay using additional evidence. The chart shows a correlation that is consistent with the structural prediction.
- The Voteview and CQ Roll Call methodologies differ slightly in how they handle absences and procedural votes. Brookings normalizes for absences; raw counts may differ slightly. The trend direction is consistent across all methodologies.
- Pre-1969 data exists in the Brookings series (back to 1953) but is not shown in this chart, which focuses on the post-LRA period. Full historical data is available upon request.

---

## Corrections

| Date | Change |
|------|--------|
| 2026-08-20 | **Inverted the chart** at editorial request, to remove the "higher is worse" reading cost across Chapter 2. Was the party-unity-vote share (rising 27% → 85%); now its complement, the bipartisan-vote share (falling 73% → 15%). Values are arithmetically identical (`100 − x`); the title is unchanged. Specific changes: (1) subtitle, legend, tooltip, and source footer reworded to the bipartisan framing, with the footer now stating the inversion explicitly; (2) callouts 1 and 2 flipped from 27%/85% party-line to 73%/15% bipartisan; (3) callouts 3 and 4 changed from average member *unity* score (72%/94%) to average member *defection* rate (28%/6%), so all four stats fall together; (4) added a caveat box defining what does and does not count as bipartisan, since the inverted framing invites a reading the measure cannot support; (5) y-axis mirrored from 20–90% to 10–80%, preserving the original's visual amplitude exactly; (6) added `house_bipartisan_votes_pct`, `house_bipartisan_trend_10yr_centered`, `house_dem_defection_rate`, `house_gop_defection_rate`, and `senate_bipartisan_votes_pct` to data.csv, retaining every original column. The 2023 data gap is unaffected and still shows as a break in the annual line. |
| 2026-07-17 | (1) Removed the 1974 reference-line marker. It was built for the earlier essay (#107) which had already named the FECA/transparency-reform argument; this chart now supports Chapter 2 (observation), which hasn't made that argument yet, so any surviving marker visually imported an unmade case. This is a chapter-wide correction — the same marker was also removed from the Legislation Decline, Moderate Decline, and Congressional Approval charts (Gridlock already had it removed previously). Secondary/supporting note: on this data 1974 also sits in a local dip (29.4%, lower than 1971 and 1973), so it wouldn't have lined up with anything visible here regardless. (2) Added a 5-year centered moving-average trend line (oxblood) alongside the raw jagged series (navy, unchanged) to make the underlying direction legible given genuine year-to-year swings of 15-35 points; added `house_trend_5yr_centered` to data.csv. (3) Added the 2024 House party-unity-votes data point (65.3%, already present in data.csv and sourced from the CQ Roll Call 2024 study cited above) to the plotted series — it had been omitted, leaving the chart jumping straight from 2022 to 2025. (4) Corrected the "avg. member unity score in 1970" callout from 74% to 72%: the 74% figure had actually been computed from 1971's House dem/gop scores (72, 76) rather than 1970's (71, 72, average 71.5% → 72%). (5) Documented 2023 explicitly as a genuine House party-unity-votes data gap (no source publishes this figure for that year) rather than leaving it silently absent; the chart now shows a visible break in the raw line at 2023, bridged by the trend line. |
| 2026-07-17 | Widened the trend line from a 5-year to an 11-year centered moving average, for consistency with the companion Legislative Gridlock chart (`book/obs01-unresolved-agenda`), whose 5-Congress centered average is ~10 calendar years. Eleven years is the annual-data equivalent of that window. Renamed `house_trend_5yr_centered` to `house_trend_10yr_centered` in data.csv and recomputed all values; updated the chart legend and tooltip label from "5-year moving average" / "5-yr average" to "10-yr average"; updated the Methodology Notes trend-line bullet accordingly. The raw annual series (navy) and all callout statistics are unaffected. |

---

## Version History

| Date | Change | Author |
|------|--------|--------|
| May 2026 | Initial publication | TSB |
| July 2026 | Corrected 1970 callout stat, added 2024 data point, added trend line, removed chart-specific 1974 marker (full detail in Corrections above) | TSB |
| July 2026 | Widened trend line from 5-year to 11-year (~10-year) centered moving average, matching the companion agenda chart's window convention (full detail in Corrections above) | TSB |
| August 2026 | Inverted the plotted series to the bipartisan share so the line falls; reworked all four callout stats; added the "what bipartisan means" caveat box (full detail in Corrections above) | TSB |

*If you believe a data point is incorrect, please open an issue at https://github.com/thestatecraftblueprint/tsb-charts or email jason@statecraftblueprint.org. We will review, correct if warranted, and document the change in this version history.*