Every figure on this site is computed by the same published rules, each with a version. Exports and citations carry the versions, so a figure can be reproduced after the rules change.
Every tally, line, cohesion, loyalty and dissent figure counts a member in the group they belonged to on the day of the vote, from the Parliament’s membership records. Members who changed group are marked ⇄ with their dates; former members are counted and listed with the day they left. Where no membership record exists the current group is used, and that is flagged. See the metric.
Named every time a line is used.
| Rule | Definition | Used for |
|---|---|---|
| Majority line | The position most of the group’s voting members took (plurality). A tie has no majority line. | Voted with the group’s majority line (group loyalty); the majority-line variant of group agreement. |
| Clear line | A position held by at least two thirds of the group’s votes cast, with at least 10 of its members voting. A group without one is divided; with fewer than 10 voting it has no clear line and is left out. | Voted against their group, agreement between groups, groups on the winning side, the dossier voting grid. |
Each metric with its version; the ⓘ next to a figure links here.
Each member is counted in the political group they belonged to on the day of the vote, from the European Parliament’s membership records. Consecutive spans in the same group are merged. A member without a membership record counts in their current group from their first recorded vote, and is flagged.
Votes cast (for, against or abstained) divided by the roll-call votes held while the member was in office — from the start of their first group membership of the term to the end of the last. Roll calls published without named positions are not counted. “No vote recorded” is not the same as absent. Members with no vote recorded at all in a period — the President chairs the sittings and does not vote — are marked, and left out of group medians and of orderings by rate. A member who changed group in the period is filed under the group they held for most of it, and not compared with one group’s median.
Abstentions divided by the votes the member cast (for, against or abstained) in the roll calls considered.
Share of a member’s votes that matched their group’s majority line — the position most of the group’s voting members took (plurality), group at the date of the vote. Votes where the group tied have no majority line and are left out. The non-attached have no group line.
A vote different from the clear line of the member’s group on the date of the vote: a position held by at least two thirds of the group’s votes cast, with at least ten group members voting — with fewer than ten voting a group has no clear line. Three cases count: opposite (for ↔ against), abstained against a for or against line, and voting for or against when the group’s clear line was to abstain (counted in the total, neither opposite nor abstained). A group without a clear line is divided, and nobody in it voted against it.
The Agreement Index of Hix, Noury and Roland (2005): (largest side − half of the other two) ÷ votes cast, from 0 (evenly split three ways) to 1 (unanimous). A bloc’s cohesion over a period is the mean over the votes where enough of its members voted; groups at the date of each vote. The two-option variant ignores abstentions. For national delegations the figure reflects the mix of groups their members sit in.
Shared votes (both recorded a position) where the two cast the same position, divided by all shared votes. Abstention counts as a position unless “ignore abstentions” is chosen. The baseline is the median agreement of all pairs of members of the two members’ groups, groups at the date: each member of a group counts only on the votes they sat in it, and for two members who sat in different combinations of groups over the period (one changed group) the medians of each combination are weighted by the votes the two shared in it. Exports add a 95 % Wilson interval.
Over the votes where both groups had a clear line (at least two thirds of their votes cast, at least ten members voting), the share where the two lines were the same; votes where either group was divided or had fewer than ten voting are left out of the count. The majority-line variant compares the plurality lines over the votes where both groups had one (not tied). Groups at the date of the vote.
For a final vote decided by a simple majority, the groups whose clear line (two thirds, at least ten members voting) matched the outcome — For when it passed, Against when it failed; divided groups and groups with fewer than ten voting are not in the set. Texts rejected although more voted for than against (an absolute majority of 361 was required) are counted separately and not read for a winning side. A group is pivotal on a vote when taking its members’ votes away would flip the simple-majority outcome; absolute-majority failures are not counted there either, since which other votes needed 361 is not recorded.
Pearson correlation between two votes over the members who voted in both, with for = +1, against = −1 and abstained = 0 (abstention treated as halfway). Votes of the same decision are left out.
Final votes: the vote that decided each decision, when it was recorded by name, agenda motions left out. All roll calls: amendments, split parts, separate votes and procedural motions as well. Roll calls are not a random sample: final votes on laws are recorded by name, most other roll calls because a group asked.
Every analysis and export carries the time the vote matrix was loaded, a data stamp, the metric versions and the code’s commit. The data stamp is a short hash of the source tables’ change stamps (row counts and last writes) — it changes whenever the data may have changed; it is not a checksum of the content. Exports also carry the filters as understood, the thresholds applied and the counts of what was left out; a CSV’s meta is served beside it as <name>.meta.json.
Below a threshold a figure is shown as a count (“3 of 7”) or left out, never as a rate. No rank numbers in member lists; ties are ties.
| Metric | Rule |
|---|---|
| Roll-call participation | A member’s rate needs 50 votes; below that the count is shown. |
| Roll-call participation | Lists ordered by a rate need 300 votes in the rate’s denominator (all roll calls) or 100 (final votes); the others follow by name. |
| Abstention rate | A member’s rate needs 50 votes cast; below that the count is shown. |
| Voted with the group’s majority line | A member’s rate needs 50 votes with a line. |
| Voted against the group’s clear line | A line counts when at least 10 of the group’s members voted. |
| Voted against the group’s clear line | Lists ordered by rate need 300 clear-line votes (all roll calls) or 100 (final votes). |
| Cohesion (Agreement Index) | A vote counts for a bloc when at least 10 of its members voted and at least a third of its seats voted (small blocs are biased upward). |
| Cohesion (Agreement Index) | Every figure — per part-session, quarter, month, the whole period, per committee or topic — needs at least 20 such votes; below that only the number of votes is shown. |
| Cohesion (Agreement Index) | Delegations under 10 seats are not ranked. |
| Cohesion (Agreement Index) | Monthly points only for all roll calls, with at least 20 votes. |
| Agreement between two members | A rate needs 50 shared votes. |
| Agreement between two members | Voting neighbours need 200 shared roll calls over at least six months in office together. |
| Agreement between two groups | A rate needs 20 votes. |
| Agreement between two groups | A clear line needs 10 of the group’s members voting. |
| Groups on the winning side and pivotal groups | A clear line needs 10 of the group’s members voting. |
| Similar votes | At least 200 shared voters; listed from r = 0.5, the ten highest. |
| Similar votes | Votes whose smaller side is under 10 % are skipped. |
| Final votes and all roll calls | Contested votes: the largest side holds less than 80 % of the votes cast. |
Every analysis result, export (meta) and citation carries:
Analysis citations pin the end date (to=), so a cited figure does not move after the next sitting. The members file has the group spans used.
A CSV export’s meta is served beside it as <name>.meta.json.
See Data & downloads.