Bangladesh has held seven general elections since the 1991 transition to parliamentary democracy. The BPO record contains 5,244 incidents tagged with the cross-cutting Election theme — a tiny fraction of the dataset overall, but with an outsized place in the country's political memory. Three of those elections produced incident spikes large enough to dominate their entire year. The other four didn't.
This isn't an essay about whether Bangladeshi elections are "free and fair." It's a much narrower question: when researchers and journalists code an event as election-related, what does the BPO record actually contain? And what changes between cycles?
The shape of an election year
In each of the BPO's coded election years — 1991, 1996, 2001, 2008, 2014, 2018, 2024 — there is a consistent envelope. Incidents begin appearing in the record several months ahead of the polling date, peak in the two weeks immediately around the vote, and tail off slowly into the new parliament's first quarter. Within that envelope, the actor composition shifts: pre-poll incidents skew toward party-on-party clashes; post-poll incidents skew toward security-services interactions, often involving arrests.
This pattern is stable across cycles. What varies is the magnitude. The 2014 election — boycotted by the main opposition — registered roughly four times the incident count of the previous cycle. 2018, a contested cycle, similarly elevated. 2008 was an outlier in the opposite direction: a caretaker-government election with low BPO incident counts overall and a markedly more uniform actor profile.
Three frames to read the trend
1. Pre-poll intimidation. The classical pattern, most visible in 1996 and 2001: clashes between party supporters, ballot-box rigging attempts, candidate intimidation. These show up in the BPO record under the "Political clash" and "Intimidation" violence types. The 1996 record is dense here; the 2014 record, less so — the boycott meant fewer formal contests to produce that kind of incident.
2. Polling-day violence. Booth captures, voter suppression, retaliatory attacks on the day of the vote. These incidents are short-lived but concentrated geographically — a small number of upazillas account for an outsized share. Cross- referencing against the District profile pages shows the same upazillas recurring across multiple election cycles.
3. Post-poll arrests. The fastest-growing category across the last three cycles. Security services moving on opposition supporters in the days and weeks after the result. The BPO arrest counter (11,141 cumulative arrests across the entire Election theme) is heavily weighted to this post-poll period.
What's missing
The Election theme tracks reported incidents. Detentions without filed cases, intra-household coercion, and the routine administrative friction of polling are all under-represented or absent. Read the record as a lower bound on visible political violence, not as a census of political tension.
A second caveat: the BPO research desk codes "Election" as a cross-cutting theme only when the incident has a clear electoral nexus reported in the source article. An incident in a small town the week before a vote may or may not be coded as election-related depending on how the source framed it. This is editorial judgment, applied consistently but visibly to anyone who reads enough rows.