Census indicators describe different dimensions of daily life. Reading them together reveals questions about access and inequality that a single national average can conceal.
- Observation coverage
- 2022 census · 2026 election comparison
- Source build date
- Not recorded by the source; see coverage above
- Website refresh policy
- Publication changes only; manual review
Published snapshot. A scheduled check does not imply that a new source observation is available.
Loading the published dataset…
Reading the evidence
01
Access is only the beginning of a service
An electricity connection says little about hours of supply, voltage quality or affordability. Internet use, permanent housing and sanitation capture other dimensions of household conditions. They should be read separately before being combined into an index: a high value in one domain cannot establish that another service works well. The explorer therefore starts with individual indicators and retains each indicator's original unit.
02
Local variation is not a national estimate
This extract was assembled for an election match, rather than as a complete census dissemination dataset. Metropolitan exclusions and missing indicators shape which upazilas appear. District shading is an unweighted mean of represented upazilas, not a district population rate. Small and large upazilas contribute equally, and the composition differs between districts. Use the underlying table to inspect coverage before comparing apparent geographic patterns.
03
An association does not explain a vote
Census conditions precede the election, but time order alone does not establish a causal effect. Geographic history, occupation, campaigning and candidate characteristics may be associated with both local conditions and voting. Aggregated results cannot establish how individual residents voted. The published regressions control for division and registered-voter count, but those controls cannot remove every alternative explanation. All party series use the same display, units and uncertainty treatment.
04
Uncertainty includes the data match
The original regression snapshot contains 449 matches for 448 unique census upazilas. Boalkhali's indicators appear against two different election records. The census explorer counts that area once; the original regression output remains explicitly labelled as inherited. Its intervals describe model uncertainty conditional on the supplied matches. They do not account for matching errors or the effects of excluding metropolitan areas. Reviewing the geographic crosswalk and rerunning sensitivity checks is necessary before treating a coefficient as a policy finding.
Methods, sources and reuse
Census percentages are preserved as published in the supplied extract. Nulls are excluded only from the particular mean or scatter plot that requires that indicator. Geographic means are unweighted and deduplicated by district/upazila. Regression coefficients and standard errors are inherited from the prototype, converted from vote-share fractions to percentage points, with approximate Student-t 95% intervals using the reported degrees of freedom. No model is silently re-estimated when explorer filters change.
- 1. BBS: Population and Housing Census 2022, National Report (Volume I)
Primary census reference. The website uses the prototype's selected area extract, not a fresh transcription of the full report.
- 2. Election results map used by the prototype
Secondary election compilation. Its joins and classifications are inherited and require independent verification.
- 3. Source extract and regression snapshot
Input lineage, original data and prototype implementation.
Download and cite this snapshot
Institute of Policy Dynamics, “Bangladesh Census: Living Conditions and Local Variation”. Coverage: 2022 census · 2026 election comparison. Cite the original data providers alongside IPD.
Download the complete published dataset (JSON)Source fingerprint (SHA-256): 21d2be66013ef3d6d63c7368b00e6407aeecfc2747a10352de196f8af604790d
Chart CSV files contain the selected observations. Original data remain subject to their providers' terms; public access alone does not establish an open licence.