Data & Methodology

1. What data powers this atlas

The atlas draws on the Multiple Indicator Cluster Surveys (MICS) for Bangladesh — large, nationally representative household surveys implemented by the Bangladesh Bureau of Statistics (BBS) with technical support from UNICEF. This project uses four rounds: 2006 (MICS3), 2012-13 (MICS5), 2019 (MICS6) and 2025 (MICS7).

Each round interviews tens of thousands of households, eligible women aged 15-49, and children under 5 (the 2019 and 2025 rounds add a full birth-history module and a 5-17 years module too), collecting fertility, family-planning, maternal and newborn care, child health, nutrition, immunization, child protection, early childhood development, WASH and household information using UNICEF's standardised MICS questionnaires, comparable across countries and over time.

2. The survey rounds in this project

Round	MICS wave	Notes
2006	MICS3	Core module set; maternal-health (MN) question numbering differs from later rounds; no full birth history or 5-17y module; no GPS.
2012-13	MICS5	Own maternal-health numbering scheme (e.g. caesarean section and birth-weighing questions sit at different codes than 2019/2025); no full birth history or 5-17y module.
2019	MICS6	Adds full birth histories (direct child-mortality estimation) and the 5-17y module.
2025	MICS7	Adds cluster-level GPS coordinates (3,149 clusters, mapped on this atlas); the standalone HIV/AIDS knowledge module was dropped from the women's questionnaire in this round.

Exact sample sizes and the indicators available differ by round — always confirm against the corresponding official MICS final report before quoting a figure elsewhere.

3. The data files (recodes) explained

MICS distributes each round as several standardised recode files. Each file is a different unit of analysis:

File	One row =	Used here for
hh.sav	a household	Water, sanitation, electricity, wealth index
hl.sav	a household member	School attendance/net attendance ratios, household headship
wm.sav	a woman 15-49	Antenatal care, skilled birth attendance, contraception, child marriage, women's education, HIV knowledge (2006-2019), breastfeeding initiation
ch.sav	a child under 5	Nutrition/anthropometry, immunization, ORS & care-seeking, exclusive/continued breastfeeding, birth registration, ECD index
bh.sav	a birth in the full birth history	Direct under-5/infant/neonatal mortality (2019 & 2025 only)
fs.sav	a child/adolescent 5-17	Child labour (2019 & 2025 only)

scripts/convert_mics.py opens the correct recode for each indicator and each round, resolves that round's own variable names and coding scheme (these drift substantially across MICS rounds in Bangladesh — see §6), applies the official sample weight, aggregates by division and district, and writes the results this dashboard reads.

4. Indicators

The atlas computes 64 indicators across 13 themes: child mortality & survival, nutrition & anthropometry, infant & young child feeding, immunization, childhood illness care, maternal & newborn health, family planning & fertility, child protection, early childhood development & education, WASH, gender/adolescents/HIV, media & ICT access, and household & wealth. Each is a survey-weighted proportion or rate. Full definitions, the exact source module, and both English and Bangla names for every indicator are in data/indicators.json and appear in the tooltip/narrative under each indicator on the dashboard.

Some indicators are not available in every round because the underlying question or module did not exist yet (full birth histories and the 5-17y module start in 2019; GPS starts in 2025) or was dropped (the HIV/AIDS knowledge module was removed from the 2025 women's questionnaire). The trend chart simply shows fewer points for those rounds rather than a fabricated value.

5. Weighting, geography & reliability flags

Sampling weights

MICS uses a stratified, two-stage cluster sample, so every record carries a sample weight (hhweight/wmweight/chweight/fsweight, already normalised to a mean of ~1.0). All estimates here are weighted so they represent the population, not just the interviewed sample.

Wealth quintiles

The MICS wealth index (windex5) is built from household assets and housing characteristics and split into five equal national groups — poorest to richest — used by the Wealth filter.

Geographic representativeness — important

MICS Bangladesh is designed to be statistically representative at the division level in every round, and generally supports district-level estimates given the achieved sample, though smaller districts carry a wider margin of error. No MICS round in Bangladesh samples enough clusters per upazila to support a direct upazila-level estimate. This atlas is explicit about that instead of pretending otherwise:

• Division and district values are real, direct estimates from the microdata, each carrying a reliability flag based on its unweighted sample size (n).
• Upazila values are a modelled visual — computed as the real parent-district estimate with a small deterministic variation, so nearby upazilas are not shown as identical, and always flagged accordingly. They are useful to see roughly where a district's estimate would sit, but must not be quoted as an official upazila-level statistic.
• The 2025 GPS layer (toggle "GPS clusters" on the map) plots the real latitude/longitude of all 3,149 MICS 2025 survey clusters — actual sampled locations, not modelled.

Flag	Meaning
Reliable	Adequate unweighted sample (n ≥ 200)
Indicative	Small sample (50-199) — read with caution
Not representative	Very small sample (< 50)
Modelled (upazila)	No direct estimate exists at this level in any round; value is derived from the parent district

6. Notable methodological choices & caveats

• Round-to-round variable drift. MICS Bangladesh renumbers whole modules between rounds (for example the maternal-health block: 2006 uses lettered MN2A-X/MN7A-Y; 2012 uses MN1/MN3/MN17-22; 2019/2025 use MN2/MN5/MN19-34). The pipeline resolves each round's own column names and coding scale explicitly rather than assuming they match.
• Children's age is taken from CAGE (age in completed months), not the derived age-in-years variables, so that month-based indicators (exclusive breastfeeding <6mo, minimum dietary diversity 6-23mo) are computed on the correct window.
• Exclusive breastfeeding requires that no other liquid or food was given "yesterday", not just that the child was breastfed — the disqualifying-question list is round-specific.
• Birth registration counts a child as registered if either a certificate was seen/reported seen or the respondent confirmed registration without showing a certificate.
• Direct child mortality (u5mr/imr/nnmr) is estimated from the full birth histories (2019/2025 only) as a synthetic-cohort rate among births in the years immediately before that survey's own reference year — a transparent approximation, not a full period life-table.
• Unmet need for family planning uses a simplified current-status approximation (fecund, non-using, wants to stop/space) rather than the full retrospective algorithm in the official MICS report — it will differ slightly from the published unmet-need figure.
• Age at first marriage/union uses the derived, universally-populated WAGEM/agem variable rather than a supplementary follow-up question that is only asked of a small sub-sample.