The boolean that halves a coverage figure

Sanitation has the same shape as water

An improved facility is necessary and not sufficient, and the field that makes the difference is a boolean nobody looks at twice.

Rung	Definition
Safely managed	Improved, not shared, and excreta safely disposed of or treated
Basic	Improved, not shared with other households
Limited	Improved but shared
Unimproved	Pit latrine without a slab, hanging latrine, bucket
Open defecation	No facility

IMPROVED_SANITATION = {
    "flush-to-sewer", "flush-to-septic", "vip-latrine", "pit-latrine-with-slab",
}

improved = households["sanitation_facility"].isin(IMPROVED_SANITATION)
shared = households["shared_sanitation"]

print(f"improved facility:  {improved.mean():.1%}")
print(f"basic (not shared): {(improved & ~shared).mean():.1%}")
print(f"limited (shared):   {(improved & shared).mean():.1%}")

improved_sanitation <- c("flush-to-sewer", "flush-to-septic",
                         "vip-latrine", "pit-latrine-with-slab")

households |>
  mutate(service = case_when(
    sanitation_facility == "open-defecation" ~ "open defecation",
    !sanitation_facility %in% improved_sanitation ~ "unimproved",
    shared_sanitation ~ "limited",
    TRUE ~ "basic"
  )) |>
  count(service) |>
  mutate(share = n / sum(n))

Rung	Households	Share
Basic	1,038	43.2%
Limited (shared)	452	18.8%
Unimproved	594	24.7%
Open defecation	319	13.3%

62.0% have an improved facility and 43.2% have basic service. Eighteen points sit on a latrine that meets the construction standard and is shared with another household.

Why sharing costs a rung

Because the reason for the standard is use, not construction. A shared latrine is used less at night, less by women and girls, and less by children — which is where the health benefit is. The classification follows the behaviour, not the concrete.

This is the same argument as the thirty minutes. In both ladders the second condition exists because a facility that is technically adequate and practically avoided produces no health outcome, and an indicator that counted it would report progress that nobody experienced.

Open defecation is the number to lead with

od = households["sanitation_facility"].eq("open-defecation")
by_district = households.groupby("district")["sanitation_facility"].apply(
    lambda s: s.eq("open-defecation").mean()
)
print(f"{od.sum()} households, {od.mean():.1%}")
print((by_district * 100).round(1))

households |>
  summarise(open_defecation = mean(sanitation_facility == "open-defecation"),
            n = n(), .by = district)

13.3%, and it is the one WASH figure that a non-specialist reads correctly without being told the definition. Lead with it, disaggregate it, and note that the district field in this file is written six ways — the cleaning course dealt with that, and grouping without normalising splits the worst district in four.

The hygiene ladder, and the observation that carries it

Basic hygiene service requires a handwashing facility with water and soap available. Three rungs, and the survey answers all three, because the enumerator observed rather than asked.

has_facility = households["handwashing_facility"].ne("no-facility")
soap = households["soap_observed"]

print(f"any facility:            {has_facility.mean():.1%}")
print(f"basic (facility + soap): {(has_facility & soap).mean():.1%}")
print(f"limited (no soap):       {(has_facility & ~soap).mean():.1%}")
print(f"no facility:             {(~has_facility).mean():.1%}")

households |>
  mutate(hygiene = case_when(
    handwashing_facility == "no-facility" ~ "no facility",
    soap_observed ~ "basic",
    TRUE ~ "limited"
  )) |>
  count(hygiene) |>
  mutate(share = n / sum(n))

Rung	Households	Share
Basic — facility and soap	819	34.1%
Limited — facility, no soap	811	33.7%
No facility	773	32.2%

67.8% have somewhere to wash their hands and 34.1% have soap there. The hygiene ladder loses half its top rung to a single observed item, which is a larger drop than either of the other two ladders.

Observation is why this is trustworthy. Reported handwashing runs far above observed handwashing everywhere it has been measured, because the question has an obviously correct answer. The column here is what the enumerator saw, and the column name should say so — soap_observed, not soap.

The three ladders cannot be averaged

summary = pd.DataFrame({
    "ladder": ["Drinking water", "Sanitation", "Hygiene"],
    "at_least_basic": [0.547, 0.432, 0.341],
    "denominator": ["all households"] * 3,
})
print(summary)

tibble::tribble(
  ~ladder,          ~at_least_basic,
  "Drinking water",           0.547,
  "Sanitation",               0.432,
  "Hygiene",                  0.341
)

Three numbers on the same denominator, which is unusual and worth noticing — they can legitimately be shown side by side. What they cannot be is combined. There is no “WASH coverage” indicator, and a mean of the three would describe no household: a household with basic water, shared sanitation and no soap is not “58% covered”, it is a household with one of three services.

Where a single composite is demanded, publish the three and the count of households that have all three, which is a real quantity and usually a sobering one.

What comes next

The ladders classify. They do not say how much water arrived, how long the queue was, or whether the water was safe to drink. The next two lessons are the measurements Sphere adds on top — and the first of them has two unit errors hiding in it.