What Phase 3 actually asserts

The sentence, in full

An area classified IPC Phase 3 (Crisis) asserts this and no more:

At least 20% of households in the area are in Phase 3 or worse, having food consumption gaps with high or above-usual acute malnutrition, or are marginally able to meet minimum food needs only by depleting essential livelihood assets or through crisis coping strategies.

Four things follow from that sentence, and each is routinely lost.

It is a threshold, not an average

The 20% rule is the whole mechanism. An area is classified in the highest phase for which at least a fifth of the population meets the criteria.

import pandas as pd

# Illustrative distribution across an area
distribution = pd.Series({
    "Phase 1 — None/Minimal": 0.34,
    "Phase 2 — Stressed": 0.44,
    "Phase 3 — Crisis": 0.18,
    "Phase 4 — Emergency": 0.04,
})
cumulative = distribution[::-1].cumsum()[::-1]
print((cumulative * 100).round(1))
print("Highest phase reaching 20%:",
      cumulative[cumulative >= 0.20].index[-1])

# Cumulate from the worst phase down; the classification is the last one
# that still reaches 20%.

In that distribution, 22% are in Phase 3 or worse, so the area is Phase 3 — even though 78% are not. The classification is about the tail, and it is designed to be.

So “the area is in Phase 3” and “the average household is in Phase 3” are different claims, and only the first is made. A report saying “the population is in crisis” has changed a statement about a fifth into a statement about everyone.

It says nothing about any household

An area classification is an area classification. It is assigned to a geography from a body of evidence about that geography, and it does not classify the households in it.

This is the most consequential misuse. A Phase 3 area is not a targeting list; the households inside it range from Phase 1 to Phase 4, and identifying which is a separate exercise using household-level indicators — the ones the first two units of this course computed.

print("Area classification → geographic prioritisation")
print("Household indicators → who inside it is eligible")
print("These are different analyses on different units.")

# Two questions, two units of analysis, two datasets.

The 4 in Phase 4 is not four times the 1

The phases are ordinal, and the intervals between them are not equal or even defined. Phase 4 is not twice as bad as Phase 2, and an area moving from 2 to 3 has not deteriorated by “one unit”.

So phases cannot be averaged, subtracted or trended arithmetically. A national figure of “mean phase 2.6” is meaningless, and the correct summary is the number of people in each phase, which is what the IPC publishes.

population = pd.Series({"Phase 1": 340_000, "Phase 2": 440_000,
                        "Phase 3": 180_000, "Phase 4": 40_000})
print(f"Phase 3 or above: {population[['Phase 3', 'Phase 4']].sum():,} people")
print(f"share: {population[['Phase 3', 'Phase 4']].sum() / population.sum():.1%}")

# People per phase. Never a mean phase.

A classification is a group’s judgement, not a computation

Nothing in this course produces a phase, and that is not a gap in the course.

A phase is assigned by a technical working group that reads the evidence table, weighs the reliability of each row, applies the reference thresholds, argues, and records both the conclusion and the level of confidence in it. The analyst’s job is to build a table that makes the argument possible, not to pre-empt it.

That is why the six-row table in the last lesson ends with a “not available” block. A working group that cannot see what is missing cannot weigh what is present.

What this analysis could and could not support

Lean season 2024, four districts

  What the evidence supports
    A deterioration is underway: six of six indicators point the same way,
    and the year-on-year maize comparison at the same point in the season
    is +22% with wage terms of trade down 49%.
    Coping and asset depletion are widespread: 48.5% of households used a
    crisis or emergency livelihood strategy.

  What it does not support
    An area phase classification. Nutritional status and mortality are the
    outcome indicators the IPC weights most heavily and neither was
    collected. This is consumption and coping evidence only.
    A household targeting list. The survey identifies indicators, not
    eligibility, and four indicators disagree about which households are
    affected by a factor of seven.

  What would settle it
    A SMART nutrition survey in the same districts, and a second market
    round in October to establish whether the peak has passed.

“What it does not support” is the section that gets an analysis taken seriously, and it is the section most likely to be deleted for length.

The habit these three courses were for

Module 4 has now applied the same three questions in three sectors.

1. What did I count, over what? Six denominators in this course alone, and no two of the four household indicators share one.
2. What else could explain it? Seasonality, which moves maize 74% within a year and would have been read as a programme effect.
3. What should someone do differently because of this? The evidence supports prioritising the districts and acting before consumption falls. It does not support a phase, and saying so is part of the answer.

The vocabulary changes by sector and the questions do not. MUAC and weight-for-height in nutrition, three functionality rates in WASH, four food security indicators here — every one of them is the same problem: a measure that looks like a fact until you ask what it counted.

What comes next

Module 4 continues with protection case management and education attendance. The instruments will be unfamiliar and the questions will not.