Calculating Food Security Scores

Module 4: Covariate Prediction

Overview

Food security — the consistent access to sufficient, safe, and nutritious food — is a key determinant of child nutritional status. The Food Insecurity Experience Scale (FIES) captures household food security through eight questions of increasing severity, from worry about food availability to going entire days without eating.

FIES produces raw scores from 0 (food secure) to 8 (severe food insecurity). However, raw scores are ordinal — the difference between scores 1 and 2 is not necessarily equal to the difference between 7 and 8. To enable valid statistical modeling, we apply the Rasch model recommended by FAO, which transforms responses into interval-scale measurements where differences are comparable across the entire range.

FIES Item Response Distribution

The eight FIES items follow FAO’s standard ordering by severity. The expected pattern — decreasing affirmative responses as severity increases — is confirmed in SIVESNU data: 65% report worrying about food (mild), while only 9% report going a full day without eating (severe).

Table 1: FIES Item Response Distribution

FIES Item Response Distribution. Food Insecurity Experience Scale (8 items)
Code	Question	Severity	Yes	No	NA	% Yes
FIES Item Response Distribution
Food Insecurity Experience Scale (8 items)
C001	Worried about not having enough food	Mild	533	282	0	65.4
C002	Unable to eat healthy/nutritious food	Mild	386	429	0	47.4
C003	Ate only a few kinds of foods	Mild	399	416	0	49.0
C004	Had to skip a meal	Moderate	168	647	0	20.6
C005	Ate less than should	Moderate	272	543	0	33.4
C006	Household ran out of food	Moderate	178	637	0	21.8
C007	Felt hungry but did not eat	Severe	176	639	0	21.6
C008	Went without eating for a whole day	Severe	74	741	0	9.1
Source: SIVESNU 2018. Items ordered by increasing severity per FAO FIES methodology.

Rasch Model Estimation

The Rasch model is fit on the eight binary FIES responses using survey weights, following the FAO methodology for the Voices of the Hungry project. Two sets of parameters emerge from the estimation: item difficulty parameters (one per question, characterizing how severe the experience reported by each item is) and person severity parameters (one per raw score, mapping each total to an interval-scale food insecurity level).

Item Difficulty Parameters

Item difficulty (β) represents the severity threshold at which each experience is typically reported. Negative values indicate experiences reported by most respondents; positive values indicate experiences reported only by severely food insecure households.

Table 2: FIES Item Difficulty Parameters (Rasch Model)

FIES Item Difficulty Parameters. Rasch model estimates (ordered by difficulty)
Rank	Code	Question	Severity	Difficulty (β)
FIES Item Difficulty Parameters
Rasch model estimates (ordered by difficulty)
1	C001	Worried about not having enough food	Mild	−3.042
2	C003	Ate only a few kinds of foods	Mild	−1.582
3	C002	Unable to eat healthy/nutritious food	Mild	−1.356
4	C005	Ate less than should	Moderate	−0.091
5	C006	Household ran out of food	Moderate	0.940
6	C007	Felt hungry but did not eat	Severe	1.003
7	C004	Had to skip a meal	Moderate	1.024
8	C008	Went without eating for a whole day	Severe	3.104
Higher difficulty = more severe food insecurity. Negative values = more common experiences.

Raw Score to Rasch Transformation

The model transforms each raw score (0-8) into an interval-scale severity parameter (θ). These parameters are used to assign individual food insecurity scores.

Table 3: Severity Parameters by Raw Score

FIES Severity Parameters by Raw Score. Rasch model transformation
Raw Score	Severity (θ)	Std. Error	Classification
FIES Severity Parameters by Raw Score
Rasch model transformation
0	−3.913	1.582	Food Secure
1	−2.963	1.233	Mild food insecurity
2	−1.746	1.014	Mild food insecurity
3	−0.806	0.936	Mild food insecurity
4	0.034	0.900	Moderate food insecurity
5	0.842	0.906	Moderate food insecurity
6	1.724	0.989	Moderate food insecurity
7	2.927	1.241	Severe food insecurity
8	3.885	1.582	Severe food insecurity
Classification follows FAO thresholds.

Validation

Score Distribution Comparison

The following table shows the population distribution across raw scores and their corresponding Rasch transformation. Most children live in households experiencing some degree of food insecurity.

Table 4: Distribution of FIES Scores: Raw vs Rasch

FIES Score Distribution. Raw scores mapped to Rasch severity parameters
Raw Score	N	% (weighted)	Rasch Score (θ)	Classification
FIES Score Distribution
Raw scores mapped to Rasch severity parameters
0	219	26.3	−3.913	Food Secure
1	123	15.3	−2.963	Mild food insecurity
2	94	12.0	−1.746	Mild food insecurity
3	103	12.6	−0.806	Mild food insecurity
4	82	10.0	0.034	Moderate food insecurity
5	59	7.0	0.842	Moderate food insecurity
6	48	6.0	1.724	Moderate food insecurity
7	41	5.1	2.927	Severe food insecurity
8	46	5.8	3.885	Severe food insecurity
Rasch scores provide interval-scale measurement; raw scores are ordinal.

Model Fit Statistics

Table 5: Rasch Model Fit Statistics

Rasch Model Fit Statistics
Statistic	Value
Rasch Model Fit Statistics
Rasch Reliability (observed)	0.744
Rasch Reliability (flat)	0.765
Complete Cases	550
Convergence	Converged!
Reliability > 0.7 indicates acceptable measurement precision.

Distribution Visualization

The histogram shows the distribution of Rasch-transformed food insecurity scores. The bimodal pattern reflects distinct food secure and food insecure subpopulations.

Histogram of Rasch-transformed food insecurity scores (theta) on the x-axis and frequency on the y-axis, with bars shaded by food security classification from food secure through severe food insecurity. A dashed vertical line marks zero, separating less severe scores on the left from more severe scores on the right. The distribution is bimodal, with one peak of food secure children and a second peak spanning the food insecure classifications, consistent with two distinct subpopulations. — Figure 1: Distribution of FIES Rasch Scores

Summary

Food Security Classification

Classification	Raw Score	Prevalence
Food Secure	0	26.3%
Mild food insecurity	1-3	39.8%
Moderate food insecurity	4-6	23.0%
Severe food insecurity	7-8	10.9%

Methodological Notes

Rasch model advantages — Transforms ordinal raw scores into interval-scale measurements with equal spacing. This enables valid arithmetic operations (means, regression coefficients) in subsequent modeling.
Survey-weighted estimation — Item parameters were estimated using survey weights to ensure population representativeness, following FAO methodology.
Item ordering validated — The estimated difficulty parameters follow the expected severity gradient: mild items (worry, diet quality) have negative β, while severe items (hunger, day without food) have positive β.

Limitations

Extreme scores — Raw scores 0 and 8 require assumptions for parameter estimation (handled by RM.weights default settings).
Local independence — The Rasch model assumes conditional independence between items given the latent trait; residual correlations may indicate violation.

Application

The Rasch-transformed FIES score enters the stunting models as a continuous predictor, carrying the full gradient of food security into the model rather than a small number of categorical strata.