6  Schema & Variable Dictionary

Status

This variable dictionary is a working draft. Confirm the final variable list, value-label codes, and the precise definition of welfare_self with the 50by35 methodology team before treating it as authoritative.

6.1 Overview

Every harmonized file uploaded and approved through the pipeline must conform to this schema. Variables are grouped into mandatory (must be present and non-missing per the rules of the variable) and optional (included when available in the source survey).

The unit of observation is the individual: one row per person, with household-level variables (e.g. welfare, hhid) repeated across all members of the same household.

6.2 Conventions

• Type refers to the Stata storage type of the variable in the harmonized .dta file: str# (string), byte/int/long (integers), float/double (reals). Read the equivalents in R (character, integer, double) via haven.
• Missing — mandatory variables may not be missing except where noted. Use Stata’s system missing (.) for genuinely unknown values; do not use numeric sentinels such as -9 or 99, or extended missing values (.a–.z).
• Labels — categorical variables must carry Stata value labels; the numeric codes below are authoritative and must match across all surveys.
• Identifiers are strings to preserve leading zeros and avoid precision loss.

6.3 Mandatory Variables

Variable	Label	Type	Unit / range	Missing	Notes
code	Country code	str3	ISO / WDI 3-letter	Not allowed	e.g. AGO, ETH
year	Survey year	int	1990–2035	Not allowed	Four-digit year of data collection
survname	Survey name	str#	—	Not allowed	Survey acronym (e.g. GEIH, EHCVM, RHS)
hhid	Household identifier	str	—	Not allowed	Unique per household within the survey; string to preserve leading zeros
pid	Person identifier	str	—	Not allowed	Unique per individual within the survey
welfare	Welfare aggregate	double	> 0, LCU, annual per capita	Not allowed	Consumption or income aggregate from the survey’s official poverty methodology
welfare_self	Self-reliance-adjusted welfare	double	≥ 0, LCU, annual per capita	Not allowed	Welfare measure used for the 50by35 self-reliance indicator
welfare_type	Welfare measure type	byte	1–3	Not allowed	1 = Consumption, 2 = Income, 3 = Expenditure
weight	Sampling weight	double	> 0	Not allowed	Individual sampling/population weight
camp	Camp/non-camp	byte	0/1	Allowed	0 = Non-camp, 1 = Camp
urban	Urban/rural	byte	0/1	Allowed	0 = Rural, 1 = Urban

6.4 Optional Variables

Variable	Label	Type	Unit / range	Notes
hhsize	Household size	int	≥ 1	Number of household members
age	Age	double	0–120 years	Decimal allowed for children under 5 (per GMD 2.0); integer otherwise
male	Sex	byte	0/1	0 = Female, 1 = Male
educat4	Highest education (4 cat.)	byte	1–4	See value labels below
empstat	Employment status	byte	1–4	See value labels below
strata	Sampling strata	int	—	Design variable for variance estimation
psu	Primary sampling unit	long	—	Design variable for variance estimation

6.5 Value Labels

Categorical variables must use these codes and labels exactly.

welfare_type

1 Consumption · 2 Income · 3 Expenditure

male

0 Female · 1 Male

urban

0 Rural · 1 Urban

educat4

1 No education · 2 Primary · 3 Secondary · 4 Tertiary

empstat

1 Employed · 2 Unemployed · 3 Out of labor force · 4 Not applicable

6.6 Validation

These variables and value-label codes are enforced by the validation scripts (Stata/validate_50by35.do, R/validate_50by35.R), which run before every harmonized upload. Any deviation from the schema triggers an error and prevents the harmonized file from being uploaded to datalibweb. The checks cover missing values in mandatory variables, negative or zero values in welfare and weight, and that welfare_self is not greater than welfare. Households where hhsize differs from the number of pid records are flagged with a warning rather than an error, since rosters can be incomplete and refugee-only samples keep a subset of household members.