Where do unmatched businesses go when trade and employment records are linked?
Show the coverage of the linked population separately from successful key matching.
The economy extends beyond the joined rows
Linking trade and employment records can reveal who benefits from services trade. Retaining only matched rows can also hide businesses with different identifiers or registration coverage. Truthound can check references and missingness in prepared tables, separately from the judgment that an analysis represents the economy.
UNCTAD's proposed linkage
UNCTAD introduced its national-data linkage report on September 11, 2026. It recommends linking tax, customs, business and employment information through common identifiers with anonymisation and confidentiality safeguards, to investigate differences aggregate trade figures cannot show.
Preserve both starting populations
Preserve each source population and period first. The businesses in customs data and those in employment registration may not start as the same set. Distinguish records with a usable common key from records actually linked, showing where the population contracts.
A join type changes the question
Choosing an inner join or retaining the source population is part of the analytical question. Keep unmatched rows in a separate register whichever approach is used. A failed match must not silently mean that a business has no activity or employment.
The reference and missingness checks
Truthound Core Python SDK 3.3.4, operating on a prepared Polars table, provides ForeignKeyValidator for supplied child/parent keys and CompletenessRatioValidator for a declared minimum non-null ratio. Owners define the expected population, identifier system and missingness meaning. Valid references do not establish statistical representativeness.
Keep the useful pandas relationship
An existing pandas 3.1.1 merge(validate=...) can retain 1:1, 1:m or m:1 key checks. Those results differ from assessing businesses outside the join. A m:m setting permits that relationship without establishing appropriate totals or population coverage. pandas can match null business keys; records without an identifier must not silently enter the successfully linked population.
Look for groups disappearing from the analysis
Where permitted, inspect coverage by an approved grouping such as business size or registration type. Lower linkage in one group raises a question about identifier changes or source coverage. Do not fill the gap with another group's rate or an undocumented guess.
Access and analysis scope differ
Technical linkability does not grant permission to analyse or publish every detail. Quality checks do not implement anonymisation or access control. Responsible owners establish the permitted aggregation and disclosure boundary, including how unmatched coverage can be described.
Publish the gap with the result
Show the analysed and excluded populations with reasons for unmatched records. Readers should not generalise a result for linked businesses to the whole economy automatically. Explaining who did not link is as important as adding another source.
Conditions
Use authorised data with established identifier meaning and confidentiality/anonymisation conditions.
Alternatives
Retain pandas 3.1.1 join-cardinality checks while reviewing unmatched populations separately.
Limitations
Reference checks do not establish population representativeness or an employment effect of trade participation.
Evidence
- Event checked
- Subject checked
- Modified
