Glossary

This glossary explains terms that appear across SynthPopCan. It is written for readers who may be comfortable with historical, social, or cultural sources but new to synthetic population methods.

Synthetic Population Terms

Synthetic population : A generated table of people, households, dwellings, or other units. The rows are not real people. They are modelled records designed to match selected features of a target population.

Synthetic household : A generated household row. In linked workflows, person rows point back to a synthetic household identifier.

Synthetic person : A generated person row. A synthetic person may carry individual attributes, such as age group or sex, and household context inherited from their synthetic household.

Linked household/person output : Two generated tables that belong together: one household table and one person table. Person rows include a household identifier so we can check that each person belongs to a generated household.

Seed records : Starting rows used by IPF. IPF changes how much each seed row counts; it does not invent missing columns.

Weight : A number saying how much a row counts. A row with weight 4.5 represents four and a half units in a weighted table. We may later convert fractional weights into integer rows.

Expanded population : A table with one row per generated unit. Expanded files are often easier for downstream tools to understand, but they can be much larger than weighted files.

Candidate rows : Plausible generated rows before final calibration or selection. In small-area workflows, model packages generate candidate households and people; calibration chooses which households are assigned to each target geography.

Validation report : A structured report that checks generated output against controls or checks whether linked household/person rows remain internally consistent.

Provenance : Evidence about where data came from and how it was transformed: source files, URLs, access restrictions, column mappings, command lines, random seeds, software versions, and validation reports.

Controls and IPF

Control : A target total the generated or weighted population should match. Controls usually come from public aggregate data, such as a Census Profile table.

Control table : A normalized file of controls that SynthPopCan can read.

Margin : One set of control totals. A sex margin might contain totals for women, men, and other published categories. A joint age-by-sex margin contains totals for each age and sex combination.

Dimension : A column or category axis used by a control. In an age-by-sex margin, the dimensions are age and sex.

IPF : Iterative proportional fitting, also called raking. IPF repeatedly adjusts row weights until weighted totals match selected margins as closely as possible.

Integerization : The step that turns fractional weights into whole counts or expanded rows. This is separate from IPF fitting and can introduce small differences from the fitted weighted totals.

Structural zero : A category combination that should not exist, such as a household with zero people.

Sampling zero : A category combination that could exist but does not appear in the seed or training sample. Sampling zeros are difficult because a model may mistake absence in the sample for impossibility.

Incompatible controls : Controls that cannot all be fitted together because they describe different universes, use mismatched categories, or imply different totals.

Model Terms

Tree model : A model that splits training rows into groups and uses the observed outcomes in those groups to generate new values.

Decision tree : A tree made from repeated questions, such as “is tenure owner or renter?” and “is household size one, two, or three or more?”

CART : Classification And Regression Trees, a common family of decision-tree methods.

Conditional-frequency model : A simpler model that groups rows by conditioning columns and samples outcomes from the observed frequencies in each group.

Forest model : A model that combines many trees. Forests can be more stable for prediction, but they are harder to explain as a single research artifact.

Support : The amount of training data behind a group, branch, or leaf. Low support means the model is relying on very few examples.

Purity : How dominant one outcome is inside a group or leaf. High purity can be normal, but high purity with low support can signal memorization or disclosure risk.

Model package : A reviewed artifact that contains enough model information and metadata to generate synthetic rows without redistributing raw microdata.

Canadian Census Geography

Province or territory : One of Canada’s top-level political geographies, such as Quebec, Ontario, or Nunavut.

Census division (CD) : A regional geography between province/territory and municipality-like units. Depending on the province, CDs may correspond to counties, regional districts, regional county municipalities, or similar units.

Census subdivision (CSD) : Usually a municipality or municipality-like unit: a city, town, village, township, parish, reserve, or unorganized area.

Census metropolitan area (CMA) : A large urban region organized around a major urban core and commuting zone.

Census agglomeration (CA) : A smaller urban region organized around a smaller core population.

Census tract (CT) : A relatively stable small urban geography, usually in CMAs and larger CAs. CTs are useful for urban case studies but do not cover all of Canada.

Aggregate dissemination area (ADA) : A grouping of dissemination areas. ADAs cover Canada and are often a useful first wall-to-wall geography for province-wide or country-wide synthesis.

Dissemination area (DA) : A small standard geography used for publishing census data. DAs cover Canada and are finer than ADAs, but controls can become sparse.

Dissemination block (DB) : The smallest standard census geography. DBs are best treated as a later placement geography, not the first place to calibrate household composition.

Target geography : The geography we are fitting or assigning output to, such as CT, ADA, or DA.

Wall-to-wall coverage : Complete coverage of a province, territory, or country. CTs are not wall-to-wall because they exist only in tracted urban areas; ADAs and DAs are wall-to-wall.

Statistics Canada and Source Terms

Statistics Canada : Canada’s national statistical agency and the source of many public aggregate tables used by SynthPopCan.

Census Profile : A large public census table that reports many characteristics for many geographies. SynthPopCan can turn selected Census Profile rows into controls.

WDS : Statistics Canada’s Web Data Service, an API for finding and downloading public data products and metadata.

PUMF : Public Use Microdata File. A PUMF contains anonymized sampled records, but it is still governed by documentation, access rules, and disclosure constraints.

Hierarchical PUMF : A PUMF structure where household, family, and person records can be linked. The 2016 hierarchical PUMF is important for linked household/person modelling.

Universe : The population a table describes. Two tables can use similar categories but still be incompatible if their universes differ.

Suppression : A source-data protection practice where some cells are withheld or altered, often because counts are too small or quality is too weak.

Rounding : A public-data protection and presentation practice where counts are rounded. Rounded controls may not add up exactly.

Special code : A value such as “not applicable,” “not available,” “valid skip,” or “not stated.” These codes should be decoded before being treated as ordinary numeric values.

Software Surfaces

Local web app : A browser interface started with synthpopcan serve. It runs locally and is meant for guided exploration, not public deployment.

CLI : Command-line interface. The synthpopcan command is the main CLI.

Beginner API : The small top-level Python API exposed as import synthpopcan as spc. It is meant for notebooks, teaching examples, and short scripts.

Advanced library : Lower-level Python modules for contributors and people building reusable research tools.

Read the Docs : The hosted documentation site for SynthPopCan: https://synthpopcan.readthedocs.io/.