Methodology

Companion note to "Canada quadrupled its defence contracting. Canadian manufacturers never saw it." — sources, deduplication, and limitations behind the figures.

Sources

Contract figures are drawn from Proactive Publication – Contracts, published by the Treasury Board of Canada Secretariat. The file covers all federal contracts above $10,000 and contained approximately 1.29 million records when retrieved on 26 July 2026.

Manufacturing figures come from Statistics Canada tables 16-10-0047-01 (manufacturers' sales by industry, monthly) and 16-10-0012-01 (manufacturing capacity utilization rates by NAICS, monthly from January 2017).

Scope

Contract records are filtered to owner_org = 'dnd-mdn', the Department of National Defence. The department field is populated across the full history of the file, whereas commodity_code is null on older records, so a broader definition based on defence commodity codes would not produce a consistent time series.

Contracts placed by Public Services and Procurement Canada on behalf of National Defence are therefore not captured. The figures understate total defence procurement and should be read as departmental contracting only.

Deduplication

The same contract appears in successive quarterly reports. Each publication receives a new reference_number while retaining the same procurement_id. Aggregating on reference_number, or not deduplicating at all, counts a contract once per publication.

The deduplication key used here is procurement_id — falling back to reference_number where blank — together with contract_date, vendor_name, contract_value, original_value, amendment_value and commodity_type. Retaining the value fields means genuine amendments, which change those values, survive, while verbatim republications are removed.

This removes 1,728 of 353,289 DND rows, or 0.49% of records. The effect on annual totals is between 0.0% and 0.4% in every year except 2024, where it is 54%: $32.02bn before deduplication against $20.79bn after. The $11.2bn Future Aircrew Training award to SkyAlyne appears twice in the file.

The concentration of defence contracting is what makes a 0.49% record-level problem into a 54% error in a single year. Anyone working with this file should deduplicate before aggregating.

Value measure

On an amendment row, contract_value restates the contract total rather than the increment, so summing it double-counts the original award. The measure used here, "new money committed," takes amendment_value where it is non-zero and original_value otherwise. Other constructions are defensible and results should be tested against them.

Classification fields

commodity_type distinguishes goods (G) from services (S). country_of_vendor gives the contracting entity's domicile.

Manufacturing data

Capacity utilization is published only at the three-digit NAICS level, so NAICS 334 cannot be decomposed using that series. Shipments are published at four and five digits, which is what makes the sectoral breakdown possible.

Shipment figures are annual averages of monthly values, unadjusted and nominal. Both series are strongly seasonal, so only full-year comparisons are used. Growth rates are calculated on a 2020 base.

Limitations

• Award values are commitments, not disbursements. A 25-year contract is recorded in full in its award year. Annual totals therefore overstate near-term economic activity, and this is the most important caveat on the size of the increase.
• country_of_vendor records domicile, not content origin. A Canadian subsidiary of a foreign prime is coded as Canadian, so the domestic figures reported here are generous rather than harsh.
• commodity_type is unaudited self-reporting by the contracting entity.
• Vendor names are not resolved to parent entities. General Dynamics Land Systems – Canada, for example, appears under two spellings differing only in spacing. Concentration statistics are correspondingly conservative.
• 2020 is a COVID trough, which flatters growth rates measured from that base. On a 2019 base the shipment figures would be lower, though the gap between industries widens.
• Values are nominal. Real growth is several points lower across all series.
• Goods classification before 2018 appears less consistent, and pre-2018 composition figures should be treated with caution.
• NAICS 3345 includes medical instruments alongside navigational and control instruments, which dilutes any defence signal. There is no finer public split.
• Nothing here identifies a causal effect. The figures cover a single department over a period containing a policy regime change, and domestic defence procurement is a small share of output in the manufacturing industries examined. The contract and shipments data are independent observations that point the same way.

Reproduction

Analysis was performed in DuckDB over the raw CSV. The contracts file is approximately 628MB, and the portal's SQL aggregation endpoint is disabled, so the file must be downloaded rather than queried through the datastore API. Scripts and derived tables are available on request.