Google Data Commons

The shortest honest description of Google Data Commons is also the most useful one: somebody took the world's public statistics and gave them a single shape. Run as a Google initiative, it harmonizes figures from the World Bank, OECD, Eurostat, the Food and Agriculture Organization, the World Health Organization, NOAA, the US Environmental Protection Agency, the Bureau of Labor Statistics, the Bureau of Economic Analysis and the U.S. Census Bureau - among hundreds of provenance sources - into one statistical knowledge graph. Every indicator becomes a statistical variable tied to an entity and a date: a variable like Manufacturing_ValueAdded_AsFractionOfGDP resolves against country/USA rather than living in one agency's spreadsheet. The consequence for anyone tracking diversified industrials is that manufacturing value added, GDP, employment, trade and population arrive for every market a conglomerate operates in, in one consistent form, instead of a dozen agency portals each speaking its own dialect.

Inside Datadory's catalog this record anchors the macro layer of Industrial Conglomerates - the measure-the-pond layer that sits beneath company filings and facility registries. Get a sample of this dataset and we will cut it to the variables, countries and years you name.

Three shipped observations, flattened for reading:

# one observation = one entity x one variable x one date x one provenance facet
variable : Count_Person
entity   : country/USA
date     : 2020
value    : 329484123

variable : Count_Person
entity   : country/DEU
date     : 2020
value    : 83160871

facetId  : 18369491376878146239
entity   : country/DEU
date     : 2020
value    : 83160871

The third row is the one worth staring at. It repeats Germany's 2020 population but tags it with a facetId - a provenance stamp identifying which source's measurement is being reported. When two agencies publish different counts of the same thing, the graph does not average the disagreement away; it keeps both, facet by facet, so a pipeline chooses its authority deliberately instead of discovering the conflict downstream. In your sample each observation lands as a typed row - variable, entity, date, value, facet - ready to filter by region, indicator or source.

Ten documented fields make up an observation response, grouped four ways. An addressing layer (byVariable, byEntity) keys results by statistical variable and by place, so one pass fans out across both dimensions without repeated lookups. A provenance layer (orderedFacets, facetId) carries the source attribution behind every number - the feature that separates this from an ordinary indicator table. A measurement layer (observations, date, value) holds the figures themselves. A plumbing layer (obsCount, earliestDate/latestDate, nextToken) reports how much returned, how far history runs and where pagination continues.

Five of the ten carry unambiguous presentation and ship with worked examples in the dictionary below. The remainder - the facet list, the observation arrays, the count and bounds fields, the continuation token - arrive populated on records but without a fixed worked example, so they fold under additional fields on request, together with the derived columns Datadory attaches on delivery: cross-facet reconciliations, place-hierarchy rollups and long-run backfills.

Geography - worldwide at the top, descending to country, state and district or county level. The United States runs deepest of all: state and county grain that most international collections never attempt, which matters when a conglomerate's footprint is measured plant by plant.

Temporal - declared per series rather than asserted site-wide. Multi-decade runs are common, census and World Bank lines reach back to the 1960s, and every provenance facet self-reports its own earliest and latest dates. Available history is therefore something you read off the data, not something a brochure promises.

Granularity - one observation per entity per statistical variable per date per provenance facet. That fourth dimension is the discipline here: aggregation always starts from the same atomic unit, so nothing is silently blended across disagreeing sources.

Against the wider catalog - 1,744 datasets, average quality score 7.81 - this slice scores 7/10, carried by verified field documentation and shipped sample rows, tempered by a scale the source describes in charts rather than published counts.

API, files, or your warehouse. Daily, weekly, or hourly.

Pick the channel your team already works in: queryable rows for live lookups on specific variables and markets, flat files sized for overnight warehouse loads, or a direct pipe into Snowflake, BigQuery or Redshift. Cadence is yours to set - and to change when the models change.

Every delivery ships the full field dictionary, sample rows for validation, and a schema that holds steady between deliveries.

• Multi-market macro baselines. Pull manufacturing value added, GDP, employment and trade indicators for every country a conglomerate touches, from one consistent shape - the denominator work that feeds the wider market sizing shelf.
• Provenance-aware benchmarking. Facet tagging makes source disagreements visible instead of hidden, so a Germany-versus-France comparison can name the authority behind each figure.
• Long-run trend construction. Series reaching back to the 1960s turn a snapshot into a structural story - deindustrialization, catch-up growth, energy transitions - with dates declared per facet.
• Feature engineering for models. Typed entity-variable-date rows join cleanly onto firm-level feeds; see the ML model training workflows built on exactly this shape.
• Citation-grade research. Every figure carries its source, which is why the citation-first teams start here; the citation-grade research playbook leans on it.

Ranked by relevance in Datadory's persona tagging. Data scientists and ML engineers (relevance 3) get harmonized macro features joined on stable entity keys - see the data scientists x industrial conglomerates workflows. Market researchers and consultants (relevance 3) replace a shelf of agency portals with one queryable shape - market researchers use cases. Developers and data-product builders (relevance 3) build on an addressing scheme that treats places and variables as first-class keys - developers builders use cases. Journalists, academics and students (relevance 3) cite figures that arrive pre-attributed to their source - journalists academics use cases. Investors and quants (relevance 2) read country fundamentals as the backdrop against which company estimates get sanity-checked - investors quants use cases.

Four honest caveats, stated plainly.

First, presence is not guaranteed. The graph imports a selection of what each provider publishes - not everything. Confirm a specific variable exists for your places before building a pipeline on it; that confirmation is precisely what a sample is for.

Second, scale is unnumbered. "Tens of thousands of variables across hundreds of provenance sources" is the honest ceiling of what the documentation states, and even that appears as coverage charts rather than published counts. Treat any precise total - ours included - as a floor.

Third, history is uneven. Depth rides with the facet: a line reaching to the 1960s in one country may begin a decade later in its neighbor. Longitudinal designs should read the declared bounds per series rather than assume a shared start.

Fourth, the grain is places, not firms. This is the macro layer of the vertical. When the question descends to named corporate groups, pair it with a filings or registry feed and join on geography and year - several sit on the same industry shelf.

Notes that pair well with this page:

• Industrial conglomerates data hub - the pooled industry view, from filings and registries to this macro layer.
• World Bank WDI - Manufacturing, value added (% of GDP) - the single-indicator original, about 265 economies deep; the graph re-serves it beside tens of thousands of sibling variables.
• Eurostat Data Browser (Industrial Production & Structural Business Statistics) - the European production layer in official SDMX detail.
• BEA GDP by Industry - the United States production account, the domestic complement to the worldwide series.
• SEC EDGAR Company Facts & Submissions API - when the question drops from economies to filers; standardized XBRL for every US-listed conglomerate.
• Persona pages - what data-science, market-research, developer, journalist and investor teams each do with this slice.