Global Entrepreneurship Monitor data, ready to use

Sources

All figures come from the Global Entrepreneurship Monitor (GEM) Adult Population Survey (APS). The individual file has one row per respondent; the national file has GEM's published figures per country and is used only to check ours.

Files downloaded from . GEM publishes no licence for these files. Country-level figures are freely available; the individual responses belong to GEM and its national teams, so this portal only lets registered users download them and never republishes the files.

The version is the date GEM puts in the file name. The download date was not recorded for these files. The original zips are kept unchanged in private storage; the SHA-256 fingerprint lets anyone check they have the same file.

Coverage

Respondents per country and year in the individual file. A dash means the country did not take part in GEM that year: it is missing data, not zero entrepreneurship, and it is never filled in.

Definitions and calculations

• Population: adults aged 18–64, weighted with GEM's weight for that age range (weight_l), as in GEM's reports. Respondents without that weight are left out of rates.
• Rate: weighted share of "yes" among valid answers to the question. "Don't know", refusals and questions not asked are not counted as "no".
• Indicators about the kind of entrepreneurship (motives, growth, innovation, customers abroad) are shares of early-stage entrepreneurs (TEA = 1), not of all adults. A motive counts when the person agrees or strongly agrees (4 or 5 on a 1–5 scale). Innovation counts when the product or process is new to the country or the world.
• TEA (Total early-stage Entrepreneurial Activity) measures how many adults are starting or running a business up to 42 months old. It measures activity, not the quality or success of businesses.
• Confidence interval (95%): Wilson interval using Kish's effective sample size, n_eff = (sum of weights)² / sum of squared weights, which accounts for the loss of precision from weighting.
• Difference between two groups, two years or two countries: p1 − p2 with standard error √(p1(1−p1)/n_eff1 + p2(1−p2)/n_eff2). It is called clear when its 95% interval does not include zero.
• Figures based on fewer than 100 valid answers are not shown in the country profiles or the comparison. In the dashboard they are marked with an asterisk.
• Peers: economies in the same income group that took part in GEM the same year (at least 5). The classification is the one in the GEM national file: World Economic Forum in 2019, World Bank from 2020. Typical range = 25th to 75th percentile.
• Average of countries in the dashboard: simple mean of the country values, not weighted by population.
• The "Compare with other countries" button in a country profile picks the two peers with the closest TEA that year. It is a starting point, not a measure of similarity; you can change the countries.
• Rounding: one decimal, half up. The dashboard adds integer sums of weights so any combination of filters gives exactly the same figure in the browser and in the Python checks.

Quality checks

Every time a year is added, the data are rebuilt from the original files and these checks run. If one fails, nothing is published.

• Rows per country and year match the original file, and two runs produce identical files, byte for byte.
• Country codes are checked against ISO 3166 and fixed where GEM uses its own (for example UK → GB).
• Every harmonized variable only takes its allowed values, and missing answers stay missing.
• Our rates are compared with the figures GEM publishes in its national file: 2,564 of 2,577 comparisons are within 0.5 points. The rest are listed below.
• Each figure in the dashboard and the country profiles is recalculated with separate code (pandas and statsmodels) and compared one by one.
• Each figure shown by the web pages is read back from a real browser and compared with those recalculations.

Known differences with GEM

In these cases some of our figures differ by more than 0.5 points from the GEM national file. They are shown with a warning in the country profiles. Germany (all years): GEM's national file probably uses another weighting. Uruguay 2021: the individual file is a revised December 2025 version and the national file is from March 2022. Luxembourg 2022: under 1 point, cause not identified. All are pending confirmation with GEM.

Limitations

• Only survey years 2019 to 2022 are available as individual data. GEM releases them about three years after fieldwork.
• GEM does not publish the strata or clusters of each national sample, so intervals account for the weights but not the full survey design. They may be somewhat too narrow.
• This is a cross-sectional survey of different people each year. It shows associations, not causes, and cannot tell whether a policy worked.
• Fieldwork in 2020 and 2021 took place during the COVID-19 pandemic, in some countries by phone or online instead of face to face. Changes in those years may partly reflect how the survey was done.
• Motive questions changed in 2019 (four new motives), so they cannot be compared with earlier GEM years.
• Countries take part in different years. Averages and rankings change with the set of countries, not only with entrepreneurship.
• A few respondents have no 18–64 weight in the original GEM file. They are left out of the rates; a check confirms these are exactly the weights missing at the source.
• The National Expert Survey (NES) on entrepreneurial conditions and other sources (World Bank, OECD) are not included yet.

Variable dictionary

Harmonized variables in the downloadable data, with the original GEM variable. The same dictionary is included in every download.

Version and reproducibility

The pipeline is deterministic Python code (pyreadstat, DuckDB, pandas) with fixed versions: the same original files always give the same results. Steps: read the original files, harmonize, build the dashboard and country profile files, and run the checks.

SHA-256 of the harmonized data in use:

Please cite: Global Entrepreneurship Monitor (GEM), Adult Population Survey, the year used, and this portal (GEM Open Data, gem-portal.pages.dev).