Download everything
Tables
Each table is one CSV file. Open a table to see what each column means.
Survey microdata
- Waves 1–3, 2018–2020, 14 countries, 42,831 respondents: CSV, codebook, variable list.
- Poland, March 2023, 2,254 respondents: Excel file, used in “At the edge of war” (accepted version).
- More project data (protests, crime, expenditures): Military Deployments project page.
How the numbers are built
Troop deployments
From the troopdata R package (Michael Flynn): Defense Manpower Data Center (DMDC) quarterly reports, with Tim Kane’s Heritage Foundation series for 1950–2005. Counts are active-duty personnel assigned to each location.
- One report per year. Each annual figure comes from one report: September if there is one, otherwise March, June,
then December. Because every column comes from the same report, the service branches add up to the total. The R package’s own annual
figure (
get_troopdata(quarters = FALSE)) instead takes each column’s largest value in the year, so its totals can be a little higher and its branches can come from different reports. - Where each figure comes from. The
sourceandqualitycolumns say whether a figure is from a DMDC report, Kane (2006), an outside estimate (for example for Iraq, Syria and Afghanistan in periods the reports do not cover), or stepwise imputation between reports. The dashboard draws estimated years lighter and names the source in its tooltips. - Quarters without a report repeat the last report and are flagged (
carried_forward = 1). Quarters reported by fewer than ten places are partial releases and are left out of the quarterly series. - Afloat (personnel at sea) is kept as its own row. It is left out of country counts, regional and global totals, and rankings, because it is not a host country.
- National Guard, Reserve and civilian counts begin in 2008. Coast Guard begins in 2008 and Space Force in 2023.
Construction spending
DoD Comptroller C-1 exhibits, FY2000 onward. The amount is appropriations; where a report leaves that blank, the next reported amount
(authorized for appropriation, authorization, total obligational authority) is used. The source reports thousands of dollars; these files hold
dollars. Constant dollars use the U.S. consumer price index (World Bank FP.CPI.TOTL) for the calendar year. Worldwide accounts with no
country (planning and design, classified locations) and U.S. locations are left out. All of FY2016 and part of FY2010 are budget requests
rather than enacted amounts (request = 1).
Public opinion
Online surveys with quotas on gender, age and income, so the counts are unweighted. Percentages leave out “don’t know / decline to answer” unless you include it. The Poland 2023 survey is matched to the 2018–2020 wording: answer labels and the “greatest drawback” options are harmonized, and its education groups are mapped from degree levels (primary, vocational and secondary to 12 or fewer years; bachelor’s to 13–16; master’s or higher to 17 or more). Groups with fewer than 50 respondents are marked.
Population and exercises
Population is World Bank SP.POP.TOTL from 1960; years after the latest release repeat it (estimated = 1).
Exercises are the Multinational Military Exercises data, 1980–2010; the dashboard counts exercises, not participant rows.
Using the files directly
The dashboard reads the same tables as JSON at data/<name>.json, stored as {"columns": [...], "rows": [[...], ...]}.
data/manifest.json lists every table, its row count and column descriptions, and the data version.
How to cite
Please cite the data source you use. Pick a source and a style: