The National Material Capabilities (NMC) dataset, published by COW Project, contains variables that are central in describing a country’s power. The quality and quantity of data vary across countries and time periods, which makes it not well-suited for time-series analyses. OWID uses two of their variables - military expenditure (milex) and military personnel (milper) - to construct this dataset. BOE exchange rates and US CPI are used to convert milex figures in constant US$. This allows comparison among countries in different years but with limitations. Finally, to extend milex series till 2016, figures from SIPRI’s Military Expenditure Database are spliced for the years 2013-2016.

Hosts of the NMC dataset, Professor Michael Greig and Professor Andrew Enterline, explain that the aim was to create a dataset which allowed for annual comparisons of the relative capabilities of states/countries in the international system. So NMC dataset is best for cross-sectional comparisons. Detailed description of original data is available here: http://www.correlatesofwar.org/data-sets/national-material-capabilities/nmc-codebook-v5-1. However, military expenditure figures have been converted in constant currency units previously such as by SIPRI, World Military Expenditures and Arms Transfers (WMEAT) and Military Expenditures and Economic Growth, Castillo et al (2001). Thus, to be transparent and acknowledge data limitations, we explain below the way OWID has generated this dataset.

Data construction:

Military Personnel (milper)

Milper values were originally given in thousands. OWID multiplies those figures by thousand to produce the variable ‘Military personnel’. NMC dataset also reports total population of the country in a given year. The variable ‘Military personnel relative to total population’ is generated by taking the ratio of ‘Military personnel’ and ‘Total population’.

Military Expenditure (milex)

NMC dataset converted expenditure figures from national currency into a standard unit. However, it reports milex in thousands British pounds sterling prior to 1914 and in thousands US dollars thereafter. The decision to report milex in such a way reflects that UK was a dominant power before 1914 and US after that. We reconstruct milex data in the following way:

We multiply milex values from NMC dataset by thousands. It gives military expenditure in British pounds from 1816-1913 and in US dollars from 1914-2012. We add to these, the milex values from SIPRI’s Database in US$ from 2013-2016.

Even though SIPRI provides milex values from 1949-2016, NMC dataset by COW Project is more complete for the overlapping years. Thus, we choose NMC dataset as our main dataset.

To convert milex into current US$ from 1816-1913, we use exchange rates from Bank of England (BOE) dataset. As a result, we get military expenditures in current US$ for the entire period 1816-2012.

Please note that we use BOE exchange rates because the original COW exchange rates are not available. While we are aware that using a different exchange rates for the period preceding 1914 introduces noise to the series, we believe that this does not affect trends or levels in any systematic way. It simply adds to the margin of error that historical estimates already had.

Then, we use US Consumer Price Index (CPI) from 1816-2016 to adjust for inflation. OWID uses 2015 as the base year i.e. price deflator in 2015 = 1. After dividing milex in current US$ by price deflator, we get military expenditure in constant 2015 US dollars.

SIPRI also provides milex data in constant 2015 US$. While the OWID reconstructed figures follow the same trend, two series are different. It is due to difference in methodologies used. SIPRI uses individual country’s CPI to first adjust for inflation, and then market exchange to convert national currency in 2015 US$. If there were perfect markets, both approaches should give same result as exchange rates would reflect changes in relative value of each currency over time.

Finally, for all three variables - Military expenditure, Military expenditure in current US$ and Military expenditure in constant US$ - we take their ratio relative to the total population to express variables in per capita terms.

Sources

Data Published By: Our World In Data (dataset constructed by Esteban Ortiz Ospina and Ruby Mittal)

Data publisher source:Data on military expenditure and military personnel (1816-2012) corresponds to National Material Capabilities (NMC) Dataset, Version 5.0, The Correlates of War (COW) Project. Data on military expenditure (2013-2016) corresponds to Military Expenditure Da

Link: http://www.correlatesofwar.org/data-sets/national-material-capabilities,https://www.sipri.org/databases/milex,http://www.bankofengland.co.uk/research/Pages/datasets/default.aspx,http://www.me