Military Recruitment

The Military Recruitment Data Set provides country-year observations on states’ method of military recruitment, whether conscription or volunteer. For state membership information, it employs the Correlates of War (COW) State System Membership List. Please cite this data set as follows, including the current version number:

Upton Institute. 2022. Military Recruitment Data Set, version 2022.

The complete data set is available to members. See the membership page for more details.

Principal Sources

Data for this data set come principally from five sources. On rare occasions, information also came from the CIA World Fact Book, US Department of State country background notes, and Library of Congress Country Studies. Country-specific sources are cited in the codebook.

  1. Prasad, Devi, and Tony Smythe. 1968. Conscription: A World Survey: Compulsory Military Service and Resistance to It. London: War Resisters’ International. [Hereafter referred to as PS.]

  2. Horeman, Bart, and Marc Stolwijk, eds. 1998. Refusing to Bear Arms: A World Survey of Conscription and Conscientious Objection to Military Service. London: War Resister’s International. http://www.wri [Hereafter referred to as WRI.]

  3. International Institute for Strategic Studies. 1959–present. Military Balance. London, UK: International Institute for Strategic Studies. [Hereafter referred to as MB.]

  4. Heyman, Charles, ed. 2002. Jane's World Armies. Southampton, UK: Hobbs. [Hereafter referred to as JWA.]

Keegan, John. 1983. World Armies, 2nd ed. Detroit, MI: Gale Research Co. [Hereafter referred to as WA.]


  • Country Code (ccode) [integer]: COW country code for the country of observation (if available). If the country does not appear in the COW2 data set, then the observation is left blank.

  • Country Abbreviation (cabbr) [string]: COW abbreviated country name (if available). A missing value means that the country-year was not in the COW State System Membership List.

  • Year (year) [integer]: The year of observation.

  • Recruitment (recruit) [dummy]: The method of recruitment; 1 = volunteer recruitment, 0 = conscription.

  • Term of Enlistment (tmenlist) [integer]: The number of months for which recruits typically join.

  • Military (mil) [dummy]: The existence of military forces; 0 = no standing armed forces, 1 = standing armed forces. A code of 0 means that the recruitment variable for the observation refers to paramilitary and police forces only.

  • Estimate (estimate) [dummy]: 1 = observation required an educated guess for the recruit variable, 0 = no educated guess required.

  • Version Number (version) [float]: The version number of the Military Recruitment Data Set.


The method of recruitment is how a state satisfies its military manpower requirements. The method of recruitment is considered to be “conscription” if the principal means of induction into the military is the use of force, be it through legal means (e.g., conscription) or extra-legal means (e.g., impressment), or where individuals cannot realistically say “no” to military service.

The method of recruitment is “volunteer” if individuals enter the military as a matter of choice. States that allow for conscientious objection can still be considered to use conscription as the method of recruitment, as long as conscription is the principal means of satisfying the military manpower requirement. States that use a selective service system (e.g., a non-universal draft that distinguishes inductees from non-inductees based on economic necessity) are considered to use conscription as the method of recruitment as long as the military manpower requirement is still typically satisfied via the draft.

Conscription is considered the principal means for satisfying the military manpower requirement as long as a non-trivial number of recruits are enlisted through force. For instance, a country in which a minority of soldiers are technically conscripts can be considered to have a conscription system—as long as the proportion of conscripts is significant—because there may be many soldiers who volunteer in order to avoid the undesirable circumstances associated with being a conscript (as in, for example, Morocco). This would especially be the case if conscripts are indeed treated poorly and if all (male) citizens must register for service liability at a certain age (as in Egypt). Methods of forced recruitment must thus be either unsystematic or inconsequential in order for a military manpower system to qualify as volunteer.

Given these rules, a country-year observation is coded 1 if the state’s method of military recruitment in that year is volunteer, and 0 if it is conscript. This variable is thus dichotomous between 0 and 1.

The typical term of enlistment refers to the period of active duty (not reserve duty) for which most soldiers initially enlist, be they conscript or volunteer. This information is easier to come by for conscript militaries, since the term of enlistment is generally fixed for these militaries. If the term of initial enlistment is variable (e.g., if naval conscripts serve for one year and army conscripts for two, or if some volunteers are inducted with shorter enlistment terms than others because of desirable skills), then the period can be averaged (in the former case) or the coder can take the most typical term (in the latter case).


  1. This codebook was originally developed in 2005, with explicit notes for areas of future improvement. The intent was that others would be able to replicate and improve on the first attempt. As of 2022, there appears to have been only one attempt to extend the data. Some of the notes below were written in 2005, even though many time series have been extended beyond that.

  2. WRI and PS make a point of indicating when military service is forced on the populace. As a result, when these sources make no mention of whether manpower requirements are filled voluntarily or not, the assumption is that recruitment is voluntary, although this assumption was generally only made when there was evidence that military service was likely voluntary (e.g., when there were two periods of conscription separated by a long period of time, or when the sources mention that a country used conscription only during wars).

  3. It was occasionally necessary to calculate the tmenlist variable using an average of the enlistment terms in different services or at different education levels (see list below). When averaging ranges to determine the tmenlist variable, the result was rounded down to the next lowest integer. In arriving at these averages, the terms of officers, NCOs, and specialists were ignored, since this variable is intended to measure the typical term of enlistment amongst the rank-and-file.

  1. To the extent possible, changes in recruitment policy were coded as occurring in the year in which the change took place, unless there was clear information that the policy change occurred in the second half of the year.

  2. When consulting MB, the first of the two years covered by an issue to constitute the relevant time period for data extraction was used. In other words, the 1999/2000 issue furnished data for 1999, and the 2000/2001 issue for 2000. Since data in MB are current as of August 1 of the first year in the date range, this did not seem problematic.

  3. When there was a discrepancy between any of the principal sources (MB, WA, WRI), I examined the differences to assess which information most closely coincided with the coding rules.

  4. Following on the explanatory notes published in the preface of each issue of MB, if there was a listing for “Terms of service” for a country, then that country was understood to employ conscription in the given year. If there was no such entry for a country (or if service was explicitly listed as voluntary under the terms of service), then the method of recruitment was considered to be volunteer. Still, the method of recruitment was coded as conscription if MB listed the term of service as either selective or voluntary and there was a non-trivial number of conscripts listed as serving in the active forces (e.g., Venezuela in the 1999/2000 issue).

  5. A country can maintain a military with conscripts in it and still receive a 1 on the recruit variable, since the coding change in the variable occurs when conscription ends, not when the last conscript leaves the service. Thus, there is a lag between the end of conscription and the end of conscript service. The reason for this decision was that it is too difficult to determine when no conscripts remain in the military for a reasonably large sample of countries.

  6. In general, changes in recruitment policies were reflected in the year in which they actually took place, instead of the year after. If, however, it was clear that one system was used in more than half of a year, then the country year was coded as using that method of recruitment for the entire year, and the change was reflected in the next year. For example, since the United States changed to volunteer recruitment on July 1, 1973 (at the exact midpoint of the year), the change was regarded as occurring in 1973 instead of 1974—the United States did not have volunteers for more than half of 1973, but for exactly half of it.

  7. Since earlier issues of MB did not have the complete explanatory notes in the preface—these came later—there were some entries that listed a term of service while it remained unclear whether the method of recruitment was conscription, selective conscription, or voluntary (for instance, Libya in 1970 and 1971). When this was the case, the giving of a service term was not taken to mean that there was conscription in that country in that year, which is the implication for MB entries in more recent editions. Thus, Libya in 1970 and 1971 received a 1 for the recruit variable, based on information in WRI. To give another example, Cuba is coded 1 for the recruit variable in 1967–1968, 1970, and 1972, even though MB gives a service term.

  8. To close small gaps in the time series, the best information on hand was used to make an educated guess as to the actual coding for a country year. For each observation filled in using this estimation method, a 1 was entered for the estimate variable. For large gaps, and small gaps at the beginning or end of series, cells were left missing data.

For country-specific notes and other details, see the codebook.