Unemployment and Military Labour Supply: A Study on Belgian Data for the Period 2005-2020

ABSTRACT

The Belgian military has been an all-voluntary force since the suspension of the conscription (1994). Characterised by a high average age and confronted with large outflows of personnel due to retirement, the Belgian armed forces are forced to substantially increase their recruitment efforts. Belgium is not only a country with a small military but it also has a very specific labour market for which policy is a responsibility of the three regions. Not only does policy differ between the different regions but so does the unemployment rate with clear differences between the north and the south of the country. This makes the country a unique case study to examine the effects of unemployment across the different regions. We therefore estimate the determinants of military labour supply by means of a mixed-level model, capturing the impact of unemployment on the application rate at the regional level. Our study confirms earlier findings in the literature, showing that changes in unemployment have an important impact on the number of candidates for joining the military. Our study does, however, not reveal a clear-cut North–South distinction (in line with the strong discrepancies between the Northern and Southern provinces), indicating the importance of other explanatory variables.

KEYWORDS:

• Military labour supply
• Unemployment rate
• Recruitment
• Belgian Military

JEL CLASSIFICATION:

• J20
• J22
• J45

Acknowledgments

We like to thank two anonymous referees of this journal and the participants of the 2022 International Conference on Economics and Security for their valuable inputs. We also thank Lieutenant Colonel Bart Van Grieten for providing us the data needed to calculate the application rates. Finally, we like to thank Dr. Ruben Balcaen for help with the data analysis and Dr. Josselin Droff for his contributions and inputs on earlier versions of the manuscript.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1. In contrast to most studies, these authors however use the number of recruited personnel as the main indicator of military labour supply. One could argue that the number of applications is more sensitive to these occupational variables whereas the number of recruits (also) largely depends on the demand side of the market. As we have no information on the drop out ratio for this study, we however cannot verify this.

2. Candidates need to meet a certain number of conditions such as age (there is a maximum age for certain jobs), nationality and for certain jobs a specific degree.

3. The Belgian military searches for candidates that are intrinsically motivated, not candidates that are applying because they are unemployed or do not have other alternatives.

4. Officers, for example, also need to take a mathematical exam and two language exams (the native language and the second Belgian language, being Dutch or French).

5. Although we cannot call Brussels stricto senso a province (as it is a region), we do so to avoid confusions.

6. We choose this age cohort, as it constitutes the main target audience for recruitment. Moreover, this allows to make valid comparisons with the data on unemployment, which measures the unemployment in the age cohort upon the age of 24 (the next age cohort for measuring the unemployment constitutes the age cohort 25-49).

7. We stress that we only retain ‘unique candidates’, filtering out candidates that apply for multiple job openings. A candidate that applies for multiple jobs is hence only counted as one.

8. Whereas there is only one province in the Northern region that has an application rate exceeding 0.5%, all southern provinces have values higher than 0.5%.

9. Bäckström (2019) discusses the potential problems of endogeneity and uses the unemployment in the age category 30-64 as an instrumental variable. He did not find evidences that the unemployment rate is endogenous, which is consistent with the view of the Swedish Armed Forces being a relatively small employer in relation to the total labour market. By means of a robustness test, we also conducted an IV estimation, using the unemployment rate in the age category 25-49. The estimations obtained were very similar to the results discussed later in this article.

10. We considered the possibility of endogeneity between the variable application rate and the variable % of military per province. However, only a fraction of youngsters that apply for a job in the military succeed the tests and effectively join the armed forces. Moreover, there is a time delay between applying for a job in the military and effectively joining the armed forces (X + 1).

11. The majority of the recruitment centre have been allocated the same number of staff, i.e. 7 persons before 2014 and 4 after 2014. The recruitment centre in Brussels has two additional staff members, the centre in Liège has one additional member and the Flemish-Brabant and the Walloon-Brabant province do not have a recruitment centre.

12. We also considered introducing a dummy variable referring to the periods when Belgian military were involved in a specific mission. This, however, appeared to be a daunting task, as the Belgian military was deployed in Afghanistan and parts of Africa during the entire period of 2005-2020. It was hence difficult to distinguish specific periods where the Belgian military was more or less active in certain geographical parts of the world.

13. This is often considered the best starting point when conducting multilevel analysis (Rabe-Hesketh and Skrondal 2012).

14. The intra-class correlation coefficient (ICC) expresses the proportion of the total variance that is explained by intersectional differences. A high value of the ICC hence indicates that the total variation in an outcome measure is associated with cluster membership (in our case living in a specific province).

15. Unfortunately, we are in this stage confronted with a drop of the ICC to 0.09. This could be explained by the number of variables included in the model, combined with the limitations in terms of the number of clusters (i.e. the 11 provinces). There is no consensus as to how large the ICC should be to allow using a mixed model structure. However, a value higher than 10% of this coefficient generally indicates a mixed model should be used (Chen and Chen 2021). We argue the value of using a mixed model for studying our research question, as the ICC is fairly close to 10 %. Moreover, the variation in the different slopes across the provinces (see further) leads us to confidently say we can consider the data as being hierarchical.

16. The province with the lowest population constitutes Luxemburg, with a little bit over 285.000 persons living in this province in 2020. The province with the highest population is Antwerp, with a population of a little bit over 185.000 persons in 2020.

17. Plotting this variable shows a clear upward trend of the number of foreign born youngsters age 17-24 over the period 2005-2020 in all Belgian provinces.

18. Certain provinces, for example, contain a large number of very small barracks (providing only very limited possibilities to be employed), whereas others provinces only contain one barrack which appears however to employ thousands of soldiers.