Compulsory Military Service and Future Earnings: Evidence from a Natural Experiment

Abstract

Using Israeli census data, and the random assignment of Arab males to military service, this study provides new evidence on the long-term effects of military service on the earnings of veterans. Among Druze men, we find an economically and statistically significant positive effect of 23% on their wages. The unskilled experience a slightly higher premium. The positive effects are large and intensify over time. Skill enhancement and usual human capital accumulation do not explain the positive effect of military service. Networking during service is proposed as a likely explanation.

JEL Codes:

• J24
• J31
• J38

Keywords:

• Military service
• labor market
• earnings
• minority
• social capital

Acknowledgments

I am indebted to Randy Filer, Janet Currie, and Ira Gang for their generous comments. Thanks also go to seminar participants of the European Association of Labor Economists, The Canadian Economics Association, and ISET, for their comments. Any remaining errors are mine.

Notes

1. The Jewish Virtual Library, http://www.jewishvirtuallibrary.org/jsource/Society_&_Culture/druze.html, last accessed 31 October 2015.

2. For example, for this study period, the observed wage of the Druze was only 86% of the wage of Jewish citizens. (For other Arabs this amounted to 68% only.)

3. Times of Israel http://www.timesofisrael.com/his-big-secret-hes-arab-muslim-and-serves-in-the-idf/, last accessed 4 June2016. See, for example, Wikipedia account of this fact: ‘… From among non-Bedouin Arab citizens, the number of volunteers for military service – some Christian Arabs and even a few Muslim Arabs – is minute, and the government makes no special effort to increase it …’ https://en.wikipedia.org/wiki/Israel_Defense_Forces, last accessed 5 June 2016.

4. Probably the nature of the ongoing Israeli–Palestinian conflict facilitated the persistence of the status quo as to military service for Israeli Arabs (for background on this conflict see Asali, Abu-Qarn, and Beenstock, forthcoming).

5. The Defense Service Law is also enforced on the Circassians (starting from 1958), who are non-Arab Muslims – but this is a very tiny segment of the Israeli population (there were only about 4,000 Circassians living in Israel in 2007. See, for example, http://www.ynet.co.il/articles/0,7340,L-3413724,00.html ; accessed 13 January 2014). In our data, Circassians cannot be identified because they are reported as ‘Muslims’ and thus cannot be distinguished from Arab Muslims. However, knowing that this community is concentrated in only two villages (Kfar-Kama and Reihaneya), one can identify Circassians by the code of their localities. It appears that there are only 46 observations of Circassians with positive income in the data: dropping these observations has virtually no effect on the results of this study.

6. The insight that can be had from other interesting peculiarities of the Israeli law can provide additional venues for future studies of the effect of military service on earnings. The exemption from service for ultra-orthodox Jews, or that for early married Jewish women, is a potential example.

7. It is worth noting that the income surveys, also maintained by the CBS, potentially provide a more detailed source of information as they cover more demographic aspects of the respondents, and provide exact values of the variables rather than intervals. They are also available on a yearly basis, unlike the census which is available only once in a decade. However, the income surveys could not be used for analysis because they miss the most important variable for identification in this study: the religion of the respondent. The income survey reports whether the individual is an Arab or Jewish, but does not specify the Arab’s religion: Druze, Muslim, or Christian.

8. We use the ‘Census Stage B’ file. The 1995 census covers the population as of 4 November 1995. It includes transitory residents and foreign workers, but these are excluded in the analyses in this study. Institutionalized individuals, and those in active military service, are not included in the census. The original sample includes 1,113,420 observations.

9. Alternatively, an imputation based on a uniformly distributed variable over the given interval was used and produced similar results.

10. The wages of Jewish workers, which are not discussed here, are remarkably higher than their Arab (Druze or non-Druze) counterparts. For an overview of the Jewish–Arab wage gap, its evolution and causes in the 1990s and the early 2000s, see Asali (2010). Notice also that Palestinians are not covered in our data, and their effect on Israeli Arab workers (including the Druze) is not covered in this study; for that see Asali (2013, 2017).

11. It might be tempting to compare the wages of Israeli Jews with those of non-serving Israeli Arabs; however, this comparison will not reveal the effect of military service because the differences are marred with other confounding variables like labor market discrimination and other wage determinants that are not comparable across these groups (Asali 2010). Nonetheless, a Jewish–Druze comparison is discussed in section ‘Military Service Effects: comparing Jewish and Druze workers’ of this study.

12. Asali (2015) discusses the likely difference in the estimated military premium experienced by members of the minority group as opposed to that experienced by members of the majority group.

13. We provide results for hourly wage beside those of log hourly wage in order to ensure that the found results are not an artifact of scale effects. That being satisfied, notwithstanding, we divert our attention to the results from the log of hourly wage, as is conventional in this literature. Also, the lower R² from the absolute hourly wage regressions attests to the fact that wages are log-linearly distributed, rather than linear, rendering the log wage-based analysis more suitable for this context.

14. As mentioned earlier, the conscription rate among the Druze is close to perfect. Therefore, essentially the reported estimates are for the intention to treat rather than the effect of treatment on the treated (but these are identical if compliance rate is exactly 100%). To get the latter, we have to divide the estimated effects by the actual compliance rate.

15. That is beyond the direct effect of individual human capital variables and the indirect effect of other external (environment) variables like foreign direct investments and the like (the latter are discussed in Asali, Cristobal-Campoamor, and Shaked (2016)).

16. Exploring the question with the top 1% and top 5% yielded similar results.

17. This is somewhat similar to Card and Cardoso (2012) who reject location or experience as mechanisms through which military benefits accrue. However, they find a slight positive effect of education as a channel for that effect.

18. See studies about the multitude effects of networking and social capital in many different settings (e.g. Andrabi, Ghatak, and Khwaja 2006; Khwaja 2001, 2009, among others).

19. One of the possible implications of networking is that the involved individuals enjoy the option of upgrading their status in the labor market by switching to more lucrative industries, through matches formed during the service. Asali (2015) provides evidence of the post-service redistribution of Druze male workers across industries, supporting this implication of networking. The evidence thus provided does not prove the existence of networking, however. It is just consistent with one of its implications, and thus fails to disprove it.

20. Asali (2010) reports large and unexplained wage gaps between Arabs (who include the Druze) and Jews in Israel.

21. Asali (2015) explores other parametric ways to study the heterogeneous time effect of service.

22. As networking induces a multiplier effect that amplifies the premiums with time (Glaeser, Sacerdote, and Scheinkman 2003).

23. For examples about the economic (intensive and extensive) effects of religion see Sacerdote and Glaeser (2001) for the effect on education; Barro and McCleary (2003) for the effect on growth; and Ewing (2000) for labor market effects.

24. If feedback effects, induced by statistical discrimination, do exist, then they might provide a challenge to this argument, however.

25. The data file is found on the CBS web site at: http://cbs.gov.il/publications12/local_authorities10/excel/p_libud.xls.