Does Affective Forecasting Error Induce Changes in Preferences? Lessons from Danish Soldiers Anticipating Combat in Afghanistan

ABSTRACT

This paper investigates how affective forecasting errors (A.F.E.s), the difference between anticipated emotion and the emotion actually experienced, may induce changes in preferences on time, risk and occupation after combat. Building on psychological theories incorporating the role of emotion in decision-making, we designed a before-and-after-mission survey for Danish soldiers deployed to Afghanistan in 2011. Our hypothesis of an effect from A.F.E.s is tested by controlling for other mechanisms that may also change preferences: immediate emotion, trauma effect – proxied by post-traumatic stress disorder (P.T.S.D.) – and changes in wealth and risk perception. At the aggregate level, results show stable preferences before and after mission. We find positive A.F.E.s for all three emotions studied (fear, anxiety and excitement), with anticipated emotions stronger than those actually experienced. We provide evidence that positive A.F.E.s regarding fear significantly increase risk tolerance and impatience, while positive A.F.E.s regarding excitement strengthen the will to stay in the military. Trauma has no impact on these preferences.

KEYWORDS:

• Risk preferences
• time preferences
• affective forecasting errors
• P.T.S.D.
• Afghanistan
• combat

Acknowledgments

We thank Dean Lillard, Vincent de Gardelle and an anonymous reviewer for their numerous and valuable comments and suggestions, which helped improve the paper. We also thank participants in the SFI Advisory Research Board Conference, the 2014 Annual Meeting of the Association of Southern European Economic Theorists (ASSET), the 6th Conference of the French Experimental Economics Association (ASFEE) and participants in the CEEM seminar in Montpellier for suggestions on an earlier version, and Marjorie Sweetko and Natalie Reid for their thorough re-reading of the English. We are grateful to the Danish Armed Forces for facilitating data collection.

Disclosure Statement

No potential conflict of interest was reported by the authors.

Supplementary Material

Supplemental data for this article can be accessed here.

Correction Statement

This article has been corrected with minor changes. These changes do not impact the academic content of the article.

Notes

1. See, e.g. Anderson and Mellor (2008); Barsky et al. (1997); Borghans et al. (2009); Dohmen et al. (2010); Golsteyn and Schildberg-Hörisch (2017); Heckman, Stixrud, and Urzua (2006).

2. For findings on stability of preferences, see Galizzi, Machado, and Miniaci (2016); Guiso, Sapienza, and Zingales (2018); Salamanca (2018); Schildberg-Hörisch (2018); Woelbert and Rield (2013). On the impact of life events, see Anger, Camehl, and Peter (2017); Bleidorn, Hopwood, and Lucas (2018); Preuß (2019).

3. A.F.E is also called impact bias in the literature (Miloyan and Suddendorf 2015).

4. As the number of women is too small for reliable statistical inference, we focus on the 355 male soldiers deployed to Afghanistan in combat and logistics units.

5. The before-mission response rate is approximately 82–86% and the after-mission rate is 99%. Danish soldiers’ average mission lasts six months. However, differing deployment periods mean that some soldiers may be absent at mission preparation, debriefing, or both – e.g. while most soldiers are deployed for six months, mechanics return after only four months. Furthermore, some of our soldiers returned earlier for medical or personal reasons, and one was killed in action (K.I.A.). Most returned to Denmark less than one month before responding.

6. Although experiencing combat could have prevented soldiers from answering both questionnaires, 75% of the wounded answered them. In Tables S.1 and S.2 in Supplementary Appendix A, we test for the bias in non-answering the ‘after’ questionnaire (attrition) and find no differences in the central variables of our analysis. While t-tests show that the variables age, seniority in the army and having children are significant, the F-test for joint significance cannot reject the null hypothesis that attrition is random. Furthermore, an attrition probit model test (Fitzgerald, Gottschalk, and Moffit 1998) cannot reject the null hypothesis that attrition is random for the variables in Tables S1 (p-value = .331) and S2 (p-value = .2147). Finally, the pooling tests (according to Becketti et al. 1988) similarly show that we cannot reject the null hypothesis that attrition is random (with respective p-values of .41 and .21). Thus we do not expect attrition to constitute a major threat for our findings.

7. As expected, the equality tests between average characteristics of exposed and non-exposed soldiers show very significant differences, ruling out the exogeneity of combat exposure. As Table S.4 in Supplementary Appendix A shows, soldiers can predict the probability of combat but not its intensity.

8. This residual-on-residual approach has since been applied in several causal inference analyses to tackle non-linearity (Banerjee and Duflo 2003), simultaneity (Graham 1999) and endogeneity issues in parametric models (Gallegati, Ramsey, and Semmler 2014), or more recently as double/debiased machine learning in non-parametric models accounting for high dimensional settings (Chernozhukov et al. 2018).

9. This approach is also in the spirit of Wooldridge (2015)’s interpretation of the control function approach to make endogenous explanatory variables appropriately exogeneous.

10. As the post-emotions (E_after) are known only for combatants, we cannot estimate a system of equations on the entire sample, with instrumental variables to account for the endogeneity issues. Excluding X from Eq. (2)(2) $\mathbf{S}\mathbf{t}\mathbf{e}\mathbf{p}\mathbf{2}:AFE\mathrm{\_}E=c_0+c_{1}Intensity+c_{2}X+\eta$(2) does not change our results (not shown).

11. To check the robustness of our results to this assumption of null A.F.E. for non-combatants, we will also estimate step 3 on combatants alone (see results in Table 3).

12. The differences between pairwise correlations and estimates in Table 3 may be driven by two factors. First, pairwise correlations could not properly account for the complex empirical strategy presented in Figure 3 (particularly the endogeneity problem and the learning effect on the anticipatory emotions). Second, estimates in equation (3)(3) $\mathbf{S}\mathbf{t}\mathbf{e}\mathbf{p}\mathbf{3}:\hat{\mathrm{e}}\left(\mathrm{\Delta }P\right)=d_0+d_1\hat{\mathrm{e}}\left(AFE\mathrm{\_}E\right)+d_{2}Combat+\xi$(3) may suffer from a lack of accuracy in equations (1) and (2(2) $\mathbf{S}\mathbf{t}\mathbf{e}\mathbf{p}\mathbf{2}:AFE\mathrm{\_}E=c_0+c_{1}Intensity+c_{2}X+\eta$(2) ) as pointed out in Wager (2021).

13. A total score of at least 50 is considered to be P.T.S.D.-positive in military populations (44 in the general population).

14. For example, a soldier was K.I.A. less than one month before the end of the I.S.A.F. 11.

Additional information

Funding

This work was supported by the Agence Nationale de la Recherche [grants RISKEMOTION, ANR-2008RISKNAT00701 and ANR-17-EURE-0020] and by the Excellence Initiative of Aix-Marseille University – [A*MIDEX]; Soldaterlegatet with financial support from Tryg [grant ID119562], Lundbeck, Novo Nordisk, and Aase and Ejnar Danielsen’s funds.