Trouble on my mind: the effect of catastrophic events on people’s worries

Abstract

Major economic, environmental, or social shocks induce uncertainty, which in turn may impact economic development and may require institutional change. Based on the idea that catastrophic events (CEs) affect people’s perceptions of reality and judgments about the future, this paper analyzes the effect of CEs on people’s worries in terms of social, economic, and environmental issues. In particular, we consider the terrorist attack 9/11 in 2001, the beginning of the financial crisis in 2008, and the nuclear disaster in Fukushima in 2011. We propose two possible mechanisms: A CE in one sphere may affect people’s worries in general (“spillover”) or it may lead to people focusing on that sphere and being less worried about other spheres (“crowding out”). We argue that the determinants of the mechanisms are related to the type of CE, that a person’s professional background moderates the influence of a CE on his or her worries, and that the subsequent development of worries is affected by whether institutional responses are contested. The analysis is based on longitudinal data of the German Socio-Economic Panel.

Appendices

Appendix A: Further results

Appendix B: Robustness checks

In order to test for sample attrition and entry effects over the entire 15-year period ranging from 1999 to 2013 (including the 2-year “9/11” preliminary period and the 2-year “Fukushima” follow-up period), we re-estimate the model using a complete sample, i.e., people interviewed in each of the 15 years. Results for the one-time level effects ($\beta_{1}$, cf. Sect. 4) are included in Table 5, which excludes all other covariates for brevity. For the “9/11” CE, magnitude and significance levels of the coefficients do not change as compared to Table 3. The same holds true for “Lehman” and “Fukushima,” with two exceptions of changing significance levels. Hence, our results are fairly robust as to sample composition over time.

Next, we check sensitivity as to the exact interview date. We introduce an artificial error (normal distribution with μ = 0 and σ = 1) to the recorded date in order to attenuate potential dependencies between CE and interview date. Table 5 shows that results are virtually robust as compared to Table 3. In addition, we rerun our analysis after removing all interviews carried out on the particular day of the respective CE with special attention to the “Fukushima” CE, where a number of 90 interviews were conducted on March 11, 2011. Results in Table 5 show that no distorting effects are to be expected.

Further, we consider the influence of the chosen two-year follow-up period for each CE. We re-estimate the model using a one-year and a three-year follow-up scenario, respectively. As to the shortened one-year follow-up period, there are three marked discrepancies compared to our previous results. For the “9/11” CE the $\beta_{1}$ level effect estimate loses its significance for environmental worries. As to the “Lehman” and “Fukushima” CEs, the estimates for social worries turn significant. We conclude that our model is most affected by a shortening of the follow-up period requiring some time to capture the nonlinear evolution of worries following a CE. As to the extended three-year follow-up period, however, results are largely in line with Table 3 (where two estimates marked by “NA” are not available due to numerical instability). All in all, a two-year follow-up period as applied in Sect. 5 appears to be a reasonable choice, as results tend to stabilize upon extension to a three-year follow-up, see also Goebel et al. (2014).

Next, note that the main effect (i.e., not interacted with CE) of (nearly) time-invariant variables such as sex or education is generally not identified in FE models (in contrast to time-varying interaction effects). As a robustness check we will include a random effects (RE) estimator instead of the FE estimator applied in Sect. 5. The RE level effects displayed in Table 5 are in broad agreement with the FE results of Table 3. Hence, we may avoid the more sensitive RE assumptions which, in addition, are clearly rejected by a Hausman test (also included in Table 5). See e.g., Wooldridge (2010) for further technical details and precise assumptions of the FE and RE approaches.

Finally, we test whether our findings are robust to an ordered probit model (using Stata’s xtoprobit) in order to challenge the cardinality assumption introduced in Sect. 4. Note that, due to the nonlinear structure of the probit model, only the coefficients’ sign and significance can be compared. With this in mind, results appear to be robust against the linear model. Due to numerical instability the result for economic worries following the Fukushima CE is not available.