News shocks at the local level: Evidence from a conditional Covid-19 containment measure.

To reduce the spread of Covid-19 whilst limiting the economic costs of containment policies, governments have introduced geographically-flexible conditional restrictions - measures targeting sub-national areas whose severity depends on the virus's local incidence rate. I analyze whether conditional measures impact transmission rates via a news-shock effect - that is, by incentivizing indirect actions in anticipation of the policies being carried out. Exploiting a natural experiment from Romania in a regression-discontinuity framework, I provide early empirical evidence in this sense: I find that the Covid-19 incidence rate fell significantly in targeted constituencies following the announcement of a conditional containment measure, but prior to the policy being implemented. My results add to a broader literature on news-driven fluctuations, wherein expectations of future policies can impact immediate behaviors. I conclude by discussing an important avenue for future research.

Introduction

To contain the spread of Covid-19, political leaders have implemented large-scale non-pharmaceutical interventions or ”lockdowns” (Flaxman et al. 2020). However, the evidence suggests that, while effective (Hartl et al. 2020), large-scale restrictions impose great socio-economic costs (Adams-Prassl et al. 2020; Pulejo and Querubín 2021; Brodeur et al. 2021). Attempting to balance the costs and benefits of restrictions (Shiva and Molana 2021), governments have turned towards less socially-disruptive measures that may still sufficiently alleviate the virus’s health-system burden (Spiegel and Tookes 2021).

Of particular note for my purposes are local conditional containment measures [LCCMs] - restrictions targeted towards specific sub-national geographical areas (e.g., constituencies), whose imposition and severity depend on the local Covid-19 transmission intensity.2 In Romania, LCCMs were widely employed — for instance, the closure of certain businesses in constituencies with more than 1.5 Covid-19 cases per one thousand residents, or mobility restrictions when the transmission intensity surpassed three per a thousand.3

Analyzing the impact of LCCMs is insightful economically given the potential channels by which these measures may reduce the transmission intensity — illustrated in Fig. 1. On the one hand, via what I term the mechanical effect, the restrictions themselves help contain the spread. For instance, local school closures reduce transmissions by limiting student interactions (Amodio et al. 2021).

Fig. 1

The Effect of Local Conditional Containment Measures — Theoretical Mechanism Decomposition.

More interestingly, however, via the news-shock effect, LCCMs incentivize individuals to take indirect, independent actions which come at a personal cost in order to contribute to a public good: preventing or removing restrictions on their local community. To exemplify, individuals may take actions such as a non-mandated wearing of face-coverings or social-distancing (Mitze et al. 2020), or may engage in behaviors such as test avoidance (Thunström et al. 2021) to reduce the (reported) local transmission rate.4

Nevertheless, while some evidence exists suggesting that LCCMs are effective (Laydon et al. 2021), the scholarship has not yet empirically determined if, independently of any mechanical impacts, the news-shock effect of LCCMs is meaningful, an effort which I argue is valuable both policy-wise – adding to our understanding of the costs and benefits of geographically-flexible restrictions, and economically – exploring how expectation changes impact the actions of forward-looking agents (Arezki et al. 2017) and how, in times of crises, individuals engage in personally costly behaviors to contribute to a local public good (Barron and Nurminen 2020).

I address this literature gap. By exploiting a natural experiment, I provide early evidence for the existence of a causal LCCM news-shock effect.

Concretely, I investigate an LCCM in Romania which mandated that a significant fraction of a constituency’s middle and high-school students could not physically attend classes if the local transmission rate exceeded one in a thousand. Crucially, no other restrictions were imposed at this threshold neither shortly before or after the policy’s announcement, and the LCCM was communicated to the public five days before schools reopened following the Easter holidays.

Exploiting these appealing features in a regression discontinuity [RD] framework, I document a sharp causal decline in the Covid-19 reported transmission intensity taking place after the announcement, but prior to the end of the holidays, when comparing constituencies situated close to the threshold. The estimated effect is economically-meaningful, suggesting that the announcement led to an average fall of 0.1–0.2 cases per a thousand residents, with the largest impact retrieved the day prior to schools restarting.

I argue that future work can build upon my findings by extending their external generalizability and identifying what actions drive the news-shock effect.

Background: Romania’s hybrid schooling policy

I briefly describe Romania’s policy which allows me to corroborate the existence of a causal LCCM news-shock effect.5

Communicated on the 29th of April 2021 by Romania’s Ministry of Education, the LCCM mandated the implementation of a hybrid online schooling system in all constituencies where the Covid-19 transmission intensity exceeded one per a thousand residents — illustrated in Fig. 2.

Fig. 2

The Hybrid Schooling Local Conditional Containment Measure — Classification of Constituencies. Note: The local Covid-19 reported transmissions rate was above (below) one per a thousand residents in treated (control) constituencies on the 29th of April 2021 — that is, at the time of the hybrid schooling policy’s announcement. Out of 3180 constituencies, 1387 are classified as treated. Map created using QGIS.

Starting from the 5th of May, students in grades five through seven, and those in grades nine through eleven would not be allowed to physically attend classes. The scale of the policy was therefore substantial, with three quarters of middle and high-schoolers, or roughly one million students being potentially affected.6 In constituencies with a lower transmission intensity, the policy imposed no restrictions.

Three reasons make this setting suitable for my purpose. First, I can exploit the existence of a clear threshold in the policy’s geographical scope to isolate the LCCM’s causal effects. Since I am interested in the news-shock channel exclusively, I use the incidence rates reported for the 29th of April to code the forcing variable in the RD specification — as discussed below.

Second, the LCCM was announced roughly five days prior to the vacation ending, thus creating a time-window in which the public was informed of the policy while no measures were yet enforced. It is this announcement-implementation time-lag that allows me to assess whether indirect actions may partially explain an LCCM’s effects.

Finally, and just as importantly, no other restrictions were enforced at the one per a thousand Covid-19 incidence threshold7 neither after or shortly8 before the schooling policy’s announcement, allowing me to isolate the LCCM’s news-shock effect from the confounding influence of other measures.

Analysis

I use a sharp RD design (Imbens and Lemieux 2008) to estimate the policy’s news-shock effect.

First, I define the forcing variable for each constituency i: where the virus’s incidence rate – capturing the fourteen-day case notification rate per a thousand residents reported on the 29th of April – is retrieved from Romania’s Health Ministry. Next, I construct my treatment variable TREAT, equal to one when MARGIN is positive (zero otherwise).

To estimate the LCCM’s (local) average treatment effect [LATE], I restrict my sample such that MARGIN [- +], where is computed using the algorithm in Calonico et al. (2014), and run:

Y captures the Covid-19 incidence rate reported for constituency i on day d. To quantify the news-shock effect, I focus on the five days between the 30th of April and the 4th of May.

is the coefficient of interest. If the policy does lead to a reduction in the transmission incidence via the news-shock effect, I expect ’s estimate to be negative in the days prior to the Easter holidays ending. For efficiency, I include a vector of controls X.9 I use heteroskedasticity-robust standard errors.

The findings are reported in Tables B1 and B2. In Table B1, I give OLS results. In Table B2, I present the RD estimates. The RD estimates are also depicted graphically in Fig. 3, alongside their corresponding 90 percent confidence intervals.

Fig. 3

Main Result — The News-shock Effect of Local Conditional Containment Measures. Note: I plot the treatment effect estimates retrieved from fitting the preferred RD specification alongside their 90 percent confidence intervals. The dependent variable is the constituency-level Covid-19 incidence rate per a thousand residents.

Overall, my results provide early evidence for the existence of a causal LCCM news-shock effect. First, I am unable to reject the zero effect null hypothesis when considering the days before the announcement, suggesting that the RD framework is contextually valid.10 More importantly, I find that the policy’s announcement led to a statistically-significant reduction in the Covid-19 reported transmission intensity in the days predating the end of the holidays on May 5th, a time-window where no restrictions were yet imposed.

The magnitude of the estimates also show that the news-shock effect is meaningful in terms of its magnitude. While no effects are retrieved on the day immediately following the announcement11 , the numbers estimated thereafter suggest that the policy led to a reduction of between 0.1 and 0.2 cases per a thousand residents via the news-shock channel, with the largest figure retrieved for the final day of holidays — estimated at a reduction of 0.191 (95% C.I. 0.034–0.349) per a thousand12 , or just under twenty percent relative to the one per a thousand incidence rate threshold at which the policy was implemented.13

Discussion

Exploiting a conditional Covid-19 containment policy announcement, I provide evidence for a news-shock mechanism partially explaining the effects of local restrictions on the virus’s transmission intensity. More broadly, my results add to a wider literature on news-driven fluctuations, which has documented how information on future policies — for instance, on fiscal choices (Barro and Redlick 2011), tax changes (Mertens and Ravn 2012), or natural resources (Arezki et al. 2017) - can lead to immediate responses. Concretely, my findings suggest that restriction-policies may incentivize indirect actions in targeted communities14 independently of their mechanical effects, contributing to our understanding of the potential benefits of geographically-flexible lockdowns.

However, the present study has a clear limitation that future work should tackle, in that the exact actions underlying the news-shock effect need to be decomposed — in particular, it is unclear whether the effects are driven by socially “desirable” (e.g., better hygiene practices) or “undesirable” (e.g., test avoidance15 ) actions, an important distinction for understanding how effective these policies are in terms of their actual objective: alleviating the virus’s societal burden.