Hard, not early: putting the New Zealand Covid-19 response in context

Abstract

A popular narrative that New Zealand’s policy response to Coronavirus was ‘go hard, go early’ is misleading. While restrictions were the most stringent in the world during the Level 4 lockdown in March and April, these were imposed after the likely peak in new infections. I use the time path of Covid-19 deaths for each OECD country to estimate inflection points. Allowing for the typical lag from infection to death, new infections peaked before the most stringent policy responses were applied in many countries, including New Zealand. The cross-country evidence shows that restrictions imposed after the inflection point in infections is reached are ineffective in reducing total deaths. Even restrictions imposed earlier have just a modest effect; if Sweden’s more relaxed restrictions had been used, an extra 310 Covid-19 deaths are predicted for New Zealand – far fewer than the thousands of deaths in some widely reported mathematical simulations.

KEYWORDS:

• Coronavirus
• deaths
• policy timing
• response stringency
• New Zealand

Acknowledgements

I am grateful to helpful comments from Michael Reddell and an anonymous referee.

Disclosure statement

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

Notes

1 See, for example a March 14 report: https://www.newsroom.co.nz/we-must-go-hard-and-we-must-go-early.

2 This point is corroborated by the report, in late September 2020, that New Zealand’s earliest case arrived from Italy (and infected others) one week earlier (on February 21) than the previously earliest reported case: (https://www.health.govt.nz/news-media/media-releases/3-new-cases-covid-19-and-6-historical-cases). At the time, arrivals from China were restricted but not those from Italy. The revised record of the earliest cases highlights the incomplete nature of data on cases and supports an approach based on deaths data.

3 Nevertheless, deaths data have some problems of over-counting when positive Covid-19 tests are linked with subsequent deaths from any cause (Williams, Crookes, Glass, & Glass, 2020).

4 Homburg (2020) uses a lag of 23 days but from March and early April when treatment protocols for Covid-19 were still being developed. Improved care since then may extend the lag. Heneghan and Jefferson (2020) say 21–28 days and the Covid-19 tracking project (https://covidtracking.com) suggests a four week lag.

5 The time-series that Homburg used ended on April 13. A critique by Felder and Robra (2020) used a longer time-series, ending on April 30, but the same basic framework. While they find that Homburg’s analysis was based on a truncated distribution because it did not observe the full trajectory of the epidemic, these authors corroborate Homburg’s finding that infections in Germany started declining before lockdown. Wieland (2020) reaches the same conclusion, and criticizes some papers by economists with contrary findings that relied on cases data which lacked information on true infection dates. The use of a six-month time-series here should overcome the main criticism of Homburg made by Felder and Robra.

6 The delta method standard error for the timing of the inflection point for the median country in Table 1 is 0.9 days. The standard error for the inflection point for New Zealand is 0.4 days.

7 Another sensitivity analysis is to use a Gompertz curve rather than a logistic, which would date the inflection point for infections in New Zealand six days later than what Table 1 shows, which is still prior to lockdown.

8 The difference in the elasticity of death rates with respect to policy stringency prior to peak infections and the elasticity with respect to policy stringency after the peak is statistically significant at the p<0.04 level.

9 I calculate leave-out means for the same six OECD regions that Sebhatu et al. (2020) use.

10 To predict death rates from a regression with log death rates as the dependent variable I use the Duan (1983) smearing estimate.