That said, Shaun’s proposed solution appears to confuse CBA with cost-effectiveness analysis. For example: “In a CBA one should look for controls that minimise the total cost (economic + health)” is consistent with CBA thinking, whereas “bring Reff as close as possible to zero” and “more economically efficient in bringing case numbers down to a desired level” is about cost-effectiveness, i.e. choosing the cheapest way to achieve an exogenously specified goal. CBA is always an attempt to maximise net social benefits, not the pursuit of other goals.
CBA, properly applied, will not lead to a universal answer as to whether a short period of harsh controls has higher net benefits than does a longer period of laxer controls. CBA analysis will be specific to circumstances, as both costs and benefits change with circumstances. Further, in addition to the factors that Shaun mentions, the analysis will be sensitive to:
A. The proportion of the population with Covid at the start of the relevant period. This is because the costs of controls apply to the whole population, whereas benefits, measured as avoided Covid-related health costs, scale with the numbers infected.
B. The expected time horizon for the realisation of benefits. As NZ learnt during 2020 and 2021, the benefits of “elimination” were truncated every time a border breach resulted in a behind-the-border outbreak. An analysis that doesn’t take this into account will likely over-estimate benefits.
C. The expected counterfactual. Evidence from overseas suggests that a significant proportion of the population adopted voluntary controls in jurisdictions with relatively lax controls, especially when A was high. (Similarly, compliance with harsh controls was patchy, especially when A was low.) This makes determining the effect of specific packages of controls on Reff tricky.
All else equal, a high A will favour harsher controls. Over time, as controls reduce A, the optimal set of controls will change towards laxer ones. So, the output of a CBA analysis should be an optimally staged path, rather than a single “best” package of measures.
A longer time horizon B would favour harsher controls. At the other extreme, a low A combined with a very high chance of an externally sourced outbreak, say 1 week from now, would make harsh controls all but pointless in the intervening period.
Admittedly, my May 2020 analysis was inadequate with respect to B and C.
Without a properly specified and conducted CBA that incorporates these additional factors, I think it is premature to offer the conclusion that “Alert Level 3 is roughly 70% more economically expensive in controlling an outbreak than Alert Level 4”, in the specific circumstances NZ faced during the two 2020 outbreaks. And even such a CBA would not support a more universal conclusion.
I read Dave Heatley’s recent piece on the pandemic with interest. I note that his suggestion here:
"* place just enough restrictions on the populace to control disease spread (i.e. manage the reproduction rate down to below 1);”
is probably not correct in general, although I would be interested in seeing his argument for this.
These kinds of issues can be addressed using a dynamical model of disease spread. In the simplest models the decay time for case numbers scales as T/(1-Reff) where Reff < 1 and T is the generation time. Thus, placing just enough controls to bring Reff below one leads to a long decay time i.e. controls have to be held in place for a long time. In contrast, using very strong controls to bring Reff as close as possible to zero will reduce the time that controls need to be in place. In a CBA one should look for controls that minimise the total cost (economic + health). Stronger controls will reduce health costs but *may* also reduce economic cost if they only need to be in place for short times. When this is the case there is no trade-off between health and economic costs and a CBA would suggest using the strongest controls practically available.
If we assume that controls have a fixed daily cost (this is the approach Treasury has taken for example) then the total cost of imposing a set of controls is the daily cost (c, say) times the number of days that the controls need to be imposed. This time period scales as T/(1-Reff), so we can compare two control regimes to see which is more economically efficient in bringing case numbers down to a desired level via the ratio (c1/(1-R1))/(c2/(1-R2)).
To use some real data, we can compare Alert Levels 3 and 4, using Treasury’s costing (~18% and 30% daily GDP respectively) and estimates of Reff from the March-April 2020 and August 2020 wild-type COVID-19 outbreaks (I use TPM’s branching process fits here of 0.77 and 0.35 respectively). This suggests that Alert Level 3 is roughly 70% more economically expensive in controlling an outbreak than Alert Level 4. Alert Level 4 also minimises cases and corresponding health costs, so a decision-maker using a CBA should prefer it to Alert Level 3.
Andrew, a reader, asked whether my optimistic prediction in the last paragraph applied only to rich people in rich countries. His question made me realise that I had failed to specify a timeframe. What I had in mind was a decade or more hence, by which time the relevant technology should be reasonably universal. Should humans be unlucky enough to experience a second novel pathogen in the 2020s, then its benefits may be more patchily realised, as Andrew suggests.
Replying to Shaun Hendy's comment, posted below.
Shaun is completely correct to criticise my point: “place just enough restrictions on the populace to control disease spread (i.e. manage the reproduction rate down to below 1)”. I have no excuse for over-summarising the much more complex position I have held since May 2020, when I advocated the use of CBA to choose Alert Levels (see https://www.productivity.govt.nz/assets/Documents/cost-benefit-analysis-covid-alert-4/92193c37f4/A-cost-benefit-analysis-of-5-extra-days-at-COVID-19-at-alert-level-4.pdf).
That said, Shaun’s proposed solution appears to confuse CBA with cost-effectiveness analysis. For example: “In a CBA one should look for controls that minimise the total cost (economic + health)” is consistent with CBA thinking, whereas “bring Reff as close as possible to zero” and “more economically efficient in bringing case numbers down to a desired level” is about cost-effectiveness, i.e. choosing the cheapest way to achieve an exogenously specified goal. CBA is always an attempt to maximise net social benefits, not the pursuit of other goals.
CBA, properly applied, will not lead to a universal answer as to whether a short period of harsh controls has higher net benefits than does a longer period of laxer controls. CBA analysis will be specific to circumstances, as both costs and benefits change with circumstances. Further, in addition to the factors that Shaun mentions, the analysis will be sensitive to:
A. The proportion of the population with Covid at the start of the relevant period. This is because the costs of controls apply to the whole population, whereas benefits, measured as avoided Covid-related health costs, scale with the numbers infected.
B. The expected time horizon for the realisation of benefits. As NZ learnt during 2020 and 2021, the benefits of “elimination” were truncated every time a border breach resulted in a behind-the-border outbreak. An analysis that doesn’t take this into account will likely over-estimate benefits.
C. The expected counterfactual. Evidence from overseas suggests that a significant proportion of the population adopted voluntary controls in jurisdictions with relatively lax controls, especially when A was high. (Similarly, compliance with harsh controls was patchy, especially when A was low.) This makes determining the effect of specific packages of controls on Reff tricky.
All else equal, a high A will favour harsher controls. Over time, as controls reduce A, the optimal set of controls will change towards laxer ones. So, the output of a CBA analysis should be an optimally staged path, rather than a single “best” package of measures.
A longer time horizon B would favour harsher controls. At the other extreme, a low A combined with a very high chance of an externally sourced outbreak, say 1 week from now, would make harsh controls all but pointless in the intervening period.
Admittedly, my May 2020 analysis was inadequate with respect to B and C.
Without a properly specified and conducted CBA that incorporates these additional factors, I think it is premature to offer the conclusion that “Alert Level 3 is roughly 70% more economically expensive in controlling an outbreak than Alert Level 4”, in the specific circumstances NZ faced during the two 2020 outbreaks. And even such a CBA would not support a more universal conclusion.
A comment from Shaun Hendy, via email:
I read Dave Heatley’s recent piece on the pandemic with interest. I note that his suggestion here:
"* place just enough restrictions on the populace to control disease spread (i.e. manage the reproduction rate down to below 1);”
is probably not correct in general, although I would be interested in seeing his argument for this.
These kinds of issues can be addressed using a dynamical model of disease spread. In the simplest models the decay time for case numbers scales as T/(1-Reff) where Reff < 1 and T is the generation time. Thus, placing just enough controls to bring Reff below one leads to a long decay time i.e. controls have to be held in place for a long time. In contrast, using very strong controls to bring Reff as close as possible to zero will reduce the time that controls need to be in place. In a CBA one should look for controls that minimise the total cost (economic + health). Stronger controls will reduce health costs but *may* also reduce economic cost if they only need to be in place for short times. When this is the case there is no trade-off between health and economic costs and a CBA would suggest using the strongest controls practically available.
If we assume that controls have a fixed daily cost (this is the approach Treasury has taken for example) then the total cost of imposing a set of controls is the daily cost (c, say) times the number of days that the controls need to be imposed. This time period scales as T/(1-Reff), so we can compare two control regimes to see which is more economically efficient in bringing case numbers down to a desired level via the ratio (c1/(1-R1))/(c2/(1-R2)).
To use some real data, we can compare Alert Levels 3 and 4, using Treasury’s costing (~18% and 30% daily GDP respectively) and estimates of Reff from the March-April 2020 and August 2020 wild-type COVID-19 outbreaks (I use TPM’s branching process fits here of 0.77 and 0.35 respectively). This suggests that Alert Level 3 is roughly 70% more economically expensive in controlling an outbreak than Alert Level 4. Alert Level 4 also minimises cases and corresponding health costs, so a decision-maker using a CBA should prefer it to Alert Level 3.
Andrew, a reader, asked whether my optimistic prediction in the last paragraph applied only to rich people in rich countries. His question made me realise that I had failed to specify a timeframe. What I had in mind was a decade or more hence, by which time the relevant technology should be reasonably universal. Should humans be unlucky enough to experience a second novel pathogen in the 2020s, then its benefits may be more patchily realised, as Andrew suggests.