Investigators' Blog

Probability of elimination for COVID-19 in Aotearoa

Probability of elimination for COVID-19 in Aotearoa


5 June 2020


Probability of elimination for COVID-19 in Aotearoa New Zealand


Executive Summary

  • Our model of COVID-19 spread estimates that after 2-3 weeks of no new reported cases, there is a 95% probability that COVID-19 has been eliminated in New Zealand.
  • A 95% probability of elimination is achieved after 10 consecutive days with no new reported cases under an optimistic scenario with high detection of clinical cases, and after 22 days under a more pessimistic scenario with low case detection.
Women remain under represented at top levels of academia

Women remain under represented at top levels of academia

New research published in the journal Education Sciences suggests that women remain disproportionately under-represented in senior academic positions within New Zealand universities.

The study has shown that from 2012 to 2017 there was little if any improvement in gender parity in senior roles at all eight or our universities –  the University of Auckland, Auckland University of Technology, University of Waikato, Massey University, Victoria University of Wellington, University of Canterbury, Lincoln University and University of Otago.

Existing gender diversity programmes appear to have had limited impact

“We’re still seeing an absence of women at the higher levels of academic employment across New Zealand universities,” says the study’s lead author Dr Leilani Walker, Te Pūnaha Matatini Associate Investigator.

“There are disproportionately fewer women in senior lecturer, professor positions and so forth, and this is in spite of various programmes that have been developed to try to improve the situation. Based on the data we have, it looks like women are proceeding up the academic promotion ladder at a slower rate than their male colleagues.”

Most of our universities, except for the University of Canterbury and Lincoln University, had equitable proportions of women in their academic work forces in 2017. However, the study found that men dominated the more senior employment roles, making up 64-69% of Associate Professors/Heads of Department and 74%-81% of Professors/Deans – from 2012 to 2017.

Consistent with previous research, gender disparities in senior university roles within New Zealand could not be explained by male and female age difference distributions.

Potential need for institutions to review their promotion processes 

These findings may provide a timely opportunity for New Zealand’s academic institutions to review and update their processes around hiring and promotion, says Dr Walker.

“We have a variety of programmes at New Zealand universities that try and help promote the careers of women into more senior positions, but it’s not really apparent in our minds whether they work,” Dr Walker says.

“We also question the extent to which just increasing the number of people present can create a culture change. Should we instead be starting to look at ways of engendering a culture tilt, rather than just getting more bodies in the room?”

“Perhaps we should be looking at existing models being used to judge success,” says Dr Walker. “The careers of female academics are often disrupted by life’s other priorities – for example, parental leave or to care for parents – and such interruptions can impact their research performance. If New Zealand universities continue to measure academic success based on the assumption of a linear, straight-forward career path, then any deviations will continue to disadvantage women.”

About the study authors

Dr Walker’s co-authors on this paper are all investigators at Te Pūnaha Matatini. They include Dr Tara McAllister (Te Aitanga ā Māhaki, Ngāti Porou), Research Fellow, School of Biological Sciences at the University of Auckland, Dr Isabelle Sin, Research Fellow, Motu Economic and Public Policy Research in Wellington, Associate Professor Cate Macinnis-Ng, School of Biological Sciences at the University of Auckland, and Kate Hannah, Deputy Director, Equity & Diversity Te Pūnaha Matatini.

Feature photo by Clay Banks on Unsplash.

Effective reproduction number for COVID-19 in Aotearoa

Effective reproduction number for COVID-19 in Aotearoa


22 May 2020


Effective reproduction number for COVID-19 in Aotearoa New Zealand


Executive Summary

  • The effective reproduction number, Reff, is the average number of secondary cases infected by a primary case, a key measure of the transmission potential for a disease.
  • Compared to many countries, New Zealand has had relatively few COVID-19 cases, many of which were caused by infections acquired overseas. This makes it difficult to use standard methods to estimate Reff.
  • We use a stochastic model to simulate COVID-19 spread in New Zealand, and report the values of Reff from simulations that gave best fit to case data.
  • We estimate that New Zealand had an effective reproduction number Reff = 1.8 for COVID-19 transmission prior to moving into Alert Level 4 on March 25 and that after moving into Alert level 4 this was reduced to Reff = 0.35.
  • Our estimate Reff = 1.8 for reproduction number before Alert Level 4, is relatively low compared to other countries. This could be due, in part, to measures put in place in early- to mid-March, including: the cancellation of mass gatherings, the isolation of international arrivals, and employees being encouraged to work from home.
A structured model for COVID-19 spread

A structured model for COVID-19 spread


15 May 2020


A structured model for COVID-19 spread: Modelling age and healthcare inequities


Executive Summary

  • We develop a structured model to look at the spread of COVID-19 in different groups within the population. We examine two case studies: the effect of control scenarios aimed at particular age groups (e.g. school closures) and the effect of inequitable access to healthcare and testing. These scenarios illustrate how such evidence could be used to inform specific policy interventions.
  • An increase in contact rates among children, which might result from reopening schools, is on its own unlikely to significantly increase the number of cases. However, if this change in turn causes a change in adult behaviour, for example increased contacts among parents, it could have a much bigger effect.
  • We also consider scenarios where outbreaks occur undetected in sectors of the community with less access to healthcare. We find that the lower the contact rate between groups with differing access to healthcare, the longer it will take before any outbreaks are detected in any groups who experience unequitable access to healthcare, which in Aotearoa New Zealand includes Māori and Pacific peoples.
  • Well-established evidence for health inequities, particularly in accessing primary healthcare and testing, indicates that Māori and Pacific communities in Aotearoa New Zealand are at higher risk of undetected outbreaks. The government should ensure that the healthcare needs of Māori and Pacific communities with respect to COVID-19 are being met equitably.
New interactive app simulates COVID-19 spread

New interactive app simulates COVID-19 spread

A New Zealand-specific interactive epidemic simulation app developed by Dr Audrey Lustig, Associate Investigator at Te Pūnaha Matatini, and hosted by the University of Auckland’s Centre for eResearch, has just been released.

Called the COVID-19 Take Control simulator, the app illustrates the effects of hygiene and physical distancing measures that all Kiwis are undertaking to control the spread of COVID-19. One of the app’s key features is that it allows the user to see the effects higher and lower collective cooperation with policies aimed at breaking the chain of transmission.

Try it out now! Check out the app here.

Effect of Alert Level 4 measures on COVID-19 transmission

Effect of Alert Level 4 measures on COVID-19 transmission


22 April 2020


Effect of Alert Level 4 on Reff : review of international COVID-19 cases


Executive Summary

  • In response to the COVID-19 pandemic, countries around the world are implementing a range of intervention measures, such as population-wide social distancing and case isolation, with the goal of reducing the spread of the virus.
  • Reff, the effective reproduction number, measures the average number of people that will be infected by a single contagious individual. A value of Reff > 1 suggests that an outbreak will occur, while Reff < 1 suggests the virus will die out.
  • Comparing Reff in an early outbreak phase (no or low-level interventions implemented) with a later phase (moderate to high interventions) indicates how effective these measures are for reducing Reff.
  • We estimate early-phase and late-phase Reff values for COVID-19 outbreaks in 25 countries (or provinces/states). Results suggest interventions equivalent to NZ’s Alert Level 3-4 can successfully reduce Reff below the threshold for outbreak.




Inequities in COVID-19 infection fatality rates in New Zealand

Inequities in COVID-19 infection fatality rates in New Zealand


17 April 2020


Estimated inequities in COVID-19 infection fatality rates by ethnicity for Aotearoa New Zealand


Executive Summary

  • There is limited evidence as to how COVID-19 infection fatality rates (IFR) may vary by ethnicity. We combine demographic and health data for ethnic groupings in Aotearoa New Zealand with international data on IFR for different age groups to estimate inequities in IFR by ethnicity.
  • If age is the dominant factor determining IFR, estimated IFR for Māori is around 50% higher than non-Māori.
  • If underlying health conditions are more important than age per se, then estimated IFR for Māori is more than 2.5 times that of New Zealand European, and estimated IFR for Pasifika is almost double that of New Zealand European.
  • IFRs for Māori and Pasifika are likely to be increased above these estimates by racism within the healthcare system and other inequities not reflected in official data.
  • IFR does not account for differences among ethnicities in COVID-19 incidence, which could be higher in Māori and Pasifika as a result of crowded housing and higher inter-generational contact rates. These factors should be included in future disease incidence modelling.
Modelling COVID-19 spread and the effects of Alert Level 4 in New Zealand

Modelling COVID-19 spread and the effects of Alert Level 4 in New Zealand


9 April 2020


A stochastic model for COVID-19 spread and the effects of Alert Level 4 in Aotearoa New Zealand


Executive Summary

  • While New Zealand case numbers remain low, tracing, testing, and rapid case isolation, combined with population-wide control methods, offer an opportunity for the country to contain and eliminate COVID-19.
  • Simulations using our model suggest that the current population-wide controls (Alert Level 4) have already had a significant effect on new case numbers (see figure below).

  • We also find that fast case isolation, whether as a result of contact tracing, rapid testing, or otherwise, can lead to containment and possibly even elimination, when combined with strong population-wide controls.
  • Slow case isolation can also lead to containment, but only as long as strong population wide controls remain in place. It is unlikely to lead to elimination.



While case numbers remain low, population-wide control methods combined with efficient tracing, testing, and case isolation, offer the opportunity for New Zealand to contain and eliminate COVID-19. We use a stochastic model to investigate containment and elimination scenarios for COVID-19 in New Zealand, as the country considers the exit from its four week period of strong Level 4 population-wide control measures. In particular we consider how the effectiveness of its case isolation operations influence the outcome of lifting these strong population-wide controls. The model is parameterised for New Zealand and is initialised using current case data, although we do not make use of information concerning the geographic dispersion of cases and the model is not stratified for age or co-morbidities.

We find that fast tracing and case isolation (i.e. operations that are sustained at rates comparable to that at the early stages of New Zealand’s response) can lead to containment or elimination, as long as strong population-wide controls remain in place. Slow case isolation can lead to containment (but not elimination) as long as strong Level 4 population-wide controls remain in place. However, we find that relaxing strong population-wide controls after four weeks will most likely lead to a further outbreak, although the speed of growth of this outbreak can be reduced by fast case isolation, by tracing, testing, or otherwise. We find that elimination is only likely if case isolation is combined with strong population-wide controls that are maintained for longer than four weeks.

Further versions of this model will include an age-structured population as well as considering the effects of geographic dispersion and contact network structure, the possibility of regional containment combined with inter-regional travel restrictions, and potential harm to at risk communities and essential workers.




Suppression and mitigation strategies for control of COVID-19 in New Zealand

Suppression and mitigation strategies for control of COVID-19 in New Zealand



25 March 2020


Suppression and Mitigation Strategies for Control of COVID-19 in New Zealand


Executive Summary

  • Suppression strategies aim to keep the number of cases to an absolute minimum for as long as possible. This requires early and effective control interventions.
  • Suppression can only delay an epidemic, not prevent it, but may buy enough time for a vaccine or treatment to become available.
  • Mitigation strategies aim to control an epidemic so that herd immunity is acquired by the population without overwhelming healthcare systems.
  • Mitigation strategies are likely to be very high risk: they are unproven internationally, potentially sensitive to uncertainty, and it may take years for herd immunity to be acquired.
  • Strategy can be switched from suppression to mitigation. For example, once successful mitigation strategies have been tested in other countries. It is likely to be difficult or impossible to switch from a mitigation to a suppression strategy.
  • A combination of successful suppression, strong border measures, and widespread contact tracing and testing resulting in containment could allow periods when control measures can be relaxed, but only if we can reduce cases to a handful.



A standard SEIR-type compartment model, parameterised for New Zealand, was used to simulate the spread of Covid19 in New Zealand and to test the effectiveness of various control strategies. Control aims can be broadly categorised as either suppression or mitigation. Suppression aims to keep cases to an absolute minimum for as long as possible. Mitigation aims to allow a controlled outbreak to occur, with the aim of preventing significant overloads on healthcare systems and gradually allowing the population to develop herd immunity.

Both types of strategy are fraught with uncertainty. Suppression strategies can succeed in delaying an outbreak, but only for as long as such control measures can be sustained. Once controls are eased or restricted, an epidemic is likely to follow as no herd immunity has been acquired. The success or failure of mitigation strategies can depend sensitively on the timing and efficacy of control measures, and require the ability to bring rapidly growing outbreaks under immediate control when needed. This is as yet untested even for a combination of national interventions including case isolation, household quarantine, population-wide social distancing and closure of schools and universities.

Although there are disadvantages to both types of approach, suppression has the advantage of buying time until a vaccine and/or treatment become available and allowing NZ to learn from rapidly unfolding events in other countries. A combination of successful suppression, strong border measures, and widespread contact tracing and testing resulting in containment could allow periods when control measures can be relaxed, but only if cases are reduced to a handful.



The appendix, containing the model specification, AVAILABLE HERE [MINOR REVISIONS MADE 30 MARCH 2020].


Creating an app to enrich the network visualisation experience

Creating an app to enrich the network visualisation experience

Te Pūnaha Matatini intern Shih-Hao (Samuel) Chen talks about his work with Nebula Data over the 2019-20 summer developing a tool that provides enhanced network analysis capabilities.

By Shih-Hao (Samuel) Chen

Networks arise in all shapes and forms in our everyday lives, and their features can provide new information to its respective topic. However, a list of nodes and edges alone is challenging to interpret, and transforming the dataset into a useful visualisation relies on inflexible third-party applications. We wanted to build a supportive, customisable tool that would enable data analysts to uncover new observations.

How does one analyse networks?

Networks are a group of nodes (things) and edges (relationships). The analysis of these networks provides answers to questions that otherwise would be difficult to answer, such as which bus stops have the most traffic, or which characters appear the most in a movie.

The question then arises – how does one analyse networks? The answer is to use a network visualisation and processing application. Gephi is a well-known application that provides this functionality, but Gephi has limitations in terms of user controllability. Our task was to produce an alternative software, but with more flexibility and control from an analytical perspective. For example, our product may offer more colouring schemes, alternative algorithms or additional forces.

When presented with a list of nodes and edges, a computer has no intuitive method of drawing the network. After all, the visualisation is purely aesthetic. The solution is to implement a force-directed algorithm that transforms the nodes into a “natural” layout. We can see the result in Figure 1:

Figure 1. Demonstration of a force-directed network in action.

The network in this visualisation is slowly untangling, allowing us to identify central nodes and clusters.

We created our application hoping for flexibility and customisation. Upon further reading, we found that the attraction/repulsion ratio between the nodes determines the resultant network shape. As shown in Figure 2 below, we can see how the visualisation algorithm can dramatise clustering by changing the forces alone.

Figure 2. Three different attraction/repulsion force ratios, and its resultant shapes.

Although these observations are informative, metrics and values help form reliable claims on these networks. By colouring the nodes based on some measure, the application presents the information in a reader-friendly way. Figure 3 below illustrates two examples of analyses that the app supports:

Figure 3. Katz (left) and Betweenness (right) Centrality, highlighted by the colouring of the nodes.

Finally, our application also provides a time-lapse feature, where we can observe how graphs change over time. The insertion/removal of nodes and edges will change the shape of the network and analysts can use these changes to make observations.

Where to next?

Although these features are incredibly useful for uncovering hidden stories of networks, there are plenty of features that could provide further insight into the relationships around us. Community detection, for example, is an exciting field that groups nodes of high connectivity together. There are also visual features, such as nodes-overlap prevention, that would improve readability.


Shih-Hao (Samuel) Chen is currently studying a Bachelors of Engineering, specialising in Software Engineering. Samuel enjoys problem-solving and is incredibly fascinated with the complex nature of algorithms.