ECOS3997 Final Report PDF

Preview

Summary

Final report...

Description

ECOS3997 FINAL REPORT 470411290 480441803

Edward Khoury Dylan Sheldon Executive Summary As it leaves a period of lockdown, the city of Melbourne must look for a more effective contact tracing system to help combat COVID-19. We propose the introduction of legislation making the COVIDSafe App the default method of ‘scanning in’ to public venues so as to centralise the collection of contact tracing data and smooth the overall process of tracing the movements of a confirmed case of COVID-19. We conduct a randomised controlled test using a control and treatment group to determine the expected impact of mandating the app. In conducting our regression analysis, we look to distinguish between the null; there is no significant effect caused by the mandate; and the alternative; the implementation of the mandate will have a significant effect on the use of the COVIDSafe App. Our results align closely with our estimated effect size, which predicts the population percentage of Melbournians utilising the COVIDSafe App rising to 66%. This report will conclude in favour of the proposed mandate.

The Problem As the City of Melbourne faces coming out of lockdown, the key concern must be keeping new cases low and manageable while reopening the city and returning to regular lifestyles. The World Health Organisation (2015) defines contact tracing as “the process of identifying, assessing, and managing people who have been exposed to a disease to prevent onward transmission”. Such an effective system is essential in limiting the transmission of any virus (WHO, 2017) and has proven effective when implemented in various COVID19 breakouts in NSW (Cockburn, 2020). The problem arises in the revelations that the Victorian government and health authorities have little confidence in their current contact tracing systems (Grattan, M. 2020). A lack of guidelines and regulation concerning what data must be gathered from potential contacts, and how this data is collected and stored, leaves large cracks in the process (Coopes, A & Bennett, C. 2020), which are filled with ineffective and archaic ‘pen and paper’ practices (Taylor, 2020). There is an over reliance on the honesty and accuracy of people’s recollections under the existing systems. Most significantly, the genesis of the contact tracing routine is compromised when individuals provide incorrect contact details, either intentionally or not (QLD Health, 2020). In areas where digital sign in systems have been adopted, the distinction between physical pen and paper is almost purely ergonomic. The onus is still very much placed on the venue itself (Federal Department of Health, 2020), and the margins of error persist since data is not centralised or managed under a single umbrella (Criddle & Kelion, 2020).

Proposed Solution

We propose the introduction of legislation requiring the COVIDSafe App become the default method of signing into venues. This would require minimal technical work. Once legislation has been introduced, an update to the app introducing a camera-linked scanner would be required. Then, similarly to existing digital sign in systems, QR codes designed for the COVIDSafe App would be issued to all necessary venues. Rather than the QR code transporting you to a webpage where you are prompted to fill in your personal contact details, this code would time-stamp and register your unique profile contained in the COVIDSafe App with the venue, in a similar fashion to how the app interacts with other apps within your vicinity. In the event of a reported case at, or concerning a patron within 14 days of visiting, that venue, all individuals who were registered within 24 hours of the infected patron would be notified via the app. Individuals who were registered within 4 hours of the infected patron would be issued an urgent notification. This proposal is rooted in both psychology and behavioural economics. When one is provided, people exhibit a tendency to use a default option (Dinner et al, 2011). By making the COVIDSafe app the default option for signing in, we are making it appear to be the ‘normal’ option. Currently, digital sign-ins are merchant driven and commercialised, and as a result most venues do not share the same system. People typically overvalue their time (Leclerc et al, 1995), and as such are likely to be frustrated by having to re-enter their details every time they use a different provider’s sign-in system. If the mandated option is also faster and more efficient, we limit the active choice and subconsciously nudge consumers towards a new ‘normal’. We are also looking at a perfect opportunity to implement our proposal. Individuals are far more likely to adopt new habits during periods of transition (Service et al, 2014), such as the current transition out of lockdown. With this in mind, the proposal should be implemented as soon as possible to give it the best chance of traction and success. As the COVIDSafe App essentially provides a digital bread crumb trail, this would allow for far more accurate tracing. Similar approaches have been tried and tested successfully, as recently as this year in South Korea and the UK. The Korean approach demonstrated that it is more effective to test small samples of the right (high risk) people with haste than to test large hordes (Park et al, 2020). Meanwhile in the UK, where a similar system employs an app to trace the movements of infected individuals, shortcomings have been identified in that people are unlikely to answer unrecognised public phone numbers in fear of spam or hoax calls (Hughes et al, 2020). Using the COVIDSafe App to sign in to venues will allow far greater tracing to occur without having to speak with the positive individual, however there may still be occasions where it is necessary to create the most comprehensive tracing profile possible. To be effective, tracing must occur within 48 hours (WHO, 2015). Thus we propose that in the event of a positive test, should the individual not be contactable via phone call after three failed attempts, a team be sent to their physical address. This can be serviced by the 1000 ADF personnel deployed in Victoria to assist in management of COVID19 (ABC, 2020). Following the initial announcement of the COVIDSafe App, there were 3 million downloads nationwide within three days (12% of the population). It then took a further month to double to the 6 million milestone (Meixner, 2020). We expect to see a similar flattening effect following the legislative announcement mandating the COVIDSafe App as the required method of signing in. Since current data is undisclosed, we apply the initial national engagement rate of 12% to Melbourne’s population of 5 million to assume 600,000 Melbournians currently have the app downloaded. If the majority of Melbournians attend venues or services on a regular basis, we expect at least 2/3rds of the population (around 3 million) to be engaged with the app following the proposal’s announcement - a raw increase

of 2.4 million. Over the month following, the download curve throughout Victoria will increase at a decreasing rate as individuals who attend less venues (or less frequently) engage with the app as suits them.

Evaluation Plan

Due to the coronavirus being a relatively new world issue, no data in relation to the impact of making the COVIDsafe app the default has been made, and as such in order to evaluate whether the intervention strategy we are proposing in fact has an impact on the usage of the app, we will be running a Randomized controlled test (RCT) using regression to infer the results. The reasoning for including a regression is because it describes parametrically how the factor of interest is affected by the intervention. In this case we are measuring the correlation and direct effect of being forced to have the app on their phone and whether the consumers will in fact use it. The regression on use of the app is going to give the right causal coefficient, coefficient of the impact of intervening and making someone download the app. We randomly assigned people to two groups: o Treatment group = people who were instructed to download the app o Control group = People who weren’t told to download the app The proportion of people who used the app in the treatment group grants an unbiased estimate of the probability of use if a random person were to be forced to use the app and the control group gives an unbiased estimate of the probability of use if a random person is not mandated to download the app. The difference between the two gives the average causal effect, where the expected impact of mandating the use of the app would lead to more people using it. Reflecting the impact on an average person independent of all other factors. Outcome Variable and Regression The outcome variable which will be assessed is the proportion of people using the app in the period of a week. We documented the data usage of the app for each participant prior to the RCT, and then required each individual to forward a screenshot of their data usage of the COVIDsafe app after a week. If the data usage had increased by 2 megabytes or more then the participant had been using the app, granting a clear indication of whether or not the COVIDsafe app had been used. In order to interpret the data we will be using the simple linear regression(SLR), with the outcome being linearly related to only one regressor. The reason for using a SLR is because when using RCT data the SLR yields an unbiased estimate of the average causal impact of mandating that the app be downloaded and is easy to interpret and work with. Additionally only one regression was used due to scarce resources in both finances and time, disallowing the capacity to analyse other regressions such as differences in demographics or geographic areas. The regression being used is whether or not an individual was mandated to download the app. For the treatment group, we required each individual to download the app (if they hadn’t already) in our presence or show photographic evidence of the download, whereas in the

control group no such requirement was made. This regression was chosen as it allows us to compare the proportion of participants who used the app in either group with a simple unbiased estimate of the effect that our intervention will have on the participants. How we will assess to see if the intervention works We will determine whether the intervention is effective by interpreting whether to accept or reject the null hypothesis. The null hypothesis (Ho) = there is no significant effect caused by the mandate on whether the COVIDsafe app will be used. The coefficient is zero. The Alternate Hypothesis (H1) = the implementation of the mandate will have a significant effect on how often the COVIDsafe app will be used. The variables being observed are: o The dependent variable, which is whether the COVIDsafe app is being used, o And the independent variable which indicates whether an individual has been instructed to download the app. If the coefficient for the mandate increases over the threshold (which is zero), then the intervention is deemed effective and the H1 will be accepted, if not then the Ho is accepted. When assessing whether to accept or reject the intervention we are at risk of two errors. A false positive error, which is if the mandate intervention wasn’t effective although the threshold was exceeded and as such the Alternative Hypothesis was accepted. Or a false negative error, which occurs when the Alternative Hypothesis was rejected because the threshold was not exceeded, although the intervention was effective, and the Ho was false. As such the following error rates will be incorporated into the calculation of our sample size to attempt to eliminate the risk. o B= 0.2 is the false negative error rate which is conventionally considered acceptable o a = 0.05 is the false positive error rate which is considered acceptable Sample Size For our sample size we decided on randomly selecting 100 participants, with 50 people in both the treatment group and the control group. Sample size has a direct impact on the reliability and usefulness of the data, whereas if the sample size is too small then the results derived from the data are irrelevant and potentially misleading. The aim is to minimise the false negative error and false positive error in our estimates. As such, errors were taken into consideration (a = 0.05, B = 0.2) to obtain reasonable certainty when running the sample size test, with at least an 80% chance of rejecting the Ho if the intervention is effective.

After running the sample size test, we were quoted as only needing 24 participants in our sample group to accurately indicate the effect that a mandate has on the number of people who use the COVIDsafe app. Such a small sample group is required due to our research in relation to the app that predicts an exponential increase in usage of the app from 32% to 66% if the intervention were applied. Despite this size test, data is always made more reliable by increasing the sample size, and as such due to us having adequate time and money to increase the number of participants, we considered it possible to increase the sample size to 100 people.

Analysis of regression and results In order to assess whether our random sample is statistically significant and whether our data is applicable to the greater population we firstly analysed the p-values. The p-values of the mandate tests the null hypothesis, determining if it has no correlation with the use of the app. Considering the p-value of the Mandate is less than 5% at 0.1128%, it is considered statistically significant, indicating there is 0.1128% chance that the Ho is correct. Therefore there is sufficient evidence to indicate that the mandate will have an effect on the population level usage of the COVIDsafe app. Furthermore, the regression coefficient estimates the parameters for the actual population. The coefficient for the mandate is 0.322, indicating a positive correlation, whereby when the mandate is applied the mean usage of the app will increase by a probability of 0.322 (1 being 100% probability). This is reflected in the increase in mean probability of usage from 0.3 in the control group to 0.62 in the treatment group, almost doubling the probability of an individual using the app. This aligns quite closely with the estimated effect size, which is expected to increase the number of users of the COVIDsafe app to 0.66. The minuteness of the difference may be attributed to human and estimation error. Furthermore, the coefficient surpassed the threshold (Ho = 0) and as such the alternative hypothesis has been accepted and the null hypothesis is rejected. Appendix 1.0

References

WHO. (2015). Implementation and management of contact tracing for Ebola virus disease. Retrieved from https://apps.who.int/iris/bitstream/handle/10665/185258/WHO_EVD_Guidance_Contact_15.1_eng.pdf ?sequence=1 WHO. (2017). Contact Tracing. Retrieved from https://www.who.int/news-room/q-a-detail/contacttracing Cockburn, P. (2020, September 8). NSW is the Gold Standard for COVID-19 Management According To the PM - Here’s Why. ABC. Retrieved from https://www.abc.net.au/news/2020-09-08/why-pm-saysnsw-is-gold-standard-in-covid-19-control/12636890 Grattan, M. (2020). Daniel Andrews’ delay prompts new questions about Victoria’s contact tracing. The Conversation. Retrieved from https://theconversation.com/daniel-andrews-delay-prompts-newquestions-about-victorias-contact-tracing-148783 Coopes, A., Bennett, C. (2020). Where did Victoria go so wrong with contact tracing and have they fixed it?. Retrieved from https://www.croakey.org/where-did-victoria-go-so-wrong-with-contact-tracingand-have-they-fixed-it/ Taylor, J. (2020, September 8). ‘Less pen and paper’: Victoria to digitise Covid contact tracing after federal criticism. The Guardian. Retrieved from https://www.theguardian.com/australianews/2020/sep/08/salesforce-to-digitise-victorias-covid-contact-tracing-after-federal-criticism Federal Department of Health. (2020). National Contact Tracing Review. Retrieved from https://www.health.gov.au/sites/default/files/documents/2020/11/national-contact-tracing-reviewnational-contact-tracing-review.pdf QLD Health. (2020). Fake names and numbers putting broader community at risk. Retrieved from https://www.health.qld.gov.au/news-events/doh-media-releases/releases/fake-names-and-numbersputting-broader-community-at-risk-2020-07-13 Criddle, C., Kelion, L. (2020, May 7). Coronavirus contact-tracing: World split between two types of app. BBC. Retrieved from https://www.bbc.com/news/technology-52355028 Dinner, I., Johnson, E.J., Goldstein, D.G., Liu, K. (2011). Partitioning default effects: Why people choose not to choose. Journal of Experimental Psychology: Applied, 17(4), 332-341. Leclerc, F., Schmitt, B.H., Dubé, L. (1995). Waiting Time and Decision Making: Is Time like Money?. Journal of Consumer Research, 22(1), 110-119. Service, O., Hallsworth,M., Halpern, D., Algate, F. Gallagher, R., Nguyen, S., Ruda, S., Sander, M., Pelenur, M., Gyani, A., Harper, H., Reinhard, J., Kirkman, E. (2014). EAST: Four simple ways to apply behavioural insights. The Behavioural Insights Team. Retrieved from https://www.behaviouralinsights.co.uk/wp-content/uploads/2015/07/BIT-PublicationEAST_FA_WEB.pdf Park, Y.J., Choe, Y.J., Park, O. (2020). Contact Tracing during Coronavirus Disease Outbreak, South Korea. Emerging Infectious Diseases, 26(10), 2465-2468. ABC. (2020, August 12). How is the Australian Defence Force assisting states during COVID-19?. Retrieved from https://www.abc.net.au/news/2020-08-12/fact-check-defence-force-coronavirus-factfile-hotel-quarantine/12522492?nw=0 Meixner, S. (2020, June 2). How many people have downloaded the COVIDSafe app and how central has it been to Australia’s coronavirus response?. ABC. Retrieved from

https://www.abc.net.au/news/2020-06-02/coronavirus-covid19-covidsafe-app-how-many-downloadsgreg-hunt/12295130 Currie, D.J., Peng, C.Q., Lyle, D.M., Jameson, B,A., Frommer, M.S. (2020). Stemming the flow: how much can the Australian smartphone app help to control COVID-19?. Public Health Research & Practice, 30(2)....