I am quite disappointed in the way the Singapore government
has handled the covid-19 situation recently. I have mentioned in previous blogs
that, apart from the issue of Covid in the foreign workers dormitories, the
government had done a good job. However, the recent developments leave me
disappointed. Data and analytics, to me, lie at the heart of dealing with
Covid19, but unfortunately, the Singapore government as high as it could be on
the analytics maturity scale. In this blog post, I will explain why.
But first some background. We have basically had 2 tools for
contact tracing, using 2 different technological approaches and focusing on 2
different things.
Initially, Singapore was officially going via checking-in
and checking-out from venues; and that includes parks, not just shops and
buildings, and shops/offices within buildings. Almost everywhere you would find
QR codes for you to scan. For people without mobile phones, they could use
their identity card to register.
What this data allows the government to do is keep track who
is in an area and when. The idea is that if someone who is infectious spent
time in the same location as you with overlapping time, then you can be sent an
appropriate message.
To me, the advantage of this is that it is a low battery
utilisation solution, but is quite blunt in the sense that if the shop/building
you went to is large, it is possible that you never crossed paths with the
infectious person. Basically, you may get a lot of false positives (people who
may be tagged as at risk of infection, but whose risk is actually very low).
Then, GovTech, the government’s technology arm, came up with
a blue-tooth based tracking system. This one basically captures the signal of
all devices in your proximity and records them. Hence you would be informed
only if you were in the same space and the same time (close and long enough to
establish the Bluetooth connection) as an infectious person.
The advantage of this is that it minimises false positives,
but if the virus lingers, increases false negatives (cases when it is thought
people are not infected but actually are). That is, if the virus lingers after
the infectious person has gone out of range of your device and you walk into
that space, and you are at risk, the blue tooth device does not capture this.
Also, this solution is more energy consuming than the
previous one. In mitigation, the government provided physical trackers that do
not rely on phone battery.
Now, to minimise the cases of false negatives, the
government also has people checking in into buildings like previously, but not
in specific shops necessarily (this is no longer stringently checked – earlier
people were at the doors to enforce). Still, this decreases false negatives
since people can be warned even if it was not ‘same time, same place’.
While I was writing this blog, it was just announced that
checking-in and out of markets and food courts will be mandatory. And, already
some shops have dedicated staff, again, to ensure checking in.
Hence, in terms of technology, Singapore is among the
leaders. So why do I say that the analytical maturity of the government is not
that high?
Analytics is not about the latest tech, collecting the most
information, it’s about collecting and analysing the data you need to answer an
issue and making sure the implementation of the analytics is done properly. And
to me the government has failed.
1 The data
collected has to be fit for purpose
The earlier QR code driven system captured who was where and
when. This means that, it is possible to say, among the people in Singapore who
was in a particular mall, or shop, or park at the same time. Furthermore, it is also easy to see who was
in a certain location a while after you left. The blue-tooth method captures
who is in close proximity, same time same place.
My question is, which one do you think captures the data
required to assess risk of covid better? Not just in terms of minimising false
positives or false negatives (this is important), but also in term of capturing
the behaviour of the virus.
If covid can only be transmitted by people right next to
you, then the blue-tooth method is better. It covers anyone near me who may be
at risk if I was infectious. However, if covid can linger in the air or on
surfaces after I have left, then the blue-tooth data is utterly inadequate. It
does not show who came after me once I was out of blue tooth range.
I understand that what we know of the virus has changed over
time, but the idea that close proximity long exposure contact is necessary for
infection has been very heavily challenged. You do not want to risk having
false negatives going about daily life normally and potentially spreading the
virus.
Now this has partially been addressed by enforcing QR
check-ins on top of blue-tooth. Stopping checking-in enforcement at shops (say
NTUC supermarket) was always strange to me.
2 People
need to be producing the data
Let’s say you have a nice method to collect relevant data,
if the data being created and collected? In my view, doing the ‘techy’ things
is the easy bit, just like running data through an algorithm is the easiest
part of ‘data science’. Getting the data that suits, and interpreting it are
critical. This is where, in my view, the Singapore government falls short.
2.1 Inadequate
Data
Minister Ong Ye Kun spelt it out, “But we strongly suspect,
and the police also strongly suspect, that the data we have using TraceTogether
and SafeEntry are not comprehensive,” he said, emphasising that there is a
limit to which they can use the data gathered through the contact tracing
platforms. “(1)
“There is a limit to which they can use the data gathered
through the contact tracing” may be interpreted in many ways:
A- Rules restrict how the data is used.
In this case this would not make sense. The data is being
used for the purpose it was contributed by every one who carries and uses the
app or token.
B- The data is not fit for purpose
As discussed, the data that the government has on hand is a
result of the process of collection. The decision was made to go blue-tooth and
risk false negatives and ignore the potential effects of transmission via
air/surfaces. This is ok if that was the conventional wisdom at the time.
However, deciding not to still enforce the checking-in at shops, in my view,
was a short-sighted move.
To this I would add the idea that there was some
over-confidence “The SNDGO said that even without check-out data, the Ministry
of Health (MOH) can estimate how long a person had been in various venues based
on SafeEntry check-ins at other locations, for example.” Where SNDGO means Smart Nation and Digital
Government Office (2) As someone who has spent some time in the field, I would
hesitate to implement some models when stakes are so high. You can use models
to estimate missing data, but it’s not a good idea to deliberately not capture
data when accuracy is important. Over-confidence is, in my view, a sign of an
organisation that is not mature.
It’s not that it is not possible to make such estimates, but
when stakes are high as in the case of covid, the costs of mistakes are large,
and when the costs are not large, it makes sense to gather actual data. For
example, if it was to calculate people’s steps, it may have been acceptable to
estimate, but really not when you are using the data to manage covid 19. And
this ties into the final aspect of what the minister said.
2.2 Insufficient Data
“We suspect and the police also strongly suspect, that the
data we have using TraceTogether and SafeEntry are not comprehensive”
This is very interesting because as early as March 2021, the
government used data to make trace together compulsory; stating that 90% of
residents had downloaded trace together voluntarily, the government made it
compulsory (3)(4).
As anyone who measures app acceptance and usage would tell
you, downloads and users are 2 very different measures. Unless the people who
did not download the app are precisely the people who are at risk of infection,
it would mean that people do not use the app, and the 90% number used was misleading.
Now, add to this the fact that checking-in and out was not
enforced, and you have easily identifiable potential cracks in the system.
The question is why would people bother to download the app
but not use it properly. I think there are 2 main reasons. And again, these
indicate an organisation that was still in fields of dream mode ‘If I built the
app, they will use’ (5) focused on the tech.
A- Trust
It takes trust to allow someone to track your movements; and
the people in Singapore willingly gave up their data to help contact tracing to
contain Covid19. However, the government eroded the trust. While when TraceTogether
was launched, people were assured the data would be used purely for contact
tracing. The government back-tracked 6 months later (6)(7).
Add to this the fact that the government recently publicised
the confidentiality of testing for people from the recent clusters. (8) This
may lead people to think that their tests are generally not confidential.
B- Privacy
Trust is not helped by the lack of privacy in cases of
Covid. Everyone in Singapore knows that the first person to go to a doctor in
the recent case is a person from Vietnam who entered Singapore via the familial
lane. Similarly, everyone also knows that one of the infected people was a
stall assistant in a particular stall at a particular hawker centre.
On one hand you have people sued under Official Secrets Act
(OSA) for leaking the number of covid cases (9) (interestingly it was the head
of the data unit who leaked the info), or when restrictions would hit (10), on
the other privacy of individuals is easily breached and publicised without
anyone taking umbrage– for example, a man on the street would not know how to
verify whether someone came into Singapore via familial lanes(11), or fellow
stall holders wouldn’t know who the infected fellow stall holder is (12).
Hasn’t the SingHealth issues with privacy of data of people in Singapore taught
us anything? Are we going to keep on making the same mistakes?
C- Education
I was having a discussion with a friend about the
TraceTogether app. We disagreed on how it should be used. My view is that it
should be on all the time, at least in public. On the other hand, she said it
was sufficient to turn it on, check-in to the buildings, and you turn it off
again. Basically, to me, using TraceTogether as you would safe-entry.
I tried to find evidence that I was right ( 😊
), but the website of TraceTogether, the FAQs, simply do not answer “how do I
use this app”. Educating users is critical if we want to capture the data
needed. Assuming that people will know, or there is no need to explain, again,
is the hallmark of a technology driven push, ignoring implementation, basically
not that high analytical maturity.
I know this friend for many years, and she is no mug. The
fact that we can disagree over something as fundamental as to how the app is to
be used likely means that there are many people who are unsure about how to use
the app. (I am sure that people who use the physical token were given proper
instructions upon collection, but when you download the app, there is no one
explaining the usage next to you)
In Sum
I believe that how the Singapore Government has dealt with
the case outlined above shows that the government’s tech arm is great at
building and making technological solutions available, however it lacks in all
the thinking surrounding technology from
- the right data is captured to address the
situation
- the users are fully aware of what data they are
sharing and how it will be used
- the users trust the system to protect their
privacy
I would add a fourth point, as I pointed out
earlier, technology/data is not magic, it is important to
o
know what can (should) be done and what cannot
(should not) (I am not talking about ethics here, but cold calculations and
decisions about false negatives, false positives, and weighing the impact)
o
address any issues, not only using technology,
but all the strategy around A Trust, B Privacy, and C Education that can
complement technology and enhance the quality and quantity of data.
Parting words
I believe all of us want to get out of restrictions and are
willing to sacrifice some privacy and go through some inconveniences to achieve
this. Like everywhere there will be some people who may not agree but the trick
is to ensure these people are a minority so the analytics can deliver results
we all want. I believe that the Singapore Government should spend more effort
on the whole solution, rather than focus on technology and treat the other
important pieces to the holistic solution as after thoughts. Only then would
the government become more mature in its use of data and become data driven,
for the sake of people, not technology.
1 https://www.straitstimes.com/singapore/health/confidential-covid-19-tests-for-those-who-visited-ktv-lounges-interacted-with
2 https://www.straitstimes.com/tech/tech-news/why-its-not-compulsory-to-do-a-safeentry-check-out
3 https://www.straitstimes.com/singapore/mandatory-use-of-tracetogether-token-or-app-for-checking-in-at-malls-workplaces-schools-to
4 https://www.straitstimes.com/singapore/politics/almost-90-per-cent-of-residents-on-tracetogether-programme
5 https://www.imdb.com/title/tt0097351/
6 https://www.straitstimes.com/singapore/vivian-balakrishnan-says-he-deeply-regrets-mistake-on-tracetogether-data-first-realised-it
7 https://www.technologyreview.com/2021/01/11/1016004/singapore-tracetogether-contact-tracing-police/
8 https://www.straitstimes.com/singapore/health/confidential-covid-19-tests-for-those-who-visited-ktv-lounges-interacted-with
9 https://www.channelnewsasia.com/news/singapore/former-moh-deputy-lead-osa-covid-19-case-numbers-leak-14615394
10 https://www.straitstimes.com/singapore/51-year-old-public-servant-to-be-charged-under-osa-with-releasing-information-about-phase
11 https://www.straitstimes.com/singapore/vietnamese-woman-who-is-first-case-of-ktv-cluster-came-here-in-feb-via-familial-ties-lane
12 https://www.straitstimes.com/singapore/health/toa-payoh-hawker-centre-closed-for-deep-cleaning-after-stall-assistant-who
