For the first time since 2004, Team Singapore failed to win a single medal at the Olympics. Much attention understandably focussed on swimmer Joseph Schooling, who had won our first ever gold at Rio 2016, for the 100m butterfly, setting an Olympic record in the process.

Following Schooling’s failure to qualify for the 100m semi-final at Tokyo 2020, finishing last in his heat, internet commentaries seemed quite polarised. Critics focussed on his age and weight, and suggested that his deferment from mandatory national service should not be extended. Some of it was abusive, for instance a meme comparing him to a whale.

Schooling’s supporters responded with a mix of encouragement for him and scorn for the “armchair critics”. Many argued that it is unfair to penalise elite sportspeople—for instance, by asking him to serve his national service immediately—off the back of a poor Olympics result.

Was the internet and social media a source of support or criticism for Joseph Schooling (and Singapore’s other Olympians)? And was the commentary fair or abusive? Did Schooling’s defenders react in a militant fashion?

Let’s look at the data. Our team has performed a simple sentiment analysis using public social media comments. We first identified eight of the most important Schooling articles on Channel News Asia, Mothership, Straits Times, The New Paper and Today, including the two commentaries by Eugene Tan and Gerard Wong that particularly irked Schooling fans. 

We then extracted the public Facebook comments on the official pages, e.g. the Mothership article posted on Facebook by Mothership.

We ran this limited data set of 5,532 comments through a custom natural language processing (NLP) model trained to have an accuracy of 70%. The NLP model is trained to categorise comments as “Supportive”, “Critical” or “Neutral” based on keywords. The 70% accuracy means that the system on average makes mistakes three times out of ten—e.g. calling a supportive comment “critical”—but, importantly, the errors are not biased towards one side or the other. 

The articles were published over one week, beginning on the day of the event (July 29th-August 4th). We tracked comments over ten days (July 29th-August 7th), which allows us to observe the evolution of sentiment. (The Google Sheet is available here.)

Of the 5,532 comments about Joseph Schooling made on Facebook, here’s what we found (click on images to enlarge them):

• Just over 50% were supportive. Just under 30% were critical and about 20% were neutral

• The share of supportive comments declined slightly over the ten days, from 51% to 45% while the share of critical comments rose slightly, from 34% to 36%
• Supportive comments peaked (56%) on August 2nd, two days after Gerard Wong’s CNA commentary, “We need to talk about why Joseph Schooling crashed in Tokyo”
• By contrast critical comments peaked (38%) on August 4th, the day of Eugene Tan’s commentary, ”Tackling the critical question of Schooling’s NS deferment after his poor Tokyo Olympics performance”

• In terms of news sources, the highest share of supportive comments appeared on posts by CNA (54%) and Straits Times (56%). The highest share of critical appeared on posts by Mothership (45%) and TNP (42%)

• Median comment length (in characters) grew over time. Median supportive comment grew from 95 to 367 characters, while median critical comment grew from 65 to 211 characters. Neutral comment length was flattish (notwithstanding a spike on August 3rd due to a flurry of Bitcoin-related spam)

• The proportion of supportive comments with at least one reply—a sign of engagement—was 11%. The proportion of critical comments with at least one reply was 10%.

That is the data (available here). Let me now offer some analysis, interpretation and commentary.

Many Singaporeans may be too sensitive to criticism. 

One thing that surprised me over the period is the amount of pushback to the criticism of Schooling. Among the more common narratives from Schooling’s defenders, including many friends, were: armchair critics shouldn’t comment, i.e. unless you’ve competed at the highest level, your views are irrelevant; and we should use only positive language and enforcement on our athletes, e.g. don’t use the word “fail”.

Many of Schooling’s defenders were driven by noble motives: a desire for constructive criticism rather than abuse; the need to encourage rather than discourage sportspeople; and a concern for mental health, all the more prevalent now in the wake of Naomi Osaka’s actions and words.

But a bit too much fuss, I think. In Singapore the concept of criticism is so immature, largely owing to post-1965 political forces. Not only should we be encouraging more criticism, but we should be seeking to improve the tenor of the debate.

My great worry is that an oversensitivity to criticism, and attendant support for the notion that “only experts should comment”, will stymie our civic, democratic and economic growth. How, for instance, is Singapore supposed to foster a risk-taking and entrepreneurial culture if we can’t even use the word “fail”?

In my view, several of our athletes failed in Tokyo. That’s totally fine. We all fail. Keep going.

Critical opinion pieces from public figures can spark both supportive and critical responses.

One of Schooling’s defenders had, based on a cursory review of comments, flagged two commentaries as potential lightning rods for criticism: Gerard Wong’s July 30th commentary, “We need to talk about why Joseph Schooling crashed in Tokyo”; and Eugene Tan’s August 4th commentary, ”Tackling the critical question of Schooling’s NS deferment after his poor Tokyo Olympics performance”.

Yet while the latter did apparently precipitate a boost in the share of critical comments, the former appears to have had the opposite effect.

More study is needed, but one thought is that a more broad-based approach to the critique (“We need to talk about why…”) can engender more supportive comments, while a focus on an issue that triggers many (“NS deferment”) inevitably invites more criticism. (Which of course could be the writer’s objective.)

The nature of engagement differs depending on the news source, with the more mainstream ones (CNA and ST) having a relatively more supportive (of Schooling) readership, and the edgier, slightly more alternative ones (TNP and Mothership) having a relatively more critical readership.

This probably seems intuitive to media watchers in Singapore, so here is one more data point to support it.

Comments both supportive and critical of Schooling occasionally veered into abuse and toxicity, though we have had to rely on manual rather than automated scanning to assess them.

While some of Schooling’s critics abused him, some of his defenders abused others. Our team tested available toxicity models from the Google Conversation AI team. Unfortunately they mislabeled nearly all comments as toxic. 

We did not have the resources to train an accurate model for abusiveness in addition to its general sentiment. (I suspect that even the most worldly of computer language models might have trouble with “parah pundek”.)

We therefore manually scanned and searched for keywords among the comments. Consider that when reading the points on abuse below. (And please review the sheet yourself and inform me of additions.) 

While Schooling was the subject of some abuse…

Though I did not see any abuse of Schooling on my own news feed, others did. The whale meme was one that caught the attention of many. 

Yet in terms of the 5,532 comments, it appears that any abuse of Schooling is quite mild. The worst comments about Schooling—in my view, having scanned them—are as follows:

“Too fat. Eat too much during lockdown.”

“Too many Milo and Bar chor Mee…”

“he knew his time not as good as before (based on Sea games and his interview) yet choses [sic] to skip NS and continue to relax one corner…”

“I’ve always known him to be a failure getting a tattoo and setting all kinds of bad examples for young kids…”

…the more abusive comments appear to have come from Schooling’s defenders…

“…how quick these pricks forget…… some of these MATs cant even qualify for the TEH TARIK games in Batu Pahat but want to comment about his preparation etc .. parah pundek – probably cant even swim to save his foreskin  have some respect … PRICKS”

“…dont care about this fxxkers la…we keep on supporting him…”

“It is the netizens giving unnecessary criticism over the years that led to his current Olympic performance. If theres one thing I felt should be done, just shut down CNA, ST, BT, Mothership forum commentary regarding all Sports in Singapore…All these comments are just pure jealousy and eye-sore from lowly netizens…”

Here is a Chinese Schooling defender addressing a Malay Schooling critic:

“[Malay name] Hey Bodoh he already got a Gold Medal .What sports ? Who ever got a gold medal before ? Dont be so jealous lah [Malay name].Your profile picture is turning green already. Jangan embarrass yourself further. Check your facts before you mulut besar. HE PRODUCED SATU Gold already bru. Jangan merepek ok”

The same Chinese Schooling defender addresses another Malay Schooling critic:

“[Malay name] Hey Uncle, you jaga your store lah . Jangan mulut besar. You the one embarrasing and disgracing yourself. You macham action only ..you jealous of Joseph ah ,he young and rich and all the girls chasing him. Joker man you, you got mirror in your house or not .? You disgrace yourself in front of your grandchildren. Malu only.”

For the two comments above, I mention the ethnicities only because this Chinese Schooling defender, who left many comments, busts out his Malay only when slapping down Malays. Nothing wrong with using different languages but, in my view, some of this stuff is racially insensitive, if not outright racist, e.g. “jaga your store”.

…and thus Schooling’s defenders may have been more responsible for poisoning public discourse. 

In discourse, whether in person or online, whether with your spouse or a stranger, it is always interesting to think about the point from which things start to spiral. At what point does toxicity creep in?

To reiterate, our team was hoping that the toxicity model would offer a more rigorous answer to this. Sadly it didn’t. We did not have the resources to train an accurate model for abusiveness in addition to its general sentiment. 

Thus all I feel comfortable saying now is that based on our limited data set and my subjective viewpoint, Schooling’s defenders may have been more responsible for poisoning public discourse.

---

When experts shut out others

Any attempt by a group of elite insiders to shut down criticism by outsiders, by “non experts”, always triggers me. The history of the world is replete with examples of insiders shutting out outsiders to the detriment of knowledge.

As a geographer (my first academic love), I like the story of Alfred Wegener, a meteorologist and amateur geologist, whose theory of continental drift was poo-poohed by professional geologists, accepted only some 50 years after first proposed.

In Singapore there is a long history of politicians using a spurious line of thinking to discredit those whose views they dislike, such as Catherine Lim.

“Politics are for politicians,” Lee Kuan Yew said in the 1990s. “Anyone else who wishes to comment on Singapore politics should first get himself or herself elected.”

A recent incident concerns Trace Together, Singapore’s contact tracing app. In March 2020, the Singapore establishment, including ministers and techies, told Singaporeans that TraceTogether data would be used for the sole purpose of Covid-19 contact tracing.

In March-April 2020, when privacy advocates and others raised concerns about the possible abuse of TraceTogether data, we were at best laughed off and at worst painted out to be conspiracy theorists out to undermine the national effort against the pandemic. 

Not a single person within the Singapore establishment raised any concerns publicly about data privacy and possible abuses. Even alternative media sites were afraid to broach this topic (an editor told me at the time). 

Never mind that public intellectuals everywhere were concerned about autocratic leaders using Covid-19 as a cover to expand surveillance. Yuval Harari articulated many of these concerns in “The World after Coronavirus”.

Rather than engage in healthy discourse, in Singapore the “experts”—politicians, civil servants and techies—and their handmaidens in the media did not want anybody criticising them.

As we now know, in May 2020, even before the ink had dried, the first abuse of the data had begun. The police requested TraceTogether data from the government for a murder investigation. The government shared the data.*

In October 2020, Vivian Balakrishnan, minister-in-charge of Singapore’s Smart Nation initiative, was first alerted to his possible mistake (misrepresentation of data use) by an ordinary citizen, who mentioned the Criminal Procedure Code (CPC).

In January 2021, Vivian admitted his mistake in parliament, ten months after the false assurance, and a full three months after he first realised it. In February 2021, Desmond Tan, minister of state for home affairs, finally admitted that in May 2020 the police had started using TraceTogether data. 

According to this now official version of events, Singaporeans are supposed to believe that Vivian, a long-time minister, is unfamiliar with the CPC and its powers. We are also supposed to believe that from May to October 2020, the minister(s) who released the data to the police in May did not see it fit to inform Vivian, who led the TraceTogether initiative. 

OK lor.

So, beware when “experts” relentlessly shut down fair criticism from “non experts”.

That does not mean that we should entertain the outright rejection of expert authority that has been sweeping across parts of the West (and Asia) over the past ten years. All this is not a reason to indulge in wild conspiracy theories, for instance like some peddled by the anti-vaxxer community.

Experts will always have their place and voice; but they should never be above fair criticism. Indeed, it is only by allowing it that we can cultivate trust in their authority. 

* To be clear, lest critics are again smeared as hopeless idealistic softies, I am not against the Singapore police’s use of available data and evidence. What I simply want is honesty, transparency and accountability. Rare commodities these days.

Back to School-ing

It is against that backdrop that I was worried tremendously by the “If you haven’t performed in elite sports, don’t say anything about Schooling” narrative. (The people in Schooling’s inner circle must have been delighted to see his self-appointed guardians scrambling to shut down criticism.) 

There is nothing uniquely Singaporean about this—some of Simone Biles’s supporters also ridiculed critics who’ve never competed at that level.

But it is only by allowing criticism of Schooling (and other Olympians) that we can nurture strong feelings of pride and loyalty towards our national sportspeople.

Of course, sport is not exactly like academia or politics. (Or maybe it is? Fiery patriotism was certainly on display in some comments.)

One might argue that sport is just another form of “entertainment”; the same way a consumer might criticise Adam Sandler’s performance in Uncut Gems (love it; watch it), shouldn’t they also be allowed to criticise Schooling?

Another lens to use is participation and personal experience. Because many people run, swim and engage in other sports, they surely feel some natural competence (and thus right to comment) in those areas. I do wonder if there are elements here of the Dunning-Kruger effect, a cognitive bias which has the effect of, if you pardon my Singlish: you do a bit, you think you know a lot. (Read a more intelligent take here; I most recently heard about it on this great Freakonomics radio podcast.)

In terms of professional critics, such as the likes of Gerard Wong and Eugene Tan, there is a longer conversation to be had about the critic who is outside the ring. This person perhaps can, and this is true for many disciplines, bring an external, multi-disciplinary perspective that the one inside will never have.

Of course not all critics are equal, and of course not all fields are as easily accessible. I’d probably trust the outsider more if they were critiquing movies rather than crypto wallets. But that’s something to be sorted out by the marketplace of ideas; not pre-emptive diktats about who has the authority to criticise. 

Two new books that I’d love to read are about Roger Federer and Lionel Messi, written by Christopher Clarey and Simon Kuper, journalists who have, respectively, followed tennis and football for years.

Surely Singapore must encourage not only our own Federers and Messis, but also our own Clareys and Kupers?

My own views on Schooling

Most importantly, a big thanks to everybody in Team Schooling who made all this possible. Shout-out especially to his dad who is battling cancer. 

In terms of money, I think we can be both appreciative of his family’s sacrifices (financial and otherwise), and yet also clear that anybody who relies on taxpayer money must be accountable to the taxpayer. 

I don’t think there’s a contradiction; I don’t think his family’s early sacrifices mean that Schooling should then later get a free pass.

The problem is with defining “accountability” when it comes to sport. Should we ditch Schooling after one failure poor performance? Of course not.

In terms of Schooling’s performance and his National Service deferment, I quite like the thoughts of former sports writer Chua Siang Yee (aside from, as you might imagine, his little dig at non-experts near the top).

With thanks: the friend who helped with the data, who prefers the simple reference of “practising data scientist”

Top image: Straits Times

---

---

If you enjoy my work, please consider making a donation so that I can keep the content free for all. Even a dollar is great.

For those in Singapore using PayNow/PayLah!, you can send the money to UEN: 53406735B (“STV Works”) or simply scan the QR code below.

For everybody else, please click here for PayPal, credit card and bank transfer options.

---

Post-script

For another day…

A free for all. What is fair criticism? Where do we draw the line? Are all comments equal?

Private versus public sphere. Apparently much vitriol against Schooling may have been in private forums, e.g. WhatsApp groups. Should we as a society care about what is said in private? Is a Telegram channel with a thousand subscribers a “private” forum?

When sport, national service and patriotism make for a toxic cocktail. From a friend: “go into the socio-psychological significance of NS. in some contexts like discussions of feminism, NS becomes like a black hole it’s impossible to pull rational debate out of.”

Bob DeLonge, a friend, shared some of the deepest critiques of my position and approach to this. Best that I just reproduce here (with his permission):

I’m not really of the view that Schooling (or any of the Olympians) are “elites”. For me “elites” are our political overlords, and those who own “the means of production”…

On a methodological level, even though sentiment analysis and the quantitative method can identify very interesting findings such as “50% in favour of Schooling, 30% negative, and 20% neutral”, one problem with this method is that it tells us little about the meaning accorded to these sentiments. Yes; they are broadly categorized according to “positive, negative and neutral”, but there is also the idea that 30% negative, although it is less than majority, is still significant….I’m not saying Schooling cares about numbers and how many people are saying what, but the backlash, even though in the 30% range has been particularly vitriolic, and this is something which sentiment analysis cannot quite capture. The “weight” of the comments if you will. I’m not sure how to measure “vitriol” but I’m sure you get my drift. This is why, even though it’s 30%, I don’t think it’s necessarily accurate to say that the sentiments were overall tilted in Schooling’s favour, because the intensity of the negative comments was quite acute…

…I do also think that we are on a battlefield. And as such, I don’t think encouraging more criticism of everything and anything is necessary very useful nor healthy. It involves a judgment call – what is useful to criticise, what is not useful, what shld be criticized privately, and what shld be criticized publicly. One of my gripes about the “left” in Singaporean public discourse is that they very often criticise each other as much as they criticize the “conservative right”. From a philosophical standpoint, perhaps this is a good practice, in the vein of Diogenes. But politically, it is impotent because it weakens the Singaporean “left” by making its infighting and lack of solidarity very apparent….

My sense is that the unhealthy intertwining between party and state in Singapore has led to a case where a lot of anti- or non-PAP supporters equate criticism of the ruling party with criticism of every other facet of life in Singapore, including Schooling’s failures at the Olympics. In order to preserve the integrity of our other public institutions, including sports, a decoupling of party and state seems prudent, but admittedly, not altogether forthcoming.

NLP methodology

1. A random sample of 521 comments across all news sources was extracted as the training data. Comments were inspected manually and labelled as ‘Supportive’, ‘Critical’ or ‘Neutral’ depending on whether they were supportive, critical, or neutral towards Schooling after his loss. ‘Neutral’ includes direct responses to other commenters not related to Schooling, and spammish comments.

2. We split the comments in a roughly 75-25 ratio into train and test sets. Sentence embeddings from each comment in the train set were extracted using Google’s Universal Sentence Encoder and used to train a classifier model via Spark NLP’s multi-class ClassifierDL annotator.

3. The model was evaluated on the test set. Precision and recall for Supportive comments are 76% and 85% respectively while precision and recall for Critical comments are 70% and 67% respectively, for an overall weighted accuracy of 70%.

4. Other embeddings such as BERT were tested but did not improve on the performance. Hence the original model was retained and used to predict the sentiment labels of the rest of the comments.