Elon Musk 'violated' Twitter NDA over bot-check sample size

Last week Elon Musk hit pause on his Twitter acquisition over the platform's "less than 5 percent" bot figure. The Register asked the microblogging website how it made the estimate and was stonewalled, but in ensuing discussions over the weekend, Musk blurted out that the sample size was 100 accounts. One Musk fan asked how …

[Anonymous Coward]

Richard Holway, an analyst at TechMarketView, mused: "You may think this is just an entertaining sideshow. But real people are getting hurt.

Interesting how he differentiates the public from "real people" isn't it.

Maybe, just maybe, Musk has a point. Especially if Twitter were relying on that sample size to swat off regulation ?

[General Purpose]

> Interesting how he differentiates the public from "real people" isn't it.

He doesn't. Why do you say he does?

[OldSod]

Pulling back the curtain

If Musk, by virtue of his due diligence in his purchase, draws back the curtain a bit on Twitter, and the revelation of what is really going on behind the curtain causes "people to get hurt" - should the messenger be blamed, or the people who were less than transparent about what was going on?

[Jellied Eel]

Re: Pulling back the curtain

Also if Twitter has been lying about it's fake/bot numbers, then it's already been harming it's investors.

[Anonymous Coward]

Re: Pulling back the curtain

I think that this is the key. Musk is entitled to know what he discovered because he is buying the company. I assume he will attempt to lower his bid based on Twitter's shenanigans but I think he sincerely wants to buy Twitter for Meta reasons.

The big problem is why aren't the current shareholders entitled to know about how Twitter has lied to them? NDAs have become too ubiquitous and too broad. They need to be curbed or, better, kicked to the curb.

[Chinashaw]

Re: Pulling back the curtain

So your entire assumption is that Musk is right (everything he has said is correct) and the entirety of Twitters management statements are lies? Based on the tweets of a man who constantly lies/amends the truth? Interesting.

[julian.smith]

Re: Pulling back the curtain

"If Musk, by virtue of his due diligence in his purchase"

Musk declined to do due diligence

The rest of your post is at a similar level of ignorance!

Matt Levine from Bloomberg has a delightful, informed analysis of Musk's daily brain farts.

Check out his Money Matters - you've got a lot to learn

You should get out more ... a lot more

[TheFifth]

Interesting how he differentiates the public from "real people" isn't it

I didn't get that from what he said at all. My reading is he's saying that it's all fun and games for us to watch from the sidelines, but this isn't just rich men waving their willies around, it's having a real impact on real people (i.e. not billionaires, not rich investment companies).

It's hurting the rank and file employees too, who are actually members of the public themselves. Just because they work at Twitter doesn't make them outside of the 'public'. I bet they're not earning massive amounts of cash either and it's not like they have any control over any of this.

You could of course argue that billionaires are 'real people' too, which I guess they are in the strictest sense of the word, but certainly not in a social and economic sense, which is what I think he, as a financial analyst, was likely getting at.

[Adrian 4]

violate ?

There's something wrong with a system that won't permit telling the truth in case it affects the market.

[Alumoi]

Re: violate ?

It's called economy.

[Anonymous Coward]

Re: violate ?

Your DVs seem a little harsh. I'm sure you are referring to being economical with the truth and nothing to do with money driving bollocks.

[Doctor Syntax]

Re: violate ?

The market is affected as much by who says it as by what was said. If he says something that gets interpreted as an attempt to weasel out of his bid or reduce the price it makes no difference whether it was true or not. Then let's consider your assumption that it was, in fact, true...

[Danny 2]

Re: violate ?

You can libel/slander someone by telling the truth in UK law.What about Musk breaking a legal agreement in order to profit from a share slide?

As a bum who can't afford shares I am far more concerned that a Tory MP revealed that you can make a meal for 30p without supporting recipes. I suppose if the meal is thirty peas, but otherwise restaurants should sue him for exposing their profit making. Even the Slavenation Army charges rough sleepers far, far more than that.

If you buy a used car from Musk then check it hasn't already been launched into outer space.

[keithpeter]

Re: violate ?

"... I am far more concerned that a Tory MP revealed that you can make a meal for 30p without supporting recipes."

That kind of bovine excrement was commented upon by George Orwell back in the 1930s in The Road to Wigan Pier I think.

https://orwellsociety.com/food-for-thought-i/

And remember it was our present government who cut local authority funding for Sure Start Centres for families - centres that provided support for very low income families with young children. Part of that support being information (actionable information, actual practical demonstrations) about nutrition and recipes. Just what that jerk was on about. Actually being done. Happening. In reality. Years ago.

Don't get me started on the short-sighted and self-defeating nature of these savings. I need to watch the blood-pressure these days.

[Danny 2]

Re: violate ?

Great link keithpeter,

When I was 15 I was confident I could live off potatoes and spam, and was certain I'd be down and out in Paris and Livingston. I was, eventually. Took me twenty years of career success before I became a bum. While I still would enjoy potatoes and spam I would not recommend it, albeit better than bread and tea. Listening to whores fucking in the Gare Du Nord is preferential to listening to friends fighting.

It's annoyingly amazing how often Orwell is partially quoted by far-right Americans to justify gun crime.

Your article lists aspirin and I used to use aspirin to 'prove' even the poor today are better off than medieval royalty. Except aspirin is just willow bark, and medieval royalty had that.

There is a line in a 1980s song I love, Hip Rebels by The Passage, "as bad as the thirties".

I often have said this is as bad as the thirties, but that's hyperbole. So far, ask me next winter."

[LybsterRoy]

Re: violate ?

or maybe just channeling his spirit?

[TheFifth]

Re: violate ?

Many years ago I did some work for a Sure Start centre in a deprived area of the South West. The work they did was amazing, especially in the area of advice and support for struggling parents. Seeing the day to day work close up was a real eye opener for how Government funded schemes like Sure Start can make a real difference to people's lives. The guy who ran it was an absolute inspiration. The removal of their funding was a travesty.

Last I heard, their old premises was earmarked for housing development.

[LybsterRoy]

Re: violate ?

Are you related to amanfrommars?

[Chinashaw]

Re: violate ?

Given he was a Labour councillor and a miner and has worked with the foodbank in question before becoming a Tory MP there is a lot more nuance to his comments than you seem to understand. He is also offering to let MP's and I imagine similarly too quick to judge people come to the food bank in his constituency to see the work they do and how he came up with that figure.

It might also be worth looking into the concerns that if people do not (across all levels of society) cook regularly but do spend money on eating out then they will end up in general with higher living costs and possibly obesity issues. Unfortunately this is very true in poorer area's for any number of reasons.

https://www.gov.uk/government/publications/health-matters-obesity-and-the-food-environment/health-matters-obesity-and-the-food-environment--2

Interesting and easy to read article from Public health England.

[Danny 2]

Re: violate ?

@Chinashaw

"I imagine similarly too quick to judge people come to the food bank in his constituency to see the work they do and how he came up with that figure."

Imagine there's no foodbanks, it's easy if you try. In fact I can clearly recall when there weren't any. This is new.

It's not just unemployed and disabled, police and nurses are using foodbanks. I personally never have and (hopefully) never will but I know people who do, and the inevitable rule of supply and demand has kicked in. There are fewer donations now and more hungry people so the shelves go bare and rationing is necessary. That whole safety net is overloaded. School meals are being downsized - never mind obese adults, kids need to eat to study.

Brain dead and tone deaf Tory MP Lee Anderson claimed 30p meals were possible to any hungry Briton, and come to his constituency to see how. I might if I could afford the fare, but I know what I will see. He based his costing on 1,300 meals lacking in nutrition. So where am I supposed to store 1,300 meals if I can't afford the electricity for a freezer or a fridge? And he didn't price in the cost of energy for cooking it at at time people are turning down potatoes at food banks because they can't afford to cook them. And do I really want to eat the same bleugh for four years.

"do spend money on eating out"

Aye, maybe, stupid folk, or people with two or three jobs who are time poor or can't afford a cooker. And yet Lee Anderson wonders where the workers have gone.

I read an article that people are going to McDonalds for the warmth. They buy their kids a 'happy meal', don't eat themselves but stay there all night for the free wifi.

Personally I negated my food costs for twenty years by growing my own and by taking waste food of supermarket skips. Cooked on a wood stove visited my parents at bleak times. Partly for environmental reasons, mostly so I could afford whisky and cigarettes. Skips are empty now, so many people are reliant on them, and I can't work an allotment due to "muscle loss due to malnutrition".

The milk of human kindness has increased in price by 65% this year.

[Mostly Irrelevant]

Well, if Musk violated the NDA then I guess the takeover is void /s.

[Marty McFly]

Scared lawyers...

Sounds like the Twitter NDA enforcement legal team is a bit worried about their relevance in a post Musk world. I imagine the call went something like this...

Legal droid: "Mr. Musk, your recent statement is in violation of your NDA agreement."

Musk: "Ooops, thank you for catching that. It is valuable employees like you that made Twitter what it is today. What is your name and who do you report to?"

Legal droid, slowly realizing the gravity of their error: -Click-

[Doctor Syntax]

"Any sensible random sampling process is fine."

Without knowing how the 100 were selected it's impossible to say whether it was a sensible sampling process.

[Alumoi]

With over 350 million users, a sample of 100 is meaningless. I know, I know, lies, damned lies, and statistics...

[sebacoustic]

no, rubbish.

The ensemble size doesn't matter. If you choose 100 _truly random_ users, and "turing-test" them (this test being 100% accurate for the sake of the argument), finding 5 robots, your result is expected to be accurate to ~sqrt(5), no matter how many million users twitter has.

"Lies damn lies..." refers to _improper use_ of statistical methods not the statistical science itself.

[Primus Secundus Tertius]

@sebacoustic

You are right that the standard deviation is ~sqrt(5) or 2.2 approx. But this means any one sample may show between 1 and 9 robots, i.e. a divergence of up to 2 standard deviations. A sample size of about 1,000 is needed for a sensible result.

[sebacoustic]

very good incentive to freshen up my stats knowledge, which is essential in this world of numerical BS.

Thanks

[keithpeter]

Thanks for posting that.

I have the same argument about pass rates in GCSE Maths every August.

Yes, the national pass rate is around 68%. No, that does not mean that in one class of 30 students out of 20 classes we get worried and point fingers at the teacher if less than 20 pass.

Sigh

[Anonymous Coward]

*fewer*

[Doctor Syntax]

Actually it's skewed. With a larger population that wouldn't show but 5 minus 3 SDs is less than zero but 5 plus 3 SDs is feasible. The ancient rule of thumb remembered from my student days is that you should look for a quantity of at least 30 so I'd want a sample of about 600 to give 30 or so bots.

But then there's the question of how the 100 (or 600) are chosen. If they're weighted towards the early days bot accounts might be rarer than if they're weighted to more recent times. Ideally the age and geographic distribution should at least resemble that of accounts as a whole. If we don't know something about the sampling process the results don't mean too much but maybe that's too much of a complication for Wall Street.

[Anonymous Coward]

Clearly bots are like weeds so a Poisson distribution is appropriate. Make up a say 10Km square out of a suitable material and throw it somewhere. Count the number of twitterers and bots in the square and repeat a few times. Finally pluck a number that you like out of the air.

Job done.

[Adrian 4]

I'm pretty crap at stats but doesn't that assume a gaussian distribution ? And how would you determine whether that existed by taking such a small sample ?

[cornetman]

Your are only partly right. If the sample is truly random then *if* it reflects the population then it should be accurate. However, the sample size affects your confidence that it does and in a way that we can calculate. If the sample size is very small then the confidence is also small. The likely variance makes the practical usefulness of the result meaningless. I think that this is what the original commentator meant. It is a little like the difference between accuracy and precision, albeit a poor analogy.

The way you choose the sample is important, sure, but the sample size is also extremely significant.

[Horst U Rodeinon]

"...ilke the difference between accuracy and precision..."

I think you mean accuracy and resolution, which can be quite different from precision which I equate with accuracy.

[Anonymous Coward]

Precision is essentially the number of digits used to report something. It has no relevance to how that thing was measured.

Accuracy is how close the measurement is to reality.

Resolution is how small the increments can be (what difference can be resolved).

So if you were to take a picture of something from a long distance, then measure that thing in the picture, the _resolution_ would be the pixel width - say 5cm. You could then report the size as 25.98632987469817cm which is incredibly _precise_ but also incredibly unlikely to be _accurate_.

[cornetman]

Well I did say that the analogy was poor.

Precision gives the illusion of accuracy which is what I was trying for. Merely asserting that the sample is "very random" sounds superficially convincing in the same that high precision looks superficially like high accuracy.

I think that you are over-thinking it.

[MachDiamond]

"The way you choose the sample is important, sure, but the sample size is also extremely significant."

It's much more than the sample size. It's also the methodology that's used to determine if an account is a bot or a human. I wouldn't be surprised if a big bot farm has people to respond to inquiries. Most companies will let you place a limit on the types of email you want to receive. If the person setting up the bot account only leaves the mandatory box for official notices, that should cut down on all of the marketing carp so any mail from Twitter is far more likely to need a look. Most of the spam I get through my web sites comes from (or says it comes from) Google and Mail.ru, those are still the best sources of unlimited numbers of free email accounts to use to secure a new Twitter account. Route all of those addresses to a couple of slaves that will respond to official mail and you're set (and not a bot by their measure).

[jgard]

'no rubbish'.... Really?

I don't understand the upvotes on this. I am afraid it is you who is talking rubbish if you claim sample size doesn't matter. That's categorically wrong, and as you say: '_improper use_ of statistical methods'.

Think about this: if sample size is irrelevant, why not just sample 1 account? Because it will provide absurdly unreliable data; given a true 5% incidence, 95% of the time you would find there are no fake accounts, 5% of the time you find they're all fake. Take 2 accounts, or 5 even and it doesn't get much better.

Sample size is crucial when you can't test a whole population, as sample size grows so does confidence in the result. It's one of the most basic principles in probability.

[Piro]

"Truly random" doesn't even remotely remove the possiblity, however ridiculously small, of selecting only bots or only real people, with such a small sample size (100). In any case, there's a non-null chance of that sample being totally unrepresentative.

[MachDiamond]

"The ensemble size doesn't matter. If you choose 100 _truly random_ users, and "turing-test" them (this test being 100% accurate for the sake of the argument), finding 5 robots, your result is expected to be accurate to ~sqrt(5), no matter how many million users twitter has."

I expect they will have some criteria for their selections or they could wind up with 100 zombie accounts.

Elon should be respecting the NDA he agreed to and signed. There are good reasons why Twitter will not want to divulge how they vet accounts since people running bots will find that information very useful. Once you've violated confidences and publicly backtracked on promises you have made, don't be surprised when the next person is far less trusting.

[Anonymous Coward]

Eh, assuming the sampling is robust it lands you on (in laymans' terms) +/-10%, which isn't great but is fine for the kind of orders-of-magnitude estimation that this figure is being used for.

Elon, of course, as an apparently well-educated technologist should know this off the top of his head. But he's not picking this fight because he has issue with the statistics. He's picking the fight because he bought Twitter for $40Bn the week before the tech stock market collapsed.

[jollyboyspecial]

What natters here is whether the test is repeatable. But Twitter don't seem to be answering that question. What they have admitted is that their data could be wrong and the real figure could be over 5%. In their position that seems to be a particularly silly thing to do. They can argue that Musk violated the NDA, but he can argue that they presented him with data that they have admited may be innacurate.

[Anonymous Coward]

The point is that as a prospective buyer he has the means and the mechanism to do that in private. That is the entire point of due diligence. Doing so through the proper channels would give him the means to access otherwise privileged information to come to his own estimates and validate any data provided by the company being acquired.

But, again, he's not doing this because he cares about the figures.

[Patched Out]

It's completely repeatable! Using the same random seed to select the 100 account samples, they always get the same answer!

[MachDiamond]

"but he can argue that they presented him with data that they have admited may be innacurate."

Did they admit the data may be inaccurate?

They shouldn't have handed him any more information than they would give to any stockholder. Twitter wasn't advertising itself for sale nor a wish to take the company private. It's a hostile takeover and as such, the buyers often have to rely on public information for their due diligence. Elon would have been in a better position if he took a post on the board where he would have been made privy to the "real" numbers and plotted a takeover at that point, or not.

[Anonymous Coward]

Statistics

I don't think that this is an order of magnitude estimate (i.e. estimating a value to within a factor of ten). If you mean to within +/-10% of 5%, being 4.5% - 5.5%, then that is also wrong.

As a numerical example, if we were to make the assumption that the true proportion of robot accounts on Twitter was actually one-in-ten, and not one-in-twenty, then the probability of a random sample of 100 accounts containing five or fewer bot accounts is actually more than five percent (the cumulative distribution function for p=0.1, k=5, n=100 when we take the number of robot accounts in the sample to be a binomially distributed random variable, which is a valid assumption, even though we are not replacing, because 100 is much less than 300 million).

As another commenter mentioned, it is also entirely possible to draw many different random samples and simply choose the most favourable and only publish that. However, a sample size of 100 for proving a prevalence in the region of five percent is not nearly large enough, even if only one random sample was taken.

I am not a fan of Musk, but his motivations for going public with his ridicule of Twitter's methods aside, if Twitter's methodology can be accurately summarized as simply sampling 100 accounts, then it is a fair criticism.

[jollyboyspecial]

Your data is only valid if you publish it to the people concerned. In other words if I were in Musk's (no doubt horribly expensive) shoes I would have wanted to see the sample and know exactly how it was selected. Normally you wouldn't need to see the whole sample, maybe just a sample of the sample, but when the sample size is so ridiculously small you may as well see all of it. And then of course I'd ask what tests were used to check whether the account was bot controlled.

[LybsterRoy]

"And then of course I'd ask what tests were used to check whether the account was bot controlled."

I was wondering how long it would be before someone asked that question. For the moment forget sample size just tell me how you know with 100% accuracy if an account is real or not.

[Anonymous Coward]

Sample size

The UK pop charts used to be independently compiled from over 1,800 shops in the sample each week. With a sample size that large, obvious fiddling (buying 20 copies of a record in one shop then 3 minutes later buying 20 more from the shop at the other end of the High Street) was spotted and that data disregarded.

It was then taken "in-house" by the record industry - sample size 3 shops, each in a different city!

[wolfetone]

We're only one tweet away from Karen Musk demanding to speak to Twitter's manager.

[sebacoustic]

1)Twitter has already announced job losses and rescinded recent job offers.

2) Recent buyers of Twitter shares might well have taken Musk at his word. Clearly that wasn't that wise as they are now nursing big losses.

Group 2 gets relatively little pity from here...

[jollyboyspecial]

Musk has a habit of saying, doing and tweeting things that influence stock prices. Whether or not he does it with that intent is open to debate. But you can't debate that he is somewhat impetuous. His intent to buy twitter may be just another example of his impetuosity.

[lglethal]

Wait a second...

After the offer had been approved by the board, shouldn't trading in twitter shares have stopped? So how does their share price drop?

If trading hadn't been stopped, quite frankly they were asking for it with Elon. This deal was never going to be smooth...

[Doctor Syntax]

Re: Wait a second...

It's up to the shareholders to accept it. The board may recommend acceptance but the share's he's offering to buy belong to the shareholders, not the board. There may also be regulators taking an interest. It's not over 'til it''s over.

[ComputerSays_noAbsolutelyNo]

What's the endgame?

So Musk offered to Twitter for roughly 54 USD per share, his "interactions" with Twitter, the world and the universe send the share price plummeting. What sense does this make?

Discounting him having shorted Twitter shares, which would be very serious money crime, what's his motivation?

The deal does not get cheaper for him, if the Twitter share loses value, since he's offered a fixed price.

The deal also isn't going to go away, since any pre-this-whole-story owner of Twitter shares is even more likely to see the deal through, because of the fixed price offered by Musk.

So, what could be his strategy?

[Anonymous Coward]

Re: What's the endgame?

If the deal falls apart, say, due to his now-complex network of co-financiers getting cold feet, then the deal will collapse. It's not guaranteed that it goes through. In this scenario he is due to pay Twitter $1Bn in cash, but that is far less of a penalty than whatever he's looking at from in the collapse of stocks, or even from the loss of value in Tesla because of the way he's using those shares to underwrite his Twitter buyout.

[SImon Hobson]

Re: What's the endgame?

But if the deal collapses because it turns out Twitter was touting false numbers ?

[Anonymous Coward]

Re: What's the endgame?

As a publicly traded company that would be a serious allegation against the company and its officers, potentially implying criminal fraud. Nobody - not even Elon - is suggesting the figures are false. He's suggesting the figures are insufficiently rigorous.

[Flocke Kroes]

Re: What's the endgame?

Musk's legal team must be dusting off their old defence:

The social network is ‘famous for invective and hyperbole’. His decision to accuse Twitter directors of fraud was a ‘gratuitous barb’ rather than an actual allegation in which ‘factual information is conveyed’.

[MachDiamond]

Re: What's the endgame?

"Nobody - not even Elon - is suggesting the figures are false."

That's what you are taking away from Elon's statements? My impression is inverted by 180°.

[DS999]

Re: What's the endgame?

Twitter using a sample size of 100 might mean they are touting meaningless numbers, but the only way they can be false is if they sampled 100 and found 10 bots but lied and said there were only 5.

[Doctor Syntax]

Re: What's the endgame?

As several commentards have already said a figure like that needs a confidence limit unless the sample size was so large as to render the limit negligible. If that were s sample of 100 of which 5 were rated as bots that's not negligible. if it were 100 bots counted in a sample of 20,000 it's looking better and maybe good enough for the purpose in hand.

[DS999]

Re: What's the endgame?

Musk is clearly claiming the sample size Twitter used was 100, so 5 bots found in that sample. If what they really did was find 100 bots in a sample of 20,000 then Musk was outright lying. Wouldn't be the first time for that though I suppose.

[iron]

Re: What's the endgame?

There is a theory that the whole Twitter deal was just a way for him to sell large quantities of Tesla shares without harming their value. Now that he has sold said Tesla shares "to purchase Twitter" he is backing out of the deal.

[Flocke Kroes]

Re: What's the endgame?

There wasn't a sane plan. Dunning-Kruger effect got him into this mess. His brain is beginning to catch up with his mouth but that will be a long and difficult journey with lots of detours. Musk is far better at learning from mistakes with rockets and cars than with social skills.

[DS999]

Re: What's the endgame?

So, what could be his strategy?

You are discounting the possibility he was extremely high when he decided to buy Twitter.

The best suggestion I've heard is that it has to do with the cryptocurrency freefall. He wasn't funding this deal by himself, he had some others and one of them was apparently some crypto mogul. Maybe that guy can no longer come up with the promised funds so either Musk has to provide more himself (which means selling more Tesla shares further depressing the price, or borrowing against them with interest rates on the rise) or he has to find some way to either back out of the deal or get Twitter to lower the price.

No matter what numbers Musk comes up with for his estimate of number of bots he won't be able to use that to back out the deal. Twitter's sample size of 100 is a joke, but Musk using the same is also. It will be impossible for him to establish any fraud on the part of Twitter's board or management regardless of what the "true" percentage of bots turns out to be because while such a small sample size is terrible statistics it wouldn't meet the threshold of proving deliberate fraud.

Twitter's board is in a tight spot if Musk says he wants a lower price or he'll back out of the deal. Yes, they can sue him for backing out of the deal (the $1 billion is only if the deal is stopped by external forces like regulators) but whether they'd be able to collect and how many years that would take is a gamble. In the meantime they'd have to contend with multiple shareholder suits from people who think taking a lower offer is better than walking away. But if they take a lower offer they will be sued by shareholders who think that's a bad idea - and those who bought shares after Musk made his offer would have a particularly strong case.

[Anonymous Coward]

So what's the percentage of MDAU over total accounts (which is the number often reported)?

[jollyboyspecial]

A sample size of 100? Really?

How many accounts do twitter claim are active? A third of a billion last I heard. I don't think that can be considered a statistically significant sample size. That would be like picking 100 people in the US and trying to extrapolate statistical information about the whole population of the country.

Of course Twitter themselves have admitted that their estimate of below 5% may not be accurate. In other words there may be more than 5% of accounts controlled by bots. Furthermore they have said they need to improve their method of estimation. Like what? Starting with a decent sample size?

The fact that they choose such a ridiculously small sample size also calls into question what method they use for working out whether an account is a bot or not.

[DS999]

I'm assuming the method they chose was to actually have a human review all the tweets. I doubt any automated method could reliably tell a decently written bot from a human, regardless of what Musk thinks (he also thinks we'll have driverless cars five years ago and he'll have a working brain interface real soon now, so his ability to evaluate blue sky technology has proven to be pretty poor)

[MachDiamond]

" (he also thinks we'll have driverless cars five years ago and he'll have a working brain interface real soon now,"

The self-driving thing has been promised for something like 7 years now. Of course, he's very confident that it will be by the end of this year.

[Doctor Syntax]

"what method they use for working out whether an account is a bot or not."

Suspend the accounts and see who or what complains?

[Dinanziame]

A sample size of 100? Really?

You're making a beginner's mistake: You're assuming Elon Musk isn't lying through his teeth.

[Howard Sway]

"It's all getting a bit beyond a joke now…"

I don't know, watching the supposed genius lose at least a billion dollars whether he backs out of the deal or overpays for the shares is going to be funny for quite some time.

[Someone Else]

From the article:

Richard Holway, an analyst at TechMarketView, mused: "You may think this is just an entertaining sideshow. But real people are getting hurt. Twitter has already announced job losses and rescinded recent job offers. Recent buyers of Twitter shares might well have taken Musk at his word. Clearly that wasn't that wise as they are now nursing big losses.

I keep recalling something about a sucker, and every minute. I'm sure Elon is counting (possibly literally) on that.

And then there was this part, also from the article:

"I always thought there were trading regulations around making price sensitive announcements. Every tweet that Musk makes moves markets – see his comments on taking cryptocurrency to buy Tesla cars etc.

Silly Richard! Shirley, you should know by now that such inconveniences as "trading regulations" don't apply to fatasses like Elon Musk.

[jgard]

Random Sample? Hmmm....

Given their less than rigorous approach to statistical methods, and the obvious incentive to 'calculate' a low bot count, it wouldn't surprise me if they ran the test repeatedly with 100 random accounts until they got the result they wanted. Just sample 100 random accounts in a loop and repeat until a sample with only 5 is found. Strictly they aren't lying as that sample was of 100 random accounts - they don't have to let on that the overall process was extremely non-random.

[herman]

CEO vs Legal

It seems that the CEO and his legal beagles have a different opinion of how the calculation works.

[Anonymous Coward]

Read the article please, the actual sample size is NOT 100 of Musk's followers.

[DenTheMan]

NDA, where have I heard that before?

Oh yes,

Getting your Tesla repaired!

[Jan K.]

"You may think this is just an entertaining sideshow. But real people are getting hurt. Twitter has already announced job losses and rescinded recent job offers. "

Probably part of the plan... those workers are way too expensive and works way too little anyway.

The most dedicated workers are of course... chinese! https://www.theguardian.com/technology/2022/may/12/elon-musk-praises-chinese-workers-for-extreme-work-culture

[Anonymous Coward]

Does he think 5% is too high or too low?

Have we already established how high a fake/bot percentage Elon Musk is actually hoping for?

It's clear the banks do not have much faith in Musk's twitter foray considering the collateral they have demanded to secure the loan is insanely high. They are taking him to the cleaners unless he can make Twitter more profitable than it currently is. A way to make it more profitable _could_ be by eradicating most fake/bot accounts. If it's indeed only 5%, as Twitter has been saying for a while now, that could be too low to make a serious dent in profitability.

Perhaps Musk is hoping it's more in the 25% range. Any ideas?

[Velodrome]

Bound and gagged

Elon Musk has signed a binding contract to buy Twitter. The contract gives Twitter the right to force him to close at the agreed upon price as long as his financing is available. There is no reason to think the financing won’t be available. He's simply trying to strong arm Twitter into taking a lower price (which is smart since tech stocks have gotten hammered in the three weeks since the offer).

Twitter's financial filings “that false or spam accounts represented less than 5%” of active users for at least eight years, and it has always qualified those statements by saying that it is an estimate and “may not accurately represent the actual number of such accounts.” Musk signed a merger agreement without doing any due diligence on that number, and nothing has changed since he signed the agreement.

To get out of the deal, Musk would need to prove to the satisfaction of the Delaware courts that the bot number has a "materially adverse effect" on Twitter's financials. An MAE is a finding that would “substantially threaten the overall earnings potential of the target in a durationally-significant manner,” the courts have said. Typically, an MAE requires a 40% decrease in long-term profitability. If 6% (or 16%) of Twitter accounts are bots, that will cause consternation, upset the SEC and possibly reduce Twitter’s future advertising revenue, but will it be ruled an MAE in Delaware? No.

The law is on Twitter's side to force him to close at the agreed upon price. If he refuses to close, and if the Twitter board has any backbone, they will ban Musk from Twitter and take him to court. That would be pretty smart by the Twitter board. Musk will lose his bully pulpit, a few billion dollars and find out he isn't above the law.

[Sirius Lee]

Re: Bound and gagged

I think you are mis-understanding the law. Any buyer has a right to continue due diligence on an acquisition. Remember the disastrous purchase of Autonomy by HP?

If Musk has a point about the credibility of the numbers provided in the prospectus then the Twitter board has a lot more than Elon Musk to worry about. If the numbers in the prospectus are wrong, then these are numbers that have been reported incorrectly to the SEC in their 10K and 10Q disclosures in which case the SEC will have to investigate. If the SEC finds material problems, there are going to be serious questions for the Twitter CEO and CFO to answer. Failure to answer the questions appropriately can lead to prison time.

What was the methodology for arriving at the claim Twitter makes? At the very least, using a sample size of 100 is not likely to be judged adequate. Any statistician the SEC cares to consult with (and they have a lot of their own) will likely agree that the sample size is too small. If the 100 sample accounts were selected randomly then may be it can have some validity. However, any reasonable person has to ask why only 100? With millions of accounts available, why not make the sample size bigger? And exactly how were the accounts selected?

Also, why report it as "< 5%"? If the 100 account sample is random then the only data Twitter has is the sample result which yields an exact, integer value. The only way the result can be less that 5% is if there were 0, 1, 2, 3 or 4 bot accounts identified in the sample. Why not report this number because its the only real data available? Any other number is PR spin.

[Velodrome]

Re: Bound and gagged

PR spin or not, Sections 4.6 and 4.7 of the contract (https://www.sec.gov/Archives/edgar/data/1418091/000119312522120461/d310843dex21.htm) indicate agreement to the numbers presented unless they have a Materially Adverse Effect on the post merger entity. That is a very high bar in the Delaware Court of Chancery. To the best of my knowledge, they have found an MAE to justify a merger termination exactly once.

Wasn't part of Musk's stated argument to take over Twitter that he would fix the bot problem? That might show that he understood the scope of the bot issue before agreeing to the terms. His rush through through the process (https://www.bloomberg.com/news/articles/2022-04-26/twitter-takeover-was-brash-and-fast-with-musk-calling-the-shots?cmpid=BBD051322_MONEYSTUFF) might come back to haunt him in court.

Nobody knows whether Twitter would prevail in a court action. These things are never certain, but the contract is clearly in Twitter's favor. The numbers were presented and accepted. Not only would it be necessary to carry the argument on statistical validity but also on a significant and ongoing financial impairment that is material. Again, a high bar.

[Winkypop]

A bull in a China shop with a greased floor

Pass the popcorn!

[man_iii]

Musk mush

Elongate is definitey going to happen but twatter legal and CEO sound like spineless yes-men doing someone elses bidding

[Clive Galway]

mDAUs

"that are sampled at random, consistently over time, from accounts we count as mDAUs [monetizable daily active users]"

So if the process they use to define an account as an mDAU inherently means that pretty much no bots get selected, then it massively skews that figure.

After all, a bot is probably not monetizable?

[MachDiamond]

Re: mDAUs

"So if the process they use to define an account as an mDAU inherently means that pretty much no bots get selected, then it massively skews that figure."

That's a lot of cheating to try and get away with for the long term.

A good metric would be an account that is logged into a certain number of times per week that follows a certain number of other accounts and the user posts a certain number of posts. If you want to look for bots, accounts that meet the above criteria but never respond in a meaningful way to another user's posts might be a redneck, err, bot.