Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

> The large size by itself does not guarantee correctness

That's true for "reasonable" poll sizes. As your poll size increases past a certain point, it does begin to guarantee correctness.

An election is a poll with the size of all eligible voters.



Indeed.

"Official" election results were giving Tsikhanoyskaya around half a million votes while over 1 million Telegram users with Belarusian phone numbers already said that they voted for her.

Se here https://t.me/telegrambelarus/9


> As your poll size increases past a certain point, it does begin to guarantee correctness.

> An election is a poll with the size of all eligible voters.

I think you're oversimplifying the situation. Clearly, sample size alone doesn't have that much of a correctness guarantee, or according to your own statement, we'd be able to trust the official results.


> Clearly, sample size alone doesn't have that much of a correctness guarantee, or according to your own statement, we'd be able to trust the official results.

People are not complaining because the election is a biased sample of the population (not possible because by definition an election is open to all eligible voters).

People are complaining because they believe the government is not truthfully reporting the actual election results.

Good polling technique cannot mitigate fraud.


Yes, this is the point I am making. Sample size is but one of many factors that influence the reliability of a poll, and it is not the only consideration for good polling technique. A large sample does not mitigate those factors.

If there is an issue with the underlying polling technique, making the sample larger does not guarantee more correctness. You simply end up with a larger set of bad data.


Even if the sample size is the same size as the population?


1. Possibly, if they are different populations.

2. This sample in the Telegram poll is ~20% the number of voters in Belarus.


So what it it wasn't random, but captured 90% of the population?


Where are you seeing 90%?

The poll itself says 59% of the 2.3m respondents were people who voted in the election.

That's 1.4m people, which is 15% of their population and 21% of their eligible voting population.


Forget the percentage - there are more people who said they voted for the second-place contender than the second-place contender's official vote count.


Yes, clearly the evidence demonstrates the election results are less than trustworthy.

But that’s not what I was taking issue with. Above, I was disputing the claim that a large sample size “guarantees correctness”

Likely, neither this Telegram poll nor the election are statistically sound in their results, for different reasons.

I am not saying that the Telegram poll has to be statistically sound to be valuable evidence.


Yeah I think Telegram, like Whatsapp, is adopted broadly enough not to bias too heavily in one direction. Definitely not 90%, and especially not if that “bias” matches the word on the street.


Biases are irrelevant. If you look a the numbers, there are more people saying they voted for the opposition on the Telegram pool than votes on the official pool. About 5 times as many.


What about the people under 18 who I assume could take the poll and also people who didn't go to vote but took a few seconds to take the poll on Telegram?

Also what about all the babushkas who most likely voted for Lukashenko but don't have a smartphone?

This is not to say that more people didn't vote for the opposition than the official numbers state. But Lukashenko still could have won.


We'll say Belarus has 1m people that are old enough to both have a smart phone and be under the legal voting age (and that's being extremely gracious). Unless you're saying literally EVERY ONE OF THEM voted in this poll AND voted for the opposition, there are STILL more people of voting age in the telegram poll who voted for the opposition than "officially" voted for the opposition. The numbers are nearly impossible to believe unless Telegram is intentionally fudging the numbers.

https://en.wikipedia.org/wiki/Demographics_of_Belarus


This is certainly damning data, but having a telephone in Belarus does not necessarily mean that person voted in the prior election. This is good evidence, but not a mathematical proof.


The people can lie to Telegram, Telegram can lie, somebody can attack the communications, somebody can attack the telephones, somebody can impersonate the numbers, all the Telegram voters can be from those 60% that didn't vote...

There are many ways that could happen. But it's pretty good evidence to add to the context, and the pile of evidence was already quite big.


Exactly, this is impossible to reconcile. Either there has been vote fraud or this people didn't actually go to the poll station. Or people can vote more than once.


Telegram? 5% of the world uses Telegram. And I bet the 50+ age demographic is heavily underrepresented in that subset. Significant sample bias shows up in US polling using mediums that are exponentially more widespread.

Nonetheless, I don't doubt the validity of these particular results, because I think we have enough corroborating evidence. We don't have to justify the rigorousness of a Telegram poll to come to that same conclusion.


People in different parts of the world differ HUGELY in what instant-messaging communications platform they use.

For instance people in the US still use text-messaging to a significant percentage. That's hard for me to believe as well.

It's no use looking at Telegram's world usage, that's for sure.


Maybe Telegram is popular in Belarus, but there is one thing that is true all over the world: rates of technology literacy and access is lower for the old/poor/rural.


What we have here is a Sample, and in statistics (of which Polling is a discipline) you require a Randomized Sample of the Population before you can draw any meaningful conclusions.

Telegram users are not going to pass any "Randomness" scrutiny. For all we know, Telegram User A asks Telegram User B to take the poll, etc. That's not random, and can introduce all sorts of statistical bias.


What kind of a statistical bias would explain having 1 million Belariusian phone numbers claiming having voted for a candidate, that officially received around 0.5 million votes total?


Adulthood necessary to vote could be one?


You’re ignoring the finite population and the actual numbers involved.


It is simultaneously possible for the results to have statistical issues while also being good enough to provide utility as evidence for drawing some conclusions. You can both be right.


Indeed


What does that have to do with the fact that the number of telegram users that voted for the opposition candidate exceeds the official number of voters that voted for the opposition candidate?


Phone number is not an unique identifier. There are plenty of people who use multiple phone numbers.

While I don't believe in Belarus official results, that Telegram evidence is not a real evidence. It's just a hint.


I very highly doubt around double the amount of ”official” voters would have a second belarussian number.


Also, they have to decide to actually use that second number to cheat in a Telegram poll significantly more often than people of the other party.

Otherwise you can only draw your error bars equally in both ways at once, and then they need to be pretty large before the numbers stop saying what they clearly are saying.


Especially if your poll is showing extremely lopsided results.

If a poll shows 80% for candidate A, you’d only need to hit 62.5% of the population to guarantee that candidate A would hit 50% of the vote even if the remaining 47.5% voted for other candidates.


"An election is a poll with the size of all eligible voters."

Not really though.

An election is ostensibly 'perfect sample of the electorate' (assuming everyone voted), which is what makes it 'good'.

It's very easy to get a 'very large sample size' that is still 'very inaccurate'.

In this case, we're talking about potential numbers larger than literal voters, which makes it interesting - but the sample size again is not the issues if we're looking at a 'poll'.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: