How seriously do you take statistics?

[sweep_shot]

I personally do not pay much attention to most statistics. I think the methods are very faulty and results are often inaccurate due to biases and insufficient sample sizes.

Statistics was showing Hillary Clinton would win by a good margin yet Trump won in 2016.

There are more examples.

How seriously do you take statistics?

 

[Buffet]

I personally do not pay much attention to most statistics. I think the methods are very faulty and results are often inaccurate due to biases and insufficient sample sizes.

Statistics was showing Hillary Clinton would win by a good margin yet Trump won in 2016.

There are more examples.

How seriously do you take statistics?

If you don't take stats seriously then you are contesting a fight with one leg tied.

The base rate, Gaussian, range of outcome, etc are extremely useful mental models. By not taking it seriously you are putting yourself at a huge disadvantage.

Now no stats were showing that Hillary will win with 100% probability. She had a higher probability, but that simply means that if you run a simulation then she would have won 60, 70 or 80 times of out of 100. Still, you had 40, 30 or 20 percent chance of her losing.

 

[Napa]

Humans are constantly required to make decisions without complete information. To make better decisions they gather data. Statistics is the tool they use to make inferences from data.

Suppose we stopped using statistics, that would mean we stopped gathering data. This would lead to outcomes such as:

1) Firms would scrap their market research departments and make products without checking how the market would receive the product. Instead they should hire soothsayers to predict product demand.

2) Pharma companies would make vaccines and people would use them without any trials.

3) Universities would close down their stats, big data, econometrics etc. departments.

etc.

 

[KB]

Statistics used well can be incredibly powerful. They can shine on the familiar an unfamiliar light, bathing it with new insights. And I am sure people can provide many examples of how powerful they can be. To use one example that I have heard the economist Tim Harford refer to, is the work of Richard Doll and Austin Bradford Hill - produced many years ago - which demonstrated a link between smoking and lung cancer. They could not have done so without a careful use of statistics.

That said, I can understand where the cynicism expressed in the opening post comes from. Statistics are ubiquitous in public life but the way they can be used should give us all pause. The media often make sensationalist statistical headlines out of complex studies; politicians often use statistics to obfuscate and mislead; we can all be selective in what statistics we choose to present and to believe in.

Being too credulous is a problem. But I don’t think the answer is to swing to the other extreme and become entirely dismissive and cynical of statistics and to rely only on our hunches. What is needed instead is a critical mindset, which really seeks to understand what is being claimed when a statistic is mentioned but also a mindset that demonstrates self-awareness, as our own emotions often influence how we react to statistical claims. In this spirit, there was a very good piece here on some rules of thumb that help in sustaining that critical mindset:

https://timharford.com/2018/03/your-handy-postcard-sized-guide-to-statistics/

 

[sweep_shot]

Humans are constantly required to make decisions without complete information. To make better decisions they gather data. Statistics is the tool they use to make inferences from data.

Suppose we stopped using statistics, that would mean we stopped gathering data. This would lead to outcomes such as:

1) Firms would scrap their market research departments and make products without checking how the market would receive the product. Instead they should hire soothsayers to predict product demand.

2) Pharma companies would make vaccines and people would use them without any trials.

3) Universities would close down their stats, big data, econometrics etc. departments.

etc.

I am not saying that all statistics are bad. I am simply saying that many statistics don't seem to give us accurate information.

There are also many stupid statistics. For example, win predictor during ICC World T20. I feel this is very misleading.

 

[sweep_shot]

Statistics used well can be incredibly powerful. They can shine on the familiar an unfamiliar light, bathing it with new insights. And I am sure people can provide many examples of how powerful they can be. To use one example that I have heard the economist Tim Harford refer to, is the work of Richard Doll and Austin Bradford Hill - produced many years ago - which demonstrated a link between smoking and lung cancer. They could not have done so without a careful use of statistics.

That said, I can understand where the cynicism expressed in the opening post comes from. Statistics are ubiquitous in public life but the way they can be used should give us all pause. The media often make sensationalist statistical headlines out of complex studies; politicians often use statistics to obfuscate and mislead; we can all be selective in what statistics we choose to present and to believe in.

Being too credulous is a problem. But I don’t think the answer is to swing to the other extreme and become entirely dismissive and cynical of statistics and to rely only on our hunches. What is needed instead is a critical mindset, which really seeks to understand what is being claimed when a statistic is mentioned but also a mindset that demonstrates self-awareness, as our own emotions often influence how we react to statistical claims. In this spirit, there was a very good piece here on some rules of thumb that help in sustaining that critical mindset:

https://timharford.com/2018/03/your-handy-postcard-sized-guide-to-statistics/

Thanks. Very balanced post.

I agree with you.

 

[drubwrubnat]

I personally do not pay much attention to most statistics. I think the methods are very faulty and results are often inaccurate due to biases and insufficient sample sizes.

Statistics was showing Hillary Clinton would win by a good margin yet Trump won in 2016.

There are more examples.

How seriously do you take statistics?

Election prediction is unsupervised learning.
The output variable is , here , is a boolean -
Clinton win or Trump win.
Even a rickshawwalla has 50% chance of being right.

The input variables imo can be state polls, previous voting history, orientation of state, demographic specific correlated variables- ( like the logarithm of age of a voter multiplied or divided by his ethnicity, religion, country of origin etc etc, ), social media activity.

Where there is a chance of error is -
a. People lying in an exit poll to appear more agreeable.
b. Human bias with a belligerent personality like Trump
c. Incorrect sampling mech. resulting in selection bias

In other words, even the simplest of null hypotheses assumes there is no interrelationships between input variables.

So any error in input data ( survey etc. ) can lead to violation of assumptions in predictive analysis.

With supercomputers it is easy to do the math, but interpretation of data is very crucial- and that will give you accurate outputs 90 % of the time.

However, my own PERSONAL opinion is -

Trump did the math on his own. He invented a new input variable - its the chemical called
TRUMP-AMPHETAMINE.
He played the sentiment of the public & secretly used analytics to manoeuvre his own campaign in a very creative way, the first time around.

The second time, this chemical was factored in by the opponents.

You read it here first

 

[drubwrubnat]

Talking about statistics.

I will never forget how the Pakistanis used reverse psychology on me on this forum to mislead me into believing the betfair odds & bet on India.
[MENTION=93712]MenInG[/MENTION] i shall have my revenge

 

[Technics 1210]

As far as election results go; 2016 changed all polling methods. Polling simply involves a process where voters are asked who they will vote for. The problem? Those who voted for Trump and Brexit didn't reveal so when asked in fear of social retribution and simply said would vote for the other. This skewed the election results thus resulting in surprise victories.

Exits polls are similar but are backed with actual votes thus more accurate than pre-election polling.

As for stats, economic stats are the least accurate and most fudged/manipulated to suit a political message.

 

[Technics 1210]

Talking about statistics.

I will never forget how the Pakistanis used reverse psychology on me on this forum to mislead me into believing the betfair odds & bet on India.

[MENTION=93712]MenInG[/MENTION] i shall have my revenge

Betting odds are just like polling data but the most accurate. India were definite favourites but not because betfair etc think so but because that was the outcome with the most money on. In other words more people thought India would win thus more money was placed on India winning. Unlike election polling data where the vote has to take place, market odds the move with the money thus more accurate and quantified in what people are thinking.

 

[drubwrubnat]

Betting odds are just like polling data but the most accurate. India were definite favourites but not because betfair etc think so but because that was the outcome with the most money on. In other words more people thought India would win thus more money was placed on India winning. Unlike election polling data where the vote has to take place, market odds the move with the money thus more accurate and quantified in what people are thinking.

If what you say is true, then can I practically be in the money ALL the time by arbitraging different oddsmakers since every single one differ in cumulative bets & volume?

 

[Technics 1210]

If what you say is true, then can I practically be in the money ALL the time by arbitraging different oddsmakers since every single one differ in cumulative bets & volume?

As I said the odds are a reflection of what people are thinking not a prediction of an outcome. And yes you can be in the money by waging bets from different bookmakers. There are many books on arbitrage betting; some of the most popular strategies are based on half-time betting in football matches.

Time delta is another favourite, especially with Test matches.

 

[drubwrubnat]

As I said the odds are a reflection of what people are thinking not a prediction of an outcome. And yes you can be in the money by waging bets from different bookmakers. There are many books on arbitrage betting; some of the most popular strategies are based on half-time betting in football matches.

Time delta is another favourite, especially with Test matches.

Agree when you say they favour the most backed option, because that gives them the best chance.

In my view , I feel that now increasingly, they do the ML runs / prediction first, and then add in their margins ( the thing youve mentioned )
This is my hunch.

 

[DeadlyVenom]

I take statistics that agree with my point of view quite seriously as they are usually produced using robust methods.

I disregard those that are not in agreement with my point of view as they are usually flawed, if the flaws aren't evident I will spend time finding them.

 

[Napa]

I am not saying that all statistics are bad. I am simply saying that many statistics don't seem to give us accurate information.

There are also many stupid statistics. For example, win predictor during ICC World T20. I feel this is very misleading.

It is easy to do a statistical test to check the accuracy of the "win predictor". There is enough data available. Just take a bunch of "win prediction probabilities" at various stages for a large number of games and see if the predicted probabilities match the actual outcomes.

Without having done such a test, you shouldn't call it "misleading".

A properly done "win prediction model" will keep getting updated based on data from recent matches. Given that international cricket is big business, I would assume they have hired competent statisticians to develop the win prediction model.

 

[sweep_shot]

It is easy to do a statistical test to check the accuracy of the "win predictor". There is enough data available. Just take a bunch of "win prediction probabilities" at various stages for a large number of games and see if the predicted probabilities match the actual outcomes.

Without having done such a test, you shouldn't call it "misleading".

A properly done "win prediction model" will keep getting updated based on data from recent matches. Given that international cricket is big business, I would assume they have hired competent statisticians to develop the win prediction model.

I am still calling that predictor misleading (and even unnecessary) because it doesn't take contexts into account.

For example, it was showing Bangladesh were favorites against West Indies at a certain stage despite the fact Bangladesh were clearly struggling.

 

[sweep_shot]

As far as election results go; 2016 changed all polling methods. Polling simply involves a process where voters are asked who they will vote for. The problem? Those who voted for Trump and Brexit didn't reveal so when asked in fear of social retribution and simply said would vote for the other. This skewed the election results thus resulting in surprise victories.

Exits polls are similar but are backed with actual votes thus more accurate than pre-election polling.

As for stats, economic stats are the least accurate and most fudged/manipulated to suit a political message.

Yeah. That makes sense.

I also do not take economic stats seriously (particularly if it comes from corrupt third world countries).

 

[Kroll]

I personally do not pay much attention to most statistics. I think the methods are very faulty and results are often inaccurate due to biases and insufficient sample sizes.

Statistics was showing Hillary Clinton would win by a good margin yet Trump won in 2016.

There are more examples.

How seriously do you take statistics?

The difference here is that those are polls based on probability which is a sub topic of stats. Statistics based on data of events that have occurred can shed import insights and provide a useful analytical tool.

For eg. successful marketing campaigns will analyse data on what has worked in the past to run a successful ad or sales funnel in the present.

 

[ahsan17]

Statistics is the best tool to make a strong argument. They represent large aggregate data, and cannot be discredited by individual anecdotes, assuming the statistical methods used are good.

There are a few things to be aware of when using statistics though, and these are things you learn in statistics classes.

1. Have a strong understanding of the statistic you are being presented, i.e. what exactly did the statistical study researched, how they gathered data, what questions were asked and how they were worded whether it's peer reviewed, the probability levels etc. In case of 2016 US elections, many of the polls were focused on popular vote and not electoral collage. Clinton did indeed win the popular vote.

2. Be very careful of how a third party is presenting those numbers to you, because very often those numbers are misrepresented, taken out of context, over simplified, etc.

There are people I know that do not take statistics seriously, or only do when it favours their POV. TBH, while I can be friends with them, I generally avoid getting into arguments with them since it's hard to reason with not accepting statistics.