No poll watching this year.

Every (primarily presidential) election cycle I used to track the polls that appear on RealClearPolitics to try to extrapolate a realistic result by filtering out polls that are what I would consider statistically unreliable.  Some were too small (high margin of error), some oversampled Democrats (not a truly random sample, therefore biased), some polled adults or registered voters instead of likely voters.  Some may have even defined likely voters oddly.  They needed to be filtered out  in order to get a clearer view of how race really looked.  It worked in 2016, I was seeing a Trump narrow win over Hillary Clinton.  I thought I had to be wrong and it turned out I was; I understated president Trump's electoral performance. 

In 2018 I looked at Senate races and my filtering method still did pretty well.  But in 2020 I did not see a Let's Go Brandon victory. In 2022 I didn't do that well in senate races either. I'm not convinced that my methodology was wrong, but I think the polling has changed in a way that is not clear to me. That could mean I am doing something that is wrong that perhaps used to work, I don't know.  But the point is, as a poll watcher, I do not trust my input data any longer.  

I've enjoyed looking at the What-If scenarios, and I may still do it this year. I like statistics and doing the analysis. But until I can figure out why the polls are different and my previous method of filtering out suspicious results can be fixed, I don't think it's productive to share my results here, at least for this election cycle.

Besides, I don't believe, as Rush Limbaugh used to say, poll watching is the best use of our time. I'll find another way to be productive and useful to the conservative cause. What does it matter if the polls show a close race and it turns out that hundreds of thousands of Let's Go Brandon votes appear at 2 a.m. in Pennsylvania for Let's Go Brandon, with a corresponding (and statically impossible) zero votes for Trump. 

But outrage over suspicious results won't change the results. What will change the result is a rejuvenated RNC that  does it's job properly. Let's hope the RNC's new leadership is up to the task.

Is ChatGPT woke?

ChatGPT is a model-based chatbot cum search engine that does a remarkable job of mimicking human style communication in answering questions posed to it.  Here's the link if you haven't seen it yet. If fed properly posed questions it can provide remarkably well made answers.  Whether ChatGPT and AI (artificial intelligence) are a good idea for humanity is a broader debate, for another time.  There are some specific problems with how the model has been trained and I'll point out one example below regarding climate change.  It appears that the model (or engine or ChatGPT) is woke, at least on this subject.

This, by the way, is not ground-breaking.  Others like Ben Shapiro have already tested it's wokeness on issues and come away at least concerned about it. 

Of course the model is not the issue.  It learns it's responses from the data made available to it.  Garbage in - garbage out.  That's not to say everything it was fed was garbage, but enough of it, on certain subjects, clearly was garbage.  Observe my 'conversation':

Me

Write an article explaining why global warming is not a man-made phenomenon using specific scientific facts.

ChatGPT

 It is a widely accepted scientific fact that the Earth's climate is warming, and that human activity, particularly the burning of fossil fuels, is a major contributor to this warming. However, some argue that global warming is not a man-made phenomenon and that natural factors, such as solar radiation and volcanic activity, play a larger role in the Earth's climate.

One argument is that variations in solar radiation, such as changes in the sun's output of energy, have a greater impact on the Earth's climate than human activity. However, studies have shown that while the sun's radiation does play a role in the Earth's climate, the warming observed in recent decades cannot be explained solely by changes in solar radiation.

Another argument is that volcanic activity is a significant contributor to global warming. Volcanoes release large amounts of greenhouse gases, such as carbon dioxide and methane, into the atmosphere. However, the amount of greenhouse gases released by volcanoes is significantly less than the amount released by human activity. Additionally, volcanic activity has been shown to have a cooling effect on the Earth's climate in the short term, due to the release of aerosols that reflect sunlight back into space.

It is also true that the Earth's climate has undergone significant changes in the past, long before human activity became a significant factor. However, the rate of warming in recent decades is much faster than any observed in the past, and the overwhelming majority of scientists agree that human activity is the primary cause of this accelerated warming.

In conclusion, while natural factors such as solar radiation and volcanic activity do play a role in the Earth's climate, the evidence overwhelmingly shows that human activity, particularly the burning of fossil fuels, is the primary cause of global warming. It is important to note that scientific facts are supported by scientific evidence and research, and that global warming is a complex issue with various factors influencing it.

Me

Make a counter-argument to your article.

ChatGPT

One counter-argument to the idea that human activity is the primary cause of global warming is that the Earth's climate is constantly changing and that the current warming trend is simply part of a natural cycle. Proponents of this viewpoint argue that the Earth has gone through periods of warming and cooling in the past and that the current warming trend is not unusual or unprecedented. They may also point to the fact that the Earth's temperature has been relatively stable for the last 9000 years and that the current warming trend is not as severe as some of the warming trends that occurred in the past.

Another counter-argument is that the models used to predict future warming are not reliable and that the warming may not be as severe as projected. Some argue that the models used to predict future warming are based on a limited amount of data and are not accurate enough to make definitive predictions about the future climate. They may also argue that the models do not take into account all of the natural factors that influence the Earth's climate, such as changes in solar radiation and volcanic activity.

Additionally, some argue that even if human activity is contributing to global warming, the potential negative impacts of reducing greenhouse gas emissions, such as economic downturns, would outweigh the potential benefits. They may argue that alternative solutions, such as adaptation or geoengineering, should be pursued instead of reducing emissions.

It is important to note that while these counter-arguments do present valid points, the scientific consensus is that human activity is the primary cause of global warming and that immediate action is needed to reduce emissions and mitigate the potential negative impacts of climate change. It is also important to consider that scientific research and facts are constantly evolving and that there are different perspectives on the issue of global warming.

[Emphasis added]

Two tries and ChatGPT tried to steer me towards global warming being a real threat; despite my ask to counter-argue it.

This does not represent a robust line of questioning, and ChatGPT does make errors and recognizes them when you point it out to it's credit (I have witnessed it in an unrelated sports chat I had).  Here's the general issue I have with the woke climate change answers as a specific issue and as a broader generalization.  People will come to ChatGPT for answers, perhaps eventually in greater numbers than Google as a search engine. But they will be unaware of potential garbage-in garbage-out issues and take ChatGPT answers as gospel truth.  This will result in a reinforcing of woke culture, just as woke culture is being pushed back for the first time.

More broadly, existing biases and misinformation will become the new common sense or the "new normal" view with an over-reliance on the tool.  Think Wikipedia issues, but on a much bigger scale.

This is not to say that I am at odds with models and improving what AI can do. I believe that improving predictive modeling, even what has been done with ChatGPT can be very useful. However, it must be done extremely diligently with a lot of forethought.  I have built statistical models during my career. When done properly, it is extremely time consuming to ensure the right data is included to avoid biases, the right type of model is used, the model must be thoroughly tested etc.  There is a lot of work involved. Every model I built required a very large amount of documentation, so it could be reviewed the in future for accuracy and robustness.

ChatGPT is not woke.  It's a model built on web data up to 2021.  If 90% of the web says global warming is real, then ChatGPT will provide answers based on that set of inputs. It is not really artificial intelligence, and neither are any models. They are statistical predictive tools at this point, nothing more.  They will likely advance beyond that at some point, but we are nowhere near that point now.  The real danger is that people may believe we are, and accept erroneous or dubious outputs from the tool as gospel truth, and make decisions accordingly.

Conservatives Against Models - a really bad idea

I work in data and analytics in my daily life.  I've seen the power of models, and while they are not perfect, they are truly powerful tools.  I've also heard on shows like Rush Limbaugh and other conservative pundits (for example on Patriot Radio on SiriusXM),  decrying models.  As a conservative I will state unequivocally, we ignore and discount models at our political peril.

While it is true that statistical models can provide wildly inaccurate results (for example millions of COVID-19 deaths, hurricane paths that make no sense, etc), that is because not the model, but of the assumptions baked into the model by the creator(s) of that model.  Garbage in, garbage out. 

But models have also successfully predicted what vintage of wine and from what region will be a particularly good vintage, or bad.  Models have extremely accurately predicted consumer behaviors to an alarmingly amazing degree.  In the latter case I know personally as I have both used and created models with stunning predictive ability.  Groups like Camebridge Analytica have thrived on the amazing amount of data available in public to predict behaviors.

Models are used for a startling amount of things.  Just because models often lead to incorrect results do not discount them - if a model is set up to predict a specific outcome in a specific way, it will do exactly that.  Garbage in, garbage out.  But to discount and denigrate models as a whole is to throw the baby out with the bathwater.

Models can predict election outcomes with razor precision if they are built properly. Now it is especially true since machine learning can lead models to self-improve over time. If we turn conservatives against models holistically, we are dooming ourselves to being out-campaigned by Democrats who understand the power of smart-campaigning leveraging model-driven insights.

This is no I'm-trying-to-keep-my-job warning.  Models are the future of everything; from airline safety to stock price predictions, to pandemics, to human health and genetics, o music popularity to farming and beyond.  I work in the industry and see how over time, people are learning to better set models up for success and the models are teaching themselves to get better over time.  That is not to suggest that they will replace humans, but that they can be leveraged as truly useful tools, if they are properly trained (and yes training models is a real thing).

Here's an example of Artificial Intelligence, leveraging modeling of behavior, learning to play hide and seek, and then, even to break the rules:

Thoughts on the police shootings in Dallas (part 1 of 2)

It's never as easy as the numbers.  This morning I was surfing around Facebook and I came across a couple of items regarding violence and the police. No doubt it's a hot topic because of the police shootings in Dallas.  First I found this picture and there's a couple of problems with it.

Firstly, the comment added at the bottom of the image, as a commenter pointed out, doesn't take into account the population percentages.  Because non-Hispanic whites represent 62.6% of the population they apparently should account for roughly that percentage in the statistic above. They actually account for 51.75% of the cases where the race was reported (including the unreported race in the denominator is misleading and I have excluded it from consideration.  They could be all white, all African-American or all Asian for example - we don't know).  The comment too is overly simplistic.

African-Americans are over-represented in the killings by police.  However, they are also over-represented in violent crimes committed, accounting for 28% of all violent crimes and 49% of all murders. Given that increased rate of violent activity it makes sense that African-Americans are over-represented in the police killings category, not necessarily because of police intent but because they could have been killed in the process of committing a violent act.

From the overly simplistic graphic above it is impossible to tell if there is a systemic bias of some kind.  Furthermore, the source of that variance in rates of violent crimes is in itself a topic for debate.  Is it born of poverty and desperation? And is that poverty a result of the disproportionate single mother households in the African American community?  Is it a racial bias on the part of African Americans that causes them to be disproportionately violent towards other races?  A chart on CNN is grossly misleading.

What is also missing is the reverse statistic regarding how many police were killed in the line of duty. The video below is also from 2014.  Not only were 108 police killed in the line of duty, how many attempted assaults or murders of police were also committed?  The video below is a reminder of the dangers that police officers face daily.  It is easy to see why, despite the overwhelming restraint police show given how many millions of incidents occur without negative consequences, that there are situations that escalate, and that mistakes can happen.  That is not to say that individual officers cannot be racist, some surely are, just as within any other population.  But to credit all police killings to race is not only overly simplistic, it's intentionally misleading.

Watch the video below and tell me you would not also be extra cautious and perhaps a bit quick to react if you were a police officer.  We ask them to protect us and to sometimes put themselves in harms way to do so.  To expect perfection is unreasonable, and still the vast majority day in and day out perform remarkably close to that level and deserve our respect.

Most will never understand.
Posted by 53 Hours on Monday, August 3, 2015

As I stated in my post title, it's never as easy as the numbers.  I say that as a numbers guy who thinks that numbers don't lie.  Benjamin Disraeli originally said "There are three kinds of lies: lies, damned lies, and statistics." Statistics are a great tool for understanding things, but dangerous when used incorrectly. It's not the numbers that are the problem, it's how the numbers get used that is the problem.  It's kind of like the same notion as guns don't kill people, people kill people.

Part 2 will follow on Monday.

Obama repeats the past, only worse

Decades ago when Social Security was enacted,  and later Medicare and Medicaid as well, planners planned based on assumptions about what I would call a static growth model.  That is to say that the success (or more correctly, sustainability)  of those programs was, and is, dependent upon two things:

1. that the population would continue to grow, or at least not shrink, and 
2. that the make up of the population would not change significantly over time.  

Of course neither of those conditions are certain and one has in fact, already proven to be wrong.

It's okay, the president went on Letterman

The Middle East is a disaster, and then there's this bit of statistical reality:

Click to enlarge/view.

Never fear America, your president appeared on David Letterman's show.  All is well.  Nothing to see here. These are not the droids statistics you are looking for.

Saturday Learning Series - Reasoning with Probabilities

It's interesting that something on probabilities would come from a philosophy tangent. Kevin deLaplante of The Critical Thinking Academy has a great mini-series on reasoning with probabilities. It might seem a bit dry to some but there are some pearls of wisdom in here. This one hits home for me big time, since I do a lot of work with statistics in my day job. I'm no statistics guru, having only taken it in university and then getting to use it again only within the last 10 years or so (after a decade out of school). But I've worked with it enough to understand the importance of statistics and the implications of what it can tell us (and a little more, but that's not relevant here).

The Dow Jones drops again

Dow Jones Industrial Averages over the last decade have been an interesting study. The closing average when Bush became President elect, November 3rd 2000 was 10,817.95 Bush had inherited a mounting recession and by the time he took office it had already dropped 230 points.

In the November following the 9/11 attacks it had dropped to 9300 (November 2nd - 9,323.54). By the 2004 it had regained some momentum closing on November 4th at 10,314.76, and by August 29th, 2007 it had reached 13,289. The Bush Presidency had seen a 22.8% growth over 7 years, or an average growth of 3.26% at that point. Those are respectable numbers in the face of a recession, 9/11, two wars and Hurricane Katrina. But the recession had already started and by August 29th 2008 it had sunk to 11,543.96. Despite a recession being underway, the market is off by 13.11% in the year, but still up over the Bush tenure by 6.7%, or an average of 0.8% per year.

Which brings us to the Obama time frame. The closing price on November 4th 2008 was 9,625.28. For the record, the Bush Presidency closed the DJIA down 11%. Obama, now President elect, begins talking about the economy and his plans for America. He has impacted the direction of the market. By Obama's inauguration January 20th 2009 the Dow closed at 7,949.09. Between Obama's election and inauguration, the DJIA down 17.4%. Is this all Obama's fault? No. Did he contribute to the sell off? Absolutely.

The closing closing price today - 6763.29. Since Obama's inauguration the market is down 14.9%. Or since he became President elect and Chris Matthews decided that Obama was already in charge and Bush should resign, it has dropped a whopping 29.7%. That is indeed bad news for America.

Bush dropped the market 1192.67 over 8 years. Obama has managed to drop it by 1185.8 in 6 weeks since his inauguration. That is a precipitous fall. President Obama talked about drastic action being required urgently - being required now. That action, as far as the Dow Jones Average is concerned, has not produced any positive immediate results. Of course, I don't recall him mentioning immediate results. But he definitely mentioned urgent several times.

What does this mean? Not much actually, 6 weeks is a very short window to view the prism of economic impact of the Presidency. And to be sure, the DJIA is not beholden to only presidential proclamation. There are many factors in play. However, for all the fierce urgency of now stuff, the market is reacting in a roughly uni-directional way; downward. The President has done nothing to stop the slide in any major way because his solutions are not the solutions the market requires. Then again, perhaps the slide in numbers the President is concerned with is not the Dow, but rather Gallup's presidential approval ratings.