Posts Tagged ‘statistics’

Correlation and Causation and Guns and Games

February 24, 2018

Seventy-two percent of the recent decline in youth violence can be attributed to video games.

I am combining and re-issuing two articles from the past (2012 and 2014) because they are again relevant, but need re-casting. They deal with the relationship between violence in video games and violence (particularly gun violence) in real life.

This is important, because the President and the Governor of Kentucky, among others, have both made that connection. They are both wrong, and to the extent that they are in a position to know they are wrong, they are both lying for political gain.

The key point, true in all science, is that if the correlation is zero, then you can’t tell me there’s a real-world relationship. And if the correlation is negative, then the relationship goes the other way — an increase in one causes a decrease in the other.

Here’s a couple of examples.

Back in 2012, the Washington Post had an article on the game/gun relationship. The TL;DR version is: There isn’t any, get over it. Here’s a helpful graphic. If there was a relationship, the gun violence levels would go up with the levels of video games. Notice how that doesn’t happen. At all.

Source: Washington Post

Source: Washington Post

The study they posted compared spending on video games in different countries, vs gun deaths in those countries. Leaving out China, a distinct social and governmental entity all its own, video game spending varies by a factor of almost three, from Germany to socially similar Netherlands. Gun deaths vary from near zero in the UK and Japan, to 0.5 per 100,000 in Canada, which is almost an outlier, because everyone else is down near 0.25. Except for the US, of course, which is a true outlier at 3.2 — for a country that spends less per capita than even Germany.

Lack of correlation creates a strong presumption of lack of causation. If I claim that solar eclipses cause plagues, and you look at two thousand years of solar eclipses and find that the overwhelming majority did not take place immediately before a plague outbreak, then it’s a pretty good bet that my hypothesis is wrong.

Science can’t really prove claims, no matter how strong the evidence, because a later test might show the claims to be wrong. What science can say is that theory x has passed every test we have set for it. What science can say is, theory y fell at the first jump, because its claims of correlation have been shown to be not true.

Want another example? OK, this from 2014, from Florida. Florida, guys.

C.J. Ferguson, at Stetson University,did a simple study* of the correlation between real world youth violence vs video game violence, using historical statistics.

Here’s the key graphic.

A good example of non-causality

As the level of video game violence goes up, the level of youth violence goes down. Based on this, you could claim that video game violence actually reduces youth violence. After all, if you’re at home playing games, you’re not out on the street, getting wenches with child, wronging the ancientry, stealing, fighting. It’s what’s called a negative correlation. Specifically, it has bivariate correlation value of -0.85. And as any statistician will tell you, this give you an R² of 0.72, which means that 72% of the decline in youth violence can be attributed to video games.

The studies are four and six years old. Politicians have staffs. Politicians have helpful outsiders providing them with facts — and in some cases, with fake news. If they chose to listen to the fake news, they are choosing to lie to the public to advance their own agenda, specifically to dispel any efforts at gun control. If they lie about this, what else are they lying about?

The Long Farewell: Intimations of Mortality

August 29, 2017

The bad news is, I’ve got multiple myeloma. The good news, such as it is, is that it looks like it’s the so-called smouldering myeloma variety, AKA dumpster fire in your bones.

TLDR: It’s blood cancer. It’s incurable. It’s controllable. I have an early stage.

MM normally starts out as MGUS, or monoclonal gammopathy of undetermined significance. MGUS converts to MM at the rate of about 1.5% per year. Mine may have been that way for ten or 15 years, undetected. Since the treatment for MGUS is hide and watch, that was fine. In fact the treatment for smouldering is still hide and watch, but at a somewhat more watchful level.

I don’t know much more than that right now. When I get back from Japan, we’ll do a full body MRI and pee in a bottle for 24hrs. Ask me again in October.

Not much available on survival statistics for MM, not because it’s an exotic disease, but because doctors are terrible at reporting statistics. The best I can find is that the survival rate for full up MM, untreated, is 7 months. Treated 5yr survival 49% . Assume median (and mean) survival is 60 months. Range is then 7 – 113, StdDev ~ 0.25*range = 28 months. This is not totally accurate, because the curve is not normal, on account of the 0-month wall on the left. The curve is skewed right by an unknown amount.

In any event, the clock doesn’t start ticking until the smouldering bursts into flame, and who knows when that will be. Still, I’m dumping my long-term Treasury Notes.

I am off to Japan in 12hrs or so, and my responses will be erratic. The family request that all messages of support and condolence be sent to the Democratic National Committee.

 

 

 

 

 

TSA can’t do math. Either

January 26, 2016

There’s a couple of TSA reports over the last year that nobody seems to have linked up, perhaps because statisticians have better things to do than read TSA reports.

Last June, the TSA IG released a report that said that TSA inspectors at airports failed to find 95% of the contraband items (guns, explosives) used to test the system. In November, reports from Congress indicated there’d been no improvement. Perhaps in an effort to get 2016 off on a better PR footing, this month, TSA reported that their seizures of firearms was up 20%, to 2,653 and that 83% of them were loaded. Let’s do some math.

We have 2,653 weapons found. Some 83%, or 2200 were loaded. Now, we know that TSA regularly misses 95% of the weapons the IG tries to smuggle on, which implies that those 2200 loaded firearms represented 5% of the ones that were actually carried on board.

1. Question for the class. How many loaded firearms successfully boarded aircraft in 2015?

Answer:

.05 * X = 2200 = loaded guns found
X = 2200/.05 = 44000 = total guns smuggled
44000 – 2200 = 41800 = guns successfully smuggled

So, 41,800 loaded guns were successfully smuggled onto airliners in the US in 2015. That’s just over one gun every fifteen minutes.

2. Take home question for the class. Of those 41,800 loaded weapons on airplanes, how many were used in hijacking attempts?

3. Critical thinking question: What does this tell you about the real threat?

Haters?

July 27, 2014

A student project at California’s Humboldt State University maps all references to “hate” words on geolocatable tweets between June 2012 and April 2013. It’s an interesting study, but the results should be used with care. Three aspects of the data collection and processing make this approach problematic, but the study deals directly with only one.

First, was the tweet really a negative? Phrases like “…queer theory says…” and “…I’m just an old cripple…” are two ways that ‘hate’ words might not be negatives. The study deals with this in a straightforward manner — the students read every tweet and applied a definitional rubric.

Second, is there any kind of processing bias? If you use raw numbers, big cities will dominate the map: Portland will generate more tweets and more hate tweets than Tilamook. To avoid this, the study categorized the data as a percentage of tweets from a given area. This throws them into another basin of attraction for errors: a small town with few tweeters will show up here if it holds even one prolific hater. For example, The Dalles is a little one-Starbucks town in northern Oregon (population 13,000, or about two cruise ships). Portland is a major metropolis (750,000 people in the county). On the map below, The Dalles stands out like a beacon in the NW, while Portland doesn’t even warrant shading.

Is The Dalles really a hotbed of hatred?

Is The Dalles really a hotbed of hatred?

Third, do haters tend to hide their geolocation more than normals do? This is a basic limitation of the data collection technique, and could only be compensated for by sampling the location of the non-geolocated tweets, an essentially impossible task. The best one might do is ask Twitter to run an equivalent study based on tweet IP address, except that that might violate Twitter’s privacy policy, and in any event is fraught with its own problems — IP-based advertising regularly offers me the opportunity to meet lonely women in the wrong part of the state, the wrong state, or even the wrong region of the country (I’m not sure I’ve ever been to Louisiana).

Still, this is an imaginative use of data available from social media, and despite its flaws it’s a worthwhile project.

Correlation and Causation and Guns and Games

December 20, 2012

The Washington Post has an interesting article on the relationship between video games and gun violence. The TL;DR version is: There isn’t any, get over it. Here’s a helpful graphic. Notice how the datapoints fail to cluster along the hypothesized trendline.

Source: Washington Post

Source: Washington Post

The study they posted compared spending on video games in different countries, vs gun deaths in those countries. Leaving out China, a distinct social and governmental entity all its own, video game spending varies by a factor of almost three, from Germany to socially similar Netherlands. Gun deaths vary from near zero in the UK and Japan, to 0.5 per 100,000 in Canada, which is almost an outlier, because everyone else is down near 0.25. Except for the US, of course, which is a true outlier at 3.2 — for a country that spends less per capita than even Germany. (more…)

Hypothesis Testing

April 10, 2011

Having just gotten on an add pictures jag (thanks Kurt), I decided to test if, besides being aesthetic, they’d encourage people to read something they might not otherwise. In the best tradition of science, I thought I’d run a quick experiment. My hypothesis, H1, was that people would be entranced by the pretty pictures, and click on the link while in a trance state. My null hypothesis, H0, was that it wouldn’t make a difference, since nobody reads this blog anyway, except a few of my friends (Hi Sandy), and HOTD fans.

I decided to use my AAVSO post (scroll down two, past Three Wolves). I originally put it up on March 24th, and over the course of the next week it got exactly two views. Should be really easy to measure growth from that baseline. So I found the picture that went with the blog post that inspired me, and updated the article to include it on the main page, right above the “Read the rest of this entry” link. Then I sat back to let the data roll in.

Result one week later: two more views. No change.

Now, no change is the assumed state of the world. Most things we do don’t really change the world, not even within our own restricted circle. If we are to accept a hypothesis as coming from a good model, we have to demonstrate that our action made a difference. In this case, it didn’t, at least, not within the parameters of our experiment. I am reminded of two aphorisms from my youth:

1. Intelligence is our last defense against wishful thinking. Replace Intelligence with Statistics and you have something applicable to the wider world. In passing, I would note that Intelligence is capitalized for more reasons than just starting a sentence. I am talking about the formal discipline.

2. How badly you want something to be true has absolutely no impact on whether it is true or not. You build your model, you draw your hypothesis, you run your test. The universe tells you if you got it right.

Google and I, we don’t either of us have a source for the above quotes.