What a great subtitle to the page. The first thing I looked for in the graph was data that supported my opinion. The subtitle called me out, and I read the rest of the page with (what I hope was) a more objective eye.
Unfortunately guns are a very polarizing topic, and reliable, unbiased analysis is difficult to come by. The magnitude of other confounding factors (income, crime, education, family history, policing policies and intensity, etc.) make an honest attempt at understanding their influence even more difficult.
>Although my opinion is "not a strong correlation between the two", and that certainly seems to be supported by the supplied data.
That position is supported by the first analysis, but not the other two. The three analyses provide three different conclusions: no correlation, correlation with increased homicide, correlation with decreased homicide.
It doesn't help that the government is literally disallowed from funding any gun related medical research. My girlfriend does rural health related research and this fact always infuriates her - I never even knew this until she had explained it to me.
I run the analytics department at Daily Burn and we were joking today about making shirts with this slogan on it.
Unbiased insight is hard enough to get in a business setting. There is a latent tendency to prove one's preconceptions. Add divisive political tactics that actively try to destroy open mindedness (which are employed by both sides of the political spectrum) and the problem is exacerbated.
This is great. It's interesting that they can support the anti-gun argument without even resorting to the typical advocacy cheat we see of excluding all but a few nations from the analysis. 143 nations isn't all of them, but it's enough to prevent cherry-picking. Yesterday I saw something about suicide that excluded Japan. Hilarious!
You seem to think that's somehow wrong. Excluding ouliers some times makes sense. If you're trying to decide whether variable x ptedicts y, it makes sense to remove data points where you have strong reasons to believe that other variables stronger than x might be at play.
No argument was made. Japan wasn't mentioned. It was something like, "these are the top ten nations measured by quality of life as judged by the international administration of judgmental assholes, and you can see that they all have lower suicide rates than USA."
When the exclusion is mentioned, and some reference to other, more decisive factors is made, your point might be valid, although in that case one has to wonder why we're not looking at the more decisive factors instead. It is rarely the case that such caveats are mentioned when trying to dress up political opinions in sociological camouflage. USA is a postcolonial nation of the Americas with a large, diverse population. As such, when considering health or violence it should be compared to other postcolonial nations of the Americas with large, diverse populations, e.g. Mexico or Brazil. It's rare for that to take place. TFA succeeds in that, so it has my compliments.
Why qualify the homocides? Even unintentional homocides are bad; the number of stories of parents shooting their kids unintentionally or kids shooting themselves or their family unintentionally are pretty legion.
I would be more interested in all gun-related homocides per capita.
Not "legion", at least in the US, ~600 fatal/year (with plenty of false reporting), specifically per Wikipedia (https://en.wikipedia.org/wiki/Gun_violence_in_the_United_Sta...) in 2013 505 vs. 11,208 homicides. Since I started tracking that in the very early '80s, it's gone down by 25% while the total population and number of guns has grown by very roughly 50%.
Realize that a lot of these press and anecdotal reports are false. Depicting an innocent (and more importantly, unprosecutable) child as the shooter lets a negligent (or perp) teen or adult evade investigation and prosecution.
There were several stories last year about a 2-year-old finding a handgun and shooting a family member with it. Given the hand strength and coordination of even your strongest 2-year-old, the idea that any could actually lift, grip (given the hand size vs. the frame size of even the smallest production handgun) aim, enable (e.g. take off safety where appropriate), and squeeze a trigger with the action necessary to make it fire straight is at best lottery odds. I don't think it's ever been demonstrated in a laboratory condition (which you would think anti-RKBA researchers must have attempted).
I could theoretically picture a kid pulling a trigger with thumb strength, but that could only hit the child or somebody directly behind him.
TL;DR and details probably left out of that article:
14 year old had some "friends" over for a slumber party. Father had left a gun outside of the safe while working on getting a holster to fit, in an area of the house where they weren't allowed to be in.
Of course they ventured there, one "friend" saw the gun, picked it up and pointed it at her head, and pulled the trigger.
It was ruled a "tragic" accident when it was at minimum manslaughter.
That's why I said it was "akin", a like case, where in a different way, it was officially scored as an accident when it was obviously was nothing of the sort.
He used 2 year olds as an example where, physically, they couldn't have done what was officially blamed on them.
Bottom line, the authorities are known to lie, in my home town they white knighted the killer and blamed the father. Be especially wary of statistics from cities with high murder rates when the politicians apply pressure to reduce the murder statistics, Chicago being only the latest example I've heard of.
Homicide is simply the act of one human being killing another. There is no context as to what happened unless it is qualified. Even with intentional homicides, it isn't necessarily bad. Someone defending themselves by committing a homicide may be considered justified.
Also, just because you hear about act A more than act B doesn't mean there is more of act A than B.
I've always felt that both sides' arguments about gun control are frustratingly uncertain, so this data does in fact support my opinion! :D
Has there been any significant effort to really comprehensively evaluate the data around gun violence and gun control? I've seen arguments on both sides that seem convincing given the data they present, but they always seem to be cherry-picked to support the "right" conclusion. Nobody is interested in evaluating first and then making policy.
Maryland, which I lived next to while in Arlington, VA for a dozen years, and Missouri where I was born, raised, and have retired to, have very similar statistics with very different gun control regimes. Just looked at racial makeup on their separate state pages, and Maryland has almost 3 times the percentage of blacks, which would not at all support a thesis that they're generically overly prone to murder with guns.
Or look at Joplin, and at the other end of I-44, which replaced Route 66, is St. Louis, just now the murder capital of the US. Night and day, same gun control laws or lack thereof, e.g. no more difficult to buy them than anywhere else in the US and easy to obtain concealed carry licenses.
Should tell you gun violence is more a symptom of other issues.
I grew up where you could go deer hunting and leave school early to do so. Rule was you had to put your gun / bullets in the principals office. Same place those of us who were in trouble sat.
Unattended. 50+ guns / couple hundred bullets. No one ever got shot nor was it ever even talked about.
When my father went to high school here in the late '40s, he and his peers would keep their long guns in their lockers for the same reason, same result with absolutely no control by the school authorities. The late Supreme Court justice Scalia, 3 years younger, related in the oral arguments of Heller how he'd carry his target rifle on the NYC subway to and from his school/JROTC? rifle range.
When I attended in the late '70s, the JROTC rifles were in a safe to which we did not have the combination (well, they were government property), now they use air rifles :-(.
It's not the comprehensive review of the literature you're looking for, but more recently in 2011, the Justice Department put out a report that might also be of interest: http://bjs.gov/content/pub/pdf/fv9311.pdf
1) the effect of firearms on homicide, if it exists, must be small
2) more firearms means more homicide...
3) unless you live in a developed country in which case more firearms means less homicide
My favorite part about this is how it demonstrates how data can "clearly support your opinion". So what's the verdict here, do guns = more homicide as it relates to most of us in a first world country? Would this change your opinion on gun control?
If guns made people safer, the United States would be the safest place in the world. Unfortunately, it's actually safer to live in any other country thats not currently at war.
p<0.14 reads "the likelihood that the observed positive correlation between intentional homicide rate and firearms per population rate is due to random effects is < 14%"
That is not that same as "statistically insignificant" which OP uses.
It is a p-value, not a Bayesian probability. A p-value is the probability of seeing data as or more extreme purely due to chance, not the probability that your data is due to chance.
In other words, X is the event of observing data as or more extreme than what you observed, H0 is the event that the relationship is due to random effects. P(E1 | E2) is the probability of E1 given E. You are claiming that P(H0 | X) < 0.14, but the evidence is that P(X | H0) < 0.14.
If you wanted to do Bayesian statistics, you could obtain P(H0 | X) = P(X | H0) * P(H0) / P(X) = P(X | H0) * P(H0) / (P(X | H0) * H0 + P(X | not H0) * (1-H0))
To do that, you need a prior probability of H0 - in other words, to do Bayesian inference you need a prior belief about how likely something is to be true and evidence will update your belief. If you are already absolutely certain of H0 or not H0, no amount of evidence will shift your view in Bayesian inference, as any amount of evidence is best explained as random variation.
"Not statistically significant" is not the same as "statistically insignificant". It only means that p-value is greater than or equal to 0.05. There can still be an effect, it just cannot be seen over noise given sample size and confidence level.
All of the statistics are misleading. The third one is particularly so, because they are weighting by s$population * s$homicides_per_100000, i.e. 100000 * homicides which makes it completely meaningless).
Residual standard error: 6561 on 141 degrees of freedom
Multiple R-squared: 0.1046, Adjusted R-squared: 0.0982
F-statistic: 16.46 on 1 and 141 DF, p-value: 8.179e-05
Residual standard error: 7591 on 69 degrees of freedom
Multiple R-squared: 0.06827, Adjusted R-squared: 0.05477
F-statistic: 5.056 on 1 and 69 DF, p-value: 0.02774
In other words, the most obvious analysis to do shows strong evidence for a positive relationship between total homicides and total firearms - the link is in the same direction but much weaker when you only look at the upper half of countries by GDP.
Correlation doesn't imply causation - it could be that in some countries, people obtain firearms because the homicide rate is high, or there could be some other confounding factor not even in the data that drives both.
If this had been an experiment and each country had been assigned a firearm ownership rate, it would have been safe to conclude that each million firearm owners cause between 55 and 68 homicides (99% confidence interval).
Weighting by number of homicides is actually reasonable. Since homicides are rare, variance of homicide rate is driven by number of homicides rather than by population. Weighting by inverse variance is pretty common.
37 comments
[ 4.4 ms ] story [ 89.7 ms ] threadWhat a great subtitle to the page. The first thing I looked for in the graph was data that supported my opinion. The subtitle called me out, and I read the rest of the page with (what I hope was) a more objective eye.
Unfortunately guns are a very polarizing topic, and reliable, unbiased analysis is difficult to come by. The magnitude of other confounding factors (income, crime, education, family history, policing policies and intensity, etc.) make an honest attempt at understanding their influence even more difficult.
Although my opinion is "not a strong correlation between the two", and that certainly seems to be supported by the supplied data.
That position is supported by the first analysis, but not the other two. The three analyses provide three different conclusions: no correlation, correlation with increased homicide, correlation with decreased homicide.
You want some money for that? Billionaire Bloomberg's pockets are wide open right now, and are funding a lot of "research".
Unbiased insight is hard enough to get in a business setting. There is a latent tendency to prove one's preconceptions. Add divisive political tactics that actively try to destroy open mindedness (which are employed by both sides of the political spectrum) and the problem is exacerbated.
What was the argument to exclude Japan?
When the exclusion is mentioned, and some reference to other, more decisive factors is made, your point might be valid, although in that case one has to wonder why we're not looking at the more decisive factors instead. It is rarely the case that such caveats are mentioned when trying to dress up political opinions in sociological camouflage. USA is a postcolonial nation of the Americas with a large, diverse population. As such, when considering health or violence it should be compared to other postcolonial nations of the Americas with large, diverse populations, e.g. Mexico or Brazil. It's rare for that to take place. TFA succeeds in that, so it has my compliments.
I would be more interested in all gun-related homocides per capita.
Realize that a lot of these press and anecdotal reports are false. Depicting an innocent (and more importantly, unprosecutable) child as the shooter lets a negligent (or perp) teen or adult evade investigation and prosecution.
There were several stories last year about a 2-year-old finding a handgun and shooting a family member with it. Given the hand strength and coordination of even your strongest 2-year-old, the idea that any could actually lift, grip (given the hand size vs. the frame size of even the smallest production handgun) aim, enable (e.g. take off safety where appropriate), and squeeze a trigger with the action necessary to make it fire straight is at best lottery odds. I don't think it's ever been demonstrated in a laboratory condition (which you would think anti-RKBA researchers must have attempted).
I could theoretically picture a kid pulling a trigger with thumb strength, but that could only hit the child or somebody directly behind him.
TL;DR and details probably left out of that article:
14 year old had some "friends" over for a slumber party. Father had left a gun outside of the safe while working on getting a holster to fit, in an area of the house where they weren't allowed to be in.
Of course they ventured there, one "friend" saw the gun, picked it up and pointed it at her head, and pulled the trigger.
It was ruled a "tragic" accident when it was at minimum manslaughter.
He used 2 year olds as an example where, physically, they couldn't have done what was officially blamed on them.
Bottom line, the authorities are known to lie, in my home town they white knighted the killer and blamed the father. Be especially wary of statistics from cities with high murder rates when the politicians apply pressure to reduce the murder statistics, Chicago being only the latest example I've heard of.
Also, just because you hear about act A more than act B doesn't mean there is more of act A than B.
Has there been any significant effort to really comprehensively evaluate the data around gun violence and gun control? I've seen arguments on both sides that seem convincing given the data they present, but they always seem to be cherry-picked to support the "right" conclusion. Nobody is interested in evaluating first and then making policy.
E.g. look at this by state and hit the column sorting buttons: https://en.wikipedia.org/wiki/Gun_violence_in_the_United_Sta...
Maryland, which I lived next to while in Arlington, VA for a dozen years, and Missouri where I was born, raised, and have retired to, have very similar statistics with very different gun control regimes. Just looked at racial makeup on their separate state pages, and Maryland has almost 3 times the percentage of blacks, which would not at all support a thesis that they're generically overly prone to murder with guns.
Or look at Joplin, and at the other end of I-44, which replaced Route 66, is St. Louis, just now the murder capital of the US. Night and day, same gun control laws or lack thereof, e.g. no more difficult to buy them than anywhere else in the US and easy to obtain concealed carry licenses.
I grew up where you could go deer hunting and leave school early to do so. Rule was you had to put your gun / bullets in the principals office. Same place those of us who were in trouble sat.
Unattended. 50+ guns / couple hundred bullets. No one ever got shot nor was it ever even talked about.
When I attended in the late '70s, the JROTC rifles were in a safe to which we did not have the combination (well, they were government property), now they use air rifles :-(.
http://www.businessinsider.com/congressional-ban-on-gun-viol...
The NRC did a review in 2004 that is probably the closest to what you're seeking: http://www.nap.edu/catalog/10881/firearms-and-violence-a-cri...
It's not the comprehensive review of the literature you're looking for, but more recently in 2011, the Justice Department put out a report that might also be of interest: http://bjs.gov/content/pub/pdf/fv9311.pdf
My favorite part about this is how it demonstrates how data can "clearly support your opinion". So what's the verdict here, do guns = more homicide as it relates to most of us in a first world country? Would this change your opinion on gun control?
That is not that same as "statistically insignificant" which OP uses.
In other words, X is the event of observing data as or more extreme than what you observed, H0 is the event that the relationship is due to random effects. P(E1 | E2) is the probability of E1 given E. You are claiming that P(H0 | X) < 0.14, but the evidence is that P(X | H0) < 0.14.
If you wanted to do Bayesian statistics, you could obtain P(H0 | X) = P(X | H0) * P(H0) / P(X) = P(X | H0) * P(H0) / (P(X | H0) * H0 + P(X | not H0) * (1-H0))
To do that, you need a prior probability of H0 - in other words, to do Bayesian inference you need a prior belief about how likely something is to be true and evidence will update your belief. If you are already absolutely certain of H0 or not H0, no amount of evidence will shift your view in Bayesian inference, as any amount of evidence is best explained as random variation.
Starting from the data defined in that article:
> d$homicides = d$homicides_per_100000 * d$population / 100000 > d$firearms = d$firearms_per_100 * d$population / 100 > summary(lm(formula = d$homicides ~ d$firearms))
Call: lm(formula = d$homicides ~ d$firearms)
Residuals: Min 1Q Median 3Q Max -12851 -2081 -1841 -978 50042
Coefficients: Estimate Std. Error t value Pr(>|t|) (Intercept) 2.114e+03 5.592e+02 3.780 0.000231 * d$firearms 9.142e-05 2.253e-05 4.058 8.18e-05 * --- Signif. codes: 0 ‘’ 0.001 ‘’ 0.01 ‘’ 0.05 ‘.’ 0.1 ‘ ’ 1
Residual standard error: 6561 on 141 degrees of freedom Multiple R-squared: 0.1046, Adjusted R-squared: 0.0982 F-statistic: 16.46 on 1 and 141 DF, p-value: 8.179e-05
> s <- d[d$gdp_ppp_per_capita > median(d$gdp_ppp_per_capita),] > summary(lm(formula = s$homicides ~ s$firearms))
Call: lm(formula = s$homicides ~ s$firearms)
Residuals: Min 1Q Median 3Q Max -4509 -2097 -1997 -1747 50543
Coefficients: Estimate Std. Error t value Pr(>|t|) (Intercept) 2.096e+03 9.202e+02 2.278 0.0258 * s$firearms 6.158e-05 2.739e-05 2.248 0.0277 * --- Signif. codes: 0 ‘’ 0.001 ‘’ 0.01 ‘’ 0.05 ‘.’ 0.1 ‘ ’ 1
Residual standard error: 7591 on 69 degrees of freedom Multiple R-squared: 0.06827, Adjusted R-squared: 0.05477 F-statistic: 5.056 on 1 and 69 DF, p-value: 0.02774
In other words, the most obvious analysis to do shows strong evidence for a positive relationship between total homicides and total firearms - the link is in the same direction but much weaker when you only look at the upper half of countries by GDP.
Correlation doesn't imply causation - it could be that in some countries, people obtain firearms because the homicide rate is high, or there could be some other confounding factor not even in the data that drives both.
If this had been an experiment and each country had been assigned a firearm ownership rate, it would have been safe to conclude that each million firearm owners cause between 55 and 68 homicides (99% confidence interval).