Summations Only

No, you don't seem to grasp the difference between real world data and ideal mathematical relationships, in that you're the one insisting that correlation is causality, that a logical inference logically implies its inverse, that broad group characteristics can be applied to individuals (the Ecological Fallacy), and other elementary errors.I appreciate that you're trying to make this a teachable moment for me, but you need to understand that this has to be a teachable moment for you about reasoning properly from statistical information and using statistical tools. For instance: No, you did it, perhaps by accident, and I know that because you posted a screencap of the software you used. Xmin and Ymin are user-editable fields in your screencap, and they're both set to -1 - thus, defining the origin at -1,-1. Whether or not you did that intentionally or unintentionally, you didn't fix it and therefore it's your mistake, and a discrepancy between your model and your description of your model. You either didn't notice, or assumed we wouldn't. But the thing is - I know more about statistics and model-fitting than you do. So I did notice because I went looking for the kind of common errors people make in stats work when they don't know what they're doing. Yes. Your failure to understand the difference between a function and its first derivative, as you did when you said:

quote:

---

But since a significant source of guns used in crimes is stolen guns, reducing the number of armed citizens should also reduce the number of armed criminals.

---

You maintain that the accusation is "false" but false in what sense? Did you not make this statement? It appears in your post, presumably not by accident. Do you contend that the statement doesn't represent you mistaking a function for its first derivative? If so you're obligated to explain how, mathematically, reducing the rate at which criminals can acquire new guns would somehow deprive criminals of the guns they already have. If you didn't think you had to do that because you didn't realize that's what was being asked, then again, my accusation holds because anyone who understands how a function differs from its first derivative would see that as an obvious and unaddressed counter to your proposition that "since a significant source of guns used in crimes is stolen guns, reducing the number of armed citizens should also reduce the number of armed criminals." I don't expect real world data to go through 0,0. You expect real world data to go through 0,0:

quote:

---

mathematically in an ideal isolated region far from any crazed Canadians shooting Americans from across the border, 0 guns must correspond to 0 gun deaths. The line can only rise from there.

---

Again, I make no expectations about your data except that it be accurately described when you attempt to describe it. But it doesn't. The line you show doesn't rise from 0,0, it rises from -1, -1. And 0 guns doesn't necessarily correspond to 0 gun deaths, for a myriad of bizarre corner-case reasons. I do have such a mechanism, which I described in my last post. Asserting that I don't is a misrepresentation.Edited by crashfrog, : No reason given.

It's actually been my contention throughout that "emotional responses" aren't particularly helpful in responding to this situation, but it's nice to arrive at least one point of agreement.

Percy writes: The software did that to make the data easy to see. Our dear esteemed crashfrog will certainly notice that there zero points to left or below the 0,0 origin. It's not just the nearest integer, it's nearest integer that covers the data span in a manner that makes it easy to see. Perhaps if we throw out the bad data point, the software would rescale everything so that the points close to 0,0 would be far away enough from 0,0 for the software to choose the 0 axis in each direction.Also, this is a 1st-order analysis getting a straight line through the data. As RAZD has pointed out in Message 1023, assuming a linear relationship is going to be conceding to an approximation of fit. For example, one might easily imagine that the 1st gun in a country could kill more than 1 person, but that the 1st gun death in a country would be unlikely to have more than 1 gun responsible (don't most firing squads have only 1 gun loaded?).

Why are you wanting to remove the USA outlier?
I didn't think outliers had to be removed - I thought they only needed to be checked that they were accurate and legitimate. If my statement above is correct, then the answer is: don't throw out the USA point.

crashfrog replies to Percy: Don't be stupid. See Message 1038. There is nothing wrong here. What is wrong is your notion of what Xmin and Ymin must be. The Singapore and Japan points are easier to see this way.

Obviously not, but the trend line rises from -1, -1 - not from 0,0 as Percy has repeatedly stated. It doesn't even pass through 0,0. And again - not that I expect it to, I expect Percy'd description of what he's showing to be accurate.I don't see how this is difficult to grasp. If you tell me you're showing me a trend line that begins at 0,0 and rises from there, then you should not show me a line that begins at -1, -1 and doesn't even pass through 0,0. Saying one thing and showing another is a misrepresentation. It's the definition of misrepresentation.

Panda writes: Ok.But notice if you remove the USA point that what you have at the lower left is a cloud of data with a bad r-squared. How can that cloud have much of any significance to say about the USA point? You can do a point-by-point analysis to see how much each point contributes to the r-squared value. Maybe Argentina should also be cast out, given that they are currently going through a violent phase compared to their normal numbers?

But then that means accepting Xongsmith's point - the USA has a far greater rate of homicides than you would statistically expect from just the number of guns we have. Which is dispositive of what the graph was presented in support of - that in the USA, the number of homicides follows a statistical correlation with the number of guns.You can't both accept that the USA is a statistical outlier in terms of the correlation between the number of guns and the number of homicides, and that in the USA the number of homicides is correlated to the number of guns. "Outlier" means that it doesn't follow the general correlation.

Just to be clear - Is it your contention in this thread that the high US homicide rate and the prevalence of guns in the US are entirely unrelated?

Yes. If you remove data points you will get a different looking graph.Personally, I would suggest plugging the data from other years in to the chart.
Maybe Percy will get the time to do that.

crashfrog writes: The equation was y = 6.8x + 5.6, so the line rises from the point {0.0,5.6}, not your {-1, -1}. If we plug in your x=-1, then the y value is -1.2 not -1. See how easy it is to make noise over nothing?It is a linear approximation of what might certainly be a very non-linear relationship. As such, the straight line derived from the data is not required to pass through the entire real world truth of {0.0,0.0}. Canadians are in the real world. If there were no guns in the entire world, then there would be, by definition, no gun deaths in the entire world. This is Percy's point. I got it. Most of us here got it. Why are you making such a big deal about this? I think you should be making more noise about the lower left cloud having any statistical relevancy to the USA point.

Not at all unrelated. I suspect a significant drive of gun sales is the desire on the part of many Americans to be better equipped to commit a homicide.

Then I am a bit bemused as to the purpose of this quibbling over statistical analysis.If you agree that high gun prevalence is related to high homicide rate what exactly are you disagreeing about here?

Then the line described by that equation and the red line on the graph aren't the same line, which is a significant problem if we're expected to accept the accuracy of the graph as it reflects the data it's ostensibly presented as reflecting. He says that's not his point:

quote:

---

if you really knew anything about statistics you wouldn't expect real world data to go through 0,0.

---

He is, in fact, saying the exact opposite of what you say he's saying. Maybe you can get together with him and help him resolve this confusion. I wasn't going to, but Percy insisted that his utter misrepresentation of the data he posted was a function of "something I don't get about statistics." If he hadn't made it personal I probably would have let it go. But to the extent he insists that I'm the one with something to learn, and he's the one to teach it, I'm going to continue to demonstrate that the reverse is true.

As I keep saying, I disagree that the statistical data proves that reducing gun ownership would reduce homicides, or that the statistical correlation between gun ownership and gun injury is proof that individuals should not ever choose to own a gun.I've made this point surely half a dozen times in the past 10 posts of mine; is there some reason you didn't understand it then?