|
Although this thread does not function under the same strict guidelines as the USPMT, it is still a general practice on TL to provide a source with an explanation on why it is relevant and what purpose it adds to the discussion. Failure to do so will result in a mod action. |
On February 23 2018 01:06 superstartran wrote:Show nested quote +On February 23 2018 00:43 DarkPlasmaBall wrote:On February 23 2018 00:15 superstartran wrote:On February 22 2018 22:30 DarkPlasmaBall wrote:On February 22 2018 20:48 superstartran wrote:On February 22 2018 19:08 DarkPlasmaBall wrote:On February 22 2018 15:49 ticklishmusic wrote:On February 22 2018 13:28 superstartran wrote:On February 22 2018 13:10 ticklishmusic wrote:On February 22 2018 12:56 superstartran wrote:
[quote]
What if he chose to drive a car through the crowd? What if he got a bomb into that crowd? You're assuming that he stays at the same range if he has a different tool. There are other tools that can be used to kill more people quickly. We saw that happened when someone ran through a crowd with a vehicle.
All evidence that I have seen points to the idea that an assault weapons ban/firearm ban in general does nothing to curtail the homicide rate within a country. Why people keep insisting on one makes no sense to me. Are you familiar with the existence of a continent called Europe? Ever since the firearm bans in UK homicide rates (total homicide not firearm homicide rates) have been essentially the same. But I guess arguing with facts is not a thing anymore. Firearm bans SHOULD have a dramatic effect on homicide rates if we are to believe the 'liberal' side of the argument, especially after a decade. However, the truth is that they have almost no effect on the overall homicide rate within a country. https://crimeresearch.org/2013/12/murder-and-homicide-rates-before-and-after-gun-bans/ A glance at the left side bar (actually, the right one too) suggests that it's not the most neutral of sources. Here's a huge meta analysis/ review of the then-existing research on firearm legislation and correlation with firearm-related injuries. Firearms account for a substantial proportion of external causes of death, injury, and disability across the world. Legislation to regulate firearms has often been passed with the intent of reducing problems related to their use. However, lack of clarity around which interventions are effective remains a major challenge for policy development. Aiming to meet this challenge, we systematically reviewed studies exploring the associations between firearm-related laws and firearm homicides, suicides, and unintentional injuries/deaths. We restricted our search to studies published from 1950 to 2014. Evidence from 130 studies in 10 countries suggests that in certain nations the simultaneous implementation of laws targeting multiple firearms restrictions is associated with reductions in firearm deaths. Laws restricting the purchase of (e.g., background checks) and access to (e.g., safer storage) firearms are also associated with lower rates of intimate partner homicides and firearm unintentional deaths in children, respectively. Limitations of studies include challenges inherent to their ecological design, their execution, and the lack of robustness of findings to model specifications. High quality research on the association between the implementation or repeal of firearm legislation (rather than the evaluation of existing laws) and firearm injuries would lead to a better understanding of what interventions are likely to work given local contexts. This information is key to move this field forward and for the development of effective policies that may counteract the burden that firearm injuries pose on populations. Also, a quick look at the Wiki article on UK firearms laws shows a decline in firearms used in crimes as well as firearm-caused fatalities, even without adjustment per capita for population growth. superstartran thinks Wiki info is pretty much the worst ever- his strong reaction to it previously was one of the reasons he was banned in the first place- so that might not be a strong selling point to him. (Of course, that entry has over 100 citations and looks to be reasonable overall. Thanks for sharing it.) I like how you were so up all on that one study that uses the wrong data for a negative binomial regression, which is a pretty amateurish mistake to do (or you do it on purpose to push an agenda). So, like I said, are you going to respond why that study used rate data instead of count data? Are you referring to MLL's post? This one? + Show Spoiler +On February 20 2018 07:49 MyLovelyLurker wrote:I guess I'm a bit late on the whole 'let's dl wikipedia into a Jupyter notebook and correlate away' thing I was planning to do this morning, and happy that American stats exceptionalism is alive and well. I want to share this study www.ncbi.nlm.nih.gov that rigorously establishes correlation between gun ownership and firearm homicide rates within the 50 States. It's based on 2012 data. I didn't see anything fundamentally wrong with its scientific methodology. Interestingly enough, looking at the bibliography shows a grand total of eighty references, many of them pointing in a similar direction. Sharing the discussion / conclusion as : 'To the best of our knowledge, ours is the most up-to-date and comprehensive analysis of the relationship between firearm ownership and gun-related homicide rates within the 50 states. Our study encompassed a 30-year period, with data through 2010, and accounted for 18 possible confounders of the relationship between gun ownership and firearm homicide. We found a robust relationship between higher levels of gun ownership and higher firearm homicide rates that was not explained by any of these potential confounders and was not sensitive to model specification. Our work expanded on previous studies not only by analyzing more recent data, but also by adjusting for clustering by year and state and controlling for factors, such as the rate of nonfirearm homicides, that likely capture unspecified variables that may be associated with both gun ownership levels and firearm homicide rates. The correlation of gun ownership with firearm homicide rates was substantial. Results from our model showed that a 1-SD difference in the gun ownership proxy measure, FS/S, was associated with a 12.9% difference in firearm homicide rates. All other factors being equal, our model would predict that if the FS/S in Mississippi were 57.7% (the average for all states) instead of 76.8% (the highest of all states), its firearm homicide rate would be 17% lower. Because of our use of a proxy measure for gun ownership, we could not conclude that the magnitude of the association between actual household gun ownership rates and homicide rates was the same. However, in a model that incorporated only survey-derived measures of household gun ownership (for 2001, 2002, and 2004), we found that each 1-SD difference in gun ownership was associated with a 24.9% difference in firearm homicide rates. 'Our study substantially advances previous work by analyzing recent data, examining the longest and most comprehensive panel of state-specific data to date, and accounting for year and state clustering and for a wide range of potential confounders. We found a robust relationship between gun ownership and firearm homicide rates, a finding that held whether firearm ownership was assessed through a proxy or a survey measure, whether state clustering was accounted for by GEEs or by fixed effects, and whether or not gun ownership was lagged, by up to 2 years. The observed relationship was specific to firearm-related homicide. Although we could not determine causation, we found that states with higher levels of gun ownership had disproportionately large numbers of deaths from firearm-related homicides.' What evidence do you have that they used the wrong data? Why do you think that rate data can't be used and that count data must be used? And I don't know what you mean by "all up on that one study". I merely said it looked solid and I was interested in hearing your response to it (although you were temp banned at the time). You just laughed it off though, which I had seen you do before. That was what prompted me to reply to your comment in the first place. Here was our conversation: On February 19 2018 10:25 DarkPlasmaBall wrote:On February 19 2018 07:05 superstartran wrote:On February 19 2018 06:42 DarkPlasmaBall wrote:On February 19 2018 05:30 superstartran wrote:On February 19 2018 05:19 DarkPlasmaBall wrote:On February 19 2018 04:00 superstartran wrote:
[quote]
That is my exact point. You and I can throw any graph out there, without actually controlling for several variables you don't get an accurate picture But your response to a graph that *did* control for that variable (OECD countries) was a graph that was purposely less accurate and *didn't* control for it. Those two graphs are not both equally inaccurate, and just because you can find a bad graph doesn't mean the good graph should be ignored. The fact that you really think that OECD graph is actually an accurate representative and can be used as evidence for correlation between number of firearms and firearm related violence is laughable at best. The author himself said that you cannot use his graphs as evidence for anything because the dude got his fucking statistics from wikipedia.There's a MUCH stronger correlation of income disparity and poverty with firearm related violence more than anything, but no one wants to talk about that. It's all about guns bro. So now, well sourced encyclopedias aren't good enough references? You realize that "lol Wikipedia" wasn't even a valid rebuttal a decade ago, right? I mean, it's fine to discuss additional variables one should control for when having this discussion, but you're acting really smug for a person invoking strawman graphs and dismissing sources that are likely to be legitimate. Are you seriously trying to use outdated data that doesn't control for various different variables as a reliable source to claim that the data shown shows a strong correlation? "Well Sourced" lmao.Wikipedia 'likely to be legitimate' Sometimes I wonder if you guys actually graduated from a university and were taught basic scientific method or you are just talking out of your own ass. User was temp banned for this post. This is the last post I'll make about sourcing, as I see you're temp banned and I'm not sure if it borders on off-topic, but encyclopedias are generally good starting points for when people want to start researching topics, since they often times have an extensive bibliography for more information. And Wikipedia is no different, in that most long entries have dozens- if not hundreds- of works that are cited, and you'll immediately know if any pages aren't well-sourced. Keep in mind that it was established back in 2005 that Wikipedia's accuracy was comparable to Encyclopedia Britannica's ( https://www.cnet.com/news/study-wikipedia-as-accurate-as-britannica/ ) and Wikipedia has only become more reliable over the past 13 years (despite the taboo that comes with public editors). When doing real research, of course you're going to double-check your sources against other sources, but starting at a Wiki entry for basic overviews and looking through the bibliography is actually a pretty good informational springboard. In other words, it's completely inappropriate to automatically dismiss statements just because they exist on Wikipedia. Furthermore, I'm well aware of the mathematics and statistics references you're making (e.g., correlation), considering I teach high school and college math and statistics. You're not the only one who understands confounding variables. And I'm trying to have a dialogue with you- not get into a dick-swinging contest. The reason I had responded in the first place to your rebuttal of someone else's post was that you had tried refuting a study that attempted to control for certain variables with your own graph that specifically didn't, and you tried saying that since you found a meaningless graph, that someone else's graph was automatically equally meaningless. I found that to be disingenuous, along with your "lol Wiki is auto-wrong and anyone who uses Wiki is stupid" philosophy, and then other people ended up citing even more studies. I don't feel like your one-liner dismissive responses to some of these studies are really all that convincing, and calling people "amateurish" or saying that you are "talking out of your own ass" isn't really strengthening your arguments. firearm bans had no discernible effect on total homicide rates. This statement that you made to ticklishmusic is controversial, as there is plenty of research that disagrees with your claim: We analyzed the relationship between homicide and gun availability using data from 26 developed countries from the early 1990s. We found that across developed countries, where guns are more available, there are more homicides. These results often hold even when the United States is excluded.
Hemenway, David; Miller, Matthew. Firearm availability and homicide rates across 26 high income countries. Journal of Trauma. 2000; 49: 985-88.
Using survey data on rates of household gun ownership, we examined the association between gun availability and homicide across states, 2001-2003. We found that states with higher levels of household gun ownership had higher rates of firearm homicide and overall homicide. This relationship held for both genders and all age groups, after accounting for rates of aggravated assault, robbery, unemployment, urbanization, alcohol consumption, and resource deprivation (e.g., poverty). There was no association between gun prevalence and non-firearm homicide.
Miller, Matthew; Azrael, Deborah; Hemenway, David. State-level homicide victimization rates in the U.S. in relation to survey measures of household firearm ownership, 2001-2003. Social Science and Medicine. 2007; 64:656-64. https://www.hsph.harvard.edu/hicrc/firearms-research/guns-and-death/ The correlations here seem to indicate: 1. As the number of guns increases, so does the number of gun deaths; 2. As the number of guns increases, so does the number of overall deaths; 3. The number of guns doesn't seem to affect the number of non-firearm deaths (i.e., deaths due to other weapons aren't increasing as the number of guns changes, disputing the idea that people will just kill with other weapons if they can't get their hands on a gun). You claim to teach statistics and understand statistics but don't know that Negative Binomial Regression Models are utilized for modeling count variables? Bro, I didn't even major in statistics and I know that. You would pretty much never utilize Negative Binomial Regression Models for rates, which is exactly what that study does. You would utilize a poisson regression model if you were using rate data, but the study doesn't do that on purpose because it would blow up their claims. There are two things in here that I'd like to contest, "bro". 1. Conversions exist between counts and rates, so just because one starts with a negative binomial regression model doesn't mean the data can't be interpreted either in terms of counts or in terms of rates. Conventionally, yes, NBR is used when you start with counts, although Poissons certainly aren't only used for rates. Furthermore, there are special cases of negative binomial regression models (e.g., Pascal and Polya distributions) that tend to be preferable over Poisson counterparts, as they can be made more accurate by accounting for different means and standard deviations. NBR and Poisson are not mutually exclusive. In fact, here's another example of the same kind of comparative modeling happening, and you'll notice that this source considers both NBR and Poisson to be comparable as count models (i.e., your claim that Poisson = rates is not necessarily true), yet also allows the data to be interpreted as rates: Also, the negative binomial model, as compared to other count models (i.e., Poisson or zero-inflated models), is assumed the appropriate model. In other words, we assume that the dependent variable is over-dispersed and does not have an excessive number of zeros. The first half of this page interprets the coefficients in terms of negative binomial regression coefficients, and the second half interprets the coefficients in terms of incidence rate ratios. https://stats.idre.ucla.edu/stata/output/negative-binomial-regression/In other words, the study isn't automatically refuted simply because they're using a different model, because you can totally interpret the data in both ways. 2. I'm curious as to what evidence you have that the other model would "blow up their claims". Please elaborate. 1) Conversions weren't used here; they clearly had an agenda and utilized the wrong model. This was not a special case. 2) When you rerun the numbers and actually convert and ensure that you're using count data and not rate data, the graph will look more like this ![[image loading]](https://crimeresearch.org/wp-content/uploads/2013/12/Screen-Shot-2013-12-08-at-Sunday-December-8-1.22-PM.png) Which basically shows a pretty normal distribution. Aka their study is a bunch of bullshit. This is exactly why so many people on the 'other side' of the argument feel like 'your' side wants to take away firearms. You have people go as far as basically fudge the numbers of a study in order to 'prove' their agenda.
OK, I wasn't necessarily going to indulge in replying but...
1. One does not simply 'visually' assess the 'Gaussianity' or the 'normality' (sic) of a distribution just by optics and zooming on a graph. Trust me on that. If this was the case, simple bread and butter tools like 2-sample tests (Kolmogorov-Smirnov, AD en.wikipedia.org) or normality tests (a ton of them... en.wikipedia.org ) would simply not exist. I see marginal distributions that look exactly the same but come from totally different and separable classes everyday. For more on this phenomenon, see Anscombe's Quartet (en.wikipedia.org). It's hence very difficult to take any claim that follows from graphing a PDF followed by a 'pretty' statement, let alone a 'bullshit' one, without a heap of scientific salt.
1bis. Anyone who refers to a distribution as 'pretty normal' doesn't seem bothered about what 'normal' in stats or probability actually means, how ubiquitous Gaussians are, and hence unlikely to be able to make correct inferences - this is an orange flag in itself.
2. You are using verbatim a random person's letter, that was refused in a scientific journal, as a full-on rebuttal to a peer review. I'm not fond of ad hominems but regardless of that person's credentials, 'peer reviewed' means several peers, ie, a full committee of scientists reviewing independently of each other and of authors' identity. Scientific work is judged by committee consensus ; any single individual's opinion is less relevant.
2bis. Every single book on advertised crimeresearch.org has a pro-guns bias - this doesn't strike me as extremely bipartisan. How can this statistically be a coincidence, if the evidence from data is so ambiguous ? Incidentally, John Lott, the owner of the website, is a well known gun advocate. (en.wikipedia.org )
3. When you replied to me 'their findings were that gun ownership increased gun homicide by 0.9%' , you left out 'for every 1% extra gun ownership'. This is an important omission and a hell of a difference ! Sensitivity analysis has been around in engineering and stats methods ever since the good old days of the Taylor expansion - more than three centuries. Entire industries are built on it. Now if you forget the 1/100 factor... Especially in a setting where we're merely attempting to establish the sign of the correlation.
4. Don't be gung-ho on NB vs Poisson - neglecting the correlation between numerator and denominator in Slutsky's lemma (which will result in mishandling skew) often can be a reasonable assumption asymptotically. The point is not worth the triple repetition you've made of it. That guy litterally says he finds the opposite sign correlation as the authors ('I found a one percentage point increase FS/S produced a 1.2% increase in firearm homicides, but the point estimate using the BRFSS survey data implied the same change produced a 1.2% decrease.). I would focus on that instead if I were you, or him.
5. Even if you were to be gung ho about NB vs Poisson, the study I quoted explicitly controls for it. ('. Use of a Poisson rather than a negative binomial model did not alter the results.'.). You're litterally saying you choose not to believe the authors, which is your prerogative.
6. With all due respect, the fact you mention a six variables model as 'laughable' shows little experience with industry-wide stats or machine learning, let alone on a small dataset. People very often work classifiers or regressors off two or three variables, after a phase of dimensionality reduction. The ubiquitous PCA, or embeddings such as t-SNE or UMAP are always best visualized in 2 or 3d and a strong control for the risk of overcooking your model. We're not training a deep network with 100 million parameters here, we're making do with small data, due to the federal ban on gun control research.
7. This is important because in particular, even establishing all relevant variables doesn't mean they can all be acted upon. When you mention poverty as an important factor in determining gun crime, you are right in that it is statistically important to establish, but impossible to simply legislate away.
I also want to make use of a couple Occam Razor's points to finish. I posted one study that you dismissed as fridge science, very much like the previous Wiki-based post. But there is in fact a corpus of relevant literature that points in the same direction - dozens of journal articles. This has all been independently peer reviewed. Will we engage in similar dismissal of all of these, ie, all editorial boards simultaneously in America are ignorant or crooks, or worse, they all don't know the first thing about their negative binomial from their Poisson ?
You seem to be a knowledgeable poster and I appeal to your judgement here. I won't do this for every study out there ; just like DPB I reckon www.hsph.harvard.edu should be quality and unbiased enough literature review for other posters to inform their own opinion.
|
Math: the number one thing that can get folks on TL to write really long, hyper detailed, involved posts that I understand 75% of without having to use google. The passion is not limited to BW.
|
Awesome post, MLL 
Anyone who refers to a distribution as 'pretty normal' doesn't seem bothered about what 'normal' in stats or probability actually means
What are the chances that superstartran simply meant that normal distributions are beautiful? 
|
On February 23 2018 04:50 Plansix wrote:
Math: the number one thing that can get folks on TL to write really long, hyper detailed, involved posts that I understand 75% of without having to use google. The passion is not limited to BW.
I was closer to maybe 13.5% but for some reason still enjoyed reading it
|
On February 23 2018 04:50 Plansix wrote:
Math: the number one thing that can get folks on TL to write really long, hyper detailed, involved posts that I understand 75% of without having to use google. The passion is not limited to BW.
Yeah, some serious knowledge was just dropped on the last couple pages. Thanks to all for the unexpected KhanAcademy.
|
On February 23 2018 04:43 MyLovelyLurker wrote:Show nested quote +On February 23 2018 01:06 superstartran wrote:On February 23 2018 00:43 DarkPlasmaBall wrote:On February 23 2018 00:15 superstartran wrote:On February 22 2018 22:30 DarkPlasmaBall wrote:On February 22 2018 20:48 superstartran wrote:On February 22 2018 19:08 DarkPlasmaBall wrote:On February 22 2018 15:49 ticklishmusic wrote:On February 22 2018 13:28 superstartran wrote:On February 22 2018 13:10 ticklishmusic wrote:
[quote]
Are you familiar with the existence of a continent called Europe? Ever since the firearm bans in UK homicide rates (total homicide not firearm homicide rates) have been essentially the same. But I guess arguing with facts is not a thing anymore. Firearm bans SHOULD have a dramatic effect on homicide rates if we are to believe the 'liberal' side of the argument, especially after a decade. However, the truth is that they have almost no effect on the overall homicide rate within a country. https://crimeresearch.org/2013/12/murder-and-homicide-rates-before-and-after-gun-bans/ A glance at the left side bar (actually, the right one too) suggests that it's not the most neutral of sources. Here's a huge meta analysis/ review of the then-existing research on firearm legislation and correlation with firearm-related injuries. Firearms account for a substantial proportion of external causes of death, injury, and disability across the world. Legislation to regulate firearms has often been passed with the intent of reducing problems related to their use. However, lack of clarity around which interventions are effective remains a major challenge for policy development. Aiming to meet this challenge, we systematically reviewed studies exploring the associations between firearm-related laws and firearm homicides, suicides, and unintentional injuries/deaths. We restricted our search to studies published from 1950 to 2014. Evidence from 130 studies in 10 countries suggests that in certain nations the simultaneous implementation of laws targeting multiple firearms restrictions is associated with reductions in firearm deaths. Laws restricting the purchase of (e.g., background checks) and access to (e.g., safer storage) firearms are also associated with lower rates of intimate partner homicides and firearm unintentional deaths in children, respectively. Limitations of studies include challenges inherent to their ecological design, their execution, and the lack of robustness of findings to model specifications. High quality research on the association between the implementation or repeal of firearm legislation (rather than the evaluation of existing laws) and firearm injuries would lead to a better understanding of what interventions are likely to work given local contexts. This information is key to move this field forward and for the development of effective policies that may counteract the burden that firearm injuries pose on populations. Also, a quick look at the Wiki article on UK firearms laws shows a decline in firearms used in crimes as well as firearm-caused fatalities, even without adjustment per capita for population growth. superstartran thinks Wiki info is pretty much the worst ever- his strong reaction to it previously was one of the reasons he was banned in the first place- so that might not be a strong selling point to him. (Of course, that entry has over 100 citations and looks to be reasonable overall. Thanks for sharing it.) I like how you were so up all on that one study that uses the wrong data for a negative binomial regression, which is a pretty amateurish mistake to do (or you do it on purpose to push an agenda). So, like I said, are you going to respond why that study used rate data instead of count data? Are you referring to MLL's post? This one? + Show Spoiler +On February 20 2018 07:49 MyLovelyLurker wrote:I guess I'm a bit late on the whole 'let's dl wikipedia into a Jupyter notebook and correlate away' thing I was planning to do this morning, and happy that American stats exceptionalism is alive and well. I want to share this study www.ncbi.nlm.nih.gov that rigorously establishes correlation between gun ownership and firearm homicide rates within the 50 States. It's based on 2012 data. I didn't see anything fundamentally wrong with its scientific methodology. Interestingly enough, looking at the bibliography shows a grand total of eighty references, many of them pointing in a similar direction. Sharing the discussion / conclusion as : 'To the best of our knowledge, ours is the most up-to-date and comprehensive analysis of the relationship between firearm ownership and gun-related homicide rates within the 50 states. Our study encompassed a 30-year period, with data through 2010, and accounted for 18 possible confounders of the relationship between gun ownership and firearm homicide. We found a robust relationship between higher levels of gun ownership and higher firearm homicide rates that was not explained by any of these potential confounders and was not sensitive to model specification. Our work expanded on previous studies not only by analyzing more recent data, but also by adjusting for clustering by year and state and controlling for factors, such as the rate of nonfirearm homicides, that likely capture unspecified variables that may be associated with both gun ownership levels and firearm homicide rates. The correlation of gun ownership with firearm homicide rates was substantial. Results from our model showed that a 1-SD difference in the gun ownership proxy measure, FS/S, was associated with a 12.9% difference in firearm homicide rates. All other factors being equal, our model would predict that if the FS/S in Mississippi were 57.7% (the average for all states) instead of 76.8% (the highest of all states), its firearm homicide rate would be 17% lower. Because of our use of a proxy measure for gun ownership, we could not conclude that the magnitude of the association between actual household gun ownership rates and homicide rates was the same. However, in a model that incorporated only survey-derived measures of household gun ownership (for 2001, 2002, and 2004), we found that each 1-SD difference in gun ownership was associated with a 24.9% difference in firearm homicide rates. 'Our study substantially advances previous work by analyzing recent data, examining the longest and most comprehensive panel of state-specific data to date, and accounting for year and state clustering and for a wide range of potential confounders. We found a robust relationship between gun ownership and firearm homicide rates, a finding that held whether firearm ownership was assessed through a proxy or a survey measure, whether state clustering was accounted for by GEEs or by fixed effects, and whether or not gun ownership was lagged, by up to 2 years. The observed relationship was specific to firearm-related homicide. Although we could not determine causation, we found that states with higher levels of gun ownership had disproportionately large numbers of deaths from firearm-related homicides.' What evidence do you have that they used the wrong data? Why do you think that rate data can't be used and that count data must be used? And I don't know what you mean by "all up on that one study". I merely said it looked solid and I was interested in hearing your response to it (although you were temp banned at the time). You just laughed it off though, which I had seen you do before. That was what prompted me to reply to your comment in the first place. Here was our conversation: On February 19 2018 10:25 DarkPlasmaBall wrote:On February 19 2018 07:05 superstartran wrote:On February 19 2018 06:42 DarkPlasmaBall wrote:On February 19 2018 05:30 superstartran wrote:On February 19 2018 05:19 DarkPlasmaBall wrote:
[quote]
But your response to a graph that *did* control for that variable (OECD countries) was a graph that was purposely less accurate and *didn't* control for it. Those two graphs are not both equally inaccurate, and just because you can find a bad graph doesn't mean the good graph should be ignored. The fact that you really think that OECD graph is actually an accurate representative and can be used as evidence for correlation between number of firearms and firearm related violence is laughable at best. The author himself said that you cannot use his graphs as evidence for anything because the dude got his fucking statistics from wikipedia.There's a MUCH stronger correlation of income disparity and poverty with firearm related violence more than anything, but no one wants to talk about that. It's all about guns bro. So now, well sourced encyclopedias aren't good enough references? You realize that "lol Wikipedia" wasn't even a valid rebuttal a decade ago, right? I mean, it's fine to discuss additional variables one should control for when having this discussion, but you're acting really smug for a person invoking strawman graphs and dismissing sources that are likely to be legitimate. Are you seriously trying to use outdated data that doesn't control for various different variables as a reliable source to claim that the data shown shows a strong correlation? "Well Sourced" lmao.Wikipedia 'likely to be legitimate' Sometimes I wonder if you guys actually graduated from a university and were taught basic scientific method or you are just talking out of your own ass. User was temp banned for this post. This is the last post I'll make about sourcing, as I see you're temp banned and I'm not sure if it borders on off-topic, but encyclopedias are generally good starting points for when people want to start researching topics, since they often times have an extensive bibliography for more information. And Wikipedia is no different, in that most long entries have dozens- if not hundreds- of works that are cited, and you'll immediately know if any pages aren't well-sourced. Keep in mind that it was established back in 2005 that Wikipedia's accuracy was comparable to Encyclopedia Britannica's ( https://www.cnet.com/news/study-wikipedia-as-accurate-as-britannica/ ) and Wikipedia has only become more reliable over the past 13 years (despite the taboo that comes with public editors). When doing real research, of course you're going to double-check your sources against other sources, but starting at a Wiki entry for basic overviews and looking through the bibliography is actually a pretty good informational springboard. In other words, it's completely inappropriate to automatically dismiss statements just because they exist on Wikipedia. Furthermore, I'm well aware of the mathematics and statistics references you're making (e.g., correlation), considering I teach high school and college math and statistics. You're not the only one who understands confounding variables. And I'm trying to have a dialogue with you- not get into a dick-swinging contest. The reason I had responded in the first place to your rebuttal of someone else's post was that you had tried refuting a study that attempted to control for certain variables with your own graph that specifically didn't, and you tried saying that since you found a meaningless graph, that someone else's graph was automatically equally meaningless. I found that to be disingenuous, along with your "lol Wiki is auto-wrong and anyone who uses Wiki is stupid" philosophy, and then other people ended up citing even more studies. I don't feel like your one-liner dismissive responses to some of these studies are really all that convincing, and calling people "amateurish" or saying that you are "talking out of your own ass" isn't really strengthening your arguments. firearm bans had no discernible effect on total homicide rates. This statement that you made to ticklishmusic is controversial, as there is plenty of research that disagrees with your claim: We analyzed the relationship between homicide and gun availability using data from 26 developed countries from the early 1990s. We found that across developed countries, where guns are more available, there are more homicides. These results often hold even when the United States is excluded.
Hemenway, David; Miller, Matthew. Firearm availability and homicide rates across 26 high income countries. Journal of Trauma. 2000; 49: 985-88.
Using survey data on rates of household gun ownership, we examined the association between gun availability and homicide across states, 2001-2003. We found that states with higher levels of household gun ownership had higher rates of firearm homicide and overall homicide. This relationship held for both genders and all age groups, after accounting for rates of aggravated assault, robbery, unemployment, urbanization, alcohol consumption, and resource deprivation (e.g., poverty). There was no association between gun prevalence and non-firearm homicide.
Miller, Matthew; Azrael, Deborah; Hemenway, David. State-level homicide victimization rates in the U.S. in relation to survey measures of household firearm ownership, 2001-2003. Social Science and Medicine. 2007; 64:656-64. https://www.hsph.harvard.edu/hicrc/firearms-research/guns-and-death/ The correlations here seem to indicate: 1. As the number of guns increases, so does the number of gun deaths; 2. As the number of guns increases, so does the number of overall deaths; 3. The number of guns doesn't seem to affect the number of non-firearm deaths (i.e., deaths due to other weapons aren't increasing as the number of guns changes, disputing the idea that people will just kill with other weapons if they can't get their hands on a gun). You claim to teach statistics and understand statistics but don't know that Negative Binomial Regression Models are utilized for modeling count variables? Bro, I didn't even major in statistics and I know that. You would pretty much never utilize Negative Binomial Regression Models for rates, which is exactly what that study does. You would utilize a poisson regression model if you were using rate data, but the study doesn't do that on purpose because it would blow up their claims. There are two things in here that I'd like to contest, "bro". 1. Conversions exist between counts and rates, so just because one starts with a negative binomial regression model doesn't mean the data can't be interpreted either in terms of counts or in terms of rates. Conventionally, yes, NBR is used when you start with counts, although Poissons certainly aren't only used for rates. Furthermore, there are special cases of negative binomial regression models (e.g., Pascal and Polya distributions) that tend to be preferable over Poisson counterparts, as they can be made more accurate by accounting for different means and standard deviations. NBR and Poisson are not mutually exclusive. In fact, here's another example of the same kind of comparative modeling happening, and you'll notice that this source considers both NBR and Poisson to be comparable as count models (i.e., your claim that Poisson = rates is not necessarily true), yet also allows the data to be interpreted as rates: Also, the negative binomial model, as compared to other count models (i.e., Poisson or zero-inflated models), is assumed the appropriate model. In other words, we assume that the dependent variable is over-dispersed and does not have an excessive number of zeros. The first half of this page interprets the coefficients in terms of negative binomial regression coefficients, and the second half interprets the coefficients in terms of incidence rate ratios. https://stats.idre.ucla.edu/stata/output/negative-binomial-regression/In other words, the study isn't automatically refuted simply because they're using a different model, because you can totally interpret the data in both ways. 2. I'm curious as to what evidence you have that the other model would "blow up their claims". Please elaborate. 1) Conversions weren't used here; they clearly had an agenda and utilized the wrong model. This was not a special case. 2) When you rerun the numbers and actually convert and ensure that you're using count data and not rate data, the graph will look more like this Which basically shows a pretty normal distribution. Aka their study is a bunch of bullshit. This is exactly why so many people on the 'other side' of the argument feel like 'your' side wants to take away firearms. You have people go as far as basically fudge the numbers of a study in order to 'prove' their agenda. OK, I wasn't necessarily going to indulge in replying but... 1. One does not simply 'visually' assess the 'Gaussianity' or the 'normality' (sic) of a distribution just by optics and zooming on a graph. Trust me on that. If this was the case, simple bread and butter tools like 2-sample tests (Kolmogorov-Smirnov, AD en.wikipedia.org) or normality tests (a ton of them... en.wikipedia.org ) would simply not exist. I see marginal distributions that look exactly the same but come from totally different and separable classes everyday. For more on this phenomenon, see Anscombe's Quartet ( en.wikipedia.org). It's hence very difficult to take any claim that follows from graphing a PDF followed by a 'pretty' statement, let alone a 'bullshit' one, without a heap of scientific salt. 1bis. Anyone who refers to a distribution as 'pretty normal' doesn't seem bothered about what 'normal' in stats or probability actually means, how ubiquitous Gaussians are, and hence unlikely to be able to make correct inferences - this is an orange flag in itself. 2. You are using verbatim a random person's letter, that was refused in a scientific journal, as a full-on rebuttal to a peer review. I'm not fond of ad hominems but regardless of that person's credentials, 'peer reviewed' means several peers, ie, a full committee of scientists reviewing independently of each other and of authors' identity. Scientific work is judged by committee consensus ; any single individual's opinion is less relevant. 2bis. Every single book on advertised crimeresearch.org has a pro-guns bias - this doesn't strike me as extremely bipartisan. How can this statistically be a coincidence, if the evidence from data is so ambiguous ? Incidentally, John Lott, the owner of the website, is a well known gun advocate. ( en.wikipedia.org ) 3. When you replied to me 'their findings were that gun ownership increased gun homicide by 0.9%' , you left out 'for every 1% extra gun ownership'. This is an important omission and a hell of a difference ! Sensitivity analysis has been around in engineering and stats methods ever since the good old days of the Taylor expansion - more than three centuries. Entire industries are built on it. Now if you forget the 1/100 factor... Especially in a setting where we're merely attempting to establish the sign of the correlation. 4. Don't be gung-ho on NB vs Poisson - neglecting the correlation between numerator and denominator in Slutsky's lemma (which will result in mishandling skew) often can be a reasonable assumption asymptotically. The point is not worth the triple repetition you've made of it. That guy litterally says he finds the opposite sign correlation as the authors ('I found a one percentage point increase FS/S produced a 1.2% increase in firearm homicides, but the point estimate using the BRFSS survey data implied the same change produced a 1.2% decrease.). I would focus on that instead if I were you, or him. 5. Even if you were to be gung ho about NB vs Poisson, the study I quoted explicitly controls for it. ('. Use of a Poisson rather than a negative binomial model did not alter the results.'.). You're litterally saying you choose not to believe the authors, which is your prerogative. 6. With all due respect, the fact you mention a six variables model as 'laughable' shows little experience with industry-wide stats or machine learning, let alone on a small dataset. People very often work classifiers or regressors off two or three variables, after a phase of dimensionality reduction. The ubiquitous PCA, or embeddings such as t-SNE or UMAP are always best visualized in 2 or 3d and a strong control for the risk of overcooking your model. We're not training a deep network with 100 million parameters here, we're making do with small data, due to the federal ban on gun control research. 7. This is important because in particular, even establishing all relevant variables doesn't mean they can all be acted upon. When you mention poverty as an important factor in determining gun crime, you are right in that it is statistically important to establish, but impossible to simply legislate away. I also want to make use of a couple Occam Razor's points to finish. I posted one study that you dismissed as fridge science, very much like the previous Wiki-based post. But there is in fact a corpus of relevant literature that points in the same direction - dozens of journal articles. This has all been independently peer reviewed. Will we engage in similar dismissal of all of these, ie, all editorial boards simultaneously in America are ignorant or crooks, or worse, they all don't know the first thing about their negative binomial from their Poisson ? You seem to be a knowledgeable poster and I appeal to your judgement here. I won't do this for every study out there ; just like DPB I reckon www.hsph.harvard.edu should be quality and unbiased enough literature review for other posters to inform their own opinion.
Simple question, why would they NOT use count data when it is available and use rate data? By standard that's what is utilized in negative binomial regressions, and the data was clearly available to them. It is also in fact the standard for most gun control studies I have seen whether they lean one way or the other; it is very uncommon for a study to go about the way they did things here.
There's no reasonable explanation as to why they performed the study the way they did.
It's the same reason why I don't bring up Kleck's self-defense gun study very often because he has some very serious issues with how he presents his arguments and how he shows his statistics (some of them don't even add up)
https://pdfs.semanticscholar.org/91da/afbf92d021f06426764e800a4e639a1c1116.pdf
Ultimately the real point is that you can't just make sweeping statements about correlation off of such stats work when the stats are so limited. Where I would agree with the other side is that there needs to be far more in depth fair research on what we can do to limit gun violence. Most studies basically rig their numbers one way or another.
Also, just because your paper was published and peer reviewed doesn't mean it can't be rigged and biased from the start. Arthur Kellerman's paper over gun control is a perfect example of that.
http://www.nejm.org/doi/full/10.1056/NEJM199310073291506#t=abstract
|
Nononono superstrain, you don't get to refute mylovelylurker's post by replying it with this crap. How can you even refute that you not only cherry pick data, but you cherry pick the data by excluding 'for every 1% extra gun ownership' from'their findings were that gun ownership increased gun homicide by 0.9%'/ if you think a peer reviewed papaer is rigged and biased what chance does your favoured paper that was refused in ascientific journal? You are basically rebutting the same statistical science you are using as an argument with the argument that it is all a giant conspiracy.
|
On February 23 2018 04:23 zlefin wrote:Show nested quote +On February 23 2018 04:08 VHbb wrote:On February 23 2018 03:45 zlefin wrote:On February 23 2018 02:16 VHbb wrote:
The idea of having (armed!) security in a school is absurd to me..
yes yes I know I'm not from the US, so I cannot understand, but isn't there a point where you step back and you realize that something is broken in the system?
I'm not claiming anything about gun control or gun laws, I know nothing about guns and I'm very happy that I don't - and that I have never even seen a gun in ~30 years of life - but you clearly have a problem!
It does not matter what's the best weapon to defend yourself against home intruder, or what is the exact definition of 'assault rifle' ... I understand you come from a different perspective, but the fact that your schools are frequently theater of shootings (shootings! I honestly find hard to picture how this can happen, how is it possible that a teenager has even the idea of picking up a gun and shooting his classmates) is so sick that I don't understand how are you discussing about how correct it is to use a binomial model or not ...
I guess part of the disconnect comes from the fact that you think of the US as a 'western culture' country, but the level of separation compared to european culture is way larger than one is lead to believe. Reading this thread is actually eye opening for me (I was probably naive myself, for sure ...) Well, since a few o fthe more dangerous US schools have metal detectors, the notion of armed guards wouldn't exactly be shocking. I'm guessing you find the notion of schools with metal detectors extreme. but there are some bad neighborhoods out there. discussing the correctness of a binomial model is important though; in order to fix things it helps to understand them. oh and @poulsen it's hundreds of millions of guns in private US hands, not tens of millions. trivia: US civilians own more guns than all the world's militaries' combined. It's really an alien concept to me in some rare cases (I can only recall one) we had police coming in a university where I live, during some student protests, and everyone found that not acceptable: the police cannot enter the university in these cases, and in general schools / universities / campus are *not* a place where there should be security. This is an other point: what if the students want to organize a (pacific) protest occupying the school, knowing that there are armed security guards there? The idea of metal detector and campus security or armed guards is just so remote and distant from the way things are here, that it's really hard to grasp why you would need it. I'm not saying you don't, or that I know better, I'm just expressing surprise! I'd assume armed security guards won't disrupt a peaceful protest. at least no more so than such a protest would be disrupted in general. they have plenty of less lethal weapons/tools for that in general if they choose to use them. i'd assume the places that have metal detectors are mostly because they're in very dangerous areas. I don't have the exact statistics; but i'd say areas wherein you'd expect several students a year at least to be the victim of an aggravated assault on campus (plus a bunch more off the school property). I can round up some numbers if you want some detail.
Hi zlefin, thank you! no need for the data, I'll research them myself
and I see your point, my comment is just an highlight of how much we see things differently (even if the end result is similar): for me seeing police or security in a school 'feels' extremely wrong. I don't want to claim that it's not necessary, just that I've been raised in a place where school safety has never been an issue of this type.
|
On February 23 2018 06:18 Aveng3r wrote:Show nested quote +On February 23 2018 04:50 Plansix wrote:
Math: the number one thing that can get folks on TL to write really long, hyper detailed, involved posts that I understand 75% of without having to use google. The passion is not limited to BW. I was closer to maybe 13.5% but for some reason still enjoyed reading it
On February 23 2018 06:33 ticklishmusic wrote:Show nested quote +On February 23 2018 04:50 Plansix wrote:
Math: the number one thing that can get folks on TL to write really long, hyper detailed, involved posts that I understand 75% of without having to use google. The passion is not limited to BW. Yeah, some serious knowledge was just dropped on the last couple pages. Thanks to all for the unexpected KhanAcademy.
Regardless of which/ whose arguments seem more convincing, I found it much more enjoyable writing about guns + math than just writing about guns
|
There's no need to refer to statistics when the logic is so clear: more guns equals more safety.
|
On February 23 2018 07:49 superstartran wrote:Show nested quote +On February 23 2018 04:43 MyLovelyLurker wrote:On February 23 2018 01:06 superstartran wrote:On February 23 2018 00:43 DarkPlasmaBall wrote:On February 23 2018 00:15 superstartran wrote:On February 22 2018 22:30 DarkPlasmaBall wrote:On February 22 2018 20:48 superstartran wrote:On February 22 2018 19:08 DarkPlasmaBall wrote:On February 22 2018 15:49 ticklishmusic wrote:On February 22 2018 13:28 superstartran wrote:[quote] Ever since the firearm bans in UK homicide rates (total homicide not firearm homicide rates) have been essentially the same. But I guess arguing with facts is not a thing anymore. Firearm bans SHOULD have a dramatic effect on homicide rates if we are to believe the 'liberal' side of the argument, especially after a decade. However, the truth is that they have almost no effect on the overall homicide rate within a country. https://crimeresearch.org/2013/12/murder-and-homicide-rates-before-and-after-gun-bans/ A glance at the left side bar (actually, the right one too) suggests that it's not the most neutral of sources. Here's a huge meta analysis/ review of the then-existing research on firearm legislation and correlation with firearm-related injuries. Firearms account for a substantial proportion of external causes of death, injury, and disability across the world. Legislation to regulate firearms has often been passed with the intent of reducing problems related to their use. However, lack of clarity around which interventions are effective remains a major challenge for policy development. Aiming to meet this challenge, we systematically reviewed studies exploring the associations between firearm-related laws and firearm homicides, suicides, and unintentional injuries/deaths. We restricted our search to studies published from 1950 to 2014. Evidence from 130 studies in 10 countries suggests that in certain nations the simultaneous implementation of laws targeting multiple firearms restrictions is associated with reductions in firearm deaths. Laws restricting the purchase of (e.g., background checks) and access to (e.g., safer storage) firearms are also associated with lower rates of intimate partner homicides and firearm unintentional deaths in children, respectively. Limitations of studies include challenges inherent to their ecological design, their execution, and the lack of robustness of findings to model specifications. High quality research on the association between the implementation or repeal of firearm legislation (rather than the evaluation of existing laws) and firearm injuries would lead to a better understanding of what interventions are likely to work given local contexts. This information is key to move this field forward and for the development of effective policies that may counteract the burden that firearm injuries pose on populations. Also, a quick look at the Wiki article on UK firearms laws shows a decline in firearms used in crimes as well as firearm-caused fatalities, even without adjustment per capita for population growth. superstartran thinks Wiki info is pretty much the worst ever- his strong reaction to it previously was one of the reasons he was banned in the first place- so that might not be a strong selling point to him. (Of course, that entry has over 100 citations and looks to be reasonable overall. Thanks for sharing it.) I like how you were so up all on that one study that uses the wrong data for a negative binomial regression, which is a pretty amateurish mistake to do (or you do it on purpose to push an agenda). So, like I said, are you going to respond why that study used rate data instead of count data? Are you referring to MLL's post? This one? + Show Spoiler +On February 20 2018 07:49 MyLovelyLurker wrote:I guess I'm a bit late on the whole 'let's dl wikipedia into a Jupyter notebook and correlate away' thing I was planning to do this morning, and happy that American stats exceptionalism is alive and well. I want to share this study www.ncbi.nlm.nih.gov that rigorously establishes correlation between gun ownership and firearm homicide rates within the 50 States. It's based on 2012 data. I didn't see anything fundamentally wrong with its scientific methodology. Interestingly enough, looking at the bibliography shows a grand total of eighty references, many of them pointing in a similar direction. Sharing the discussion / conclusion as : 'To the best of our knowledge, ours is the most up-to-date and comprehensive analysis of the relationship between firearm ownership and gun-related homicide rates within the 50 states. Our study encompassed a 30-year period, with data through 2010, and accounted for 18 possible confounders of the relationship between gun ownership and firearm homicide. We found a robust relationship between higher levels of gun ownership and higher firearm homicide rates that was not explained by any of these potential confounders and was not sensitive to model specification. Our work expanded on previous studies not only by analyzing more recent data, but also by adjusting for clustering by year and state and controlling for factors, such as the rate of nonfirearm homicides, that likely capture unspecified variables that may be associated with both gun ownership levels and firearm homicide rates. The correlation of gun ownership with firearm homicide rates was substantial. Results from our model showed that a 1-SD difference in the gun ownership proxy measure, FS/S, was associated with a 12.9% difference in firearm homicide rates. All other factors being equal, our model would predict that if the FS/S in Mississippi were 57.7% (the average for all states) instead of 76.8% (the highest of all states), its firearm homicide rate would be 17% lower. Because of our use of a proxy measure for gun ownership, we could not conclude that the magnitude of the association between actual household gun ownership rates and homicide rates was the same. However, in a model that incorporated only survey-derived measures of household gun ownership (for 2001, 2002, and 2004), we found that each 1-SD difference in gun ownership was associated with a 24.9% difference in firearm homicide rates. 'Our study substantially advances previous work by analyzing recent data, examining the longest and most comprehensive panel of state-specific data to date, and accounting for year and state clustering and for a wide range of potential confounders. We found a robust relationship between gun ownership and firearm homicide rates, a finding that held whether firearm ownership was assessed through a proxy or a survey measure, whether state clustering was accounted for by GEEs or by fixed effects, and whether or not gun ownership was lagged, by up to 2 years. The observed relationship was specific to firearm-related homicide. Although we could not determine causation, we found that states with higher levels of gun ownership had disproportionately large numbers of deaths from firearm-related homicides.' What evidence do you have that they used the wrong data? Why do you think that rate data can't be used and that count data must be used? And I don't know what you mean by "all up on that one study". I merely said it looked solid and I was interested in hearing your response to it (although you were temp banned at the time). You just laughed it off though, which I had seen you do before. That was what prompted me to reply to your comment in the first place. Here was our conversation: On February 19 2018 10:25 DarkPlasmaBall wrote:On February 19 2018 07:05 superstartran wrote:On February 19 2018 06:42 DarkPlasmaBall wrote:On February 19 2018 05:30 superstartran wrote:
[quote]
The fact that you really think that OECD graph is actually an accurate representative and can be used as evidence for correlation between number of firearms and firearm related violence is laughable at best. The author himself said that you cannot use his graphs as evidence for anything because the dude got his fucking statistics from wikipedia.
There's a MUCH stronger correlation of income disparity and poverty with firearm related violence more than anything, but no one wants to talk about that. It's all about guns bro. So now, well sourced encyclopedias aren't good enough references? You realize that "lol Wikipedia" wasn't even a valid rebuttal a decade ago, right? I mean, it's fine to discuss additional variables one should control for when having this discussion, but you're acting really smug for a person invoking strawman graphs and dismissing sources that are likely to be legitimate. Are you seriously trying to use outdated data that doesn't control for various different variables as a reliable source to claim that the data shown shows a strong correlation? "Well Sourced" lmao.Wikipedia 'likely to be legitimate' Sometimes I wonder if you guys actually graduated from a university and were taught basic scientific method or you are just talking out of your own ass. User was temp banned for this post. This is the last post I'll make about sourcing, as I see you're temp banned and I'm not sure if it borders on off-topic, but encyclopedias are generally good starting points for when people want to start researching topics, since they often times have an extensive bibliography for more information. And Wikipedia is no different, in that most long entries have dozens- if not hundreds- of works that are cited, and you'll immediately know if any pages aren't well-sourced. Keep in mind that it was established back in 2005 that Wikipedia's accuracy was comparable to Encyclopedia Britannica's ( https://www.cnet.com/news/study-wikipedia-as-accurate-as-britannica/ ) and Wikipedia has only become more reliable over the past 13 years (despite the taboo that comes with public editors). When doing real research, of course you're going to double-check your sources against other sources, but starting at a Wiki entry for basic overviews and looking through the bibliography is actually a pretty good informational springboard. In other words, it's completely inappropriate to automatically dismiss statements just because they exist on Wikipedia. Furthermore, I'm well aware of the mathematics and statistics references you're making (e.g., correlation), considering I teach high school and college math and statistics. You're not the only one who understands confounding variables. And I'm trying to have a dialogue with you- not get into a dick-swinging contest. The reason I had responded in the first place to your rebuttal of someone else's post was that you had tried refuting a study that attempted to control for certain variables with your own graph that specifically didn't, and you tried saying that since you found a meaningless graph, that someone else's graph was automatically equally meaningless. I found that to be disingenuous, along with your "lol Wiki is auto-wrong and anyone who uses Wiki is stupid" philosophy, and then other people ended up citing even more studies. I don't feel like your one-liner dismissive responses to some of these studies are really all that convincing, and calling people "amateurish" or saying that you are "talking out of your own ass" isn't really strengthening your arguments. firearm bans had no discernible effect on total homicide rates. This statement that you made to ticklishmusic is controversial, as there is plenty of research that disagrees with your claim: We analyzed the relationship between homicide and gun availability using data from 26 developed countries from the early 1990s. We found that across developed countries, where guns are more available, there are more homicides. These results often hold even when the United States is excluded.
Hemenway, David; Miller, Matthew. Firearm availability and homicide rates across 26 high income countries. Journal of Trauma. 2000; 49: 985-88.
Using survey data on rates of household gun ownership, we examined the association between gun availability and homicide across states, 2001-2003. We found that states with higher levels of household gun ownership had higher rates of firearm homicide and overall homicide. This relationship held for both genders and all age groups, after accounting for rates of aggravated assault, robbery, unemployment, urbanization, alcohol consumption, and resource deprivation (e.g., poverty). There was no association between gun prevalence and non-firearm homicide.
Miller, Matthew; Azrael, Deborah; Hemenway, David. State-level homicide victimization rates in the U.S. in relation to survey measures of household firearm ownership, 2001-2003. Social Science and Medicine. 2007; 64:656-64. https://www.hsph.harvard.edu/hicrc/firearms-research/guns-and-death/ The correlations here seem to indicate: 1. As the number of guns increases, so does the number of gun deaths; 2. As the number of guns increases, so does the number of overall deaths; 3. The number of guns doesn't seem to affect the number of non-firearm deaths (i.e., deaths due to other weapons aren't increasing as the number of guns changes, disputing the idea that people will just kill with other weapons if they can't get their hands on a gun). You claim to teach statistics and understand statistics but don't know that Negative Binomial Regression Models are utilized for modeling count variables? Bro, I didn't even major in statistics and I know that. You would pretty much never utilize Negative Binomial Regression Models for rates, which is exactly what that study does. You would utilize a poisson regression model if you were using rate data, but the study doesn't do that on purpose because it would blow up their claims. There are two things in here that I'd like to contest, "bro". 1. Conversions exist between counts and rates, so just because one starts with a negative binomial regression model doesn't mean the data can't be interpreted either in terms of counts or in terms of rates. Conventionally, yes, NBR is used when you start with counts, although Poissons certainly aren't only used for rates. Furthermore, there are special cases of negative binomial regression models (e.g., Pascal and Polya distributions) that tend to be preferable over Poisson counterparts, as they can be made more accurate by accounting for different means and standard deviations. NBR and Poisson are not mutually exclusive. In fact, here's another example of the same kind of comparative modeling happening, and you'll notice that this source considers both NBR and Poisson to be comparable as count models (i.e., your claim that Poisson = rates is not necessarily true), yet also allows the data to be interpreted as rates: Also, the negative binomial model, as compared to other count models (i.e., Poisson or zero-inflated models), is assumed the appropriate model. In other words, we assume that the dependent variable is over-dispersed and does not have an excessive number of zeros. The first half of this page interprets the coefficients in terms of negative binomial regression coefficients, and the second half interprets the coefficients in terms of incidence rate ratios. https://stats.idre.ucla.edu/stata/output/negative-binomial-regression/In other words, the study isn't automatically refuted simply because they're using a different model, because you can totally interpret the data in both ways. 2. I'm curious as to what evidence you have that the other model would "blow up their claims". Please elaborate. 1) Conversions weren't used here; they clearly had an agenda and utilized the wrong model. This was not a special case. 2) When you rerun the numbers and actually convert and ensure that you're using count data and not rate data, the graph will look more like this Which basically shows a pretty normal distribution. Aka their study is a bunch of bullshit. This is exactly why so many people on the 'other side' of the argument feel like 'your' side wants to take away firearms. You have people go as far as basically fudge the numbers of a study in order to 'prove' their agenda. OK, I wasn't necessarily going to indulge in replying but... 1. One does not simply 'visually' assess the 'Gaussianity' or the 'normality' (sic) of a distribution just by optics and zooming on a graph. Trust me on that. If this was the case, simple bread and butter tools like 2-sample tests (Kolmogorov-Smirnov, AD en.wikipedia.org) or normality tests (a ton of them... en.wikipedia.org ) would simply not exist. I see marginal distributions that look exactly the same but come from totally different and separable classes everyday. For more on this phenomenon, see Anscombe's Quartet ( en.wikipedia.org). It's hence very difficult to take any claim that follows from graphing a PDF followed by a 'pretty' statement, let alone a 'bullshit' one, without a heap of scientific salt. 1bis. Anyone who refers to a distribution as 'pretty normal' doesn't seem bothered about what 'normal' in stats or probability actually means, how ubiquitous Gaussians are, and hence unlikely to be able to make correct inferences - this is an orange flag in itself. 2. You are using verbatim a random person's letter, that was refused in a scientific journal, as a full-on rebuttal to a peer review. I'm not fond of ad hominems but regardless of that person's credentials, 'peer reviewed' means several peers, ie, a full committee of scientists reviewing independently of each other and of authors' identity. Scientific work is judged by committee consensus ; any single individual's opinion is less relevant. 2bis. Every single book on advertised crimeresearch.org has a pro-guns bias - this doesn't strike me as extremely bipartisan. How can this statistically be a coincidence, if the evidence from data is so ambiguous ? Incidentally, John Lott, the owner of the website, is a well known gun advocate. ( en.wikipedia.org ) 3. When you replied to me 'their findings were that gun ownership increased gun homicide by 0.9%' , you left out 'for every 1% extra gun ownership'. This is an important omission and a hell of a difference ! Sensitivity analysis has been around in engineering and stats methods ever since the good old days of the Taylor expansion - more than three centuries. Entire industries are built on it. Now if you forget the 1/100 factor... Especially in a setting where we're merely attempting to establish the sign of the correlation. 4. Don't be gung-ho on NB vs Poisson - neglecting the correlation between numerator and denominator in Slutsky's lemma (which will result in mishandling skew) often can be a reasonable assumption asymptotically. The point is not worth the triple repetition you've made of it. That guy litterally says he finds the opposite sign correlation as the authors ('I found a one percentage point increase FS/S produced a 1.2% increase in firearm homicides, but the point estimate using the BRFSS survey data implied the same change produced a 1.2% decrease.). I would focus on that instead if I were you, or him. 5. Even if you were to be gung ho about NB vs Poisson, the study I quoted explicitly controls for it. ('. Use of a Poisson rather than a negative binomial model did not alter the results.'.). You're litterally saying you choose not to believe the authors, which is your prerogative. 6. With all due respect, the fact you mention a six variables model as 'laughable' shows little experience with industry-wide stats or machine learning, let alone on a small dataset. People very often work classifiers or regressors off two or three variables, after a phase of dimensionality reduction. The ubiquitous PCA, or embeddings such as t-SNE or UMAP are always best visualized in 2 or 3d and a strong control for the risk of overcooking your model. We're not training a deep network with 100 million parameters here, we're making do with small data, due to the federal ban on gun control research. 7. This is important because in particular, even establishing all relevant variables doesn't mean they can all be acted upon. When you mention poverty as an important factor in determining gun crime, you are right in that it is statistically important to establish, but impossible to simply legislate away. I also want to make use of a couple Occam Razor's points to finish. I posted one study that you dismissed as fridge science, very much like the previous Wiki-based post. But there is in fact a corpus of relevant literature that points in the same direction - dozens of journal articles. This has all been independently peer reviewed. Will we engage in similar dismissal of all of these, ie, all editorial boards simultaneously in America are ignorant or crooks, or worse, they all don't know the first thing about their negative binomial from their Poisson ? You seem to be a knowledgeable poster and I appeal to your judgement here. I won't do this for every study out there ; just like DPB I reckon www.hsph.harvard.edu should be quality and unbiased enough literature review for other posters to inform their own opinion. Simple question, why would they NOT use count data when it is available and use rate data? By standard that's what is utilized in negative binomial regressions, and the data was clearly available to them. It is also in fact the standard for most gun control studies I have seen whether they lean one way or the other; it is very uncommon for a study to go about the way they did things here. There's no reasonable explanation as to why they performed the study the way they did. It's the same reason why I don't bring up Kleck's self-defense gun study very often because he has some very serious issues with how he presents his arguments and how he shows his statistics (some of them don't even add up) https://pdfs.semanticscholar.org/91da/afbf92d021f06426764e800a4e639a1c1116.pdfUltimately the real point is that you can't just make sweeping statements about correlation off of such stats work when the stats are so limited. Where I would agree with the other side is that there needs to be far more in depth fair research on what we can do to limit gun violence. Most studies basically rig their numbers one way or another. Also, just because your paper was published and peer reviewed doesn't mean it can't be rigged and biased from the start. Arthur Kellerman's paper over gun control is a perfect example of that. http://www.nejm.org/doi/full/10.1056/NEJM199310073291506#t=abstract
If your response to two lenghty posts written by people clearly knowledgable in the subject is "but you see, I think the study was done in a way that I think is wrong, and also all studies are rigged anyway", then I can't take you seriously any longer. Damn shame, because for a moment it seemed like you had something interesting to say.
|
On February 23 2018 16:32 PoulsenB wrote:Show nested quote +On February 23 2018 07:49 superstartran wrote:On February 23 2018 04:43 MyLovelyLurker wrote:On February 23 2018 01:06 superstartran wrote:On February 23 2018 00:43 DarkPlasmaBall wrote:On February 23 2018 00:15 superstartran wrote:On February 22 2018 22:30 DarkPlasmaBall wrote:On February 22 2018 20:48 superstartran wrote:On February 22 2018 19:08 DarkPlasmaBall wrote:On February 22 2018 15:49 ticklishmusic wrote:[quote] A glance at the left side bar (actually, the right one too) suggests that it's not the most neutral of sources. Here's a huge meta analysis/ review of the then-existing research on firearm legislation and correlation with firearm-related injuries. [quote] Also, a quick look at the Wiki article on UK firearms laws shows a decline in firearms used in crimes as well as firearm-caused fatalities, even without adjustment per capita for population growth. superstartran thinks Wiki info is pretty much the worst ever- his strong reaction to it previously was one of the reasons he was banned in the first place- so that might not be a strong selling point to him. (Of course, that entry has over 100 citations and looks to be reasonable overall. Thanks for sharing it.) I like how you were so up all on that one study that uses the wrong data for a negative binomial regression, which is a pretty amateurish mistake to do (or you do it on purpose to push an agenda). So, like I said, are you going to respond why that study used rate data instead of count data? Are you referring to MLL's post? This one? + Show Spoiler +On February 20 2018 07:49 MyLovelyLurker wrote:I guess I'm a bit late on the whole 'let's dl wikipedia into a Jupyter notebook and correlate away' thing I was planning to do this morning, and happy that American stats exceptionalism is alive and well. I want to share this study www.ncbi.nlm.nih.gov that rigorously establishes correlation between gun ownership and firearm homicide rates within the 50 States. It's based on 2012 data. I didn't see anything fundamentally wrong with its scientific methodology. Interestingly enough, looking at the bibliography shows a grand total of eighty references, many of them pointing in a similar direction. Sharing the discussion / conclusion as : 'To the best of our knowledge, ours is the most up-to-date and comprehensive analysis of the relationship between firearm ownership and gun-related homicide rates within the 50 states. Our study encompassed a 30-year period, with data through 2010, and accounted for 18 possible confounders of the relationship between gun ownership and firearm homicide. We found a robust relationship between higher levels of gun ownership and higher firearm homicide rates that was not explained by any of these potential confounders and was not sensitive to model specification. Our work expanded on previous studies not only by analyzing more recent data, but also by adjusting for clustering by year and state and controlling for factors, such as the rate of nonfirearm homicides, that likely capture unspecified variables that may be associated with both gun ownership levels and firearm homicide rates. The correlation of gun ownership with firearm homicide rates was substantial. Results from our model showed that a 1-SD difference in the gun ownership proxy measure, FS/S, was associated with a 12.9% difference in firearm homicide rates. All other factors being equal, our model would predict that if the FS/S in Mississippi were 57.7% (the average for all states) instead of 76.8% (the highest of all states), its firearm homicide rate would be 17% lower. Because of our use of a proxy measure for gun ownership, we could not conclude that the magnitude of the association between actual household gun ownership rates and homicide rates was the same. However, in a model that incorporated only survey-derived measures of household gun ownership (for 2001, 2002, and 2004), we found that each 1-SD difference in gun ownership was associated with a 24.9% difference in firearm homicide rates. 'Our study substantially advances previous work by analyzing recent data, examining the longest and most comprehensive panel of state-specific data to date, and accounting for year and state clustering and for a wide range of potential confounders. We found a robust relationship between gun ownership and firearm homicide rates, a finding that held whether firearm ownership was assessed through a proxy or a survey measure, whether state clustering was accounted for by GEEs or by fixed effects, and whether or not gun ownership was lagged, by up to 2 years. The observed relationship was specific to firearm-related homicide. Although we could not determine causation, we found that states with higher levels of gun ownership had disproportionately large numbers of deaths from firearm-related homicides.' What evidence do you have that they used the wrong data? Why do you think that rate data can't be used and that count data must be used? And I don't know what you mean by "all up on that one study". I merely said it looked solid and I was interested in hearing your response to it (although you were temp banned at the time). You just laughed it off though, which I had seen you do before. That was what prompted me to reply to your comment in the first place. Here was our conversation: On February 19 2018 10:25 DarkPlasmaBall wrote:On February 19 2018 07:05 superstartran wrote:On February 19 2018 06:42 DarkPlasmaBall wrote:
[quote]
So now, well sourced encyclopedias aren't good enough references? You realize that "lol Wikipedia" wasn't even a valid rebuttal a decade ago, right? I mean, it's fine to discuss additional variables one should control for when having this discussion, but you're acting really smug for a person invoking strawman graphs and dismissing sources that are likely to be legitimate. Are you seriously trying to use outdated data that doesn't control for various different variables as a reliable source to claim that the data shown shows a strong correlation? "Well Sourced" lmao.Wikipedia 'likely to be legitimate' Sometimes I wonder if you guys actually graduated from a university and were taught basic scientific method or you are just talking out of your own ass. User was temp banned for this post. This is the last post I'll make about sourcing, as I see you're temp banned and I'm not sure if it borders on off-topic, but encyclopedias are generally good starting points for when people want to start researching topics, since they often times have an extensive bibliography for more information. And Wikipedia is no different, in that most long entries have dozens- if not hundreds- of works that are cited, and you'll immediately know if any pages aren't well-sourced. Keep in mind that it was established back in 2005 that Wikipedia's accuracy was comparable to Encyclopedia Britannica's ( https://www.cnet.com/news/study-wikipedia-as-accurate-as-britannica/ ) and Wikipedia has only become more reliable over the past 13 years (despite the taboo that comes with public editors). When doing real research, of course you're going to double-check your sources against other sources, but starting at a Wiki entry for basic overviews and looking through the bibliography is actually a pretty good informational springboard. In other words, it's completely inappropriate to automatically dismiss statements just because they exist on Wikipedia. Furthermore, I'm well aware of the mathematics and statistics references you're making (e.g., correlation), considering I teach high school and college math and statistics. You're not the only one who understands confounding variables. And I'm trying to have a dialogue with you- not get into a dick-swinging contest. The reason I had responded in the first place to your rebuttal of someone else's post was that you had tried refuting a study that attempted to control for certain variables with your own graph that specifically didn't, and you tried saying that since you found a meaningless graph, that someone else's graph was automatically equally meaningless. I found that to be disingenuous, along with your "lol Wiki is auto-wrong and anyone who uses Wiki is stupid" philosophy, and then other people ended up citing even more studies. I don't feel like your one-liner dismissive responses to some of these studies are really all that convincing, and calling people "amateurish" or saying that you are "talking out of your own ass" isn't really strengthening your arguments. firearm bans had no discernible effect on total homicide rates. This statement that you made to ticklishmusic is controversial, as there is plenty of research that disagrees with your claim: We analyzed the relationship between homicide and gun availability using data from 26 developed countries from the early 1990s. We found that across developed countries, where guns are more available, there are more homicides. These results often hold even when the United States is excluded.
Hemenway, David; Miller, Matthew. Firearm availability and homicide rates across 26 high income countries. Journal of Trauma. 2000; 49: 985-88.
Using survey data on rates of household gun ownership, we examined the association between gun availability and homicide across states, 2001-2003. We found that states with higher levels of household gun ownership had higher rates of firearm homicide and overall homicide. This relationship held for both genders and all age groups, after accounting for rates of aggravated assault, robbery, unemployment, urbanization, alcohol consumption, and resource deprivation (e.g., poverty). There was no association between gun prevalence and non-firearm homicide.
Miller, Matthew; Azrael, Deborah; Hemenway, David. State-level homicide victimization rates in the U.S. in relation to survey measures of household firearm ownership, 2001-2003. Social Science and Medicine. 2007; 64:656-64. https://www.hsph.harvard.edu/hicrc/firearms-research/guns-and-death/ The correlations here seem to indicate: 1. As the number of guns increases, so does the number of gun deaths; 2. As the number of guns increases, so does the number of overall deaths; 3. The number of guns doesn't seem to affect the number of non-firearm deaths (i.e., deaths due to other weapons aren't increasing as the number of guns changes, disputing the idea that people will just kill with other weapons if they can't get their hands on a gun). You claim to teach statistics and understand statistics but don't know that Negative Binomial Regression Models are utilized for modeling count variables? Bro, I didn't even major in statistics and I know that. You would pretty much never utilize Negative Binomial Regression Models for rates, which is exactly what that study does. You would utilize a poisson regression model if you were using rate data, but the study doesn't do that on purpose because it would blow up their claims. There are two things in here that I'd like to contest, "bro". 1. Conversions exist between counts and rates, so just because one starts with a negative binomial regression model doesn't mean the data can't be interpreted either in terms of counts or in terms of rates. Conventionally, yes, NBR is used when you start with counts, although Poissons certainly aren't only used for rates. Furthermore, there are special cases of negative binomial regression models (e.g., Pascal and Polya distributions) that tend to be preferable over Poisson counterparts, as they can be made more accurate by accounting for different means and standard deviations. NBR and Poisson are not mutually exclusive. In fact, here's another example of the same kind of comparative modeling happening, and you'll notice that this source considers both NBR and Poisson to be comparable as count models (i.e., your claim that Poisson = rates is not necessarily true), yet also allows the data to be interpreted as rates: Also, the negative binomial model, as compared to other count models (i.e., Poisson or zero-inflated models), is assumed the appropriate model. In other words, we assume that the dependent variable is over-dispersed and does not have an excessive number of zeros. The first half of this page interprets the coefficients in terms of negative binomial regression coefficients, and the second half interprets the coefficients in terms of incidence rate ratios. https://stats.idre.ucla.edu/stata/output/negative-binomial-regression/In other words, the study isn't automatically refuted simply because they're using a different model, because you can totally interpret the data in both ways. 2. I'm curious as to what evidence you have that the other model would "blow up their claims". Please elaborate. 1) Conversions weren't used here; they clearly had an agenda and utilized the wrong model. This was not a special case. 2) When you rerun the numbers and actually convert and ensure that you're using count data and not rate data, the graph will look more like this Which basically shows a pretty normal distribution. Aka their study is a bunch of bullshit. This is exactly why so many people on the 'other side' of the argument feel like 'your' side wants to take away firearms. You have people go as far as basically fudge the numbers of a study in order to 'prove' their agenda. OK, I wasn't necessarily going to indulge in replying but... 1. One does not simply 'visually' assess the 'Gaussianity' or the 'normality' (sic) of a distribution just by optics and zooming on a graph. Trust me on that. If this was the case, simple bread and butter tools like 2-sample tests (Kolmogorov-Smirnov, AD en.wikipedia.org) or normality tests (a ton of them... en.wikipedia.org ) would simply not exist. I see marginal distributions that look exactly the same but come from totally different and separable classes everyday. For more on this phenomenon, see Anscombe's Quartet ( en.wikipedia.org). It's hence very difficult to take any claim that follows from graphing a PDF followed by a 'pretty' statement, let alone a 'bullshit' one, without a heap of scientific salt. 1bis. Anyone who refers to a distribution as 'pretty normal' doesn't seem bothered about what 'normal' in stats or probability actually means, how ubiquitous Gaussians are, and hence unlikely to be able to make correct inferences - this is an orange flag in itself. 2. You are using verbatim a random person's letter, that was refused in a scientific journal, as a full-on rebuttal to a peer review. I'm not fond of ad hominems but regardless of that person's credentials, 'peer reviewed' means several peers, ie, a full committee of scientists reviewing independently of each other and of authors' identity. Scientific work is judged by committee consensus ; any single individual's opinion is less relevant. 2bis. Every single book on advertised crimeresearch.org has a pro-guns bias - this doesn't strike me as extremely bipartisan. How can this statistically be a coincidence, if the evidence from data is so ambiguous ? Incidentally, John Lott, the owner of the website, is a well known gun advocate. ( en.wikipedia.org ) 3. When you replied to me 'their findings were that gun ownership increased gun homicide by 0.9%' , you left out 'for every 1% extra gun ownership'. This is an important omission and a hell of a difference ! Sensitivity analysis has been around in engineering and stats methods ever since the good old days of the Taylor expansion - more than three centuries. Entire industries are built on it. Now if you forget the 1/100 factor... Especially in a setting where we're merely attempting to establish the sign of the correlation. 4. Don't be gung-ho on NB vs Poisson - neglecting the correlation between numerator and denominator in Slutsky's lemma (which will result in mishandling skew) often can be a reasonable assumption asymptotically. The point is not worth the triple repetition you've made of it. That guy litterally says he finds the opposite sign correlation as the authors ('I found a one percentage point increase FS/S produced a 1.2% increase in firearm homicides, but the point estimate using the BRFSS survey data implied the same change produced a 1.2% decrease.). I would focus on that instead if I were you, or him. 5. Even if you were to be gung ho about NB vs Poisson, the study I quoted explicitly controls for it. ('. Use of a Poisson rather than a negative binomial model did not alter the results.'.). You're litterally saying you choose not to believe the authors, which is your prerogative. 6. With all due respect, the fact you mention a six variables model as 'laughable' shows little experience with industry-wide stats or machine learning, let alone on a small dataset. People very often work classifiers or regressors off two or three variables, after a phase of dimensionality reduction. The ubiquitous PCA, or embeddings such as t-SNE or UMAP are always best visualized in 2 or 3d and a strong control for the risk of overcooking your model. We're not training a deep network with 100 million parameters here, we're making do with small data, due to the federal ban on gun control research. 7. This is important because in particular, even establishing all relevant variables doesn't mean they can all be acted upon. When you mention poverty as an important factor in determining gun crime, you are right in that it is statistically important to establish, but impossible to simply legislate away. I also want to make use of a couple Occam Razor's points to finish. I posted one study that you dismissed as fridge science, very much like the previous Wiki-based post. But there is in fact a corpus of relevant literature that points in the same direction - dozens of journal articles. This has all been independently peer reviewed. Will we engage in similar dismissal of all of these, ie, all editorial boards simultaneously in America are ignorant or crooks, or worse, they all don't know the first thing about their negative binomial from their Poisson ? You seem to be a knowledgeable poster and I appeal to your judgement here. I won't do this for every study out there ; just like DPB I reckon www.hsph.harvard.edu should be quality and unbiased enough literature review for other posters to inform their own opinion. Simple question, why would they NOT use count data when it is available and use rate data? By standard that's what is utilized in negative binomial regressions, and the data was clearly available to them. It is also in fact the standard for most gun control studies I have seen whether they lean one way or the other; it is very uncommon for a study to go about the way they did things here. There's no reasonable explanation as to why they performed the study the way they did. It's the same reason why I don't bring up Kleck's self-defense gun study very often because he has some very serious issues with how he presents his arguments and how he shows his statistics (some of them don't even add up) https://pdfs.semanticscholar.org/91da/afbf92d021f06426764e800a4e639a1c1116.pdfUltimately the real point is that you can't just make sweeping statements about correlation off of such stats work when the stats are so limited. Where I would agree with the other side is that there needs to be far more in depth fair research on what we can do to limit gun violence. Most studies basically rig their numbers one way or another. Also, just because your paper was published and peer reviewed doesn't mean it can't be rigged and biased from the start. Arthur Kellerman's paper over gun control is a perfect example of that. http://www.nejm.org/doi/full/10.1056/NEJM199310073291506#t=abstract If your response to two lenghty posts written by people clearly knowledgable in the subject is "but you see, I think the study was done in a way that I think is wrong, and also all studies are rigged anyway", then I can't take you seriously any longer. Damn shame, because for a moment it seemed like you had something interesting to say.
MLL still hasn't answered why they would use count data when the count data was available; it clearly does skew the results somewhat when you are using rate data in a way that is favorable to a gun control claim. This isn't even "I think it's wrong" it goes against most statistical studies and MLL knows this. He's just arguing semantics at this point.
DPP's point about using negative binomial regressions with rate data is generally in regards to exceptions, not the general rule. This particular study is not an exception.
|
On February 23 2018 21:42 superstartran wrote:Show nested quote +On February 23 2018 16:32 PoulsenB wrote:On February 23 2018 07:49 superstartran wrote:On February 23 2018 04:43 MyLovelyLurker wrote:On February 23 2018 01:06 superstartran wrote:On February 23 2018 00:43 DarkPlasmaBall wrote:On February 23 2018 00:15 superstartran wrote:On February 22 2018 22:30 DarkPlasmaBall wrote:On February 22 2018 20:48 superstartran wrote:On February 22 2018 19:08 DarkPlasmaBall wrote:
[quote]
superstartran thinks Wiki info is pretty much the worst ever- his strong reaction to it previously was one of the reasons he was banned in the first place- so that might not be a strong selling point to him. (Of course, that entry has over 100 citations and looks to be reasonable overall. Thanks for sharing it.) I like how you were so up all on that one study that uses the wrong data for a negative binomial regression, which is a pretty amateurish mistake to do (or you do it on purpose to push an agenda). So, like I said, are you going to respond why that study used rate data instead of count data? Are you referring to MLL's post? This one? + Show Spoiler +On February 20 2018 07:49 MyLovelyLurker wrote:I guess I'm a bit late on the whole 'let's dl wikipedia into a Jupyter notebook and correlate away' thing I was planning to do this morning, and happy that American stats exceptionalism is alive and well. I want to share this study www.ncbi.nlm.nih.gov that rigorously establishes correlation between gun ownership and firearm homicide rates within the 50 States. It's based on 2012 data. I didn't see anything fundamentally wrong with its scientific methodology. Interestingly enough, looking at the bibliography shows a grand total of eighty references, many of them pointing in a similar direction. Sharing the discussion / conclusion as : 'To the best of our knowledge, ours is the most up-to-date and comprehensive analysis of the relationship between firearm ownership and gun-related homicide rates within the 50 states. Our study encompassed a 30-year period, with data through 2010, and accounted for 18 possible confounders of the relationship between gun ownership and firearm homicide. We found a robust relationship between higher levels of gun ownership and higher firearm homicide rates that was not explained by any of these potential confounders and was not sensitive to model specification. Our work expanded on previous studies not only by analyzing more recent data, but also by adjusting for clustering by year and state and controlling for factors, such as the rate of nonfirearm homicides, that likely capture unspecified variables that may be associated with both gun ownership levels and firearm homicide rates. The correlation of gun ownership with firearm homicide rates was substantial. Results from our model showed that a 1-SD difference in the gun ownership proxy measure, FS/S, was associated with a 12.9% difference in firearm homicide rates. All other factors being equal, our model would predict that if the FS/S in Mississippi were 57.7% (the average for all states) instead of 76.8% (the highest of all states), its firearm homicide rate would be 17% lower. Because of our use of a proxy measure for gun ownership, we could not conclude that the magnitude of the association between actual household gun ownership rates and homicide rates was the same. However, in a model that incorporated only survey-derived measures of household gun ownership (for 2001, 2002, and 2004), we found that each 1-SD difference in gun ownership was associated with a 24.9% difference in firearm homicide rates. 'Our study substantially advances previous work by analyzing recent data, examining the longest and most comprehensive panel of state-specific data to date, and accounting for year and state clustering and for a wide range of potential confounders. We found a robust relationship between gun ownership and firearm homicide rates, a finding that held whether firearm ownership was assessed through a proxy or a survey measure, whether state clustering was accounted for by GEEs or by fixed effects, and whether or not gun ownership was lagged, by up to 2 years. The observed relationship was specific to firearm-related homicide. Although we could not determine causation, we found that states with higher levels of gun ownership had disproportionately large numbers of deaths from firearm-related homicides.' What evidence do you have that they used the wrong data? Why do you think that rate data can't be used and that count data must be used? And I don't know what you mean by "all up on that one study". I merely said it looked solid and I was interested in hearing your response to it (although you were temp banned at the time). You just laughed it off though, which I had seen you do before. That was what prompted me to reply to your comment in the first place. Here was our conversation: On February 19 2018 10:25 DarkPlasmaBall wrote:On February 19 2018 07:05 superstartran wrote:
[quote]
Are you seriously trying to use outdated data that doesn't control for various different variables as a reliable source to claim that the data shown shows a strong correlation?
"Well Sourced"
lmao.
Wikipedia 'likely to be legitimate'
Sometimes I wonder if you guys actually graduated from a university and were taught basic scientific method or you are just talking out of your own ass.
User was temp banned for this post. This is the last post I'll make about sourcing, as I see you're temp banned and I'm not sure if it borders on off-topic, but encyclopedias are generally good starting points for when people want to start researching topics, since they often times have an extensive bibliography for more information. And Wikipedia is no different, in that most long entries have dozens- if not hundreds- of works that are cited, and you'll immediately know if any pages aren't well-sourced. Keep in mind that it was established back in 2005 that Wikipedia's accuracy was comparable to Encyclopedia Britannica's ( https://www.cnet.com/news/study-wikipedia-as-accurate-as-britannica/ ) and Wikipedia has only become more reliable over the past 13 years (despite the taboo that comes with public editors). When doing real research, of course you're going to double-check your sources against other sources, but starting at a Wiki entry for basic overviews and looking through the bibliography is actually a pretty good informational springboard. In other words, it's completely inappropriate to automatically dismiss statements just because they exist on Wikipedia. Furthermore, I'm well aware of the mathematics and statistics references you're making (e.g., correlation), considering I teach high school and college math and statistics. You're not the only one who understands confounding variables. And I'm trying to have a dialogue with you- not get into a dick-swinging contest. The reason I had responded in the first place to your rebuttal of someone else's post was that you had tried refuting a study that attempted to control for certain variables with your own graph that specifically didn't, and you tried saying that since you found a meaningless graph, that someone else's graph was automatically equally meaningless. I found that to be disingenuous, along with your "lol Wiki is auto-wrong and anyone who uses Wiki is stupid" philosophy, and then other people ended up citing even more studies. I don't feel like your one-liner dismissive responses to some of these studies are really all that convincing, and calling people "amateurish" or saying that you are "talking out of your own ass" isn't really strengthening your arguments. firearm bans had no discernible effect on total homicide rates. This statement that you made to ticklishmusic is controversial, as there is plenty of research that disagrees with your claim: We analyzed the relationship between homicide and gun availability using data from 26 developed countries from the early 1990s. We found that across developed countries, where guns are more available, there are more homicides. These results often hold even when the United States is excluded.
Hemenway, David; Miller, Matthew. Firearm availability and homicide rates across 26 high income countries. Journal of Trauma. 2000; 49: 985-88.
Using survey data on rates of household gun ownership, we examined the association between gun availability and homicide across states, 2001-2003. We found that states with higher levels of household gun ownership had higher rates of firearm homicide and overall homicide. This relationship held for both genders and all age groups, after accounting for rates of aggravated assault, robbery, unemployment, urbanization, alcohol consumption, and resource deprivation (e.g., poverty). There was no association between gun prevalence and non-firearm homicide.
Miller, Matthew; Azrael, Deborah; Hemenway, David. State-level homicide victimization rates in the U.S. in relation to survey measures of household firearm ownership, 2001-2003. Social Science and Medicine. 2007; 64:656-64. https://www.hsph.harvard.edu/hicrc/firearms-research/guns-and-death/ The correlations here seem to indicate: 1. As the number of guns increases, so does the number of gun deaths; 2. As the number of guns increases, so does the number of overall deaths; 3. The number of guns doesn't seem to affect the number of non-firearm deaths (i.e., deaths due to other weapons aren't increasing as the number of guns changes, disputing the idea that people will just kill with other weapons if they can't get their hands on a gun). You claim to teach statistics and understand statistics but don't know that Negative Binomial Regression Models are utilized for modeling count variables? Bro, I didn't even major in statistics and I know that. You would pretty much never utilize Negative Binomial Regression Models for rates, which is exactly what that study does. You would utilize a poisson regression model if you were using rate data, but the study doesn't do that on purpose because it would blow up their claims. There are two things in here that I'd like to contest, "bro". 1. Conversions exist between counts and rates, so just because one starts with a negative binomial regression model doesn't mean the data can't be interpreted either in terms of counts or in terms of rates. Conventionally, yes, NBR is used when you start with counts, although Poissons certainly aren't only used for rates. Furthermore, there are special cases of negative binomial regression models (e.g., Pascal and Polya distributions) that tend to be preferable over Poisson counterparts, as they can be made more accurate by accounting for different means and standard deviations. NBR and Poisson are not mutually exclusive. In fact, here's another example of the same kind of comparative modeling happening, and you'll notice that this source considers both NBR and Poisson to be comparable as count models (i.e., your claim that Poisson = rates is not necessarily true), yet also allows the data to be interpreted as rates: Also, the negative binomial model, as compared to other count models (i.e., Poisson or zero-inflated models), is assumed the appropriate model. In other words, we assume that the dependent variable is over-dispersed and does not have an excessive number of zeros. The first half of this page interprets the coefficients in terms of negative binomial regression coefficients, and the second half interprets the coefficients in terms of incidence rate ratios. https://stats.idre.ucla.edu/stata/output/negative-binomial-regression/In other words, the study isn't automatically refuted simply because they're using a different model, because you can totally interpret the data in both ways. 2. I'm curious as to what evidence you have that the other model would "blow up their claims". Please elaborate. 1) Conversions weren't used here; they clearly had an agenda and utilized the wrong model. This was not a special case. 2) When you rerun the numbers and actually convert and ensure that you're using count data and not rate data, the graph will look more like this Which basically shows a pretty normal distribution. Aka their study is a bunch of bullshit. This is exactly why so many people on the 'other side' of the argument feel like 'your' side wants to take away firearms. You have people go as far as basically fudge the numbers of a study in order to 'prove' their agenda. OK, I wasn't necessarily going to indulge in replying but... 1. One does not simply 'visually' assess the 'Gaussianity' or the 'normality' (sic) of a distribution just by optics and zooming on a graph. Trust me on that. If this was the case, simple bread and butter tools like 2-sample tests (Kolmogorov-Smirnov, AD en.wikipedia.org) or normality tests (a ton of them... en.wikipedia.org ) would simply not exist. I see marginal distributions that look exactly the same but come from totally different and separable classes everyday. For more on this phenomenon, see Anscombe's Quartet ( en.wikipedia.org). It's hence very difficult to take any claim that follows from graphing a PDF followed by a 'pretty' statement, let alone a 'bullshit' one, without a heap of scientific salt. 1bis. Anyone who refers to a distribution as 'pretty normal' doesn't seem bothered about what 'normal' in stats or probability actually means, how ubiquitous Gaussians are, and hence unlikely to be able to make correct inferences - this is an orange flag in itself. 2. You are using verbatim a random person's letter, that was refused in a scientific journal, as a full-on rebuttal to a peer review. I'm not fond of ad hominems but regardless of that person's credentials, 'peer reviewed' means several peers, ie, a full committee of scientists reviewing independently of each other and of authors' identity. Scientific work is judged by committee consensus ; any single individual's opinion is less relevant. 2bis. Every single book on advertised crimeresearch.org has a pro-guns bias - this doesn't strike me as extremely bipartisan. How can this statistically be a coincidence, if the evidence from data is so ambiguous ? Incidentally, John Lott, the owner of the website, is a well known gun advocate. ( en.wikipedia.org ) 3. When you replied to me 'their findings were that gun ownership increased gun homicide by 0.9%' , you left out 'for every 1% extra gun ownership'. This is an important omission and a hell of a difference ! Sensitivity analysis has been around in engineering and stats methods ever since the good old days of the Taylor expansion - more than three centuries. Entire industries are built on it. Now if you forget the 1/100 factor... Especially in a setting where we're merely attempting to establish the sign of the correlation. 4. Don't be gung-ho on NB vs Poisson - neglecting the correlation between numerator and denominator in Slutsky's lemma (which will result in mishandling skew) often can be a reasonable assumption asymptotically. The point is not worth the triple repetition you've made of it. That guy litterally says he finds the opposite sign correlation as the authors ('I found a one percentage point increase FS/S produced a 1.2% increase in firearm homicides, but the point estimate using the BRFSS survey data implied the same change produced a 1.2% decrease.). I would focus on that instead if I were you, or him. 5. Even if you were to be gung ho about NB vs Poisson, the study I quoted explicitly controls for it. ('. Use of a Poisson rather than a negative binomial model did not alter the results.'.). You're litterally saying you choose not to believe the authors, which is your prerogative. 6. With all due respect, the fact you mention a six variables model as 'laughable' shows little experience with industry-wide stats or machine learning, let alone on a small dataset. People very often work classifiers or regressors off two or three variables, after a phase of dimensionality reduction. The ubiquitous PCA, or embeddings such as t-SNE or UMAP are always best visualized in 2 or 3d and a strong control for the risk of overcooking your model. We're not training a deep network with 100 million parameters here, we're making do with small data, due to the federal ban on gun control research. 7. This is important because in particular, even establishing all relevant variables doesn't mean they can all be acted upon. When you mention poverty as an important factor in determining gun crime, you are right in that it is statistically important to establish, but impossible to simply legislate away. I also want to make use of a couple Occam Razor's points to finish. I posted one study that you dismissed as fridge science, very much like the previous Wiki-based post. But there is in fact a corpus of relevant literature that points in the same direction - dozens of journal articles. This has all been independently peer reviewed. Will we engage in similar dismissal of all of these, ie, all editorial boards simultaneously in America are ignorant or crooks, or worse, they all don't know the first thing about their negative binomial from their Poisson ? You seem to be a knowledgeable poster and I appeal to your judgement here. I won't do this for every study out there ; just like DPB I reckon www.hsph.harvard.edu should be quality and unbiased enough literature review for other posters to inform their own opinion. Simple question, why would they NOT use count data when it is available and use rate data? By standard that's what is utilized in negative binomial regressions, and the data was clearly available to them. It is also in fact the standard for most gun control studies I have seen whether they lean one way or the other; it is very uncommon for a study to go about the way they did things here. There's no reasonable explanation as to why they performed the study the way they did. It's the same reason why I don't bring up Kleck's self-defense gun study very often because he has some very serious issues with how he presents his arguments and how he shows his statistics (some of them don't even add up) https://pdfs.semanticscholar.org/91da/afbf92d021f06426764e800a4e639a1c1116.pdfUltimately the real point is that you can't just make sweeping statements about correlation off of such stats work when the stats are so limited. Where I would agree with the other side is that there needs to be far more in depth fair research on what we can do to limit gun violence. Most studies basically rig their numbers one way or another. Also, just because your paper was published and peer reviewed doesn't mean it can't be rigged and biased from the start. Arthur Kellerman's paper over gun control is a perfect example of that. http://www.nejm.org/doi/full/10.1056/NEJM199310073291506#t=abstract If your response to two lenghty posts written by people clearly knowledgable in the subject is "but you see, I think the study was done in a way that I think is wrong, and also all studies are rigged anyway", then I can't take you seriously any longer. Damn shame, because for a moment it seemed like you had something interesting to say. MLL still hasn't answered why they would use count data when the count data was available; it clearly does skew the results somewhat when you are using rate data in a way that is favorable to a gun control claim. This isn't even "I think it's wrong" it goes against most statistical studies and MLL knows this. He's just arguing semantics at this point. DPP's point about using negative binomial regressions with rate data is generally in regards to exceptions, not the general rule. This particular study is not an exception.
Are you fucking seriously asking why rate data is preferred over count data??!
|
I think it has been clearly established that SST is not really arguing in good faith. I for one am not willing to invest any more effort into him.
He quite clearly attempts to shift the burden of proof entirely upon you, never provides any evidence whatsoever for what he believes in (He doesn't even actually state an exact position), and will increase the burden of proof ever further until nothing can possibly be satisfactory. First it was "studies, any studies". Studies were provided. Than it was "these studies don't control for enough variables". Further studies were provided. Now it is "Why are they using rate data over count data". If someone provides a study with count data or explains why that is the case in that specific study, he will find some other thing to nitpick about it.
|
On February 23 2018 22:41 Simberto wrote:
I think it has been clearly established that SST is not really arguing in good faith. I for one am not willing to invest any more effort into him.
He quite clearly attempts to shift the burden of proof entirely upon you, never provides any evidence whatsoever for what he believes in (He doesn't even actually state an exact position), and will increase the burden of proof ever further until nothing can possibly be satisfactory. First it was "studies, any studies". Studies were provided. Than it was "these studies don't control for enough variables". Further studies were provided. Now it is "Why are they using rate data over count data". If someone provides a study with count data or explains why that is the case in that specific study, he will find some other thing to nitpick about it.
im surprised so many people are still trying. i caught on pretty quick
|
On February 23 2018 23:12 evilfatsh1t wrote:Show nested quote +On February 23 2018 22:41 Simberto wrote:
I think it has been clearly established that SST is not really arguing in good faith. I for one am not willing to invest any more effort into him.
He quite clearly attempts to shift the burden of proof entirely upon you, never provides any evidence whatsoever for what he believes in (He doesn't even actually state an exact position), and will increase the burden of proof ever further until nothing can possibly be satisfactory. First it was "studies, any studies". Studies were provided. Than it was "these studies don't control for enough variables". Further studies were provided. Now it is "Why are they using rate data over count data". If someone provides a study with count data or explains why that is the case in that specific study, he will find some other thing to nitpick about it. im surprised so many people are still trying. i caught on pretty quick
The point isn't so much to convince the hold-out arguing in nit-picky bad faith rather than it is to show the audience that hiding behind arbitrary tenets of stats only works if no one else understands them. Both MLL and DPB put a lot of work into their posts and the thread is better for it
|
i think we should ban all assault rifles
wait are you saying assault is a verb not a gun? well i guess i lost this debate haha see ya next time! got me good there!
On February 23 2018 15:16 Doodsmack wrote:
There's no need to refer to statistics when the logic is so clear: more guns equals more safety.
May I please join you in your imaginary dream world where there are no accidental gun deaths?
|
On February 24 2018 10:32 ninazerg wrote:i think we should ban all assault rifles wait are you saying assault is a verb not a gun? well i guess i lost this debate haha see ya next time! got me good there! Show nested quote +On February 23 2018 15:16 Doodsmack wrote:
There's no need to refer to statistics when the logic is so clear: more guns equals more safety. May I please join you in your imaginary dream world where there are no accidental gun deaths?
You see had there simply been someone there to gun down the toddler before they accidentally shot their friend/family member/self it all could have been prevented.
|
United States42004 Posts
On February 24 2018 10:39 GreenHorizons wrote:Show nested quote +On February 24 2018 10:32 ninazerg wrote:i think we should ban all assault rifles wait are you saying assault is a verb not a gun? well i guess i lost this debate haha see ya next time! got me good there! On February 23 2018 15:16 Doodsmack wrote:
There's no need to refer to statistics when the logic is so clear: more guns equals more safety. May I please join you in your imaginary dream world where there are no accidental gun deaths? You see had there simply been someone there to gun down the toddler before they accidentally shot their friend/family member/self it all could have been prevented.
What if the toddler was the good toddler with a gun who was ready to intervene when the bad toddler with a gun was about to kill itself.
|
Honestly I don't think owning guns is a big problem, yes there are countries doing perfectly ok with guns.
But US is clearly not doing ok and this alone should be enough to ban guns.
|
|
|
|
|
|
EPT
