Bayesian Analysis of US Mass Shooters.

Jeremiah

Research Objective:

National discourse has been greatly focused on the laws governing gun ownership. While
most Americans agree that background checks could be done more efficiently for gun ownership,
they are hotly divided on the topic of raising the legal age of gun ownership. While the scope of
such a change should be considered in a wider view, the current public focus is on mass shooters.
So, my aim here is to gain a fuller understanding of the possible potential impact of raising the
age limit for gun ownership to the age of 21 in relation to the occurrence of mass shootings in
America. An event will be considered a mass shooting if there are at least three victims involved
and a victim will be defined as someone killed or injured by the shooter(s). The central focus of
this analysis will be the mean age range of a mass shooter, the likelihood future mass shooting
will be carried out by an individual under the age of 21 and the likelihood the weapon used will
be acquired legally. The purpose of this analysis is to provide information and not to conclude a
course of action.

https://drive.google.com/file/d/1eO41TW5RNI4irDic4F2UeYMnu70BUkC_/view

Feel free to share your opinions and if there are other statistics from that data set you'd like to see, it should not be very hard for me to punch something up.

fdrake

↪Jeremiah

Would be nice if you incorporated the age groups from the data directly with a normal random effect, also an age group / shooting site interaction is unmodelled variation in the data, so that's a limitation you could improve on if you're bored. For the latter an uninformative prior probably wouldn't behave so well due to the amount of data; but that's a whole can of worms you might not want to get into for a project. These are probably out with the scope of the module, though.

Over all it looks good. I would've liked to see a small discussion of what data you think you'd need to start drawing tentative qualitative conclusions about the effect of changing the ownership age. If you're dealing with these evidence based policy type questions, ideally you'd be consulted before the data was collected, so it's nice to have an idea of what you'd need to be of use to the policymaker.

Jeremiah

↪fdrake

To be honest if I look at these data some more, or similar data it will be in a classical framework. I need to practice Bayesian methods, as to be well rounded, but I disagree with it philosophically.

I did consider doing something with the location in a model, but I am not sure that will happen. I think I need to move this inquiry to border terms and look at gun violence over all. The focus on mass shootings was to appeal to current public discourse, but I don't feel it is the right place to look for the full picture. Gun violence overall is what we should be looking at. When I find a decent data set I'll start looking at that.

fdrake

↪Jeremiah

Aye. It's good to be able to speak both languages, sometimes as a matter of historical contingency or innate affinity something can be much easier in Bayes than otherwise (spatiotemporal modelling with gaussian variables for the former and 'pooling' of categorical effects for the latter). Regardless, frequentist methods (also including empirical Bayes) are more generally accepted outside of statistics and are a vital part of paper publication rituals.

UCL has a huge dataset from US census data with income and ethnic composition variables over some of the years in your study - if it's possible to aggregate the two together you could ask some really interesting exploratory questions.

I'd link to it but I can't remember where I found it. Probably on the machine learning repository.

fdrake

↪Jeremiah

I was thinking about how you wrote things. Unless you have a tiny word limit, you could spend more time demonstrating your understanding. Why do your trace plots show convergence? Why did you choose Gibbs sampling over other MCMC methods? Why did you use simple linear regression rather than multilinear regression etc...

Edit: from the style of it, it doesn't seem to be an executive summary style assignment, is why I'm including these things.

Edit2: also learn Latex. It's a lot better for formatting mathematics.

Jeremiah

↪fdrake

It was written with Latex.

fdrake

↪Jeremiah

Fair enough! It looked a lot like .docx -> pdf.

Jeremiah

↪fdrake

Ya, I used R Markdown, which allows you to add in Latex code. Then I exported to a Word Doc, but when hosting a Word Doc on Google Drive, for reasons I don't know, it does not display the Latex when viewed in the browser, so I converted to a PDF for the Google Drive link.

fdrake

↪Jeremiah

Strange. Will remember to avoid that route in the future.

Jeremiah

↪fdrake

Well I appreciate the input thus far, perhaps you could help with a dilemma in another data set I am looking at.

fdrake

↪Jeremiah

What's the dilemma?

Jeremiah

↪fdrake

It is gun violence data, and the values in each observation is formatted as such:

0::Female||1::Male||2::Male||3::Male

The number indicates the participant in the event, short of splitting each participant into their own column for each related variable, I am not sure how else to address this formatting.

fdrake

↪Jeremiah

Look up regular expressions in R. It has a lot of text processing power.

Jeremiah

↪fdrake

Thanks, I'll start reading.

fdrake

↪Jeremiah

strsplit might be a good start.

Jeremiah

↪fdrake

I am already familiar with strsplit, that's what I was intending to use to split them into separate columns if I had to. I was just wondering if there was a different way, as if I split them into their own columns then any modeling or graphs I make needs to account for ages in multiple different columns. Maybe I'll just split them and work through it.

Bayesian Analysis of US Mass Shooters.

Welcome to The Philosophy Forum!

Categories

More Discussions