31:T5c7,America leads the world in [mass shootings](https://www.wsj.com/articles/u-s-leads-world-in-mass-shootings-1443905359). And over time they have become more frequent. What I wanted to know was how much deadlier have assault weapons made them. But first let's start with the definition of a mass shooting. For this analysis I used data from [The Violence Project](https://www.theviolenceproject.org). They use the Congressional Research Service's definition of a mass shooting - "_a multiple homicide incident in which four or more victims are murdered with firearms—not including the offender(s)—within one event, and at least some of the murders occurred in a public location or locations in close geographical proximity (e.g., a workplace, school, restaurant, or other public settings), and the murders are not attributable to any other underlying criminal activity or commonplace circumstance (armed robbery, criminal competition, insurance fraud, argument, or romantic triangle)._" This is the most conservative of the definitions and why you may notice there are so few mass shootings listed compared to other databases. This database goes back to 1966 with one of the first modern day mass shootings at the [University of Texas clock tower](https://en.wikipedia.org/wiki/University_of_Texas_tower_shooting). However, when I broke down weapons into categories there was a trend that jumped out at me - the use of high-capacity magazines. _(Mouse over for details)_32:T5a2,html`

Weapon Comparison	Adjusted p-value
Assault - High Capacity Magazine	0.9322789
Assault - Handgun	0.0021474
Assault - Multiple Weapons	0.8264963
Assault - Shotgun	0.5039609
Assault - Rifle	0.5219983
High Capacity Magazine - Handgun	0.0000009
High Capacity Magazine - Multiple Weapons	0.0190770
High Capacity Magazine - Rifle	0.0180191
High Capacity Magazine - Shotgun	0.0240454
Handgun - Rifle	1.0000000
Handgun - Shotgun	1.0000000
Multiple Weapons - Handgun	0.0376475
Multiple Weapons - Rifle	1.0000000
Multiple Weapons - Shotgun	1.0000000
Rifle - Shotgun	0.7588873

`33:T4f8,It's challenging to visualize the differences between the groups because the data is not normally distributed, making it hard to clearly spot patterns. Many statistical tools depend on the normal distribution, or [bell curve](https://en.wikipedia.org/wiki/Normal_distribution), of data in order to work. One of the clever tricks Dunn's Test uses to address this is to instead rank the data. In this case we would take the shooting with the least amount of casualties and rank it first. Then the next largest number of casualties would be ranked second. We continue this process until all the data has been ranked. Dunn's test then makes comparisons across all the groups to see where there is a statistical difference, which is what you saw in the chart above. In the graph below I visualized the ranks. With this method the differences start to become a bit more clear. You can see, for example, that the bulk of the casualties by handgun are ranked lower that most of the other weapons. Where as the ranks of casualties involving high capacity magazines is higher. You can toggle the graph to change to y-axis from ranking to the number of casualties. These patterns are still there, but it much harder to compare. _(You can also mouse over this for details as well)_34:T5b0,chart = { const svg = d3.select(DOM.svg(width, height)); var beeswarm = capacity.map(d => ({name: d.magazine, value: d.rank, rank: d.rank, casualty: d.casualty, city: d.city, state: d.state, year: d.year, killed: d.killed})); var circle = d3.rollup(capacity, v => v.map(g => g.rank), d => d.magazine); svg.append("g") .selectAll("g") .data([...circle]) .join("g") .attr("class" , "bins") .attr("transform", d => `translate(${x(d[0]) + x.bandwidth()/2},0)`) .selectAll("circle") .data(d => dodge(beeswarm.filter(n => n.name == d[0]), radius * 4 + padding)) .join("circle") .attr("cy", d => d.y) .attr("cx", d => d.x) .attr("r", radius) .attr("fill", d => color(d.data.name)) .attr("stroke" , "white") .attr("opacity" , 0.9); svg.append("g") .call(xAxis); svg.append("g") .call(yAxis); const tooltip = svg.append("g"); svg .selectAll("circle") .on("touchmove mousemove", function(event, d) { tooltip.call( callout,`Location: ${d.data.city}, ${d.data.state} Year: ${d.data.year} Rank: ${d.data.rank} Casualties: ${d.data.casualty} Killed: ${d.data.killed}` ); const [mouseX, mouseY] = d3.pointer(event, this); tooltip.attr("transform", `translate(${mouseX + x(d.data.name) + x.bandwidth()/2}, ${mouseY})`); }) .on("touchend mouseleave", function() { tooltip.call(callout, null); }); return svg.node(); }35:T561,Most uses of high-capacity magazines were with an assault weapon. When I compared assault weapons that used high-capacity magazines to those that didn't there was no statistically significant difference. I should note here that for all of the comparisons I removed the Las Vegas shooting of 2017 because it was such an outlier. The only difference it made in analysis was in comparing the magazine sizes of assault weapons. When I keep it in the dataset there was a significantly higher number of casualties with assault weapons that used high-capacity magazines than those that didn't. Take it out and that difference disappears. This is what outliers do, they skew data. One of the ways to deal with that is to remove it, which is what I decided to do. This is also about the limit to how much I can subset the data. When I try to compare other weapons and their use of high-capacity magazines there just isn't enough data. But with the data I do have a picture starts to emerge. Mass shooters are learning from one another. They are seeing what causes the most carnage and they are adapting. And what they have learned is that firing off as may rounds in as little time as possible optimizes destruction. The good news is that this insight can helps us minimize that damage. We will not be able to stop every mass shooter, but we can reduce the harm as much as possible.36:T59a,function update (input) { if (input == "casualties") { var beeswarm = capacity.map(d => ({name: d.magazine, value: d.casualty, rank: d.rank, casualty: d.casualty, city: d.city, state: d.state, year: d.year, killed: d.killed})); var circle = d3.rollup(capacity, v => v.map(g => g.casualty), d => d.magazine); y.domain([0, d3.max(capacity, d => d.casualty)]), d3.selectAll(".y_axis").transition().duration(1000).call(d3.axisLeft(y)), d3.selectAll(".label").transition().duration(1000).text("Casualties"), d3.selectAll(".bins").data([...circle]).selectAll("circle").data(d => dodge(beeswarm.filter(n => n.name == d[0]), radius * 4 + padding)).transition().duration(1000).attr("cy", d => d.y).attr("cx", d => d.x) } else { beeswarm = capacity.map(d => ({name: d.magazine, value: d.rank, rank: d.rank, casualty: d.casualty, city: d.city, state: d.state, year: d.year, killed: d.killed})), circle = d3.rollup(capacity, v => v.map(g => g.rank), d => d.magazine), y.domain([0, d3.max(capacity, d => d.rank)]), d3.selectAll(".y_axis").transition().duration(1000).call(d3.axisLeft(y)), d3.selectAll(".label").transition().duration(1000).text("Rank"), d3.selectAll(".bins").data([...circle]).selectAll("circle").data(d => dodge(beeswarm.filter(n => n.name == d[0]), radius * 4 + padding)).transition().duration(1000).attr("cy", d => d.y).attr("cx", d => d.x) } }37:T550,dodge = (data, radius) => { const radius2 = radius * 2; const circles = data.map(d => ({y: y(d.value), data: d})).sort((a, b) => a.y - b.y); const epsilon = 1e-3; let head = null, tail = null; //Returns true if cricle (x,y) intersects with any circle in the queue function intersects(x,y) { let a = head; while(a) { if (radius2 - epsilon > (a.x - x)**2 + (a.y - y)**2){ return true; } a = a.next; } return false; } //Place each circle sequentially for(const b of circles) { //Remove circles from the queue that can't intersect the new circle b while(head && head.y < b.y - radius2) head = head.next; //Choose the minimum non-intersecting tangent if(intersects(b.x = 0, b.y)){ let a = head; b.x = Infinity; do { let x1 = a.x + Math.sqrt(radius2 - (a.y - b.y)**2); let x2 = a.x - Math.sqrt(radius2 - (a.y - b.y)**2); if(Math.abs(x1) < Math.abs(b.x) && !intersects(x1, b.y)) b.x = x1; if(Math.abs(x2) < Math.abs(b.x) && !intersects(x2, b.y)) b.x = x2; a = a.next; }while(a); } //Add b to queue b.next = null; if(head === null) head = tail = b; else tail = tail.next = b; } return circles; }38:[{"id":"9759af2bbba9823b5d6b4dcf1d6ddf62565565104f93ccedabc82e63d7e5c57178ecde6b4269456831b79aa8f8011adaeb66390f3a945f7754abbdd10a335747","href":"https://static.observableusercontent.com/files/9759af2bbba9823b5d6b4dcf1d6ddf62565565104f93ccedabc82e63d7e5c57178ecde6b4269456831b79aa8f8011adaeb66390f3a945f7754abbdd10a335747","name":"Sheet 1-mass@3.csv","type":"text/csv","lastModified":1659487843602.427,"size":26212}] 2f:{"title":"Mass Shootings In America","cells":[{"id":0,"value":"# Mass Shootings In America","pinned":false,"hidden":false,"mode":"md"},{"id":2909,"value":"$31","pinned":false,"hidden":false,"mode":"md"},{"id":256,"value":"addTooltips(\n Plot.rectY(\n data,\n Plot.binX(\n { y: \"count\" },\n {\n x: \"year\",\n thresholds: 10,\n fill: \"magazine\",\n title: (d) => `${d.magazine}`\n }\n )\n ).plot({\n width: width,\n height: 500,\n x: {\n tickFormat: d3.format(\".0f\"),\n ticks: [\"1960\", \"1970\", \"1980\", \"1990\", \"2000\", \"2010\", \"2020\"],\n label: \"Year\",\n labelAnchor: \"center\"\n },\n y: { ticks: 10, grid: true },\n color: { legend: true }\n })\n)","pinned":false,"hidden":false,"mode":"ojs"},{"id":2933,"value":"Just about any weapon can use a high-capacity magazine, with the exception of a revolver and a cannon and probably a few others I'm forgetting. These magazines allow a firearm to carry more ammunition, allowing a shooter to fire more rounds before having to reload the weapon.\n\nAny weapon with a high capacity magazine made a significant difference in the number of casualties, expect when compared an assault weapon. Both an assault weapon and high-capacity magazines resulted in a similar amount of carnage.\n\nBelow is a comparison table of weapons. I used what's called a [Dunn's Test](https://www.statology.org/dunns-test/) to compare them. I'll spare you the messy datails, but any p-value of 0.05 or below is considered statistically significant, and has been highlighted. ","pinned":false,"hidden":false,"mode":"md"},{"id":2943,"value":"$32","pinned":false,"hidden":false,"mode":"ojs"},{"id":2998,"value":"$33","pinned":false,"hidden":false,"mode":"md"},{"id":3006,"value":"viewof drop = select ({\n options: [\n {label: \"Rank\" , value: \"rank\"},\n {label: \"Casualties\" , value: \"casualties\"},\n ],\n value: \"rank\",\n})","pinned":false,"hidden":false,"mode":"ojs"},{"id":1377,"value":"$34","pinned":false,"hidden":false,"mode":"ojs"},{"id":3198,"value":"$35","pinned":false,"hidden":false,"mode":"md"},{"id":3010,"value":"$36","pinned":false,"hidden":false,"mode":"ojs"},{"id":3048,"value":"update(drop)","pinned":true,"hidden":false,"mode":"ojs"},{"id":2859,"value":"$37","pinned":false,"hidden":false,"mode":"ojs"},{"id":1359,"value":"x = d3.scaleBand()\n .domain(capacity.map(d => d.magazine))\n .range([margin.left, width - margin.right])\n .padding(0.2)","pinned":false,"hidden":false,"mode":"ojs"},{"id":1367,"value":"y = d3.scaleLinear()\n\t .domain(d3.extent(capacity, d => d.rank))\n\t .range([height - margin.bottom, margin.top])\n .nice()","pinned":false,"hidden":false,"mode":"ojs"},{"id":1363,"value":"xAxis = g => g\n .attr(\"transform\", `translate(0,${height - margin.bottom})`)\n .attr(\"class\" , \"text\")\n .call(d3.axisBottom(x))","pinned":false,"hidden":false,"mode":"ojs"},{"id":1374,"value":"yAxis = g => g\n .attr(\"transform\", `translate(${margin.left},0)`)\n .attr(\"class\" , \"y_axis\")\n .call(d3.axisLeft(y))\n .call(g => g.append(\"text\")\n .attr(\"class\" , \"label\")\n .attr(\"transform\", \"rotate(-90)\")\n .attr(\"y\", 15)\n .attr(\"x\" , -height/2+margin.top)\n .attr(\"text-anchor\", \"middle\")\n .attr(\"fill\", \"black\")\n .text(\"Ranking\"));","pinned":false,"hidden":false,"mode":"ojs"},{"id":2022,"value":"callout = (g, value) => {\n if (!value) return g.style(\"display\", \"none\");\n\n g\n .style(\"display\", null)\n .style(\"pointer-events\", \"none\")\n .style(\"font\", \"12px sans-serif\");\n\n const path = g.selectAll(\"path\")\n .data([null])\n .join(\"path\")\n .attr(\"fill\", \"white\")\n .attr(\"stroke\", \"black\");\n\n const text = g.selectAll(\"text\")\n .data([null])\n .join(\"text\")\n .call(text => text\n .selectAll(\"tspan\")\n .data((value + \"\").split(/\\n/))\n .join(\"tspan\")\n .attr(\"x\", 0)\n .attr(\"y\", (d, i) => `${i * 1.5}em`)\n .style(\"font-weight\", (_, i) => i ? null : \"bold\")\n .text(d => d));\n\n const {x, y, width: w, height: h} = text.node().getBBox();\n\n text.attr(\"transform\", `translate(${-w / 2},${15 - y})`);\n path.attr(\"d\", `M${-w / 2 - 10},5H-5l5,-5l5,5H${w / 2 + 10}v${h + 20}h-${w + 20}z`);\n}","pinned":false,"hidden":false,"mode":"ojs"},{"id":1587,"value":"color = d3\n .scaleOrdinal()\n .domain(capacity, d => d.magazine)\n .range(d3.schemeDark2)","pinned":false,"hidden":false,"mode":"ojs"},{"id":1361,"value":"margin = ({top: 10, right: 5, bottom: 80, left: 25})","pinned":false,"hidden":false,"mode":"ojs"},{"id":1406,"value":"radius = 6","pinned":false,"hidden":false,"mode":"ojs"},{"id":1409,"value":"padding = 10","pinned":false,"hidden":false,"mode":"ojs"},{"id":1365,"value":"height = 500","pinned":false,"hidden":false,"mode":"ojs"},{"id":698,"value":"capacity = d3.filter(data, d => d.weapon != \"Unknown\" && d.case != 154);","pinned":false,"hidden":false,"mode":"ojs"},{"id":27,"value":"data = FileAttachment(\"Sheet 1-mass@3.csv\").csv({typed: true})","pinned":false,"hidden":false,"mode":"ojs"},{"id":1283,"value":"import {addTooltips} from \"@mkfreeman/plot-tooltip\"","pinned":false,"hidden":false,"mode":"ojs"},{"id":3002,"value":"import {select} from \"@jashkenas/inputs\"","pinned":false,"hidden":false,"mode":"ojs"}],"files":"$W38","stdlib":"1","titling":"auto"}