I was thinking about my history of training martial arts and how it’s worth/benefit would be analyzed from a purely dispassionate/rationalist perspective. Many (possibly majority) of people start learning martial arts for self defense: i.e. to (hopefully) learn how to protect themselves if they are ever in a life or death violent situation. How well does learning martial arts actually do this? Can we analyze it from a quantitative point of view?

One simple way we could analyze would be to use the following equation:

$$[\text{Value}] = [\text{Benefits}]-[\text{Costs}]$$

Pretty simple. If our total benefits are greater than our total costs, it will be a positive value. So what are we exactly measuring here? Lets say time, as in lifespan. If someone attacks me and I die, then my total lifespan is shorter than if I successfully defend myself and am able to live longer. The associated cost in time would be the time spent training martial arts.

So how do we calculate or at least estimate the increased lifespan? I would formulate it like this:

$$[\text{Gained Lifespan}]=(Y-A)p$$

Where \(Y\) is the average lifespan for your demographic (currently 79 years in the U.S.), \(A\) is your current age (for sake of argument we’ll say 20 years old), and \(p\) is probability that you encounter a situation where your martial arts training makes the difference between life and death or serious injury.

However, that equation isn’t *quite* right. It is the gained lifespan if I were attacked and am able to save myself right when I am 20 years old, in which case I would gain 59 years of life. But if I were attacked when I was 78 years old, on average I would only gain one additional year of life. So we need to take it from the average age I would be, which is just the half-way point between \(Y=79\) years and \(A=20\) years, so the corrected equation is:

$$[\text{Gained Lifespan}]=\frac{Y-A}{2}p$$

As for \(p\), we currently have no idea what that probability could be, so let’s leave it for now.

The time cost is pretty easy to calculate. Let’s say you study martial arts for 10 hours/week for 10 years. The total time spent learning is then:

$$[\text{Time Cost}]=C=(10\; \text{hr}/\text{wk})(52\; \text{wk}/\text{yr})\left( \frac{1 \;\text{day}}{24\; \text{hr}} \right)\left( \frac{1\; \text{yr}}{365\; \text{day}} \right)(10\; \text{yr}) = 0.59\; \text{yr}$$

At this point we know every term in the equation to calculate \([\text{Value}]\) but the probability \(p\). What we can do though is calculate what \(p\) would have to be for us to break even, i.e. the point where the* *the expected or average gained lifespan is equal to the time we spent learning the martial arts in the first place. That’s simply setting \([\text{Value}]=0\) and solving for \(p\). Doing so we get:

$$p=\frac{2C}{Y-A} = \frac{2 (0.59\;\text{yr})}{79 \;\text{yr} -20 \;\text{yr} }=0.02$$

So our final probability is \(p=0.02\) or 2%. *If *our martial arts training is able to save our lives when we are in a life-threatening situation, then it’s worth it to spend 10 hrs/week training for 10 years *if* there is at least a 2% of us being killed by violence before we die of natural causes.

So what *is* the chance of us being killed? In the US, the CDC states that currently there are 5.8 homicides per 100,000 people per year. So assuming that homicides are completely random (which of course they aren’t, but for the sake of argument we’ll go with that simplification), then we have a \(p=0.000058=5.8 \times 10^{-5}\) probability or a 0.0058% chance of being killed by homicide in any given year.

But we need to know the total probability of it happening over the course of \(Y-A=59\) years, not just one! How do we calculate that? We can’t just add them all up 59 times and say the total probability is \(P=59p\), since you can’t just add up probabilities like that. That’s like saying if you roll a 6-sided die six times that you’re guaranteed to roll 6, or if you flip a coin twice you’re guaranteed to get a heads. And we can’t multiply \(p\) by itself 59 times either, \(P=p^{59}\) is the probability of someone being killed *ever* year for 59 years! Of course you can only be killed once (that we know of), but even if you could be killed more than once it becomes an astronomically small chance.

What we do is instead calculate the probability that we *aren’t* killed in 1 year, multiply that by itself 59 times to get the probability we aren’t killed at all over 59 years, and then subtract that by 1 to get the probability that we would be killed over 59 years.

$$P = 1-\left( 1-p \right)^{Y-A}=1-\left( 1- 5.8 \times 10^{-5} \right)^{59} = 0.0034$$

Or a 0.34% chance of a person being killed over the course of 59 years. That’s pretty high (3 people in 1000), but it’s still much smaller than the 2% required for our martial arts training to be ‘worth it’ from a cost/benefit analysis perspective.

However all this analysis invites an obvious question: would this 10 hrs/week of training for 10 years actually give me the skills to protect myself if I were to be violently attacked by someone? Does it even make any difference?

In terms of a deadly assault where someone is attacking you with a knife or a gun, I’m afraid the answer is: probably not. Statistically it’s hard to evaluate this. Lots of martial arts *claim* to be able to teach people to defend themselves against an attacker with a knife or a gun (and I’ve trained some of them myself), but unfortunately this isn’t very realistic. If someone has a gun on you and they intend to kill you, it’s *very* difficult to survive. Similarly if someone has a knife on you and intends to slash and stab you, your chances of getting out without a serious or fatal injury is very difficult.

This second scenario is actually pretty easy to test in relative safety with a partner. Put on a fairly close-fitting shirt and pants (that can get stained), and the attacker has a large permanent marker to simulate a knife. The attacker’s goal is to mark your body with the magic marker as many times as they can, and the defender’s goal is to take the marker away from them without getting any marks on them. You’ll find it’s pretty much *impossible*, even for a skilled martial artist, to be able to successfully defend themselves even against a relative beginner without several marks on their clothing, often in very dangerous areas (torso, thigh, wrist, neck, etc.)

You can do the same type of thing with a paintball gun and some protective equipment, but it can get a lot more messy. The results are pretty much the same though: the odds of the defender *not* being killed or seriously injured/maimed are very small.

Some of the best realistic conversations on this I’ve seen on the internet are by a couple of guys that evaluate some of the *least* realistic fight scenes on film: Logan Lo and Chad Vasquez at the Scenic Fights YouTube channel. These guys have a lot of experience doing full-contact martial arts with full resistance and with weapons, and they have a lot of great commentary on what is or is not realistic in a combat situation, and the realistic portrayal of fighting someone with or without a weapon.

Whether you should fight back or not can depend on a lot of different factors. A drunk belligerent guy in a bar pushes you? De-escalate and walk away. Someone with a gun or a knife demands your wallet? Just give them your wallet. Several dudes with baseball bats start approaching you menacingly? Just run and hope you can get somewhere safe. But someone with a gun tries to get you into a car? Statistically the odds of you getting out alive aren’t good, you should probably resist and fight back. Someone takes a swing at you? Combat has already started, it’s hard to get out of it at that point and you need to defend yourself. So basically, context matters.

So how likely is an assault to become a murder? Is there any data that we can use to estimate this? A couple of searches got me to these two sites: number of aggravated assaults in the US in 2020, and number of homicides in the US in 2020.

Here is a python Jupyter notebook with the data and plots:

```
import matplotlib.pyplot as plt
import numpy as np
assaults = {"Personal":74403, "Handgun":69423, "Knife":61924, "Blunt":39643, "Firearm":39548, "Asphyxiation":10640,
"None":8346, "Unknown":7772, "Rifle":4892, "Shotgun":2948, "Handgun(automatic)":2862, "Other Firearm":2646,
"Firearm(automatic)":943, "Fire":865, "Poison":485, "Drugs":456, "Rifle(automatic)":287, "Explosives":250,
"Other Firearm(automatic)":82, "Shotgun(automatic)":38, "Other":29497}
assaultsKeys = assaults.keys()
assaultsValues = assaults.values()
#plt.bar(assaultsKeys,assaultsValues)
#plt.xticks(rotation='vertical')
y_pos = np.arange(len(assaultsKeys))
plt.figure(figsize=(8,10))
plt.barh(y_pos,assaultsValues, height=0.8, align='center')
plt.yticks(y_pos,labels=assaultsKeys)
plt.gca().invert_yaxis()
#plt.xscale('log')
plt.title('US Assault Crimes in 2020')
murders = {"Handgun":8029, "Firearm":4863, "Knife":1739, "Personal":662, "Rifles":455, "Blunt":393,
"Shotgun":203, "Other guns":113, "Narcotics":113, "Fire":106, "Asphyxiation":71, "Strangulation":58,
"Poison":16, "Drowning":5, "Explosives":4, "Other":983}
murdersKeys = murders.keys()
murdersValues = murders.values()
y_pos = np.arange(len(murdersKeys))
plt.figure(figsize=(8,9))
plt.barh(y_pos,murdersValues, height=0.8, align='center')
plt.yticks(y_pos,labels=murdersKeys)
plt.gca().invert_yaxis()
#plt.xscale('log')
plt.title('US Homicide Crimes in 2020')
```

Text(0.5, 1.0, 'US Homicide Crimes in 2020')

You see that in general, there are 10x more assaults than there are homicides of any category, but the categories don’t all exactly match. Also some of the naming conventions are a little strange:

- Personal: means with hands, fists, kicking, etc. Basically assault
*without*a weapon. - Blunt: a blunt object of some kind
- Guns: the lists have
*lots*of different classifications for guns, and the two lists don’t even agree. - Asphyxiation vs strangulation: no explanation given
- None: what does that mean? No weapon used? In that case how is it different from Personal?
- How does one commit aggravated assault with drugs, fire, or explosives?

Most of the weird stuff is really small in numbers, so we can mostly ignore it on a statistical basis. I decided to do a bit of grouping and renaming in order to make the two lists agree a bit more and make more sense. From a self-defense perspective, we can simplify it to just six categories:

- Unarmed battery (punching, kicking, etc.)
- Firearm (all kinds, no need for 10 different categories here)
- Knife
- Blunt
- Strangle (also an unarmed attack, but trying to choke instead of punch/kick)
- Other (everything else)

By combining and sorting into those categories, we also need to add the assaults *and* the homicides together to get the total number of attacks for each category: presumably the homicides aren’t being double-counted as assaults as well.