Deterrence theory classically suggests that the effectiveness of a punishment in reducing future occurrences of a behavior is based on how certain someone is punished, how fast they are punished, and the severity of the punishment. In practice, the certainty of detection and punishment appears to have the strongest impact, and severity the weakest. This is good news for digital spaces, where certainty can often be more easily controlled than real world situations.
- Focus most on making your detection system as certain as possible. The more likely someone perceives or feels that a behavior will be punished, the less likely they are to do it.
- Deliver punishments as close as possible to the time at which the behavior you want to discourage occurs. The longer the time between the detection of a behavior and the arrival of a punishment (or reward), the weaker the link between the punishment and the behavior.
- Altering the severity of punishment is the weakest tool in your toolkit.
Imagine two situations:
- You have been playing a game and abusing people over chat for several months now. Over hundreds of games, it’s been fun to see their reactions and relieve stress by shouting and cursing at them. Today, you go to login and discover you have been banned from playing the game for a month.
- You just finished playing a game where you were abusing people. It was fun to see others’ reactions and helped relieve some stress from your day. You immediately go to play another game and receive a warning message about your behavior in the last game. The message says your behavior was inappropriate, and you will be punished if you do it again.
Which do you think would be more effective? In the first situation, while a month-long ban removes the player from the ecosystem during that time (assuming they don’t get around the ban), they have already ruined hundreds of games and also learned it takes quite some time before they get caught. They may also continue to behave badly in someone else’s game while they can’t play yours. Furthermore, during their ban they are also not experiencing any positive examples of play. So, despite this punishment being quite severe, it may not have much of an impact on changing their future behavior.
In the second situation, the player is still there, still in the game, which does carry a risk. But they also now feel like you are going to catch them after just one game of bad behavior. They know that you are on to them and will catch them again; they have received feedback that what they were doing was wrong. According to classical deterrence theory, this approach of certain and swift detection should have the biggest impact in reducing future negative behavior, even with low severity punishments (e.g., warnings) attached.
What is classical deterrence theory?
Classical deterrence theory is a utilitarian theory hypothesizing that the effectiveness of a punishment in deterring (reducing or eliminating) future action is determined by:
- How certain it is that someone will be punished.
- How swiftly after doing the behavior they will be punished.
- How severe the punishment is.
These ideas apply both specifically to the individual being punished as well as to society as a whole, which sees others being punished or becomes aware of how punishment works in some other way.
Certainty is the key, the critical component
Simply put, if someone feels that they will be caught for a behavior, they will be much less likely to engage in that behavior.
In the real world, there are many complications to certainty of punishment for behaviors, with many negative and criminal behaviors going completely undetected. Complex competing factors often limit how certain and swift a punishment can be delivered. Law enforcement and punishment in the real world is imperfect, complex, and carries significant costs in terms of time and money. These difficulties often lead to political attempts to address such behaviors by increasing the severity of punishments instead (which is much easier to do and also often popular with voters). However, such attempts to increase severity without addressing certainty have been shown to be ineffective or even to increase recidivism.
Thankfully, in a digital environment where behavior is being mediated via systems the owner of the environment controls, the certainty of detection becomes more possible. If you are a designer of such a system, certainty of detection is where you should focus your efforts if you want to deter negative behavior.
Another way to think about certainty is to frame it as consistency. A high consistency of being caught produces a high feeling of certainty and trust in a system. How well can you tune your detection to make sure that you are as certain as possible to catch a negative behavior? How can you signal to bad actors (or even your currently good actors who may have a bad day later) that if they do wrong they will be caught?
But isn’t this based on utilitarian / rational models of human behavior? People aren’t rational!
Absolutely. These principles of deterrence were originally based in rational terms, with the expectation that you could trade certainty, swiftness, and severity off against each other to influence a rational actor who is weighing up costs and benefits (i.e., that to compensate for reduced certainty, you could increase severity). This has proven to not be the case.
However, that is largely a problem for severity. That certainty (and importantly perception of certainty) has the largest impact does not rely on rational models of human behavior. People don’t have to be running the odds in their head constantly and acting like Homo economicus. Rather, the more feelings- and perception-based approach to decision making in Homo sapiens means that if a person feels they will be detected, then that still impacts their behavior.
Importantly, this means that features like notifications when someone you reported has been punished, prominent reporting buttons, and good behavior pledges can reduce negative behavior by boosting the perception of certainty of punishment (ultimately, this does need to be backed by good, actual certainty).
On the flip side, a low perception of certainty can cause trust issues for your players. They may, for example, stop using reporting systems if they perceive that they “do nothing.”
Swiftness helps with clarity and certainty
How soon after a behavior occurs that a punishment arrives helps establish links between the behavior and the consequence (reward and swiftness have a similar relationship). This is a basic usability principle. If I pull a lever in a game and nothing happens for 10 minutes, then I am unlikely to pull that lever again (or any other similar lever I find). The same goes for punishments. If weeks go by before I am punished for something, it becomes less clear exactly what I did, which means I can’t avoid doing it again.
The speed at which a punishment occurs also impacts how certain it is perceived to be. A warning coming right after I do a behavior feels more certain due to the strengthened clarity of behavior and outcome.
There are times when you may want to delay punishment, though, such as when you are still investigating a cheat or exploit and do not want to signal you are onto it prematurely. In these cases, you are trading off a swift punishment for future certainty of detection.
Severity is the weakest factor
You can’t compensate for 98% of behaviors going undetected by immediately banning the 2% that do get caught. A low detection rate, with a high severity, not only won’t have much (or any) deterrence effect, but it is also ineffective over your ecosystem. Yes, the one offender is now gone (if they don’t get around your ban), but they likely ruined many others’ experiences before you caught them. By focusing instead on increasing the certainty you detect people, you will have a better outcome both in terms of reforming the individual and reducing the impact on your population as a whole.
Does this mean you shouldn’t ban people for doing things like cheating? No, not necessarily. Banning cheaters does help reduce the financial incentives behind making cheats, but it is more effective to increase the certainty of detection of cheating. This can prevent people from even trying in the first place, which has a much larger impact on the financial incentives behind cheat selling.
What about escalating systems, where future recidivism results in more severe bans? Well, if you are trying to compensate for low certainty, these are largely ineffective. If your certainty is high, then these can quickly reform someone or get someone who isn’t reforming out the door. If certainty is low, however, threatening bigger punishments in the future is unlikely to have as big an impact as you would like.
Use your design to be certain a behavior won’t occur
Most of this article has been talking about enforcement and punishment. Another approach can be to use your design to make certain that a behavior simply cannot occur. This can be much more effective than any other approach. For example, removing friendly fire to eliminate issues with friendly fire and making voice chat opt-in rather than opt-out can reduce exposure to antisocial voice communications. Variations on this approach can be used to limit or remove negative behaviors and encourage positive ones.
For more information and design examples see Anti-Social Behavior in Games How Can Game Design Help?
What does it look like to have a well-designed punishment system in terms of good certainty and swiftness? Signs of success can include:
- High perception of detection — Players should feel like if they are antisocial they will be caught.
- Trust in reporting systems — Players should feel like others are being caught, and that reporting systems are worth engaging with.
- Clarity of what is being punished — Someone who is being punished should know what it is they are being punished for, and those who use a reporting system should receive feedback when someone they have reported is punished.
- Low negative behavior occurrence — Antisocial behavior rates should decrease.
- Low recidivism — Once punished, there should be a low rate of a player repeating the behavior for which they were punished.
Some strategies may seem great, but have already proven to not be.
Beware the downsides of punishments
This article is focused on how to be effective with punishment, but you should also watch out for the downsides of using punishments as a tool. Specifically, they can cause resentment and avoidance. You want people to avoid exhibiting negative behavior, but you generally don’t want them to avoid playing your game (or games you may make in the future). Furthermore, punishments like bans can, at times, face legal challenges, as you are restricting access to a product.
This isn’t to say don’t use punishments. Just be aware of these issues and consider ways you can use rewards and other techniques (e.g., designing to remove or restrict behaviors as mentioned above) to encourage positive behavior.
Now what?
With this information in mind, I hope that you can look at the systems you have and see where you can increase the certainty of detection.
Further Reading
- McLeod, S. (2025). Deterrence theory (Criminology). Simply Psychology.
- Forrester, L. & Ruiz, C. (2024). Classical theories of criminology: Deterrence. In N. Connell, L. Forrester, & S. Friend (Eds.), Introduction to Criminology and Criminal Justice.
