They say nothing in life is certain other than death and taxes, but there something else I’ve found I can count on from experience: sending out invites for a party on social media only to receive a few affirmative responses and a whole slew of “maybe”.
I know my friends and my livelier acquaintances mean well, but it is of little use when you need to give a headcount to a caterer. I can simply take my best guess or go with the old wisdom that 60% of invitees will show, but is that really the best I can do?
Probably Maybe
With my background in data science, I can’t help but employ a quantitative approach to this kind of problem. There is an entire class of predictive algorithms called probabilistic models that are perfectly suited for this type of predicament. Planning a party is complicated enough without throwing machine learning and programming into the mix so let’s look for a quick, easy method.
What is a Probabilistic Model?
Let’s take a high-level look at what a probabilistic model does. These algorithms are intended for making predictions when there are many uncertain parameters that will impact the overall outcome. These parameters can interact with and each other, and sometimes they will even influence each other. For anyone interested in diving in deeper on these, I recommend this article: Probabilistic Models in Machine Learning.
“Guest-imate”
The two-step formula goes like this:
- For every guest, put a percent chance you think they will show up next to their name.
- Add those percent chances together.
That’s it! It sounds overly simple. Let’s run through an example of how it works.
You would write your close friends’ names down and give them a higher percent odd of showing up like 0.75. Keep in mind if they live close and whether they usually show up for your invites. Give your new best friend you sang Jay-Z’s Empire State of Mind with a chance of 0.1. For your friends who are less reliable at sticking to plans, go 0.25. If you truly have no idea for some people, give them a 0.5; it’s the mathematical equivalent of “this person is a wildcard”. Even for people who RSVP yes, I give them a 0.9 because it’s likely one out of ten people will not show up for a more casual type of party.
All you need to do next is add up these percentages. As a short example, let’s assume we’re throwing a surprise party for a friend. There are ten potential guests. Only three have responded but by assigning odds to everyone based on previous parties and their relationship to the birthday boy or girl. Unreliable Eugene probably won’t show (and probably won’t even RSVP) and good friend Neddie NextDoor still hasn’t responded but is likely to show up. All combined, the estimate comes out to 4.65 so we can comfortably say four or five guests will show. We’ll tell the restaurant we’re expecting five.
| Guest | Odds | Response |
| Bobby Best Friend | 0.9 | Yes |
| Unreliable Eugene | 0.1 | No |
| Wildcard Winnie | 0.5 | No |
| Augie Acquaintance | 0.25 | No |
| Alley Acquaintance | 0.25 | No |
| Fiona Family | 0.9 | No |
| Freddie Family | 0.9 | Yes |
| Neddie NextDoor | 0.75 | No |
| Nicky NoShow | 0 | Yes |
| Kyle Karaoke | 0.1 | No |
| Total Estimate | 4.65 |
While this is just a small example with ten, it becomes more valuable when you have a larger party and error is more costly. I recently threw a surprise party for my spouse at a local restaurant and used this method. I’m happy to report the results. I invited 55 people. When I gave odds to each guest and added them up, I had an estimate of 27.4 people. I confidently gave the restaurant manager a figure of 28 guests. Ultimately, the number of guests at the party was 26. Questions about my spouse’s popularity notwithstanding, this was a great outcome. I was only over by two rather than seven people (more than three times the overestimation)!
The next time you throw a party and need to estimate the number of attendees in advance I strongly encourage you to add a touch of probabilistic modeling by trying this out.
