Good games and mighty models

Good games and mighty models

Good games and mighty models 558 344 Exploristics

By Kimberley Hacquoil, Exploristics CDSO

During a 3-day conference, statisticians, programmers and data scientists within the pharmaceutical industry were invited to play their cards right on Exploristics’ bespoke app which was built in-house. With the chance of becoming the daily champion, the competition was fast and furious. Kate, David and Willem were victorious, beating off heavy competition to be crowned our daily winners. But how did they do against our simulations?

Statisticians versus simulations

As statisticians who love the insights that simulations offer, we couldn’t resist carrying out a simulation model to see how well our winners actually did. We ran 100,000 simulations which had a very simple algorithm to decide higher or lower:

  • a guess of higher if the card was less than an eight card
  • a guess of lower if the card was greater than an eight card (ace is high)
  • a 50:50 chance of choosing higher or lower for an eight card

We assumed that no one was card counting and so we were blown away by the result below. Our daily winners beat the simulation model 92%, 97% and 99% of the time. This is a fantastic result and shows how statisticians really can beat the odds!



Pairs and peeves

Whilst watching people play the game, I was surprised by the number of pairs that kept coming up. Unfortunately, “you get nothing for a pair, not in this game”. However, I’m not sure why I was so surprised, the probability of this happening from our simulation model is ~20%. The other thing that struck me was that people got a little frustrated if they went out early (within the first 2 cards). This was again something which seemed to occur more frequently than I would have guessed. In the simulations, this happened 50% of the time, compared to an observed 52% by the competitors. So, if you did play and went out early, don’t worry you are not alone!

Game on

I was so pleased with how well the game worked and how much engagement we got from people. We are already planning some enhancements, and maybe next year the statisticians can increase their “probability of success” by card counting or utilising prior knowledge and experience. Watch this space and “don’t touch the pack, we’ll be right back!”. Well, next year at least.

Read more:

Developing the higher/lower app game

Exploristics attends the PSI2023 Conference in London

Exploristics attends the PSI2022 Conference in Gothenburg