Night boosts winning odds. Tim Hortons’ Roll up the Rim contest, a fixture since 1985, remained unchanged for 25 years until 2020. The shift to digital due to the pandemic altered the game. Entries, acquired through purchases, transitioned to the loyalty app for later use, turning into “digital rolls” in the renamed contest, Roll up to Win.
My recent breakthrough in optimizing winning chances, and making headlines nationwide, involved leveraging a simple tactic: playing during off-peak times. Despite its simplicity, achieving an 80% success rate in Tim Hortons’ annual contest demanded data, perseverance, and a substantial coffee intake to uncover the ideal strategy.
Night boosts winning odds: Digital element changes the odds
In 2020, I emphasized how “digital rolls” brought strategy into the game, with a major tip to boost your chances: play during low-traffic times.
Identifying the least crowded playing times might seem straightforward — opt for the middle of the night when most Canadians are asleep. Yet, in a country spanning six time zones, pinpointing the ideal time is a complex calculation.
Previously, I estimated 4:30 a.m. Eastern as the sweet spot, balancing not too late or too early. However, relying on estimation is still uncertain, and to determine the ultimate prime time for free coffee, data became crucial.
This year, Tim Hortons provided precisely that.
Getting data from the app
When I logged into the app at the start of the competition on March 6, a prominent message caught my eye: “Over 308,619 prizes already given away!” It was an enticing call to join in — so many winners already! — and it held valuable information.
After waiting for five minutes and refreshing the page, the message altered: “Over 309,949 prizes already given away!” Another 1,330 prizes had been claimed.
I began periodically refreshing the page, noting down the time and the number of prizes distributed. My hypothesis was straightforward: the prizes awarded should correspond with the number of participants. I aimed to create a model of player behavior for Roll up to Win and determine the optimal time to participate.
While gathering online data using software is a common practice in scientific research, automated processes are typically prohibited in contests like this one. Roll up to Win was no exception, so I manually collected the data.
Throughout the day — and even during the night — I refreshed the page to track prize counts. However, as I had other commitments, there were gaps in my logs. In statistical terms, this meant having what’s called missing data.
It’s typical to encounter missing data in practical analysis. Instances comprise surveys that are partially completed or unreturned, patients who fail to attend medical appointments, and incidents involving misplaced or corrupted data files.
Night boosts winning odds: Statistical challenges
This situation can pose statistical challenges depending on the nature and reasons behind the gaps in our records. For instance, a patient might miss a medical appointment due to illness, while another might miss it due to car trouble. These distinct scenarios offer varied insights, necessitating different approaches.
My issue with missing data was relatively straightforward. I aimed to address the gaps that occurred when I wasn’t actively monitoring the contest, such as during sleep or travel.
By examining the available data, I searched for trends. I noticed that the highest number of prizes was won between 9 a.m. and 1 p.m. Eastern time, with the lowest occurring around 3 a.m. This pattern repeated daily, and I used this information strategically.
In bridging mathematical models to real-world scenarios, statisticians often rely on assumptions. I assumed that player behavior patterns would exhibit consistency from day to day. While this assumption was relatively strong—there was some indication of a slightly later start on Sunday mornings—it seemed a reasonable approach for my specific problem.
Weighting the data
I aggregated the data from each day and applied a method called weighting. Days with a higher number of logged observations were accorded greater significance or weight in my analyses. Employing statistical techniques, I connected the data points to outline the broader pattern of player behavior.
My data-led estimate pinpointed the best time to play at 3:16 a.m. Eastern, an hour earlier than my usual playtime, and the worst time at 11:46 a.m. Although statistical uncertainties exist, playing around these times maximizes winning chances.
To confirm my findings, I conducted a test to verify if fewer wins at 3 a.m. were due to fewer available prizes rather than fewer players. Splitting 60 rolls, I played 30 around 3:16 a.m. and the rest before lunch, resulting in 23 wins early and five later, supporting my theory.
Originally planning a local interest story on applying statistics, my involvement snowballed unexpectedly. It led to appearances on various radio stations and national TV shows like CTV’s Your Morning and CBC’s The National.
In the interviews, an interesting concern emerged: if everyone starts playing at 3:16 a.m., won’t the strategy backfire?
Indeed, this scenario taps into a concept from game theory: when everyone follows the same strategy, it can become counterproductive.
However, I doubt everyone will wake up for a chance at free coffee, so the strategy might still hold. I plan to track the data until the contest’s end on April 9 to see if adjustments are needed. It might mean a few more late nights, but I’ve stocked up on caffeine for that!
Read the original article on theconversation.
Read more: Researcher Solves 60-Year-Old Game Theory Problem.
