While it may not be Halloween time or even remotely close to Halloween time, I will be diving into the unending debate on the best Candy. Full disclaimer: I took the data set and some calculations from FiveThirtyEight, but created all the visuals and performed some of the more complex analysis. Second disclaimer: the following post will contain some hot takes and some straight up bad takes on candy. The following information is not representative of the views of Statistically Significant™ or Owen Wing.
First to create the data set, FiveThirtyEight created a program that generated random matchups of two candies and let a person choose which candy they liked more. Using this method, FiveThirtyEight sampled 8,371 different IP addresses who voted on about 269,000 randomly generated matchups. When I refer to the “win percentage,” I am referring to the amount of people who chose that candy, over the other candy it was matched with. Below is a table of the top 10 candies, out of the 85 tested, ranked by win percentage.
The worst performers include a lot of candies you would probably need to look up to recognize, but notable names include jawbusters, ring pops, candy corn, lemonheads and pixie sticks.
To take a deeper look into what makes a likable piece of candy, the candy’s components (chocolate, fruit, caramel, peanuts & nuts, nougat, crispy, hard candy bar, multi-piece) were evaluated on a binary scale (either it had it or it did not). From here, the average value added to the win percentage by having the component was calculated. Below is the average value added by each component. The most valued trait of candy is chocolate, which adds an expected 19.9% win percentage vs. a candy that doesn’t have chocolate. Additionally, you can see, a candy being hard on average has a negative impact on its win percentage.
Important to note the expected values that candies have based purely on these numbers have a .72 r value and a .51 r squared value with the actual results. What this means, is you can determine slightly over half of the variation in win percentage of the candy based purely on its components. To take this one step further, I compared the expected win percentage of each candy, based purely on its components versus its actual win percentage. What this told me was the difference between how much people theoretically should like the candy versus how much people actually do like it. One could consider this a type of underperformance ranking, meaning the candies with high differences underperform their raw inputs, but instead I am going to call this, an overrated ranking, meaning some people like candies more than they should based on their input. This is where some of the hot takes come in.
Based on the calculations, Starbursts are the most overrated candy, with an expected win percentage of 60.1 and an actual win percentage of over 67. Reese’s cups, skittles, reese’s minis, nerds and twix were also overrated. Below is a rank of the most underrated candies. You probably have not heard of most of them, but hey, that’s why they are underrated. Also of note, Snickers Crisper has the highest expected win percentage of all candies because it has chocolate, caramel, peanuts, crisped rice and is in the shape of a bar.
If you are curious about the raw data, the excel doc is posted below.