Starting words, Wordle
According to the NYT, few players use a consistent starting word when playing Wordle. Of the 28% of players (with more than 10 games played) who have used the same starting word 90% of the time, the words are: ADIEU, AUDIO, STARE, RAISE, and ARISE.
Full disclosure: I’m pretty sure I’m in this sample group and RAISE is my starting word.
Many months ago at the outset of the Wordle craze, long before the NYT bought it, I wrote a Wordle solver — a computer program that will play Wordle against itself and keep lots of fun statistics about those games.
To folks in reading this in the far future: This is the sort of activity the pandemic forced us into.
I’ll save the details of how the Wordle solver works for another post (short version: it does same math as it would if it were solving Mastermind), but I was curious how these words would perform against each other. So I took my solver, looked up a few other sources for “best starting words” and simulated 100,000 games with each of them.
Would my RAISE come out on top?
| Word | Win Percentage | Average Number of Rounds | ± Rounds | |
|---|---|---|---|---|
| 1 | slant | 86.65% | 4.9680 | 1.51 |
| 2 | trace | 85.37% | 5.0417 | 1.57 |
| 3 | crate | 85.36% | 5.0340 | 1.57 |
| 4 | crane | 84.93% | 5.0744 | 1.60 |
| 5 | piano | 84.65% | 5.1039 | 1.55 |
| 6 | slate | 84.02% | 5.0526 | 1.62 |
| 7 | audio | 83.74% | 5.1774 | 1.56 |
| 8 | stare | 82.95% | 5.1180 | 1.66 |
| 9 | snare | 82.62% | 5.1358 | 1.67 |
| 10 | roate | 82.59% | 5.1500 | 1.69 |
| 11 | tears | 82.43% | 5.1214 | 1.71 |
| 12 | yuppy | 82.12% | 5.4342 | 1.35 |
| 13 | adieu | 81.84% | 5.2540 | 1.64 |
| 14 | soare | 81.21% | 5.2077 | 1.71 |
| 15 | raise | 80.01% | 5.2551 | 1.79 |
| 16 | quest | 79.98% | 5.3882 | 1.67 |
| 17 | arise | 79.72% | 5.2750 | 1.81 |
No. No it didn’t. It came in a very (very) sad 15th place out of 17.
So I guess I need to change my starting word to SLANT? With a 6% better win rate, it seems silly not to.
To define what’s going on in that table a little better — “Win Percentage” is the number of times the solver was able to guess the right word by the 6th guess (inclusive), and “Average Number of Rounds” is the average number of rounds it took to solve the puzzle (I let the program keep solving past 6 rounds just so that it doesn’t get discouraged).
Which all adds up to “RAISE” is a bad choice for a starting word. Old habits die hard though, so despite it being suboptimal I know I’m going to have a hard time changing.
I should note, all of these games are played on “hard mode” — which is to say every guess must use all of the information from all of the previous guesses. If the solver knows a word has an “E” in it, it won’t guess “COUNT” for example.
Maybe the next experiment with the solver should be removing the hard mode restriction? What happens if, instead of just specifying the first guess, the first two guesses are scripted to be words that contain different letters from one another? For example, like “RAISE” as a first guess and “COUNT” for the next guess. Would that increase the win percentage? Would the number of rounds to solve the puzzle go up because it’s more often than not throwing away one extra guess?