Excel is a fun tool. First, I took the data that Darius Kazemi (google him, you cur) scraped from GameRankings in the range of 1997-2010 and posted. And now I present (from that data) a way to map the skewed 75%-is-average critic rating to a 0%-100% scale. Certainly someone has done this before, but it took me ten minutes so what-the-hay.
First, the full range. Take a critic score on the x-axis, follow it up to the line and the y-coordinate will be the converted rating. The second image below is a zoom in on the “sweet spot” from a critic rating of 70% to 100%. Look at how vertical it gets in the 80% range! This means the reviews bunch up in this range. A perfectly flatly distributed group would be a simple diagonal line from 0%, 0% to 100%, 100%.
Here, we zoom in on the meat of the data: the 70%-100% critic range:
And for those who like tables:
| Critic | Converted | Critic | Converted | Critic | Converted | Critic | Converted |
| 37% | 0.0% | 53% | 4.3% | 69% | 27.7% | 85% | 83.0% |
| 38% | 0.1% | 54% | 5.0% | 70% | 30.4% | 86% | 86.3% |
| 39% | 0.2% | 55% | 5.9% | 71% | 33.5% | 87% | 89.0% |
| 40% | 0.3% | 56% | 6.7% | 72% | 36.0% | 88% | 91.1% |
| 41% | 0.4% | 57% | 7.8% | 73% | 39.3% | 89% | 93.2% |
| 42% | 0.5% | 58% | 8.9% | 74% | 42.8% | 90% | 95.1% |
| 43% | 0.6% | 59% | 9.6% | 75% | 45.8% | 91% | 96.4% |
| 44% | 0.8% | 60% | 11.1% | 76% | 49.7% | 92% | 97.5% |
| 45% | 1.0% | 61% | 12.3% | 77% | 53.9% | 93% | 98.5% |
| 46% | 1.2% | 62% | 13.9% | 78% | 57.5% | 94% | 99.2% |
| 47% | 1.4% | 63% | 15.3% | 79% | 61.1% | 95% | 99.6% |
| 48% | 1.7% | 64% | 17.3% | 80% | 65.0% | 96% | 99.8% |
| 49% | 2.0% | 65% | 19.2% | 81% | 68.9% | 97% | 99.9% |
| 50% | 2.6% | 66% | 21.4% | 82% | 73.1% | 98% | 100.0% |
| 51% | 3.1% | 67% | 23.3% | 83% | 77.0% | 99% | 100.0% |
| 52% | 3.6% | 68% | 25.2% | 84% | 80.2% | 100% | 100.0% |
Since Darius only pulls games with 20 reviews or more, he misses my critical failure Superman Returns which currently would fall below 0%.
[...] This post was mentioned on Twitter by Darius Kazemi, Scott Madin. Scott Madin said: RT @tinysubversions: Hey, @zhiwiller used my review dataset to normalize review scores from 0%-100%. http://bit.ly/ibadEa That 80 you go … [...]
What distribution of scores does the new mapping have in the 0-100% range?
Somewhat flatly distributed. That’s the idea. Same number of 90%s as 10%s. Am I reading your question incorrectly?