The demographics data is currently represented graphically as a deviation from the internet average. This is going to get a bit complicated, but here's how it works:
As an example, let's say that the visitors to the average website are 40% male and 60% female. Now imagine that we're interested in the demographics for a specific site, called theirsite.com. If the visitors to theirsite.com are 10% male and 90% female, then we get this:
Male visitors to theirsite.com comprise 10% / 40% = 25% of our expectation.
Female visitors to theirsite.com comprise 90% / 60% = 150% of our expectation.
So, if we set our expected value to be 100 (this represents the internet average), then an underrepresented demographic (male) could receive a score between 0 and 100. Zero would mean this demographic attribute is not present, and 100 means the demographic attribute is present at exactly the expected amount.
In the theirsite.com example, the expected male demographic would be 40%. Since only 10% of theirsite.com's visitors are male, we would give this site a score of -25.
If a demographic is overrepresented, the score will range from 100 to infinity. Since we can't represent infinity very well graphically, we will "normalize" this range and enforce a maximum value of 200.
Now, the graph itself is showing a site's deviation from the internet average. In effect, it is a ratio of ratios. There is no guarantee that the sum of those ratios should be 100%.
So if you try to break the sum down into actual numbers and a score, this is what you get for theirsite.com:
Male: 10% / 40% = .25 --> a score of -25
Female: 90% / 60% = 1.5 --> a score of +50
The -25 and +50 scores do not sum to 0.