Why Design Percentiles are not Dimension Percentiles

When you set a Dimension percentile, such as 97.5th percentile, you exclude from the design the 2.5 percent of people who are bigger in that dimension.

Having been excluded, those 2.5 percent stay excluded, whatever else you specify later.

People have different proportions, so people who are 97.5th percentile in one dimension are various other percentiles in their other dimensions.

So when you specify a second dimension, you are only selecting from among 97.5% of the original Group, AND some of the those 97.5% will be bigger than 97.5th percentile in the second dimension. So, as a logical necessity, you exclude some more people with the extra dimension that you specify.

The process is repeated for every extra dimension which has to fit the design, so some types of design, which have to fit in several different ways at once, can end up fitting many fewer people than intended.

For example if you specify three 97.5 percentile dimensions, all of which vary independently in people (a worst-case scenario), you end up with an upper Design Percentile of only 92.7%. If you specify five such dimensions, this becomes 88.1%, and if the same thing happens for your Smallest User dimensions, your design will only fit 76% of its intended users, instead of 95%. If you specify 5th to 95th Dimension percentiles the situation worsens: the design fits 54% - barely half - of the user group.