We saw few exponential values in people traits- like “4.94049245958194E-13” or signifying very low values. Does it make sense for people to have so low values. Are these values valid? or AI could not understand those traits of people? Can you explain this.
The values in the CSV files are estimated percentiles of the traits. We calculate over 100 traits.
A very low value of a trait could mean two things:
If there is not enough data collected, then the low value could simply mean that not enough signals are gathered for that trait.
If there are sufficient amount of data collected (see below), a low value means the person’s trait is relatively low compared with other people. A very low number close to 0 is entirely possible.
Almost all of our traits reach very high reliability level (i.e. the value completely stabilizes, collecting more data will not change the value) when 3000 words are collected, many reach high reliability level at 1000 words mark, a few traits become stable after only 100 words are collected.
In addition, not all the traits have the same quality. Currently, two thirds of our traits have very good quality in term of well behaved statistical distribution, i.e. close to a normal (Guassian) distribution. The rest one third are less ideal, mainly lack discrimination.
So it depends on which traits we are talking about. At some point we will publish reliability and validity results of our traits analytics, when we collect enough evidence for validity.
Should I eliminate those traits which give low exponential values like “4.94049245958194E-13”? I understand close to 0 is ok but “4.94049245958194E-13” is like very from Zero and 2 people can’t be compared if one has such low value and other has decent value of say 73.
4.94049245958194E-13 is the scientific notation for 0.00000000000049049245958194, so it is close to 0.
Two people can be compared with these values. If the other person is 0.73, then it is significantly larger than the one with the value close to zero.