Decision Mechanics

Insight. Applied.

Statisticians publish guidance on p-values

By

American Statistical Association building

The American Statistical Association has taken the unprecedented step of issuing guidance on the use of p-values. Its statement was prompted by concerns that misuse of p-values was driving bad science.

Basic and Applied Social Psychology has refused to accept papers containing p-values.

A p-value is “informally” defined in the statement as

…the probability under a specified statistical model that a statistical summary of the data (for example, the sample mean difference between two compared groups) would be equal to or more extreme than its observed value.

Doesn’t really help much, but you have to applaud the attempt. After all, even scientists don’t really understand p-values.

Google’s self-driving car causes crash

By

Google self-driving car

The New Scientist reports that one of Google’s autonomous cars drove into a bus on 14 February 2016.

Apparently Google’s cars have been involved in 18 accidents in Mountain View since it started testing in 2010. All have been other vehicles striking a stationary or slow moving Google car. However, in the latest incident the AI decided to pull out into the path of the slow-moving bus—i.e. the AI was at fault.

What interested me was the explanation given by Google

Our car had detected the approaching bus, but predicted that it would yield to us because we were ahead of it

As the driver is software, it’s possible to to examine the logs and know exactly how and why the accident occurred. Obviously, things are rarely as clear-cut when human drivers are involved in accidents. This makes it a lot easier to determine where the problem lies and, more importantly, to reduce the chances of it happening again. A single incident produces a bunch of useful data.

The ability to audit the behavior of autonomous cars is going to be one of their big selling points.