Re: Take a look at error bars of recall

[Chris Warburton]

I think I've figured out the gut feeling I had: there are *two* standard
deviations we might use! The "analytic" standard deviation depends only
on the mean value (i.e. the 'p * (1 - p)' formulae). This seemed
unsatisfying since it doesn't take into account the spread of our data.

Turns out that it's *also* perfectly legitimate to use the usual formula
(square root of average squared deviation from the mean) to get the
*sample* standard deviation. Presumably it took me so long to find that
this is the case precisely because it doesn't depend on the model
(binomial, bernoulli, normal, whatever), so there's nothing special
about using it for our binomial/bernoulli case.

Many thanks to this stack exchange answer:

https://stats.stackexchange.com/questions/112827/analytic-or-sample-standard-deviation-with-binomial-data

I've plugged the sample standard deviation into the graphs and it has a
much more convincing shape (i.e. it has a non-negligible width!)