GJOpen currently leaves resolved bins open on questions such as WTI Jan 1 Close with 1 out of 3 closed bins and Syrian Refugees in EU with 2 out of 4 closed bins.

To get the correct current forecast of a user who may have put credibility into closed bins, it is necessary to assume the bin is closed and apply a simple conditional probability calculation to adjust the forecasts on the other bins, so that the total probability of remaining bins adds up to 100.

Questions that started with a single bin should not be adjusted because the total probability of such questions can lie between 0 and 100.

This leads to the following Python function to adapt a forecast *F={bin1:f1,…,binN:fN} *for an N-bin question into an M<N-bin question in which the *N-M *bins *C={binA,binB,…}* have been closed:

`import numpy as np`

def adapt_forecast(F, C):

if len(F)==1:

return F

else:

open=[x for x in F if x not in C]

bet = np.sum([F[x] for x in open])

scale = 100.0/bet

return {x:0 if x in C else scale*F[x] for x in F}

For example, here is a prediction from user doublehanded who made just one forecast on this question (which is typical, most users in a question will make just a single prediction):

**(de facto closed).**0% Less than 560,000**(de facto closed).**55% Between 560,000 and 710,000, inclusive- 45% More than 710,000 but less than 1 million
- 0% 1 million or more

We should then read this as follows:

- doublehanded gets marginal Brier scores of 2 for bin 1 and 0.605 for bin 2.
- The adapted forecast of doublehanded is now
- 100% More than 710,000 but less than 1 million
- 0% 1 million or more

which is the result of `adapt_forecast({1:0,2:55,3:45,4:0}, [1,2])
`

## One thought on “Conditional probability: adapting forecasts on questions with resolved bins on GJOpen”