ConfidenceLevelsForProportions

To-hit probabilities can be determined from hit ratios. Assuming we hit p*N times out of N, the 95% confidence interval would be (from http://www.dianthus.co.uk/statistics/confidence.htm):

p +/- 1.96*sqrt(p(1-p)/N)

To get 95% confidence in an error of +/- (100*e)%, we'd need:

e >= 1.96*sqrt(p(1-p)/N)

or
N >= 3.8416*(p - p^2)/e^2

Probability  +/- 0.5%    +/- 1%    +/- 2%    +/- 5%
0.10            13830      3457       864       138
0.20            24586      6147      1537       246
0.30            32269      8067      2017       323
0.40            36879      9220      2305       369
0.50            38416      9604      2401       384
0.60            36879      9220      2305       369
0.70            32269      8067      2017       323
0.80            24586      6147      1537       246
0.90            13830      3457       864       138

The following table of z-values (from http://people.hofstra.edu/faculty/Stefan_Waner/RealWorld/finitetopic1/confint.html) can be replaced for the 1.96 above to calculate other confidence sample sizes (for more on z-values see http://www.fourmilab.ch/rpkp/experiments/analysis/zCalc.html):

Z      confidence level
1.282  80%
1.645  90%
1.960  95%
2.326  98%
2.576  99%
3.090  99.8%
3.291  99.9%


Last modified: Tue, 4 Jul 2006 06:40:09 GMT
Edit this page, versions of this page
Search for references to this page