Re: When to use Fisher?

Why not always use Fisher (for 2x2 tables)?

Fisher's test always gives the exact P value. The chi-square test is
based on an approximation of discretely distributed chi-square values
by the continuous (chi-square) distribution. To make the approximation
better the Yates' continuity correction could be used. However, the
"corrected" P value is usually too high. It is definitely a good idea
to avoid the chi-square test when the numbers in the contingency table
are very small. When the numbers are large, the P values reported by
the two tests will be very similar, but even in this case I cannot see
any advantages of chi-square over Fisher, unless you do calculations by
hand or with a calculator.

HTH,
Vadim Pliner

Bruce Weaver wrote:
> Tim Witort wrote:
> > I have always used Fisher when my sample size is less than
> > 30, but I have also heard that, if certain counts in the
> > 2x2 matrix cells are 5 or less, Fisher should also be used?
> > For example, with a matrix like this, should Fisher also be
> > used? The total sample is 38, but two cells have 5 or fewer
> > observations.
> >
> > Outcome 1 Outcome 2
> >
> > Group A 23 8
> >
> > Group B 5 2
> >
> > -- TRW
>
> You should be looking at *expected* frequencies, not observed. In your
> table, the expected frequency for the bottom right cell is 7*10/38,
> which is a little less than 2. So according to most rules of thumb, the
> chi-square approximation would not be very good. Here's a summary of
> the conditions under which the chi-square approximation breaks down.
>
> www.angelfire.com/wv/bwhomedir/notes/chi_square_assumptions.txt
>
> --
> Bruce Weaver
> bweaver@xxxxxxxxxxxx
> www.angelfire.com/wv/bwhomedir

.

Relevant Pages

  • Re: diehard and ent results quesion
    ... See Knuth for the chi-square statistic. ... That doesn't sound like what Ritter was ... multiple trials directly in the chi-square test, ... > as a solution to testing random generators. ...
    (sci.crypt)
  • Re: Chi squared -> p-value - Any formula??
    ... The corresponding approximation for the complementary cdf (i.e., ... the upper tail area) of a chi-square variable with 1 d.f. ... The claimed accuracy of the approximation for the standard normal cdf is somewhat better than that of the approximation Ken gave, and is also simpler to type, so I decided to try your approximation. ... the maximum fractional error in the p-values was 2.91%. ...
    (sci.stat.math)
  • Re: Fishers Exact Test and Chi-square
    ... 'best' value to calculate but chi-square is ... Chi-square approximation isn't great because the chi-square ... distribution is continuous but the cell counts are discrete. ... There can be no single exact distribution because the probabilities ...
    (sci.stat.consult)
  • Re: Chi-Square Question
    ... > I would be grateful of some advice please. ... and have been trying to use Chi-Square test in ... You can still compute Chi-square, though it will be hard to interpret. ... in such cells may dominate the overall ch-squared). ...
    (sci.stat.math)
  • Re: Fishers Exact Test and Chi-square
    ... 'best' value to calculate but chi-square is ... Chi-square approximation isn't great because the chi-square ... distribution is continuous but the cell counts are discrete. ... This is why conditional tests can be exact. ...
    (sci.stat.consult)