The sample-discrimination procedure asks listeners to identify from wh
ich of two possible distributions a given set of samples has been draw
n. Typically, discriminability (d' for a single sample) and informatio
n integration (inferred from the increase in d' after n samples) are n
ot as good as predicted by the observation-integration model. This res
earch examined whether information integration could be improved by pr
esenting n pairs of samples from the two distributions rather than pre
senting all n samples from one distribution then all n samples from th
e other (pooling samples). Pooled and paired presentations (n = 1, 2,
3, 4, 5, 6, 8, 12, and 16) were examined for ''easy'' and ''hard'' con
ditions (large and small separations between means of the two distribu
tions) for both duration discrimination and frequency discrimination.
For all four of these conditions, higher d's were always obtained for
the pooled presentations. Pairing the samples failed to improve perfor
mance and, for the ''hard'' conditions, instead made information integ
ration worse. (C) 1995 Acoustical Society of America.