Do I need to convert my attribute data to discrete data for a one tail t-test I am testing a comparison of therapeutic efficacy among 2 drug therapies. The variables are based on yes (1) or no (2) data. I was trying to keep the comparison simple by testing a one tail 2-sample t-test. 
 A: Suppose we have data for 100 randomly chosen patients taking each of the two drugs. Specifically, suppose we have 81 out of 100 Yes's for Drug A and 95 out of 100 Yes's for Drug B.
Then a 'test of two proportions' in Minitab shows a significant difference between the two drugs:
Test and CI for Two Proportions 

Sample   X    N  Sample p
1       81  100  0.810000
2       95  100  0.950000

Difference = p (1) - p (2)
Estimate for difference:  -0.14
95% CI for difference:  (-0.227959, -0.0520415)
Test for difference = 0 (vs ≠ 0):         # two-sided alternative
   Z = -3.05  P-Value = 0.002

This is Minitab's implementation of the test discussed in
the NIST handbook. This test uses a normal approximation of the difference between the two sample proportions.
Minitab also shows results from Fisher's Exact test, based on the hypergeometric distribution:
Fisher’s exact test: P-Value = 0.004

See Wikipedia for details of this test. In R the hypergeometric P-value can be computed as follows:
2*phyper(5, 100, 100, 24)
[1] 0.003871905

Both of the tests above were done as two-tailed, and could have been done as one-tailed. One-sided tests may be appropriate if we hypothesized before seeing data that Drug B is better.
Moreover, as @Abdoul Haki comments, you could make a $2 \times 2$ contingency table
of Yes's and No's for Drugs A and B, and do a chi-squared
test, obtaining very nearly the same P-value (without Yates correction) as in the normal test above.  Using R:
TABL = matrix(c(81,19,95,5), nrow=2)
chisq.test(TABL, cor=F)     # 'cor=F` suppresses 'Yates correction'

        Pearson's Chi-squared test

data:  TABL
X-squared = 9.2803, df = 1, p-value = 0.002316

Note: While formally a one-tailed test in terms of the chi-squared distribution (rejecting for large values of the test statistic), this test is inherently two-sided because of the squaring in the formula for the test statistic. So the three P-values above are comparable.
