| Author |
 Topic  |
|
|
bronzing
Germany
Posts |
 Posted - 07/26/2005 : 06:16:50 AM
|
Origin Version (Select Help-->About Origin): 7G SR4
Operating System: Windows XP Home Edition SP2
Hi there,
i'm not sure if it's legitimate to use the t-test with the following problem:
I'm analysing the proportions of mutant and wild-type DNA.
In every sample the proportion of mutant DNA is measured.
Making 45 samples we recorded a mean proportion of mutant of 29.1% with a standard deviation of 3.5%.
I want to test if this proportion is significantly different from 50% mutant and 50% wild-type.
Usually I would use the t-test. But the distribution here can't be the t-distribution, because it is restricted from 0 to 100%.
What kind of test am I supposed to do?
Your help is highly appreciated.
Bronzing |
|
|
minimax
348 Posts |
Posted - 07/26/2005 : 9:46:36 PM
|
Hi bronzing,
You maybe have a misunderstanding on (One-Sample) t-test. T-test on a
variable does not mean that the variable should obey t-distribution. It indeed requires the variable obey normal distribution. The t-test gives a Student's t statistic t=sqrt(n)*(mean(y)-u)/sd(y). Assuming that the null hypothesis (H0: mean = u) is true and the population is normally distributed, the t statistic has a Student's t distribution with n-1 degrees of freedom. Therefore, your case require that the proportion obeys normal ditribution, not t-distribution. Hence, you can first test whether the proportion is normally distributed using menu Statistics-Normality test(Shapiro-wilk). Once the proportion passes normality test, you can use
t-test; otherwise, nonparametric tests, which will be integrated in Origin version8, such as the sign rank test should be used.
Moreover, note that the one-sample t-test is appropriate in this situation because the standard deviation of the population from which the data arise is unknown. When you know the standard deviation of the population, use the One-Sample Z-Test. |
 |
|
| |
Topic |
|
|
|
Hypothesis testing with proportions