0
$\begingroup$

Matlab ttest "returns a test decision for the null hypothesis that the data in x comes from a normal distribution with mean equal to zero and unknown variance, using the one-sample t-test."

Can anyone clarify what exactly is output by the command ttest when run on a vector of numbers $a_1, a_2, \ldots, a_n$? OK to answer in a form such as "it outputs the value $x$ that maximizes $y$."

I realize this is probably a trivial question for the experts, but I can't easily find a clear answer.

Improve this question
$\endgroup$
1
$\begingroup$

Matlab's ttest takes your vector of data and performs a Student's (one-sample) t-test on it, assuming that:

  • the population mean you're testing against, $\mu_{0}$, is zero
  • $n$ is equal to length(x)
  • the level of statistical signficance, or Type I error, you're willing to accept is 5%; you can change the amount of Type I error you're willing to accept in the arguments of the function

The $t$-test calculates the mean of the data in x (i.e., $\bar{x} =$ sum(x)/length(x)), and its sample standard deviation, $s$, typically with the formula

\begin{align} s = \sqrt{\frac{1}{n - 1}\sum_{i = 1}^{n}(x_{i} - \bar{x})^{2}}, \end{align}

which corrects for the fact that $s$ estimates the true standard deviation of the population from which x samples.

Then, the $t$-statistic is

\begin{align} t = \frac{\bar{x} - \mu_{0}}{s/\sqrt{n}} = \frac{\bar{x}}{s/\sqrt{n}}, \end{align}

because $\mu_{0}$ is assumed equal to zero. The documentation doesn't say, so I assume that the test is a bidirectional $t$-test, which means that ttest returns 1 if $t$ is greater than tinv(0.95, length(x)) or less than tinv(0.05, length(x)) (these are the t-statistics corresponding to a 5% level of significance; it should be the case that tinv(0.05, length(x)) equals -tinv(0.95, length(x))). Otherwise, ttest returns 0.

Improve this answer
$\endgroup$
3
  • $\begingroup$ Thanks. Can you say what is tinv (again not terribly clear to me), and how is the p-value p calculated in a call [h, p] = ttest? $\endgroup$ – Manu Dec 14 '13 at 13:27
  • $\begingroup$ The student's t distribution is a probability distribution with a parameter $\nu$ (called the "degrees of freedom") If x=tinv(0.95,nu), then for a random variable $t$ with the Student's t distribution with $\nu$ degrees of freedom, $P(t<=x)=0.95$. In other words, x is the 95th percentile of this probabilitiy distribution. $\endgroup$ – Brian Borchers Dec 15 '13 at 5:46
  • $\begingroup$ Or if you prefer: ttest returns either 1 or zero. The array or vector fed to ttest is analyzed in order to ascertain if the vector came from a population that was normally distributed. Or not. Return of 1 means "yes" (95% CI), zero means no. Zero can also mean that the vector did not satisfy the assumption that the mean be zero. Other arguments to ttest have different meanings. This is for H = ttest(x); $\endgroup$ – jim mcnamara Dec 15 '13 at 18:15

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.