1
$\begingroup$

I realize that this question is most likely not too interesting and challenging to discuss here, it's rather a matter of different notations, but I would like to clarify that if possible.

When conducting Z/t tests, I got used to the convention that $z_{\frac{\alpha}{2}}$ is such a value that $P(Z>z_\frac{\alpha}{2}) = \frac{\alpha}{2}$. It also follows that $-z_{\frac{\alpha}{2}} = z_{1-\frac{\alpha}{2}}$. The same is true for the t-test and is just a matter of swapping the letters.

Visually speaking, this is just such a value that the area to the right equals $\frac{\alpha}{2}$.

I noticed, however, that sometimes these cutoff values are introduced as the inverse CDFs and then the order of the values on the X-axis is reversed, of course.

What is interesting, some authors mix the two and then their notation gets a bit confusing. See, for example, below:

https://online.stat.psu.edu/stat500/lesson/6b/6b.1

$t_{0.005}$ is positive here in the solution, but the illustrations place $t_{1-\alpha}$ on the right of the X-axis, which doesn't make a lot of sense to me unless there is this notation failure.

What is the more widely accepted notation? Well, I understand that this is not a big deal at all and it's always clear from context which notation the author prefers, but I just stumbled upon the online book above and this inconsistency made me wonder.

Thanks!

Improve this question
$\endgroup$
3
  • 3
    $\begingroup$ I get the feeling that half of statisticians think each way is right and think the other half is fools for believing otherwise. $\endgroup$
    – Dave
    Mar 22, 2021 at 14:12
  • $\begingroup$ It just surprises me that there can be a second opinion on a thing like this one. In the book above, this duality only resulted in somewhat confusing illustrations. Honestly speaking, currently they are just wrong $\endgroup$
    – Don Draper
    Mar 22, 2021 at 14:14
  • $\begingroup$ This is indeed a big problem in statistics, especially for students. I am a strong proponent of a consistent notation for quantiles of the (continuous) distribution of $X$: $q_{\alpha}$ is always the value so that $\alpha = \text{Prob}[X\leqslant q_{\alpha}]$. Or for other distributions: $q_{\alpha}=F^{-1}(\alpha)$ for some suitably defined inverse of the cumulative distribution function $F(x)$ of $X$. $\endgroup$
    – StijnDeVuyst
    Mar 22, 2021 at 15:45

1 Answer 1

Reset to default
2
$\begingroup$

Traditionally, printed tables for distributions t, chi-squared, and F have used the subscript notation you show to indicate right-tail probability. However, software such as R often uses a quantile function (inverse CDF) to find the value with a certain percentage below.

For example, in R, code qt(.975, 24) returns $2.063899,$ called 'quantile 0.975'. You can probably find something like $2.064$ on line 24 of your t table under header $0.025$ (sometimes called a 'percentage point').

As far as I know, textbook authors have not yet developed a standard notation for this value that is a good match for both printed tables and software output.

I am not seriously proposing this, but maybe it should be something like $t_{0.025} = t_{24; 0.025} \approx 2.064$ and ${}_{0.975}t = {}_{0.975}t_{24} = 2.063899.$

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

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