3
$\begingroup$

Alright, I'm trying to figure out how to calculate a critical value using t-distribution in Microsoft Excel... ex. a one-tailed area of 0.05 with 39 degrees of freedom: t=1.685

I know the answer, but how do I get this? I've tried TDIST() TINV() and TTEST() but they all give me different answers. This web calculator: http://www.danielsoper.com/statcalc/calc10.aspx always gives me what I'm looking for but I cannot manage to get Excel to do the same.

Any help would be greatly appreciated!

$\endgroup$
1
  • $\begingroup$ Hi Tim. SE has a site dedicated to statistics questions at stats.stackexchange.com $\endgroup$ Nov 19, 2010 at 20:20

4 Answers 4

4
$\begingroup$

The right function is TINV(), but the problem is that TINV() assumes that you're inputting the two-tailed probability, not the one-tailed probability. In other words, TINV(p,d) outputs the value of $x$ such that $P(|X| \geq x) = p$, where $X$ has a $t$ distribution with $d$ degrees of freedom.

Thus if you want TINV() to give you the critical value associated with a one-tailed probability, you have to double the probability first to account for the other tail. Thus, for example, TINV(0.1,39) yields an output of 1.684875122, which is the critical value you want.

$\endgroup$
2
$\begingroup$

Non-answer: Export the data to CSV and use software that is suited for statistical work, e.g. R.

$\endgroup$
2
$\begingroup$

For reference, in EXCEL 2010 there are now two functions: T.INV(), which gives the "left-tailed inverse" (i.e. the inverse of the cumulative distribution function), and there is T.INV.2T(), which assumes as an argument the probability of both tails.

$\endgroup$
1
$\begingroup$

As Mike suggested either you can input your alpha level (nominal alpha) as .10 (2*.05) to get the one tailed p value or you can divide the given probability (actual alpha) by two. Both approaches do the same thing. Remember that taking all other things constant, the probability of recting a true null is easier for a one tailed test compared to the two tailed test. One tailed test yields a smaller p value compared to the two tailed test. That is why you should dived the calculated p (actual alpha) by 2. Same goes for SPSS. Hope it helps :) Gulshah

$\endgroup$

You must log in to answer this question.