-1
$\begingroup$

I'm using Excels T.TEST and LibreOffices TTEST functions, and their documentation is rather lacking, so I figured I ask some things here. I'm not exactly sure if this is the right site, but I can't find something more fitting and I think I have the basics (what goes in what input field) right…

What exactly do they return? It's not explicitly stated that it is the p-value, but I would assume so.

What is the difference between Excel and LO, or Excels old TTEST? The documentation for Excels old function just states that there is a new one that may provide improved accuracy and whose names better reflect their usage. I would assume for for the standard I'm-clueless-about-statistics-I-just-need-significance situation it should not matter, but I'd rather be safe than sorry.

Is a Type 3 T-test a Welch-test? According to Wikipedia, I should use said Welch-Test, when my samples standard deviations are not identical, and the documentation says Type 3 is for unequal variance (heteroscedastic). Or is that yet another test?

Can I use strangly formated inputs? My input sheets are formed rather weird:

    0.8 0.7 0.9 0.8        0.6 0.7 0.8
        0.6 0.9 0.7            0.8 0.9
            0.7 0.8                0.7
                0.6

And I just use the whole A1:D4 and G1:I3 parts, with a lot of empty cells (as in T.TEST=(A1:D4;G1:I3;2;3)). Does that matter or do I have to reformat, for example in simple rows without empty cells?

0.8 0.7 0.9 0.8 0.6 0.9 0.7 0.7 0.8 0.6
0.6 0.7 0.8 0.8 0.9 0.7

and thus T.TEST=(A1:J1;A2:F2;2;3)?

  • 0
    What are you trying to test? $t$-test tests a hypothesis about mean of data (that it is same as some value or mean of another sample). You are taking about correlations. That is, about a statistics of your data, not about the data itself. – 2017-01-26
  • 0
    The correlations are the data. I have a bunch of them for different locations and I am trying to see if the difference between the locations is significant or not, or rather, if my two locations are different enough to be seen as different locations and not just two subsets of the same region. – 2017-01-26
  • 0
    I think you are asking about Excel when you should be asking about statistical models. // A 1-sample t test is for testing whether the mean of a sample is consistent with a hypothetical mean of a (roughly) normal population. 'Welch' and 'pooled' tests are two kinds of 2-sample t tests for using means and SDs of 2 samples to see whether the two populations from which the samples were drawn might be equal. ('Welch' is used when you don't know that the 2 _pop_ SDs are equal.) From what you say about your objective, I doubt using _your_ 'correlations as data' properly lead to _any_ kind of t test. – 2017-01-27

1 Answers 1

-1

Let me see if I understand. You have $N$ locations. For each location $i\in\{1,\ldots,N\}$, you have $n$ observations (about something). You are trying to see, for a pair $(i,j)$ of locations, if the locations are 'different enough' to treat them separately. That is why you calculated correlation between $i$ and $j$, $\rho_{ij}$. And you are trying to decide whether $\rho_{ij}$ is significantly different from zero and want to use $t$-test for that.

If this is so, then $\rho_{ij}\sqrt{\frac{n-2}{1-\rho_{ij}^{2}}}\equiv t'$ is $t$-distributed with $n-2$ degrees of freedom. Wiki on Pearson correlation coefficient is helpful here. From this getting $p$-value should be easy with the Excel's T.INV/T.INV.2T/T.DIST/T.DIST.2T functions. Namely, plugging $t'$ into T.DIST.2T provides $p$-value for the test which has $\rho_{ij}=0$ as a null and $\rho_{ij}\neq0$ as an alternative.

  • 0
    You got the first part right, but I am not correlating the locations with each other. I'm trying to test if my n_i observations (which happen to be correlation coefficients) are from the same population as my n_j observations (which is another amount of correlation coefficients). I would rather only explain the data and its causes, but the assignment says you have to prove significance, because you apparently have to… and thus I am trying to understand the t-test, which is supposedly the thing to use. – 2017-01-26
  • 0
    @JC_CL At this point I am completely confused as to what your data are. Unless you can pose the question in a comprehensible manner, it is going to be impossible to answer it. – 2017-01-26
  • 0
    I tried to change the question. I guess the type of data does not matter, since I am only trying to understand some parts of the Excel functions, that are not explained well enough for me in the documentation. – 2017-01-26