Crawford-Howell t-test

Hello,

I need to perform a Crawford-Howell t-test between single cases and normative groups of controls.

I found this code for it, but it is in R.

Does anyone have any code for it in Matlab, or is able to convert the code to matlab code?

Thanks!

[the cyclist]

The original paper seems to be freely available here, if anyone is interested.

Answer 1

[the cyclist]

The MATLAB Statistics Toolbox has a built-in function ttest2(). The result of using this function on a single case vs controls, for example,

[h p ci stats] = ttest2(5,[1 2 3])

gives me the same result (to several decimal places) as the R code in that link.

(This is just a bit puzzling to me, since ttest2() is just a standard t-test, and you are suggesting that Crawford-Howell is something other than that. But, the results are what they are.)

[the cyclist]

I was able to replicate the example on the bottom right of page 483 of the Crawford-Howell table:

kase = 33;
x = 26.457514;
kontrol = [50-x repmat(50,1,13) 50+x]; % This has mean 50 and standard deviation 10, as specified in the paper's example.

[h p ci stats] = ttest2(kase,kontrol)

Answer 2

[Star Strider]

I don’t have ‘R’, so I transliterated it as best I could understand it. (I won’t post it as a File Exchange contribution, since I have no desire to pay US$39 to download the paper to be sure I got it correct.)

It runs. I will let you determine if it gives correct answers:

function [tval,degfre,pval] = CrawfordHowell(kase,kontrol)
    lenk = length(kontrol);
    tval = (kase - mean(kontrol))./(std(kontrol).*sqrt((lenk+1)./lenk));
    degfre = lenk-1;
    pval = 2*(1-tcdf(abs(tval),degfre));
end

[t,d,p] = CrawfordHowell(1, rand(1,10))     % Test

I used ‘kase’ and ‘kontrol’ because ‘case’ is a reserved word in MATLAB and I just decided to have fun with ‘kontrol’.

[the cyclist]

StarStrider's function:

rng('default')
[t,d,p] = CrawfordHowell(1, rand(1,10))

and the built-in MATLAB function:

rng('default')
[h p ci stats] = ttest2(1,rand(1,10))

give the same result.

[Star Strider]

I appreciate the original paper. I didn’t find it in my search.

I’m not advocating anything — just doing my best to answer the question! All I did was convert the R function, and test it to be sure the code worked before I posted it.

I’m not surprised that ttest2 and the Crawford-Howell function give the same results. Apparently, Crawford and Howell thought there was some need for their approach in neuropsychology applications and got it published.

[AwedBy Matlab]

Thanks a lot to both of you for your help!