Troubleshooting an error in var when running ttest

I'm trying to do a 1-sample t-test on a vector of data that has 116 elements in it, and I keep getting an error.

To troubleshoot, I tried a random vector, x. Here is the output:

x=randn(100,1)
[h,p,ci]=ttest(x,0.1)
Error using var
Too many input arguments.
Error in nanvar (line 42)
y = var(varargin{:},'omitnan');
Error in nanstd (line 35)
y = sqrt(nanvar(varargin{:}));
Error in ttest (line 136)
sdpop = nanstd(x,[],dim);

I tried this on a 2019a install on another computer and got the same error.

I've never seen this type of error, but it doesn't appear to be t-test. What would var be looking for in this instance?

[the cyclist]

I don't get that error from running

x=randn(100,1);
[h,p,ci]=ttest(x,0.1);

It's puzzling to me, and off the top of my head I can't think of an explanation.

However, I'd like to point out that using that syntax, you would not be doing a one-sample t-test (even if it worked correctly). Rather, that syntax is doing a paired t-test of your vector x against a vector of 0.1 values, i.e.

y = repmat(0.1,[100 1]);

I'm not sure what you intended, or the purpose of the 0.1 there. Maybe you meant that that is the value of alpha, the significance threshold?

[Michael Wright]

Thank you for answering this so quickly!

That's true, but if I run the example from the Mathworks website (which would be a one-sample ttest):

load stockreturns

x = stocks(:,3);
[h,p,ci,stats] = ttest(x)

I get the same error. It doesn't seem to matter how I run it, I get the same error.

Are you using 2020a?

I might try to reinstall...

Thanks again for the response...

UPDATE:

I tried it on my Windows PC running 2018b and it seemed to work (I normally work on a Mac running 2020a). Am going to try 2018 on the Mac and 2020a on the PC to see if there may be a version issue (they all have the same install base; Statistics toolbox and the Curve Fitting toolbox)

UPDATE 2:

Seems to be working fine on Windows Installs; 2018b, 2019a, and 2020a on my Mac give the error above...

2018b on the Mac was a clean install and right out of the chute gave the var error...

[the cyclist]

I'm running R2020a on a Mac running latest Catalina OS, and

load stockreturns
x = stocks(:,3);
[h,p,ci,stats] = ttest(x)

runs fine for me.

Do you think there is any chance you are accidentally running some other ttest function? What does

which ttest -all

give as output? For a typical install, it would be

/Applications/MATLAB_R2020a.app/toolbox/stats/stats/ttest.m

[Michael Wright]

I get the same; there are no other files conflicting.

I tried running it on Matlab Online and it worked fine...

I'm the admin on this Mac, so I'm not sure why this would be an issue, or why it's var throwing the error. I've never heard of nanvar before, so this is truly a headscratcher...

If I run var on just a random vector (x=1:100;, e.g.), that works fine...

I'm running R2020a on a Mac with Mojave (10.14.6) but, as I said, if I go back to 2018b (which shouldn't have a problem on Mojave) it doesn't work there either.

Again, my thanks for thinking on this issue!

[Adam Danz]

I'd also check that you're not shadowing the other functions mentioned in the error stack.

• var
• nanvar
• nanstd

You might be inadvertently using a different var() function.

[Steven Lord]

I agree with Adam Danz. I think the likeliest scenario is that you have a var.m that's taking precedence over the built-in var function. What does this show?

which -all var

[Michael Wright]

Thanks, Adam and Steven,

Yes, that was it; I downloaded a homemade toolbox of functions to analyze neural data, and one of the functions in it is called var.

Renaming that var fixed the problem; I will adjust the dependencies of files in that toolbox to call the renamed var function...

Thank you all so much; can't believe I missed that.

Cheers!

[Adam Danz]

If your instructor is the author of the toolbox, it may be helpful to let him/her know about this so it can be avoided in the future. var() is a very common function so I can imagine that you're not the first one to have this problem with that toolbox.

[Michael Wright]

That's a good point; I'm not sure the person who made it even supports it anymore (he worked in a colleague's lab over a decade ago), but I think I can track him down!

Cheers