Latin square design

[Charley Beller]

I'm using a latin square design with groups containing six elements each.  Currently when the experiment runs Ibex chooses to display only even or odd members of the groups.  That is, within a single run of the experiment only one of 1,3,5 from each group will be shown, while in another run only one of 2,4,6 will be shown.

  Is there a way that a single run of the experiment can display one of 1-6 within each group?  (If not I can hack a solution by reversing the order of half of the groups in the input.)

Thanks,

Charley

[Alex Drummond]

Hi Charley,

This seems like a strange behavior. Do you have an example data file
which demonstrates this problem?

Alex

[Luka Crnic]

Hi Alex,

I'm also facing an issue with Latin square. 

I am running two experiments in one (one is a filler for the other) and I have different number of conditions in the two experiments.

This means that I get two kinds of groups with differing number of types, e.g., groups of the first experiment have 6 types, groups of the second experiment have 4 types.

If I do the experiments separately, I get a correct number of experimental sentences. If I combine them, I get a lower number of sentences than expected (I didn't delve into what sentences are missing).

Do you know what I might be doing wrong? Is there a way of circumventing this issue?

Many thanks & best,

Luka 

[Alex Drummond]

Hi Luka,

again, I need to see an example data file to solve issues like this.

Alex

[Charley Beller]

Hi Alex,

  Here's a copy of the data file.  It's set up to run on mechanical turk, but does the same thing if run straight from ibex farm.

I appreciate any insight you have,

Charley

--
Charley Beller
(410)-243-0103
http://mind.cog.jhu.edu/grad-students/beller/

[Luka]

I solved this problem, trivial error on my part. Thanks, Luka

[Alex Drummond]

Ok, I can reproduce the issue now, looking into it...

Alex

[Alex Drummond]

Hi Charley,

It's because you are only including in your shuffle sequence half of the items that you've defined. The latin square logic applies first, so Ibex first selects on item from every group of 6 (with '*' indicating the selected item):

Group 1: *a b c d e f

Group 2: a *b c d e f

Group 3: a b *c d e f

Group 4:  a b c *d e f

Group 5:  a b c d *e f

Group 6:  a b *c d e *f

Group 7: *a b c d e f

Group 8: a *b c d e f

Group 9: a b *c d e f

Group 10:  a b c *d e f

Group 11:  a b c d *e f

Group 12:  a b *c d e *f

...

This is how selection would proceed with the counter set to 0. As the counter increments, the offset increments (so e.g., with the counter set to 2, 'c' would be chosen from Group 1, 'd' from Group 2, etc.)

What you are then doing is selecting items from half of these groups. Given the way the data file is arranged, you are actually selecting one item from every other group. Hence, depending on the value of the counter, you get either all a c and e or b d and f.

So basically, you may get weird latin square effects if you include items in the data file which you do not then include in your experiment. I should make this more clear in the docs.

Alex

[Charley Beller]

Thanks a lot Alex, that should be an easy fix.

Charley