[VoxBo] g design questions

[Charan Ranganath, Ph.D.]

Thanks Dongbo. This raises a new question--say if you have two 
covariates, A & B, in your diagonal set, but there are a different 
number of elements in each covariate (say, 40 trials for A and 20 trials 
for condition B), and your goal is to compare conditions A and B. Are 
there any implications of mean scaling the diagonal set in terms of 
comparing A and B?

Also, what are the implications of NOT mean scaling the covariates?

CR

Dongbo Hu wrote:
> On Thu, 14 Sep 2006, Charan Ranganath wrote:
> 
>> _Hi, I've realized that the g-design widget has evolved quickly and that
>> _there are now a number of parameters now available that I don't fully
>> _understand. Consequently, I've been going consistently with the default
>> _settings, but I want to see what they mean and when to use each setting.
>> _
>> _I'll start with 3 questions:
>> _(1) When adding a diagonal set, you have a choice between "do not scale"
>> _and "scale by trial count". Do not scale is checked by default. When
>> _would you want to scale an element of a diagonal set by a trial count,
>> _and how exactly does it work?
>> _
> 
> This option (and the next one) are added to make "add diagonal set"
> operation more consistent with "add contrast", because diagonal set is
> basically the covariates with contrast matrix of "1 0 0 ... " (depends on
> how many unique keys are available in condition function).
> 
> What "scale by trial count" option does is exactly saem as the one in "add
> contrast" interface. It counts the number of elements that equals 1, then
> divide the 1's by this counter number. For example, If one of the
> covariates in diagonal set is:
> 
> 1
> 0
> 0
> 1
> 0
> 1
> 
> After "scale by trial count", it will become:
> 
> 0.333
> 0
> 0
> 0.333
> 0
> 0.333
> 
> (Because the number of 1's in the covariate is 3, so each element of 1 is
> divided by 3.)
> 
> Before this option was introduced in diagonal set, diagonal set covariates
> were not scaled by trial, so it is turned off by default.
> 
>> _(2) The next choices when adding a diagonal set are "leave offset" or
>> _"center and normalize," the latter of which is set by default. What does
>> _this mean, and what are the implications of using one or the other?
>> _
> 
> Again, this option is added to make it consistent with "add contrast"
> operation. What it does is actually mean-center the covariates. Since the
> older version mean-centers diagonal set silently, it is turned on by
> default.
> 
>> _(3) At the bottom right of the g-design widget window, there is another
>> _check box that says, "mean center all but intercept before saving". I
>> _believe this is checked by default. What are the implications of doing
>> _this, and when would you NOT want to do it.
>> _
> 
> This option is added because we noticed that most of the users will want
> to mean center all covariates except intercept before finishing the G
> design. The earlier version didn't do it automatically. This may cause
> a compatibility problem, but mean-center operation only shifts coavariates
> up or down, so it should be ok in most cases.
> 
>> _Finally, just to verify backward compatibility, do the default settings
>> _correspond to what was set by default in earlier versions of VoxBo?
>> _
>> _Thanks in advance for your wisdom, CR
>> _
>> _--
>> _Charan Ranganath, Ph.D.
>