Dear Daniel,
> More recently we have performed a H2O-PET study to investigate the
> functional neuroanatomy of prepulse inhibition (PPI) of the startle
> reflex in normals, a measure of sensory gating or flitering (used in
> schizophrenia rsearch). This paradigama implicates that a weak prepulse
> (PP)given shortly before a loud pulse (P) reduces the eyblink response
> compared to the pulse alone condition. In humans the blink reflex
> component of the startle reflex is measured using EMG of the
> orbicularis occuli muscle.
>
> The design involves the following 3 conditions, replicated 3 times,
> pseudorandomized
> NS (== background noice),PP (==weak prepulse),P(==load pulse alone)
>
> covariates measured with each condition:
> NS : eyeblink amplitude( ==0),PP: eyeblink amplitude; P: eyeblink amplitude
>
> design used: NS P PP NS PP P NS P PP
>
> we have the following questions:
>
> 1) since brain activation using the starle pardigma is suggested to
> habituate under the pulse alone and possibly also under the
> prepulse-pulse conditions, we wanted to use the coresponding eyeblink
> amplitudes as "covariates" to weight each single scan or do we have to
> use the amplitude measures as "confounding" variables to weight each
> scan for each person so that we still can use contrasts?
>
> Or do we have to weight the contrasts using the corresponding average
> of the amplitudes measured during each scan?
In fact I would not use the amplitudes measures at all. These
represent dependent variables that are caused by adaptation. If you
model adaptation the behavioural measure need not enter into the model
at all. You could use the measures to contrain the nature of the time
x condition interactions but given you only have 3 replications I would
simply model three separate conditions (i.e. 9 conditions per subject
corresponding to a 3 x 3 design (NS P PP) x (1st 2nd 3rd). I think you
have already done this.
> 2) To determine the activation/deactivation pattern between the first P
> and the second P alone condition we used the following contrasts:
>
>
> design: NS P PP NS PP P NS P PPO
>
> contrasts: 0 1 0 0 0 -1 0 0 0
>
> is that correct?
Yes - this is a simple main effect of time under level P.
> 3) what is the difference if we would have only used the first and second
> pulse alone scans?
>
> design: P1 P2
> contrasts: 1 -1
>
> Does a setting of a contrasts to 0 implicate that this scan is omitted
> or is it still "somehow" included in the calculation of the total
> variance across scans?
Yes all scans contribute to the error variance estimation. The two
approaches imply different statiical models. The original (9
condition) is more powerful if sphericity assumptions hold.
> 4) What is generally used for H20-PET : is it proportional scaling or ANCOV=
> A
Subject-specific AnCova (unless there are large [> 20%] variations in
global counts).
> 5) What number for extended threshold is normally used? And what does
> it implicate?
See the appropriate help facilities. The default is 0.
I hope this helps - Karl
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|