GPA -> Difference -> Regression/PCA
Take the difference after GPA and before data reduction or non-parametric
testing. E.g.,
Cevidanes, Lucia H. S., Alexandre A. Franco, Guido Gerig, William R.
Proffit, Dennis E. Slice, Donald H. Enlow, Helio K. Yamashita, Yong-Jik
Kim, Marco A. Scanavini, and Julio W. Vigorito. “Assessment of Mandibular
Growth and Response to Orthopedic Treatment with 3-Dimensional Magnetic
Resonance Images.” *American Journal of Orthodontics and Dentofacial
Orthopedics* 128, no. 1 (July 1, 2005): 16–26.
doi:10.1016/j.ajodo.2004.03.032.
-ds
On Wednesday, August 19, 2015 at 8:48:42 AM UTC-4, lv xiao wrote:
>
> Dear all,
>
> *My example:*
>
> I have 150 participants in my sample. At time point 1 (eg., at birth), I
> collected some background information (including both categorical and
> continuous variables) from the participants. Images of the participants
> were taken at time point 2 (eg., 6 months) and time point 3 (eg., 12
> months).
>
> *My aim:*
>
> I wish to investigate whether those background information collected at
> birth predict participants' shape changes from 6 to 12 months.
>
> *Proposed methods:*
>
> Perform between-group PCA (BGPCA) in steps below:
>
> Step 1: Group the shape data by time point. One group for 6 months data,
> and the other for 12 months data;
> Step 2: Perform BGPCA (in PCAGen8) and extract the most significant
> principal components (PCs) from scree plot;
> Step 3: Extract the PC scores ("Save PCA Scores" button in PCAGen8)
> Step 4: Since each participant appeared in both groups, I can calculate
> the difference of each participant's PC scores between the two groups (time
> points) for each extracted PC;
> Step 5: Each participant is then associated with one score reflecting
> his/her shape change along a particular PC (eg., PC 1). This score is then
> used as the dependent variable so that conventional
> statistical analyses such as simple linear regression and ANOVA, can be
> performed.
>
> *My question:*
>
> Q 1: Is the steps above appropriate?
> Q 2: Since the same participants are in both groups (groups differ only in
> the chronological order the shapes were measured), the two groups are
> correlated instead of being independent. Will such
> correlation affect the correctness of BGPCA in this example. Is there any
> way out if correlation is indeed a problem?
>
> Thank you very much!
>
> Best regards,
> Patrick
>
--
MORPHMET may be accessed via its webpage at http://www.morphometrics.org
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
