ASA Connect

 View Only

  • 1.  Proc Traj

    Posted 11-13-2017 09:42
    Greetings all,

    Does anyone know how to interpret the B coefficients from Proc Traj in terms of Trajectory shape?  Basically, how does one describe the shape. My second issue is with STATA.  I have done Proc Traj both in SAS and STATA but my results are very different.  Any explanation?   

    Briefly, 550 with mild TBI were followed over time and were asked to report symptoms on a 22-questionnaire.  Outcome of interest is any severe/very severe symptom rated as >=3 on a scale of 0-4.

    Any help is welcomed!

    Hamid


    ------------------------------
    Hamid Ferdosi MD MPH
    ------------------------------


  • 2.  RE: Proc Traj

    Posted 11-14-2017 08:30
    Hello Hamid, 
    This reference may help.
    https://www.researchgate.net/profile/Bobby_Jones2/publication/266822262_Proc_TRAJ_A_SAS_Procedure_for_Group-Based_Modeling_of_Longitudinal_Data/links/551815fc0cf2f7d80a3d2779/Proc-TRAJ-A-SAS-Procedure-for-Group-Based-Modeling-of-Longitudinal-Data.pdf

    ------------------------------
    Kirk Ketelsen
    Clinical Asst Professor
    Boise State University
    ------------------------------

    Original Message


  • 3.  RE: Proc Traj

    Posted 11-15-2017 09:51
    Group-based trajectory analysis has a substantial literature,
    including the following paper,
    Daniel S. Nagin, Analyzing developmental trajectories: a
    semiparametric, group-based approach. Psychological Methods 1999;
    4(2):139-157
    and a book:
    Daniel Nagin, Group-Based Modeling of Development. Cambridge, MA:
    Harvard University Press, 2005.

    The methodology assumes that trajectories can be well approximated by
    polynomials, but I have not seen empirical evidence that supports such
    a narrow assumption. It is reasonable to expect trajectories to be
    nonlinear, but many likely nonlinear patterns are not well
    approximated by polynomials. In applications, it would be a good idea
    to start with some sort of exploratory analysis and let the data
    reveal potential trajectories (as well as subjects whose trajectories
    seem to be "outlying").

    In the present application, the outcome seems to be dichotomous. If
    so, the B coefficients are in the logit scale.

    David Hoaglin


    Original Message


Related Content