+ Reply to Thread
Results 1 to 4 of 4

Thread: multivariate analysis?

  1. #1
    Points: 1,320, Level: 20
    Level completed: 20%, Points required for next Level: 80

    Posts
    9
    Thanks
    4
    Thanked 0 Times in 0 Posts

    multivariate analysis?




    I need to run a statistical analyis to determine which variables are associated with outcome.

    its medical study outcomes like alive/dead, and independent variables like , height, sex, age, etc .

    So I have 2 separate data sets - firstly one where the dependent outcome variable is continuous data with multilpe independent variables some continous and some catagorical.

    the second data set is very similar but the dependent variable is catagorical data.

    I believe that I need a multivariate analysis - perhaps a logistic regression ?

    Thanks ,

    Ed


    one dependent variable continuous data
    multiple independent variables both continuous and categorical

    one dependent variable categorical
    multiple independent variables both continuous and categorical

  2. #2
    Fortran must die
    Points: 58,790, Level: 100
    Level completed: 0%, Points required for next Level: 0
    noetsi's Avatar
    Posts
    6,532
    Thanks
    692
    Thanked 915 Times in 874 Posts

    Re: multivariate analysis?

    While the term multivariate, I now know, means analysis with multiple dependent variables a lot of people (like me) associate this term with more than one independent variables (because the authors I read used it that way). Be careful not to get confused by responses by posters who misunderstand what you mean by multivariate.

    I am not entirely sure what you are trying to do. If you want to know how independent predictors are related to a continuous dependent variable than linear regression works (or ANOVA). If you have a categorical dependent variable than, depending on how many categories it has and if they are ordinal or nominal, you would use some form of logistic regression.

    I can't help with models simualtaneously predicting more than one DV as I don't know these methods. I do know that MANOVA and canonical correlations are two such approaches.
    "Very few theories have been abandoned because they were found to be invalid on the basis of empirical evidence...." Spanos, 1995

  3. The Following User Says Thank You to noetsi For This Useful Post:

    edmundfitz (10-07-2013)

  4. #3
    Points: 1,320, Level: 20
    Level completed: 20%, Points required for next Level: 80

    Posts
    9
    Thanks
    4
    Thanked 0 Times in 0 Posts

    Re: multivariate analysis?

    Thanks for your reply

    What I'm trying to do is to appropriately analyse the following

    I have a data set of 250 patients which looks at 2 specific outcomes in relation to overall outcome and function after injury

    Outcome 1 is binary - alive or dead
    Outcome 2 is ordinal - hand function scoring 0 to 30

    The variables at play are different in their data type , some are

    Categorical data - male vs female
    Ordinal data - 1 vs 2 vs 3 operations
    Continuous data- age

    There are others but theses are examples

    So should I then run a linear regression for the continuous data variables , and a logistic regression for the categorical ones?


    I don't need to combine the outcomes , I'm just looking to state things like -- young age was associated with higher hand function , or more operations is a predictor of increase chance of death



    Is that clearer ? I'm sorry I don't have the best grasp of stats and so often don't use the correct lingo but I'm working on it !

    Thanks

    Ed

  5. #4
    Fortran must die
    Points: 58,790, Level: 100
    Level completed: 0%, Points required for next Level: 0
    noetsi's Avatar
    Posts
    6,532
    Thanks
    692
    Thanked 915 Times in 874 Posts

    Re: multivariate analysis?


    It looks like then you want to predict two outcomes. Outcome 1 is binary so you use binary logistic regression. Outcome two is ordinal so you could use ordered logistic regression with two proviso. First, there is a condition for ordered logistic regression that applies only to it in logistic regression, that it does not make a difference how you group categories, this is tested by the score test of proportional odds assumption in SAS. If you do not pass this you can either make your variable binary or use multinomial logistic regression. Second with 30 distinct levels ordinal regression bogs down commonly. With that many levels your dependent variable is likely interval like so you can use linear regression in practice.

    If you really want to know how a set of IV influence two DV and they in turn another DV you would be better off doing Confirmatory Factor Analysis (one of the branch of structural equation models) although that takes a while to learn.

    What your independent variables are (the form they take) is not critical although you should make categorical variables into dummies.
    "Very few theories have been abandoned because they were found to be invalid on the basis of empirical evidence...." Spanos, 1995

  6. The Following User Says Thank You to noetsi For This Useful Post:

    edmundfitz (10-07-2013)

+ Reply to Thread

           




Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts






Advertise on Talk Stats