Announcement

Collapse
No announcement yet.
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • #16
    Hanne,

    I would be interested to know what solution you employed for this in the end, as I have stumbled across this post while trying to work out why my 4 component PCA and EFA results are so different in STATA, despite so many sources suggesting the results are often fairly similar. I now understand it, thanks to this posting. The solution I have employed is to:

    Run my pca in stata and rotate the solution in STATA using the following code:

    pca item1 item2 ..... item27, comp(4)
    rotate, varimax

    This gives me the 27x4 matrix of components (eigenvectors), the rotated version of this matrix and the 4x4 rotation matrix

    Run the code which normalises my matrix of eigenvectors, such that it is a matrix of loadings (and the sum of squares = eigenvalue)
    estat loadings, cnorm(eigen)

    As noted through this post, this normalises the unrotated matrix from the previous step. There doesn't seem to be a simple way to normalise the rotated matrix of eigenvectors or, equivalently, rotate the normalised matrix. I therefore copied both the normalised matrix, and the 4x4 rotation matrix from the previous step into Excel and carried out the rotation outside STATA. I am planning to get SPSS installed to see if the results I have are replicated by SPSS.

    Did you do something similar and/or does this make sense to you?

    Thanks.

    Comment


    • #17
      As a late addition to this discussion, I recently created the following example, following from Stata documentation, that shows the similarities and differences between the results of the pca and factor, pcf commands in simple circumstances.
      Code:
      sysuse auto, clear
      // factor analysis using principal components factors
      factor trunk weight length headroom, pcf
      // principal components analysis, default all components 
      pca trunk weight length headroom
      // principal components analysis, single component
      pca trunk weight length headroom, components(1)
      In general the output is identical, except that the pca reports eigenvalues while the factor analysis reports loadings.

      In posting this I saw for the first time the question from Catherine Murphy posted above. Without understanding what her objective is, I would be inclined to replace
      Code:
      pca item1 item2 ..... item27, comp(4)
      rotate, varimax
      with
      Code:
      factor item1 item2 ..... item27, pcf comp(4)
      rotate, varimax

      Comment


      • #18
        Hi, I'm new to this field. I'm doing my final year research that relevant to trade with gravity model estimates. I identified severe correlation between independent variables. When I regress data using ppml one independent variable(distance) or two dummy variables (country dummies) have dropped.After applying principle component analysis dropping of independent variable(distance) has stopped and dropping of dummy variables is existing. I want to know reason for this and should I apply PCA only for independent variables? should it apply also to dummy variables?

        Comment


        • #19
          Shanika: Your question has nothing to do with the title of the thread. Please start a new thread. But I don't think you give enough information to allow helpful comment. It's not even clear that PCA here has any advantages over thinking as an economist which variables should be omitted to make maximum sense.

          Comment


          • #20
            OK . Thank you Mr.Nick. I like to ask what will be the reason for omitting of variables. Is it adversely affect for the estimation?

            Comment


            • #21
              Please start a new thread, repost the question, and add more information.

              Comment


              • #22
                Ok Mr.Nick. Thank you very much for the reply

                Comment

                Working...
                X