Announcement

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

  • OR and CI interpretation

    Dear all, I have difficulties of interpretation of odds ratio (OR) & 95% Confidence interval ( 95% CI), my dependent variable is childhood diarrhea and one of my independent variable is Birth order. Afrer I regress multilevel mixed effects logistic regression, the AOR (95% CI) becomes 2.76e-07 (3.8e-137,2.0e+123). What does this indicate? what is the value of" e" ? How can I interprete this ? Sorry for all, I am a new Stata user and a new member for Statalist forum.
    With Kind regards, Hassen
    Last edited by Hassen Ali; 22 Jun 2018, 02:01.

  • #2
    The notation 2.76e-07 means 2.76 * 10-7. So your result here is an adjusted odds ratio that is very close to zero and whose 95% CI is approximately from 0 to infinity. In other words, this odds ratio is, for practical purposes, not estimable from the data you have used. You don't provide example data or complete output, but the most common situation in which this would happen is if the variable this is the odds ratio of, is an indicator variable that is nearly always 0 or nearly always 1. If you -tab- that variable, you will probably see extremely lopsided results.

    Comment


    • #3
      This is usually named ‘scientific notation’. Values can be very large or very small, with several numbers after the dot, hence the use of it as an abridged alternative. You may wish to search for ‘scientific notation’ and get a further grasp with the topic.
      Best regards,

      Marcos

      Comment


      • #4
        Dear Clyde Schechter & Marcos Almeida Thank you very much for your assistance and guidance!! I wish you have an endless happiness and success in your life! I miss you guys.
        Dear Clyde, Because of I am using mobile phone, I haven't sent the sample or complete output of my data. Sorry for all my mistakes that I made.
        One Question is,can this outputs be usable? or they should be avoided from the output?
        With Best Wishes,Hassen


        Comment


        • #5
          can this outputs be usable?
          The results are almost certainly not useful: they are scarcely better than saying that the OR is somewhere between 0 and infinity, which, of course, you know to be true of any variable in any model before you even gather any data. It is difficult for me to imagine any situation where the degree of narrowing down from the 0 to infinity range that bounds all odds ratios provided by this result would be enough to be useful.

          However, the output may be useful in a different sense. It is extremely unusual to find data that is like this in the real world. You have to work very hard to contrive an example that comes out like this. So the output may be useful as a warning sign to you that there is something wrong with your data! If the variable here is x, and if your outcome variable is y, run -tab x y- to see what is going on and figure out whether this result looks plausible. I would say it is likely that there is something wrong with your data.

          Added: Bear in mind that in Stata, dichotomous variables are coded 0 = no, anything other than 0 = yes. If your variable x or y is coded 1 = yes 2 = no (as is often the case in data imported from other statistical packages like SAS) then in fact your variable x or y is taken by Stata to be true in all cases, which is probably not what you intend. Another possibility is that one of these variables is coded 1 = yes and no is represented as missing value (a common phenomenon in data imported from Excel) and this, will result in all of the values that you intend as no to be either omitted from the analysis altogether, or, for those commands that include them, interpreted as yes.
          Last edited by Clyde Schechter; 23 Jun 2018, 23:51.

          Comment


          • #6
            Dear Clyde, Thank you very much!
            I cannot fully express my appreciation for all of the guidance and support you have shown me to solve my many challenges. Your support has opened numerous doors and opportunities for me that I never would have had. There is no doubt I will always remember your kindness.
            As I described in #1, my dependent variable is childhood diarrhea status which was coded as 0=yes (having diarrhea),1=no and one of the independent variable is Birth order which was coded as 1="first born",2="2-3",3=">=3". So, has it had effect on analysis as I have coded birth order started from "1" rather than "0"? I am using Stata 14.0/SE.
            With Best Wishes, Hassen

            Comment


            • #7
              My advice is: no doubt, the number 1 shall be reserved to your outcome variable, not to the lack of outcome. But there must be something weird with the modeling itself, for in your message only one depvar and one indepvar were mentioned, as it were a ‘regular’ logistic model. However, you mentioned your model as a multilevel logistic mixed effects model. On account of that, if you want to know what further happened to the data, the best approach is presenting exactly what you typed and exactly what you got as a result. A short example of the data will also be helpful. You may use - dataex - for that matter.
              Best regards,

              Marcos

              Comment


              • #8
                I would just add my endorsement of Marcos Almeida's suggestions at this point. Without seeing an example of the data along with the actual command and the actual output, it is not possible to provide additional advice.

                Comment


                • #9
                  Dear Clyde and Marcos, Thank you very much for your nicely assistance and guidance!!
                  Sorry, I am using Huawei mobile. So, I can't send example from my data.
                  Again, Thank you very much for your kindly help me!!! I miss you, guys.
                  With Best Wishes, Hassen

                  Comment

                  Working...
                  X