Announcement

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

  • log transformation

    My data set contains both positive and negative values , how can we convert it into log form.

  • #2
    Logarithms are not defined for zero arguments and (as far as statistical applications go generally) are not defined usefully for negative arguments. So, a short answer is that you can't do that. A better answer on what you should do depends on your data and on why you think you should use logarithms at all, so please explain.

    Comment


    • #3
      i want to transform the data into log just to achieve normality of the data. my question is that, i am going to use panel ARDL, so is it necessary to have data normally distributed or i go without it.

      Comment


      • #4
        I am no kind of expert on ARDL to say what it is best for it. I have never used it but Sebastian Kripfganz might have helpful comments. Usually in statistics, it can help if conditional distributions are roughly normal, but marginal normality is rarely a requirement, or even an ideal.

        If a variable is variously negative, zero or positive usually the sign has meaning and is best preserved. If you show the results of

        Code:
        summarize whatever, detail 
        
        inspect whatever
        where whatever should be replaced by the variable name in question, there might be more specific advice,

        Comment


        • #5
          sir, can i have your email id

          Comment


          • #6
            It's easy to find an email for me, but I don't encourage that here. If your data are confidential or sensitive, they wouldn't be made less so by sending me a copy, and I can't imagine a good reason otherwise to deprive anybody interested of further sight of this thread.

            Comment


            • #7

              . summarize LaFayAnimal , detail

              LaFay: Animal
              -------------------------------------------------------------
              Percentiles Smallest
              1% -.097255 -.125772
              5% -.077227 -.097255
              10% -.0566905 -.0923973 Obs 125
              25% -.0269601 -.0839156 Sum of wgt. 125

              50% .0013215 Mean .0070494
              Largest Std. dev. .062757
              75% .0129267 .1626
              90% .0923692 .1656 Variance .0039384
              95% .1309 .1695881 Skewness 1.795134
              99% .1695881 .346236 Kurtosis 9.382075
              what do you make of it

              Comment


              • #8
                Is this puzzle corner? Are you "Guest" returned?

                Either way, this variable is summarized so that we have the four smallest, the four largest, and estimates of 1, 5, 10, 25. 50, 75, 90, 95, 99% points of 125 values. Using the stated percents in probability form and using (rank - 0.5) / 125 for points with ranks 1,2,3,4 and 122,123,124,125, we can fake a normal quantile plot.

                Click image for larger version

Name:	fake_normal.png
Views:	1
Size:	41.1 KB
ID:	1738952


                (The surgery required would have been easier if you had used CODE mark-up. Please read the FAQ Advice through before your next question.)

                As suggested also by skewness and kurtosis results. the distribution is right-skewed. longer-tailed than normal, and there is a modest outlier.

                Even if this is the response or outcome variable for some model that has an ideal condition of normal errors, I still wouldn't seek to transform it before modelling myself.

                What does zero mean here any way? Is it some reference level with meaning?

                Comment

                Working...
                X