0

I am apply log-normalization to my non-normal data. However, when I apply log-normalization, the 0 values is converted into inf, which makes it impossible to apply some basic statistics on the data (e.g., correlation). Therefore, I add the score of 1 to all data values. However, I am afraid this is not a very valid approach. I wonder what is a common approach to deal with such situations.

renakre
  • 755
  • 1
  • 9
  • 25
  • 4
    The short answer is no. – Michael R. Chernick Feb 12 '17 at 13:01
  • @MichaelChernick thanks! But then what would be a better approach? – renakre Feb 12 '17 at 13:02
  • There is no need to normalize What is your model and what data do you have? – Michael R. Chernick Feb 12 '17 at 13:07
  • I have number of discussion post by each student. However some students have 40 posts whereas some others have 1 or 3 or 10 posts. I thought I should apply log-normalization to get a more normally distributed data. Am I wrong in doing so? – renakre Feb 12 '17 at 13:10
  • 2
    State what you are doing with the data that requires normality. And note that if you have a zero it is not very logical to log that variable. – Frank Harrell Feb 12 '17 at 13:46
  • @FrankHarrell I am trying to calculate the correlation among two variables that can have any value (no range). – renakre Feb 12 '17 at 13:48
  • Perhaps you should ask instead "If one of my variables is skewed can I still use the Pearson product moment correlation?". – mdewey Feb 12 '17 at 14:04
  • 2
    Please tell us your problem. So far, after @MichaelChernick and FrankHarrell asked, you've given about half the problem. My blog post [how to ask a statistics question)[http://www.statisticalanalysisconsulting.com/how-to-ask-a-statistics-question/) may help. – Peter Flom Feb 12 '17 at 14:06
  • "Trying to apply some basic statistics" is a bad goal. State your research question. – Peter Flom Feb 12 '17 at 14:08
  • Another possible duplicate: [How should I transform non-negative data including zeros?](http://stats.stackexchange.com/q/1444/7290) – gung - Reinstate Monica Feb 12 '17 at 14:13
  • 2
    Sounds like a problem for Spearman or other rank correlation measure and be done with it. – Frank Harrell Feb 12 '17 at 14:53

0 Answers0