I am trying to analyze a set of nonnegative continuous non-integer data (i.e. the data points are not counts) that are mostly between 0 and 3 whose distribution is highly right-skewed even after log transformation. I am thinking that one possibility may be hurdle model to model the zeros and positive data points separately, but could anyone please suggest other possible choices? Covariates include categorical and non-negative continuous variables.

  • 2
    $\begingroup$ If it's not a big secret, what are the values? Measurements of something? Times? Ratios of things? Compositional data/proportions? Why did you try to transform it? If there are zeroes, why would you take logs? $\endgroup$ – Glen_b Mar 31 '14 at 23:34
  • $\begingroup$ I've added the zero-inflation tag, but you'll probably want to remove that if your data don't contain lots of true zeros. The gamma distribution may help if the data resemble a Poisson distribution (for counts). Also, it doesn't sound like you've explained your purpose sufficiently since you're mentioning covariates but not what you want to do with them. $\endgroup$ – Nick Stauner Apr 1 '14 at 0:19
  • $\begingroup$ @ Glen_b: it's quality-control data to compare between the standard and the re-calibrated photo imaging machines. Attributes of the machines are the covariates, and outcome is the absolute value of difference in color intensity between the same spot of two identical photos printed using the standard and the re-calibrated machines. Calibration is good so a lot of times the difference is zero. Other times the difference may be small, so the outcome variable is a vector that looks like (0, 0.001338568, 0, 0, 0, 0, 0, 0, 0.002826946, 0, 0, 0, 0...) $\endgroup$ – HueSX Apr 1 '14 at 3:01
  • $\begingroup$ just ran into this--gamma regression? www-m4.ma.tum.de/fileadmin/w00bdb/www/czado/lec8.pdf $\endgroup$ – HueSX Apr 1 '14 at 4:37

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.