1

My research includes a binomial variable as DV and numerous continuous variables as IV. From descriptive analysis I can tell that distribution for most IVs is not normal and couldn't find any transformation that can solve the problem. I applied log, 1/x, sqr, sqrt, ... . Here is the histogram of one of the main variables. What type of analysis suits these data (my choice was logistic regression) and how can I solve the normality issue?

Thanks enter image description here

user171254
  • 11
  • 3

1 Answers1

6

No assumptions are made or needed about the marginal distribution of the independent variables in logistic regression. You can safely not worry about this.

Matthew Drury
  • 33,314
  • 2
  • 101
  • 132
  • 1
    Furthermore, your data look rather normally distributed. – Pere Jul 26 '17 at 09:32
  • @Pere why do you say it looks normally distributed? The y axis range in the hundreds, there's holes in the data (zero) that certainly would not fluctuate statistically to 70 or 80 or 200. Also, the y axis label says "frequency" and that makes me suspicious that the actual uncertainty on the expected number of entries in each bin is not what you would expect from a Poisson distribution, but much smaller. – famargar Jul 26 '17 at 10:38
  • It might not be exactly normal, but I'd say that it doesn't seem far away from normal. As @SmallChess said, normality of predictor is not required in regression, but even when normality is required (e.g. normality of residuals), an small deviation from normal is not harmful. It's hard to say from looking, but I'd daresay that skewness and kurtosis of that sample is not very different from normal. – Pere Jul 26 '17 at 12:01