At trying to change the skew of my data, do I do all reverse, square root, logarithm, and inverse? or can I just go as a far a I need in order to reach a normal distribution?
Also what is considered to be a bad skew, is a skew beyond negative or positive 2.5 a bad skew?
Also what is considered to be a bad skew, is a skew beyond negative or positive 2.5 a bad skew?
posted by Unknown @ 12:22 PM
3 Comments:
At 10:18 AM,
T.J. said…
Skew is assessed by superimposing a normal curve over the histogram, not so much numerically. (Numerical tests are for small samples.)
Generally, you do not do more than one transformation like sqrt or log. (Reverse scoring is different). A log transform can be mathematically adjusted for a "stronger" transformation. Also, dichotomizing would make a normal distribution by definition, if I'm thinking of it right.
Have you looked at outliers as well?
At 10:45 AM,
Unknown said…
Yeah I have no outliers, there is something that the book recommended I do where I add an extra point to my highest numbered varaible then go to SPSS do a Lg10(biggest value - socre).
At 6:23 AM,
Mari Clements said…
Reverse scoring is only necessary for negative skew. (See post above.)
What you are describing above is, in fact, a log transformation. (Log10, to be exact.)
Post a Comment
<< Home