Transforming DV
Hello,
My DV has a slight negative skew. I am only selecting some levels for one of my IVs so I was wondering if I need to fix the skew with the cases I'm looking at or do I fix the skew with all the cases of DV (all 140 participants instead of 118)? The reason I ask is because if I decide to delete outliers the SD will change, which will affect what values are outliers.
My follow up question is for a negative skew what does it mean to reverse the score? And what does it mean to account for change in direction in interpretation?? On the slides it says we can dichotomize, but I thought that is only acceptable for IVs not DVs?
posted by anonymous @ 3:24 PM
1 Comments:
At 6:07 AM,
Mari Clements said…
Fix outliers prior to addressing skew.
As we discussed in class way back then, reversing the score means recoding values such that the higher scores are the lower ones. The easiest way to do this is to use "transform --> compute" and create a new value that is the difference score of one value more than the highest value in the sample minus all current values.
So, for instance, if you had a measure whose highest score in the data set were 4, you would use
transform
compute
newvariable = 5 - oldvariable
In this way, then, all 4s would become 1s, all 3s would become 2s and so on.
All assumptions should be evaluated on the data you actually use in the analysis.
Finally, also as we discussed that long ago day in class, values that are dichotomized as 0 and 1 take on properties that make them acceptable to treat as continuous variables. Dr. Kim will say much more about this in the spring.
However, note that considering this kind of transformation is only appropriate with SEVERE (not slight) skew.
Post a Comment
<< Home