So, the book doesn't recommend the need for transformation if there is a problem with normality (p.46). I think they give a reason for this, but if it's their reason, i don't understand it. They even say that "many regression textbooks" do find this type important. I know transforming is largely subjective...but what is the rule of thumb for this?
Thanks in advance for the clarification!!!
Subscribe to:
Post Comments (Atom)
3 comments:
It seems to me that the authors' reason for saying that the assumption for normality is least important is because the regression itself is not greatly affected by non-normality (i.e. you can still trust the results and use the regression equation meaningfully). They mention that most textbooks err on the side of caution and would go ahead and check the assumption and transform it as necessary. These authors say that it is not worth the time. The following paragraph explores ways of exploring normality if it is something of interest in the study (for theoretical or experiement reasons). In terms of a rule of thumb, I would check with Mari if you want a more specific one. I would always check the normality of the residuals because you want to know that information before you proceed with your study. You can make your decision about transforming based on how bad the skew is (and how badly a transformation might affect other assumptions that may or may not be violated). For this class, you will definitely need to consider checking this assumption a standard part of your research process (i.e. for the final project).
Even better news is that Kris's upcoming lecture about outliers, leverage, and influence will answer the question. He will cover how you determine the degree to which deviations from normality influence your results.
And as Kelly said, checking assumptions will be an important part of your final project...
Post a Comment