02-29-2016 11:58 PM
I really need help because I have been looking for answers for my questions everywhere.
First I want to analyze my expenditure data which I know it is very highly skewed, as a result Procsurveyreg wasn't an option. Now I have been reading multiple articles about the issue of skewed data and how to deal with it. Now from my understanding there is what called Generalized Linear modeling (GLM). The GLM compared to the ordinary linear regression is more flexible such that the dependent variable (in my case the expenditure outcomes) may be of either a normal or non-normal distribution.
Three main components make up the GLM:
1) the dependent variable distribution known as the exponential family;
2) the independent variables which maybe linear in their relationship with the dependent variable
; and 3) a link function (http://www.sagepub.com/sites/default/files/upm-binaries/21121_Chapter_15.pdf). I learnt also there is what is called modified park test to tell you exactly about the three main component mentioned above.
After this brief introduction my questions are:
1- Is there a SAS procedure along with documentation explaining how to conduct GLM and park's test? I am not looking in the documentation how statistics are computed rather I am looking for an easy explanation about the procedure.
2- I have a SUEVEY data, so I would like to know a procedure in which I can include the strata, cluster and weight.
P.S. I am fully aware of Paper 1657-2014 (http://support.sas.com/resources/papers/proceedings14/1657-2014.pdf), however, I have to admit as a student that it is above my ability to comprehend.
Help is very needed.
03-01-2016 12:20 AM
Thank you for your fast reply
I read that errors must be normaly distributed, however, how to check for that? I only was able to know how to check for normality for my dependent variable (see attachment)
Now regarding the log-transformation, I considered it until I read back transformation introduce a bias in the analysis. But if you have the equation and refernce for back transformation I might consider it again.