|
|
GMM for skewed positive dependent var... |
|
Message/Author |
|
SF Wang posted on Friday, August 08, 2014 - 1:20 pm
|
|
|
Dear professors, when we do GMM for positive extremely-skewed dependent variables, should we do transformation so that the dependent variable is roughly normally distributed? The variable we'd like to model varies from 0 to 18, but most scored from 0-2. Thanks. |
|
|
I am not fond of making variable transformations, but instead prefer to make the model accommodate the variables they way they are. You should investigate why/how the DVs are so skewed. Do they have floor or ceiling effects in which case 2-part GMM might be more suitable? Or, are they skewed without floor/ceiling effects in which case the new option Distribution = skewt can be used. |
|
Back to top |
|
|