zero coefficient added for missing dummy var
Posted: Tue Aug 29, 2017 8:28 am
Hi,
I have set missing values on all variables in my dataset to be 99. My model contains some categorical variables. As an example, I have a binary variable with yes/no responses and reference category 'yes' but the model is adding a parameter estimate and standard error (both zero) for 'omitted or invalid' for this variable. (This happens sometimes even when there are no missing data on a variable).
Example:
ReadScoreij = B0j - 34.461(6.669)Noij + 0.000(0.000)Omitted or invalidij + eij.
Can I safely ignore the "omitted or invalid" parameter estimate and standard error (as missing data are dealt with using Realcom Impute) and proceed to use a t-test (or intervals and tests when two or more dummy vars) to determine significance? If MlwiN adds n dummies including one for 'omitted or invalid', is it reasonable to assess significance on the basis of n-1 functions in intervals and tests?
Many thanks
Lorraine
I have set missing values on all variables in my dataset to be 99. My model contains some categorical variables. As an example, I have a binary variable with yes/no responses and reference category 'yes' but the model is adding a parameter estimate and standard error (both zero) for 'omitted or invalid' for this variable. (This happens sometimes even when there are no missing data on a variable).
Example:
ReadScoreij = B0j - 34.461(6.669)Noij + 0.000(0.000)Omitted or invalidij + eij.
Can I safely ignore the "omitted or invalid" parameter estimate and standard error (as missing data are dealt with using Realcom Impute) and proceed to use a t-test (or intervals and tests when two or more dummy vars) to determine significance? If MlwiN adds n dummies including one for 'omitted or invalid', is it reasonable to assess significance on the basis of n-1 functions in intervals and tests?
Many thanks
Lorraine