I am trying to apply the binary model with dynamic panel data from Woolridge (2005) in the case of the SHARE dataset.
From Woolridge (2005):
The log-likelihood equation of our model assumes complete data availability for all cross-sectional units across all time periods. When confronted with unbalanced panels and specific sample selection mechanisms, we can still apply the identical conditional log-likelihood to a subset of observations, creating a balanced panel. Let \(s_i\) act as a selection indicator: \(s_i = 1\) signifies data availability for all time periods, including \(y_{i0}\), and 0 otherwise. Should \(\{y_{i1}, \ldots, y_{iT}\}\) and \(s_i\) exhibit independence given \(\{y_{i0}, z_i\}\), the Maximum Likelihood Estimation (MLE) using the balanced panel remains consistently estimable.
The SHARE dataset is an unbalanced panel that I balanced through the use of the variable wave participation present in the dataset. However, in my variable of interest y there are still present missing values, however when running the xtprobit with the lagged variable y the model gives me results. I wanted to know how Stata is handling missing values in the xtprobit function? Does it mean that my results are not consistent? Is this like having not balanced panles?
Thanks a lot
From Woolridge (2005):
The log-likelihood equation of our model assumes complete data availability for all cross-sectional units across all time periods. When confronted with unbalanced panels and specific sample selection mechanisms, we can still apply the identical conditional log-likelihood to a subset of observations, creating a balanced panel. Let \(s_i\) act as a selection indicator: \(s_i = 1\) signifies data availability for all time periods, including \(y_{i0}\), and 0 otherwise. Should \(\{y_{i1}, \ldots, y_{iT}\}\) and \(s_i\) exhibit independence given \(\{y_{i0}, z_i\}\), the Maximum Likelihood Estimation (MLE) using the balanced panel remains consistently estimable.
The SHARE dataset is an unbalanced panel that I balanced through the use of the variable wave participation present in the dataset. However, in my variable of interest y there are still present missing values, however when running the xtprobit with the lagged variable y the model gives me results. I wanted to know how Stata is handling missing values in the xtprobit function? Does it mean that my results are not consistent? Is this like having not balanced panles?
Thanks a lot
