Dear all,
I am estimating a model for a count-dependent variable by QMLE with fixed effects. My dataset ranges from 2000-2020.
I include one of my independent variables with two lags. (The theoretical consideration is that my outcome variable (count of certain patents) will be affected not only by the IV at t=0 but also by its lags due to the general time-intensive nature of innovation processes).
The independent variable is only significant when I include these lags. If I only include the contemporaneous value, my IV is not significant. Furthermore, the signs of the coefficients for my IV and its lags are different (positive coefficient for the contemporaneous X and negative for Xt-1 and Xt-2, X and Xt-1 are both significant, Xt-2 not). Running the regression solely with the lagged variable Xt-1 or Xt-2 not controlling for the other values also results in an insignificant coefficient.
My first intuition would be that I have to include both (X and Xt-1) in my model to avoid omitted variable bias and also due to the theoretical considerations mentioned above. However, I was also thinking about the issue of suppression as my contemporaneous X loses its significance when I do not control for Xt-1 and vice versa. Might that be an issue that I should be worried about?
I have also created the first difference of X using the "d."-command. This new variable is significant which might be an indicator that the growth rate of my IV should be the focus of interest.
Thank you very much.
Patrick
I am estimating a model for a count-dependent variable by QMLE with fixed effects. My dataset ranges from 2000-2020.
I include one of my independent variables with two lags. (The theoretical consideration is that my outcome variable (count of certain patents) will be affected not only by the IV at t=0 but also by its lags due to the general time-intensive nature of innovation processes).
The independent variable is only significant when I include these lags. If I only include the contemporaneous value, my IV is not significant. Furthermore, the signs of the coefficients for my IV and its lags are different (positive coefficient for the contemporaneous X and negative for Xt-1 and Xt-2, X and Xt-1 are both significant, Xt-2 not). Running the regression solely with the lagged variable Xt-1 or Xt-2 not controlling for the other values also results in an insignificant coefficient.
My first intuition would be that I have to include both (X and Xt-1) in my model to avoid omitted variable bias and also due to the theoretical considerations mentioned above. However, I was also thinking about the issue of suppression as my contemporaneous X loses its significance when I do not control for Xt-1 and vice versa. Might that be an issue that I should be worried about?
I have also created the first difference of X using the "d."-command. This new variable is significant which might be an indicator that the growth rate of my IV should be the focus of interest.
Thank you very much.
Patrick
