Hi,
I want to estimate a model in Stata which uses Irish Labour Force Survey data. The Irish LFS is a quarterly survey where households are surveyed for five quarters, and then replaced, meaning every quarter a fifth of respondents are replaced. Thus there is 80% sample overlap between each quarter, leading to sample error auto-correlation between the quarters. This is of course a problem as it means the data is not truly panel data, and it is not independent cross sectional data either. I want to estimate my model using each quarter as an independent cross-section, but accounting for the serial correlation in the errors, as creating panel data is not feasible.
First of all, I'm open to hearing any ideas anyone has on carrying out such an analysis - maybe I'm mistaken and this is theoretically unsound.
Secondly, an article on this topic noted that the auto-correlation could be modelled by a moving average process of order five. I understand the very basics of generalised least squares, and know that Stata can handle GLS for auto-correlation processes of AR(1), but I'm having trouble finding out if this is possible for moving average processes at all, never mind the order. I believe R can handle it through its GLS function, but I'm not familiar enough with R to carry out the analysis using it - any help would be appreciated.
Thanks,
Evan
I want to estimate a model in Stata which uses Irish Labour Force Survey data. The Irish LFS is a quarterly survey where households are surveyed for five quarters, and then replaced, meaning every quarter a fifth of respondents are replaced. Thus there is 80% sample overlap between each quarter, leading to sample error auto-correlation between the quarters. This is of course a problem as it means the data is not truly panel data, and it is not independent cross sectional data either. I want to estimate my model using each quarter as an independent cross-section, but accounting for the serial correlation in the errors, as creating panel data is not feasible.
First of all, I'm open to hearing any ideas anyone has on carrying out such an analysis - maybe I'm mistaken and this is theoretically unsound.
Secondly, an article on this topic noted that the auto-correlation could be modelled by a moving average process of order five. I understand the very basics of generalised least squares, and know that Stata can handle GLS for auto-correlation processes of AR(1), but I'm having trouble finding out if this is possible for moving average processes at all, never mind the order. I believe R can handle it through its GLS function, but I'm not familiar enough with R to carry out the analysis using it - any help would be appreciated.
Thanks,
Evan
