Hi,

I've been provided with a lot of help by someone whom I engaged with through this network, whom has helped me to run an Interrupted Time Series Analysis using a standard Generalised Least Squares approach. However, I have also tried to use an alternative approach and wondered if anyone could offer any views/advice in regards to its validity.

I have a univariate time series consisting of 48 observations [quarters of a 12 year data set]. Each observation is a record of the area of defined 'green space' land that was subject to development [as m2 per km2] in each quarter. I want to quantify the effect of policy change after a two-year lag and have accordingly built an ITS as outlined.

However, I have also taken an alternative approach in which I fit a Dynamic Linear Model separately to the pre-policy period [as a separate data frame] , from which I forecast a counterfactual. having fit a separate dlm to the post-policy period I analysed the intervention effect as the absolute difefrences between the modelled post-policy period and predicted counterfactual (with 95% confidence bounds).

The second method gives similar results to the first, but picks up the trend in the pre-policy period better than the GLS method. However, I'm not sure whether modelling them separately would be seen as valid by anyone who knew what they were doing with ITS!

Does anyone have any advice? I wondered whether there is a way of replicating the use of ARIMA methods for ITS using a dynamic linear model, but am not entirely sure of how the ARIMA approach would work? Again, any advice would be hugely appreciated.

I have attached a copy of my derived data out of interest.