Hi,
I'm analysing time series data using a few different methods, including Dynamic Linear Models.
For such I am using the dlm package in R. However, I don't feel like I really understand it and am worried about what I've done.
I have followed the 'Nile' example within the data and used a Random Walk plus noise, identifying relevant parameters using the MLE function. After using the Kalman smoother I have produced a model that plots well. However, I wondered if there was any way through which to report results similar to a standard regression [such as r2, coefficients of the model etc]?
I would also like to use a segmented DLM as a method to support an Interrupted Time Series. Traditionally [in regards to a segmented regression] you could compare slope and intercept of the two regression lines or extrapolate out and calculate the difference between the observed and predicted values at a particular time. As such, I wondered if anyone knew how to derive relevant data from the dlm package [or am I just misunderstanding it...]?
I need to be able to outline some statistical results alongside my plot, but am not sure of what could be referred to?
If anyone could help it would be hugely appreciated?
If you describe accurately your problem that what you want? exactly what problem you are facing during modeling in R, then probably I will help you out.
Thanks Suraj,
I was completely lost with it and beginning to panic a bit!
I have two short, non-stationary time series - one consists of 12 observations [annual] and one subset of 48 quarters. Said data represents the area of land converted from green space to built environment in m2/Ha.
I created a DLM in regards to each time series, which plots well and shows a good upward trend. I used the dlmMLE function on a Random Walk plus Noise model to find relevant dV and dW values, then ran the model with both dlmFilter and dlmSmooth. As I said, the plot is ok, but leads to my first problem.
How do I report [statistically] the model? (I'm relaly only used to linear regression where I could refer to r2, coefficients, residuals or p-values). I just wondered how you would report the results of a dlm? Some studies I've noticed just refer to the estimated figures, but wondered whether such was appropriate?
My second question [sorry for being a pain] relates to whether the dlm could be used in an interrupted time series. An ITS [very briefly] segments a time period into two periods, which are assessed and compared separately. Whilst most research using them talks aabout the use of ARIMA, they all tend to just default to OLS regression - in which they refer to the difference between the two periods as either the difference between the slopes and intercepts (which I assume I can't do with a dlm) or continue the first segment regression into the second [as a counterfactual scenario] and measure the difference between the two at particular points.
I attempted a similar method with my data, using the dlmForecast, but [I think due to my data] just get a static, straight line prediction. I wondered whether you had any ideas?
Sorry its such a long question, I am just out of hope at the moment!
Thanks if you can help at all
hello,
What type of data are you working with? e.g Financial Time Series Data
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Regression