Thank you in advance for your suggestions and assistance!
I'm trying to analyze grade-level enrollment data for 15 unique grade levels (pre-school through G12) over five academic years. I have a few variables added to the model (applications by grade, withdrawals, new students, beginning of year enrollment, end of year enrollment, etc.).
Using Gretl, I believe this sort of analysis would use the "Panel" data structure with a "stacked Time Series" where the grade level is the unit, and the school year is the time variable. Does this sound right? After trying it out, it seems to work quite well and I've been able to eliminate a few variables to strengthen the model. However, I cannot seem to add any more observations to test the forecast. I am able to execute an in-sample analysis against the last school year and that works quite well.
However, when trying to set the range for a future forecast (after adding observations for 1 year ahead), Gretl won't allow me to change/use the added observation when trying to set the "forecast range" parameter. I have a feeling my underlying data is not structured correctly for Gretl. My columns/rows looking something like this:
| Grade | School Year | Sections | BOY Enrollment | EOY Enrollment |...
101 2016-2017 4 50 55
102 2016-2017 4 58 56
...
Should Gretl panel data structure be used for something like this? Thanks again for any advice/help you're able to offer.
Gaylon,
I do not know what caused the problem with forecast, but it looks like you have not structured your dataset in the way you wished. You wrote that you have 5 years while in your Model screenshot you have Time-series length=15 and 4 cross-sectional units included.
Best,
Błażej
You are correct and apologies for the confusion. In truth, I have tried to use Panel data structure stacked time series, stacked cross sectional, and index. In my mind, the most logical setting would be either Panel > Stacked Time Series or Panel > Index where Grade is the "unit" and School Year is the "time" variable. I have tried both options and the result is 15 cross-sectional units (Grade level) and 4 Time-series values (School Year). The data range is 1:1-15:4.
However, even when that seems correct, when I try to add observations (Data > Add observations...), I only get the option of 4, 8, 12, etc. If I choose 4, for example, a new cross-sectional unit is added which makes the data go to 16:1, 16:2, 16:3, and 16:4 (see attached). What I need it to do instead is to add one more Time-series (i.e., 5) for each of the 15 cross-sectional (Grade levels) units.
If I reset the data and make Grade level the Time-series unit and School Year the cross-sectional unit (which sounds backwards to me), the data then goes to 1:01-4:15. Surprisingly, with this structure when I go to Data > Add observations, it defaults to 15, which in theory looks correct since the data range then goes to 1:01-5:15. However, after running the OLS, etc. and trying to forecast, the forecast range doesn't allow me to go beyond 4:15 (the "up" arrow is greyed out) even though I have ensured that the sample range includes the newly added observation (see screenshots).
So, I'm stuck. All methods eventually come up with the same "best fit" model, but I cannot test the model against any sort of future forecast. I have successfully tested in-sample forecast (set the forecast range to discard the latest School Year), but I'm stuck after that. After a "add observations" do I need to go in and edit the values?
Thanks again for your help. I've attached a couple screenshots to give you an idea.
- Badges
- Science topic
- Similar topics
- Geoscience
- Asia
- India