I am very new to time series/panel/longituginal data. From my understanding Panel data = multi-object time series and I have some panel data in long format (objects have multiple rows corresponding to different time steps). The objective is to predict/classify if a person is going to leave the job due to factors of the job (hrs worked etc). The data looks something like this:

person month # of hrs worked Quit Other Feats
Mary 2 30 0
Mary 3 20 0
Mary 4 80 0
John 1 100 1
John 2 10 1
John 3 60 1
John 4 40 1
John 5 100 1
Cate 3 11 0
Cate 4 19 0
Cate 5 29 0
Cate 6 14 0
... ... ...

Where "Quit" is the y or target variable. I am curious how I can predict using this data if an existing person will quit and if a new person will quit. Can I just feed this into a random forest and expect it to do well? (similar to this) Do I need to transform it to wide format and keep 1 row per person where all the longitudinal aspects are in the columns? I have tried feeding this into a random forest and it gave a good accuracy and AUC but I am not sure if this is the correct way to solve this problem? I have seen people use RNNs to solve this.

Similarly, what about regressing/forecasting on this data? Say instead of a binary outcome I have dates of when a person quit, what techniques exist to regress on this kind of data? Can I just run a linear regression model on this data? What about pooled OLS or GEEs?

One last question is how is this different from classifying on time-series data instead? They both have a time component but panel data has multiple objects.



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.