# How to model an irrevocable decision in time series data?

#### nielsselling

##### New Member
Dear stats wizards,

Since I am not a statistical guru whatsoever, I really need some help here. Basically, I have time-series data of corporations. The time-series span 2016-2018. The dependent variable refers to a yes/no decision that a corporation can make. The decision can be made at any point in time between 2016-2018 and the decision is irrevocable. As an illustration, let us say that the population consists of American industrial corporations with no plants in Mexico. Let us further say that the decision was whether or not to build a plant in Mexico (of course, this decision is not irrevocable but, for the sake of argument, let us pretend that it is). I want to understand which factors are driving this decision. The data would look something like this:

Year Firm DV Ind.var1 Ind.var2
---------------------------------------------------
2016 FirmA 0 ... ...
2017 FirmA 1 ... ...
2018 FirmA 1 ... ...
2016 FirmB 0 ... ...
2017 FirmB 0 ... ...
2018 FirmB 0 ... ...
2016 FirmC 1 ... ...
2017 FirmC 1 ... ...
2018 FirmC 1 ... ...

The challenge I face is twofold:

(1) If the DV for FirmX is 1 at time t, it will be 1 for all subsequent years.

(2) I believe that positive decisions (DV=1) made in 2017 or 2018 are driven by different factors than those made in 2016 (we can think in terms of "leaders" and "followers"), so I want to be able to take this into account as well.

I would be super grateful for any help I can get.

Best regards, Niels

#### noetsi

##### No cake for spunky
I am not sure how you can model a decision. It was made for a reason. you just find out what that reason was - the data is not going to tell you since it could be made for many reasons (including entirely irrational ones). I don't really understand what your dependent variable is here - consider that when you read these comments.

I doubt time series is the best way to model this. I have not seen it used for this. Markov chains or decision trees might be better although I have not worked with them for decades. You might want to look them up. Again I doubt you can explain a decision by any means.

You can also run segmented regression where there is an intervention variable coded 0 before some point and 1 from then on. This will show the impact of the decision, although not why it occured.