Understanding the response curves

[ArrunAnanda]

Hi there,

I need some help understanding the response curves. For this question I'm using the dt_simulated_weekly sample data provided by Meta in the package.

Let's TV spending as an example - I want to know from where is the 256k coming from that is pointed on the line?

1. Is it a transformed value?
2. If it's a transformed value, then is it indicating the spend in the model window which is 92 weeks (2016-11-21 and 2018-08-20) or the whole time period of the data (2015-11-23 and 2018-08-20) which is 144 weeks?

Thanks!

[kyletgoldberg]

Hi - Good questions. The 256k comes from the mean non-zero spend days.

You can see the code for the underlying data set here -

Notice that the x-values going into the saturation function come from the adstocked spend values. So yes they are transformed

[ArrunAnanda]

Thanks for the quick response. Now, I understand. @kyletgoldberg

What is the timeframe that is considered for mean non-zero spend days? Is it the rolling window 2016-11-21 and 2018-08-20 or the whole time period of the data 2015-11-23 and 2018-08-20?