Fivethirtyeight's Trump approval rating tracker

#1
(I'm new here - hi all! - so I'm not really sure how things work here or whether I'm putting this in the right place. Hoping for the best here!)

I've made a project of trying to reproduce fivethirtyeight.com's Trump approval rating calculation in an Excel file (their Trump approval page is here). And I know I'm very close, but I'm stuck without having gotten it 100%. I'm hoping someone is familiar enough with their calculation methods to help get me across the finish line.

In theory, fivethirtyeight provides all the information a person would need to do what I'm trying to do. They explain their calculation method in some detail here. And they post what I take to be the raw data used to do the calculation; the raw data is updated regularly - I assume every time new poll entries are added to the running list of polls making up their approval average. Raw data link here.

But here's the thing. Two of the variables fivethirtyeight uses in their calculation, poll weight and adjusted approval numbers for a given poll, change on a daily basis. Not for all polls; only for the ones that come out most regularly (e.g. Rasmussen, YouGov, Morning Consult; but, some less frequent polls also have changing numbers). For example, the weighting for a certain Rasmussen poll might be 0.55 on the day it comes out, but only 0.135 the next day - and it keeps dropping, slightly, every day after that (not forever, but I'm only concerned about the first 30 days, since their calculation model for a given day is based on the previous 30 days). Likewise, that same poll might be adjusted downward by 0.535 the first day, but then show up as adjusted downward by 0.541 the next, and other values on later days. This can all be seen in their posted raw data from different days.

In the first place, I'd love to know fivethirtyeight's rationale for doing this. It's not mentioned in their calculation explainer at all, and it seems kind of counterintuitive.

And more importantly, I need to know how exactly these changing numbers figure into their calculation. What numbers are used to determine the current calculated Trump approval number? Is it based on the weights and adjustment factors for each of the previous 30 days as they existed on each of those days? Or is it based on the weights and adjustment factors for the previous 30 days as listed in the raw data at that moment? Or is it some combination - one thing for the weight numbers, and another thing for the adjustment factor numbers?

Apparently it's not the second option (it's "based on the weights and adjustment factors for the previous 30 days as listed in the raw data at that moment") - I have meticulously plugged in the most recent numbers in the spreadsheet I've created, and consistently get quite close to fivethirtyeight's published approval numbers, but I don't quite get a match. Of course, there could be a flaw in my spreadsheet; but getting clarity on my question is part of me figuring that out.''

By the way, fivethirtyeight was nice enough to post an email address on their site for questions or concerns about their approval model. But so far they have not been nice enough to respond to the email I sent their asking this question :) I emailed them a month or so ago about it, so I've given up hope for a response.

Thanks in advance for any help or insight anyone can offer!
 
#2
This is a nice post. I am sorry but I don't have any answers to the questions.

But it would be nice to se if someone has got some R code for this. I guess that many have tried to create that. (I am sorry @Mechitar, but excel is not so popular here.)