FiveThirtyEight's model is based on a combination of state and national polls.

Judged as of 2016 against FiveThirtyEight's published 2016 presidential model methodology. True: the methodology states that the forecast uses both state-level and national polls (for example, national polls feed the trend line adjustment). Mostly true: it uses both, with other inputs such as economic or demographic data, mainly in the polls-plus version, that the claim does not mention. Mixed: national polls play only a marginal role. Mostly false: the model relies almost entirely on one kind of poll. False: the model does not use both state and national polls.

Set before the research began.

1. E1 [PDF] Deciphering the Signal and the Noise - Ben Tanen Ben Tanen (Tufts University term paper) · 10 May 2017

   “FiveThirtyEight’s model included almost every state and national poll that was published by any major polling house throughout the 2016 election cycle (Silver 2016b).”

2. E2 The Contest to Predict the Presidential Winner in Florida Was as Close as the Race Itself Sayfie Review (Featured Column by Dr. Aubrey Jewett, UCF) · 12 January 2017

   “FiveThirtyEight starts with the actual results of published polls but then “the model weights each poll by its sample size, how recently it was conducted, and the historical accuracy and methodology of the polling firm.””

3. E3 The Contest to Predict the Presidential Winner in Florida Was as Close as the Race Itself Sayfie Review (Featured Column by Dr. Aubrey Jewett, UCF) · 12 January 2017

   “FiveThirtyEight also calculates a “Polls Plus Forecast” which includes state demographic, economic and historical information into the Polls Only model”

4. E4 Five Thirty Eight Model: Trump Wins Georgia, But Clinton Wins Nationally in a Landslide GeorgiaPol (blog) · 30 June 2016

   “The final model, polls plus, uses economic data and past voter behavior to make assumptions about how the election will turn out.”

E1, an independent paper citing Silver's 2016 methodology, says the model took in almost every state and national poll published during the 2016 cycle, so it used both kinds of poll. E3 and E4 show that the polls-plus version also used economic, demographic and historical data, and the claim does not mention these inputs. The resolution criteria rate this situation as mostly true.