Trump won in 2016 and the Republicans carried both houses despite Trump’s underwater approval ratings before the election. This suggests a model that relies on such a dubious statistic may yield flawed predictions.
The real problem is likely the same one that bedeviled political polling way back. In 1936 polls based on phone surveys predicted an Alf Landon landslide. The results were completely wrong. Why? I think the standard answer is that those who had phones were a very biased sample.
I took an informal poll and asked 25 people I know if they answer their phones when they don’t recognize the number. Not one said they did. This suggests the possibility that those participating in phone surveys are a biased sample, every bit as non- predictive as the group in 1936.
Before we can believe any current prognosticator, we need to know what they predicted in 2016 and what changes they made in response to the election result. The smart money got it wrong in 2016- why believe them now?