A horse race based on name recognition

From the most recent Harvard Harris Poll. This information is basically useless in the time frame of the 2020 election. They are still figuring out who is even going to be on the field; they certainly haven't begun competing with one another in any meaningful way. Besides, the public isn't paying attention and will not for many months yet. And of course - 1,536 registered voters and that large a number of candidates. Way too low a sampling rate for a national election.

Look at the line up

Biden 44, Sanders 14, Warren 5, Harris 9, Buttigieg 2, O'Rourke 3, Booker 3, Klobuchar 2, Yang 0, Inslee 0, Gabbard 0, Castro 0.

Note, only the top four are even outside the likely margin of error. Buttigieg, O'Rourke, Booker, Klobuchar, Yang, Inslee, Gabbard, Castro have numbers so low they are within the margin of error.

This looks like a variance on the old rule of thumb in forecasting a time sequence.

If you have no observed trend line or no data on which to base a forecast, then your most accurate forecast is going to be to take the performance of the past time period (say, a month) and replicate it forward. If X increased by 3% last month, then forecast 3% for next month. Over time, that crude forecasting algorithm will be the most accurate you can come up with. And a surprisingly accurate one considering how crude it is.

The political version would be - If you have no pertinent data, then forecast the competitive position based on name recognition alone.