Soltas has good data, but the wrong conclusions regarding Trump and trade:

The idea: To see if Trump and Sanders benefited from protectionist sentiment, I will compare their respective primary vote shares in a given county against how badly affected that county was by the decline of manufacturing employment.

-Here’s the thing, though. Soltas doesn’t really do that. Detroit’s manufacturing share of employment has held up better than that of San Francisco between 1990 and 2016. Detroit’s actual manufacturing employment has held up only slightly worse than that of San Francisco during the same period.

Edit: Soltas has told me he’s measuring the decline in manufacturing share of employment using percentage points, not percent, so this example isn’t quite right for my purpose. Nevertheless, if de-industrialization in Michigan mostly results from an influx of good service-sector jobs into an area, it’s kinda hard to argue that it would be correlated with Trump support. Manufacturing as a percentage of employment does not seem to be correlated with Trump support.

Yet, San Francisco does not look like a place devastated by manufacturing decline, its residents desiring to re-create an economic boom via protectionism. Detroit does. The obvious reason for why that is is that in San Francisco, those lost manufacturing jobs were replaced by far better tech ones. In Detroit, this occurred to a far lesser extent. When Soltas wrote

The protectionist-backlash explanation of Trump and Sanders does not seem to hold up

he apparently either forgot or never realized this.
Michigan has its own internal San Franciscos (e.g., Oakland, Kalamazoo, and Washtenaw counties). And it has its wastelands, both Black and White-majority. In other words, Ted Craig’s explanation in the Marginal Revolution comments is probably correct. Instead of the share of manufacturing in employment, it is much better to look at total employment, which shows a decisive divergence between San Francisco and Detroit from 2004 onward. Changes in personal income would work even better. The best correlations of Michigan county factors and Trump support I’ve found yet are decline in child population from 2000 to 2010:

unemployment rate:
Screenshot (241)
and average weekly wages: