How sensitive is economic sentiment measured by the University of Michigan survey to partisanship? Clearly, support among Democrats and Democratic leaners is higher under the Biden administration.
figure 1: Michigan College Democrats/Lean Democratic (blue circle), Republican/Lean Republican (red circle), and Shapiro-Sudhof-Wilson SF Federal Reserve News Sentiment Index, February 2021 to February 2024. OLS Regression Line (Slope, R2) Source: University of Michigan, SF Fedand the author's calculations.
Not only is the sensitivity of Democrats/Democrat leanings greater, the R2 of sentiment regression on news sentiment is also higher. In other words, it's a bit difficult to tie Republican/Republican-leaning sentiment to a text-based news index. Using the 5% significance level (F test), this constant is statistically significant, while the slope coefficients are not significantly different.
I also show a non-parametric smoother (nearest neighbor, viewport=0.7):
figure 2: Michigan College Democrats/Lean Democratic (blue circle), Republican/Lean Republican (red circle), and Shapiro-Sudhof-Wilson SF Federal Reserve News Sentiment Index, February 2017 to February 2024. Nearest neighbor fit line (window = 0.7) source: University of Michigan, SF Fedand the author's calculations.
Note that this relationship was even stronger during the Trump administration, before the pandemic:
image 3: Michigan College Democrats/Lean Democratic (blue circle), Republican/Lean Republican (red circle), and Shapiro-Sudhof-Wilson SF Federal Reserve News Sentiment Index, February 2017 to January 2020. OLS Regression Line (Slope, R2) Source: University of Michigan, SF Fedand the author's calculations.
During the Trump administration, Republican/Republican-leaning economic sentiment as measured by the Michigan survey was essentially completely unexplained by news sentiment as measured by the Shapiro-Sudhoff-Wilson Index. (Using HAC robust standard errors, the Democratic/Lean Democratic slope coefficient is statistically significantly different from zero.)
