Vincent Geloso@VincentGeloso
My graduate student Patrick Crawford introduced me to "partial residual plots" as a tool for showing relationships. Its so simple in elegance.
The partial residual plot below shows the conditional relationship between newspaper competition and rebellion after controlling for everything else. On the x-axis is market concentration (HHI: higher = fewer newspapers), and on the y-axis is the residualized level of rebellious activity.
Basically, the entire logic of the graph is for visualization. The regression has already controlled for other variables. What remains is the variation in rebellion intensity (depvar) attributable to the independent variable of interest. "Centered" just means both axes are normalized so that 0 corresponds to the sample mean. Positive values imply more rebellion than average (conditional on controls) (and vice-versa).
Now, its not exactly a new estimation method per se. But it does allow for a nice visualization of results that I had not seen and may even allow you to play with different functional forms with the residuals.
I know we should not care about "does it look cute" but I disagree with that sort of. I mean, at one point, we have to convince and convincing is really about cutting on inputs needed to (honestly) push a person from disbelief to acquiescence. That graph below is cutting down on inputs.
