This graph shows the estimate that Kenny Shirley and I have of support for the death penalty by sex and race in the U.S. since 1955:
We also found that capital punishment used to be more popular in the Northeast than in the South, but now it’s the other way around.
Here’s the abstract to our paper:
One of the longest running questions that has been regularly included in Gallup’s national public opinion poll is “Do you favor or oppose the death penalty for persons convicted of murder?” Because the death penalty is governed by state laws rather than federal laws, it is of special interest to know how public opinion varies by state, and how it has changed over time within each state. In this paper we combine dozens of national polls taken over a fifty-year span and fit a Bayesian multilevel logistic regression model to individual response data to estimate changes in state-level public opinion over time. Such a long span of polls has not been analyzed this way before, partly because doing so requires a suitable model for the overall national time trend of death penalty public opinion, which is challenging to formulate.
In the context of the death penalty example, we develop here a suite of methods, largely graphical, for manipulating and understanding a fitted hierarchical model. In the death penalty problem we resolve the issue of modeling the national trend of support by using redundant parametrization and a structured prior distribution for the yearly effects. The resulting model can be fit using standard MCMC techniques, but the output of the model-fitting process is difficult to analyze immediately, as it is for many large hierarchical Bayesian models. The fitted model analyses we discuss in this paper include computing finite population contrasts and average predictive comparisons, and plotting posterior intervals of within-group standard deviations to compare different sources of variation within the data. We discuss inferences about the changing nature of death penalty support across time, states, and demographic groups that could not be made without using a variety of advanced tools for model understanding.
The analysis is an example of Mr. T (multilevel regression and poststratification over time).
P.S. To clarify the graph above: The parallelness of the lines (that they all jump up and down together) arises from our additive model. The true trends could not possibly be so clean. We did, however, look at residuals over time by sex and ethnicity and did not see any big patterns, so I think the picture above is basically accurate.
P.P.S. The estimates that we have that are readily available right now are for a slightly more detailed set of interactions than state*year. We computed interval estimates for the probability of support for an individual in each of the (51, 54, 2, 2, 5, 4) (states, years, race, sex, degree, age) cells. From this, you could post-stratify (if you have census data that gives you cell probabilities for these variables) to get state*year effects.