Predicting United States Policy Outcomes With Random Forests

Reviewed the research Rana mentioned in her column your friend fwd you. Before my summery I wanted to note I've attached a Noah Smith Bloomberg column from last summer that did a good job debunking this sort of analysis (from another research team which McGuire/Delahunt base their paper off) pointing out that there is a great deal of alignment btw the middle classes' policy preferences and the rich. Smith, "...More importantly, theauthors found a strong correlation between the preferences of this affluent group and those of the middle class -- in other words, the policies that are enacted don’t often seem to buck the will of the broad public.Additionally, correlation doesn’t equal causation -- politicians might do things that they think will benefit the overall economy, and that only coincidentally happen to please the affluent. It’s also likely that Gilens and Page are just wrong. A number of other research teams have done their own studies, and most concluded that it’s the middle class, not the rich, who tend to hold sway in democratic politics. A 2015 paper by political scientist Peter Enns, for example, concludes that “policy ends up about where we would expect if policymakers represented the middle class and ignored the affluent.”…”

The Research finds that the views of the top 10% and a handful of interest groups are predictive of policy changes, the views of lower income groups have less predictive power. Granted that pushes back at the criticism Smith made of the previous paper, that the elite and middle class are generally aligned on policy preferences. Here is McGuire/Delahunt bottom line, "...We offer two main findings: (i) Policy outcomes on holdout sets can be predicted with approximately 70% balanced accuracy (vs 50% chance baseline) using only a few feature categories from the Gilens dataset: preferences of high-income earners, a subset (as few as 14 out of 43) of individual IGs’ preferences, and policy area labels. Holdout test sets can be predicted with approximately 70% balanced accuracy by models that consult only the preferences of those in the 90th income percentile and a small number of powerful interest groups, as well as policy area labels. These results include retrodiction, where models trained on pre-1997 cases predicted “future” (post-1997) cases. The 20% gain in accuracy over baseline (chance), in this detailed but noisy dataset, indicates the high importance of a few distinct players in U.S. policy outcomes, and aligns with a body of research indicating that the U.S. government has significant plutocratic tendencies. (ii) The feature selection methods of RF models identify especially salient subsets of interest groups (economic players). These can be used to further investigate the dynamics of governmental policy making, and also offer an example of the potential value of RF feature selection methods for inference on datasets such as this one...."
They look at the preferences of the P90 in terms of income. "....Voter preferences (P90): Preferences of different income tranches were obtained from national surveys of the general public, where participants were asked whether they favored or opposed a proposed policy change. Preferences along the income distribution were then imputed at various income percentiles, viz 90th, 50th, and 10th (hereafter P90, P50, P10). Gilens and Page noted that P90 was a (rough) indicator of the preferences of even higher income percentiles (e.g. P99). One of their key findings was that, among voter preferences, only P90 impacted case outcomes.Our models used P90 as the sole voter preference feature (use of P50 or P10 as features degraded model accuracy)..."
Later they note that the suspect the P90 is a proxy for say the P99, "... We suspect that the 90th income percentile serves as a proxy for the opinions of the extremely wealthy and that our results would likely be better with, say, the opinions of the 99th income percentile. However, chasing down the opinions of the top 1 percent via a random population survey would be a massive effort.... The results indicated, among other things, that the very wealthy held policy preferences that were much more conservative in such important domains as social welfare programs, taxation, and regulation of the economic system. Their research also showed that the very wealthy were more politically active, with higher rates of financial contributions to, and contact with, public officials compared to the general population. Within this elite cohort, roughly two-thirds contributed an average of approximately $4,600 to political organizations and campaigns. While this survey method yielded very important understanding, we must also consider that the sharpest increase in correlation between policy outcomes and high-income earner opinions may occur somewhere in the high end of the top 1 percent. Trying to increasingly hone in on the stated opinions of those in, say, the 99.9th percentile has a few potential issues. First, it would likely involve a large, possibly prohibitive, amount of effort. Second, the results may improve correlation with - but not definitively address - the likely lodestar variable affecting policy outcomes, which is the transfer of large amounts of money to policy makers from the wealthiest sources focused intensely on particular policies. For example, the biggest corporations who can influence in their own right or members of the billionaire class throwing around their financial weight in the political sphere..."
Shawn McGuire and Charles Delahunt, "Predicting United States Policy Outcomes With Random Forests," Institute For New Economic Thinking, October 27, 2020, https://www.ineteconomics.org/uploads/papers/McGuire-and-Delahunt-predictingPolicy_INET_25oct2020.pdf