Bayesian methods are taking the world by storm. For those of you unfamiliar with Bayesian inference, if you ever thought the idea of calculating a bunch of coin-flips in math class was silly, this is the theory for you. If someone told me that they had a string of ten straight heads, you would say, wow, that is highly unlikely, yet the framing of many introductory frequentist statistical approaches asks you to suspend your straight-forward judgement to understand calculations in the frame of multiple stochastic events. The coin flip model of frequentism doesn’t do justice to the probabilities in the world. Issues with hypothesis testing have revealed that merely disproving the null hypothesis isn’t enough to really say much. Events are often framed as random occurrences with binary outcomes. Effect sizes are ignored in favor of discussions of the absolute qualities of candidates and proposals.
I am especially interested in this in the context of public debates about Democratic nomination. There seems to be a thought among supporters of Hillary Clinton that she has a superior ability to “get things done” in parliamentary politics. Thus a vote for Clinton is preferable to a vote for longtime Senator Bernie Sanders because of the possible goods that might be possible if Clinton were both to be elected and the preferable bill passed.
To evaluate the actual effect size we would need some conception of the posterior probability of the event (let’s call it the Jobs Bill) without a Clinton election. What factors structure our calculation? Control of Congress, current economic conditions, the alternative to Clinton as President. If we assume then that the most likely scenario is that Clinton wins the Presidency and the Democrats take the Senate while the Republicans hold the House, the chance to pass the jobs bill is near zero. Unless there is some Frank Underwoodesque level of manipulation happening, there is no meaningful difference in the probability of legislation passage between the candidates.
Doesn’t this explain why the second season of House of Cards becomes less satisfying? The underlying dynamic of the show was that some unethical subterfuge could adjust effect sizes, now Frank can create entire new worlds of probability from whole cloth.
If we take a broader view of the network of people that pass legislation it becomes clear that there are far more important factors controlling effect size than the President, controlling for the party of that President, of course. This harkens back to the idea of uniqueness in policy debate. This idea mainly is used in the discussion of disadvantages to affirmative policy proposals. In order for a disadvantage to be unique, a team must establish the posterior probability both of the sequence of events that is likely to cause a negative impact and the probability of that impact. Historically, policy debaters have been willing to take a model of probability that over-emphasizes the effect size of the proposal at hand. Much like in our consideration of the probability change between Clinton and Sanders, it is important for the analysis of policy to understand the ways in which policy calculations can be come silly if we fail to map the probabilities of events in a meaningfully. This is why calculations that include nuclear war are so common in policy debate, even as effect sizes retreat toward zero and events have near zero impacts on posterior probabilities. It is possible of course, that a single Federal sex education policy might cause waves of centralization in the United States that will be taken as a political model in a distant country that causes ethnic conflict. It seems much more likely that at each step the relatively small effect size should be effectively mapped. Real policy making is increasingly turning toward better mathematics.
Why shouldn’t we care so much about our ideas that we exaggerate them? As we allow the arbitrage of effect sizes agency begins to hypertrophy. Fetishistically things that are inevitable and arbitrary and hard seem to become simple, the solution to many problems and cause of others is attributed to the powerful agent. Isn’t this one of the biggest problems with the liberal policy making establishment today? All too often ostensibly progressive politicians are willing to accede to the plans of rich people on the basis of their wealth and success, their personalties match the visage of the hyper-agental figures that politics so desires. More melodrama. Less impact.
So no, except in some very specific scenarios the choice between democratic presidential candidates doesn’t effect the probability of legislation passage, unless we are willing to understand those probabilities as ontological qualities of the candidates themselves. Policy falls away at that point, this is melodramatic posturing.