I use the term "soft information" quite a lot, but I think it's unclear to readers what I mean, and what "hard information" would by contrast mean.
"Hard information" (think "hard data") is often taken to imply true or accurate or reliable information. That is very much not what I intend when I use the term. Usually when I am writing about hard and soft information, I am advocating greater reliance upon soft information, or lamenting our institutions' inability to make use of it. That'd be pretty dastardly, if by hard information I meant good information and by soft information I meant bad.
My definition of "soft information" comes out of the banking literature I once studied as an abortive finance grad student. It's kind of a puzzle, I was taught, that banks even exist. Finance types, or at least finance types of the still very neoliberal aughts, treated markets as the best information aggregators the humans ever invented. Loans are marketable. When you and I want a loan, why don't we just issue a kind of personal commercial paper, and let potential lenders bid on the debt, establishing our market interest rate?
Okay, you and I are small potatoes, so maybe "frictions" in issuing marketable securities explain why personal loans don't work this way. But businesses take bank loans too, sometimes pretty big ones!
The traditional resolution of this puzzle is "soft information". A bank has intimate information about the businesses who bank with it. It can observe a whole arc of history, how revenues flow and how stable or unstable they are, how reliably the business pays its bills and what kind of cushion it maintains. Bankers often know the principals of their business customers personally, and develop opinions about their character, which J.P. Morgan himself, um, characterized as the most important information in a credit decision.
Some of the information bankers superintend can get aggregated into a credit score — hard information! — but much of it is simply not accessible to the rest of the market. Evaluations of character in particular depend upon personal contact, and are ideally drawn from longstanding, continuing, human-to-human relationships.
So, the theory went, banks exist, or at least exist as lenders to business, because bankers can outperform markets in pricing debt, because they have superior information. Bankers have access to all the same "hard information" other lenders would have, but also to this "soft information", available only to them.
If soft information is information only available to a particular party by virtue of its relationship with the subject of the information, by virtue of its situatedness, then hard information is information whose content and provenance is accessible to everyone, information that is unsituated.
A score issued by a credit bureau is "hard information" because every potential lender can access it and observe the same value. Although the reporting that informs the score is private, the procedure by which the scores are computed is broadly public. Sure, details might be proprietary. Everybody understands that structuring the computation had to involve making choices that were discretionary, that might have been made differently. If you disagree with those choices, you might argue the scores are inaccurate. But the credit bureaus make those choices, and promise to perform their scoring consistently. Lenders accept the scores, and take them as a pretty good proxy for creditworthiness.
More than that, because everybody sees the same information, "hard information" can serve as the basis for public conversations in a way that a banker's soft information cannot. Civil servants and the interested public can deliberate and conclude that 620 should be the minimum score for a borrower whose loan will be purchased by a government-sponsored agency. We perceive this criterion as "objective". Since we will all have the same "hard information" about a given borrower, we can all evaluate whether they are eligible. J.P. Morgan might object that crucial aspects of character are not captured by the credit score. But we'd have a hard time structuring a national bureaucracy around information accessible only to people with personal relationships to the borrower.
But J.P. Morgan was undoubtedly right! If we could in some trustworthy and dispassionate way incorporate the experience of people who know the borrower best, we'd find plenty of cases where it would in fact be wise to lend to a person with a low credit score, and be unwise to lend to a person with a high score. But "we" constituted as a large bureaucracy — public or private! Bank of America as much as Fannie Mae! — have no such trustworthy means. "We" do not know the borrower. "We" do not know and have little reason to trust individuals from whom the borrower might solicit character references. Independently scouting such evaluators would be costly, and the problem of evaluating our putative evaluators' character is no less difficult than evaluating the character of the borrower.
So, in public discussion and in large bureaucratic contexts, we necessarily rely on hard information even though we understand it to be inaccurate. We acknowledge the information condensed into a credit score is imperfect. Nevertheless we often treat it as though it is the thing itself, rather than at best a noisy proxy of a thing we wish to measure. If there are two otherwise similar borrowers, one has a credit score of 550, the other has a credit score of 650, we lend to the latter because she is "creditworthy", and deny the former because she is not.
The fundamental character of hard information is not "accuracy" or "quality", but its susceptibility to serving as the basis of social convention. We need to talk about things, to reason together, to decide together. Almost everything of interest to us is not directly accessible to our senses. Your and my "gut" evaluations would not be consistent with one another, could not provide a basis for consensus in our decisions. Often we have no perfect or even good measures of the thing of interest. So we invent something, as least-bad as we can. We make the arbitrary choices we have to make to define and structure a public measure, because something is better than nothing.
Fabricated hard information like this is everywhere. What's the "inflation rate"? The US Bureau of Labor Statistics does its very best — really! But it necessarily makes arbitrary choices, applies hedonics to commodities with quantifiable measures of quality but omits them for goods without, etc. The Consumer Price Index it publishes gives us "hard information" by which we publicly reason about inflation. But different choices might have been made than those BLS makes, and those might yield different numbers and very different conclusions, especially over long periods of time.
Has the "real income" of the median household grown over the last two decades? Under one set of numbers yes. Under another measure which, say, took into account the effect of income segregation on the affluence of neighbors as a form of housing quality, it might not have. Neither measure would be incontrovertibly more or less "accurate" than the other. But it's likely that one would become conventional, would predominate in public debate, would come to be taken as "the fact of the matter".
The susceptibility of hard information to convention, to standing in for fact, means that it often serves to engender a very authoritative kind of groupthink. The information, however well-intended its construction, will along some dimensions, for some purposes, prove inaccurate and inadequate.
Are Americans happy with their health insurance? After UnitedHealthcare CEO Brian Thompson was murdered, I saw incredulous responses on social media pointing to polling that said 81% of respondents rated their insurance excellent or good.
It's not a secret in everyday American life that a lot of people really hate health insurance companies. Insurers are perceived as making tragic situations even more miserable by virtue of the bureaucracy they impose on the sick and those who are forced to advocate for them. They are perceived as contributing to the death (murder?) of patients upon whose treatments they impose delays. Your cancer won't wait for prior approval. It's metastasizing while they mull.
Yet prior to Thompson's murder, this thing that everybody knows was largely inaccessible in policy-oriented conversation. Are people unhappy with their health insurance? The polling says no. You can bring in your anecdotes, how mad your cousin was, how your friend died after the imaging was delayed. But you could come up with as many stories of people having great experiences! The hard information says 81% of patients are happy with their insurance.
Thompson's murder has brought attention to some obvious-in-retrospect considerations. When we are judging the satisfactoriness of something, the intensity of dispreference of a dissatisfied minority can matter. Dissatisfaction on behalf of others' experience rather than ones own can be relevant. (There's a literal survivorship bias in polling customers' own health insurance satisfaction. A disproportionate fraction of people who personally suffered awful experiences are no longer around to be polled.)
We can construct measures that try to address these things. But we didn't before. Our conventional hard information blinded us. Data-driven groupthink laughed off claims that the experience of the insured was a burning problem. The policy community focused on universality rather than quality of coverage, and praised itself for the success of ACA on that score.
If hard information risks a kind of upscale groupthink, should we rely upon soft information instead?
Unfortunately reliance on soft information yields its own pathologies. When we rely upon information accessible only to particular people by virtue of their situation or perceived expertise, we cannot evaluate the quality of decisions. We open the door to unfairness in decisionmaking, sometimes intentional, often even when decision makers are doing their very best. Relying on soft-information decisionmaking risks corruption.
Most obviously, if someone is going to be trusted to make consequential decisions based on special information and judgment that others cannot evaluate or verify, who's to say the decisions aren't motivated by self-interest? On the basis of purported soft information, I lend the bank's money to my own friends and family more freely and at lower rates than to others. Perhaps by informal arrangement my own prosperity rises with that of my friends. Perhaps what I actually know, through my vaunted soft information, is my friend will make speculative investments with the money and cut me in on any gains, while she'll fail to repay the bank if things don't work out. If my cut will be big enough, the upside might be worth risking my job.
One might try to police conflicts of interest, ensure that whenever decisiomakers might have, or might even be perceived to have, a personal stake in the outcome, of course they should not be trusted to judge. But then you might condemn yourself to the worst of both worlds. You are relying upon a decisionmaker's purported special information while insisting she only decide under circumstances where in fact she lacks any special information. A perfectly ethical loan officer will have the bank lend to a person she has known forever and considers a friend in preference to others with "objectively" similar creditworthiness, if part of what she knows is that the friend would be mortified ever to default on a loan and would move heaven and earth not to. That's not a violation of the trust the bank has placed in her. On the contrary.
A worst-kept secret of finance is that the entire project of investment banking is built on managing, rather than eschewing, conflicts of interest. Top M&A firms do not compete on price, but on the thickness of their rolodexes and the trust they have engendered in the investment community. Firms looking to raise capital understand that bankers will offer access to good customers on sweetheart terms, at prices where there is still money to be made. They calculate the "money on the table" they lose from underpricing will be more than compensated by the number and wealth and quality of investors the schmoozy bankers bring into the deal. Perhaps this works out best for everyone. Firms raise more money than they would have via any alternative channel, and investors expect and get good deals from bankers they trust. Perhaps not, and one or both sides of the transaction gets told all this as a kind of fairy tale while they get screwed. The trouble with soft information is it's hard from the outside to tell. Often it's hard from the inside to tell!
In the United States, we reflexively equate "nepotism" with corruption. Opportunities should be distributed according to degrees, or exam scores, or some qualification that can be expressed on a resumé as verifiable, hard information.
But that's too easy, and dangerous. Nepotism can be wise, and not only for corrupt reasons. I may know about competences and character traits among family and close friends that no test score can easily capture. I might actually know my nephew is well-suited the job, even if his resumé is inferior to other candidates I know less about. If the job is high stakes and I set aside my soft information to go with the best resumé, I risk a consequential mistake.
The difficulty is that outsiders — and, in many cases, the decisionmaker herself — cannot easily distinguish between corruption and wise use of soft information. If I think, I love my sister and my nephew so I'll give him a great job, then hiring him is corrupt. If I know my nephew is ethical and hardworking and unusually insightful about exactly the sort of problems he will face, then hiring him may be wise. Of course, I may persuade myself of the latter when I am motivated by the former. And you, on the outside, have no way to tell which is which.
The "good government" impulse to avoid this challenge by relying on hard information only is not sufficient. Hard information is also imperfect. If I hire the best resumé from Harvard, I can justify the choice and no one can blame me, but it may well be entirely the wrong choice, if what we actually care about are outcomes. People who style themselves meritocrats like to imagine that talent is both abundant and objectively demonstrable. In many contexts, neither is true. Being precious about appearing above corruption can itself be a corrupt choice. I protect my own reputation at the expense of the cause I am pledged to serve.
Even when a decisionmaker uses soft information impeccably, when she makes the best choice possible given her full information set, soft information decisionmaking is susceptible to structural corruption.
I know, I know. That sounds like "critical race theory". Chris Rufo is going to come and destroy something else I love. But the mechanism is so straightforward even Chris could understand it.
A banker strives to make credit decisions as best she possibly can, with no bias whatsoever in terms of race or religion or nationality or other irrelevances. But our banker has her life and her own social circle. It is quite likely that her family, friends, and near acquaintances are not distributed among races and religions and nationalities in proportion to their frequency in the population.
When our banker receives a loan application from someone in or near her social circle, she has meaningful soft information that can help guide her choice. When she receives the same application from outside her social circle, she does not. Given two candidates who are in fact creditworthy, it's quite possible she will be able to recognize the quality of the candidate within her social circle, but be unsure of the candidate outside of it. Soft information decisionmaking introduces a structural bias towards choices that the decisionmaker has special information about, even though there may be equal or superior choices elsewhere. Our banker is making the best choices she possibly can, using all of the information at her disposal, but from the outside it will appear that she is biased towards the races, religions, and nationalities prevalent in her own social circle. It's the structure of her acquaintanceship, rather than any invidious discrimination, that introduces the bias.
So, where do we go from here? Relying solely on hard information does not in fact capture all of the good information that ought to go into making a decision, and leaves decisionmakers susceptible groupthink, mistaking conclusions drawn from flawed but conventional information sources for the best decisions. Relying freely on soft information renders it almost impossible to police corruption, and leaves even honest decisionmakers susceptible to personal or structural bias.
What to do? I hope to think more about this in future pieces, but a difficult truth is answers will themselves be context dependent. It takes soft information to design roles and incentives under which decisionmakers can fully exploit all the information at their disposal, neither falling back lazily to the most easily defended hard-information criteria, nor abusing (consciously or otherwise) the trust placed in their soft-information judgment.
What I hope I can convey for now is "evidence-based" or "data-driven" is not in fact a recipe for high-quality decision-making. High quality decisions in practice require recourse to information that cannot meet a standard of public evidence, and skepticism of information that appears to meet such a standard. That this is hard to discipline doesn't render it any less true.
2025-01-24 @ 10:05 AM EST