Much of armchair social science is pondering "puzzles," i.e., social behavior that doesn't fit well with simple theories, such as yesterday's cashless Christmas wish-list puzzle. A few standard theories can be easily invoked to explain most any social puzzle:
Christmas gifts are given in place of money because they are representative of the gifts given by the maji. We exchange gifts in remembrance of Jesus' birthday.
I have some thoughts about the Christmas gift phenomenon, if it is not too late to chime in. I think that a lot of people got pieces of the puzzle right, but I wanted to identify an additional component - we want to consume things ourselves, but at the same time, personal consumption looks bad because those are resources we are NOT using to signal our commitment to others. The gift-as-signaling theory definitely plays a part, in that we can be genuinely pleased with a surprisingly apt gift, but that doesn't explain why people give out gift lists. However, if the goal is to maximize our own consumption, while ALSO appearing to send commitment signals to others, we can mutually benefit from informed gift giving.
My guess is that there is also a prospect theory story going on here - we value resources that we earn for ourselves a lot more than resources given to us. We WANT to increase consumption without incurring the mental penalty of spending our money. Getting a gift, we can frame it as an unearned gain and consume it more readily, while spending our own money is framed as a loss. The mutual gifting process essentially allows us to get around our biased utility function.
The other components that people have mentioned definitely factor in as well - particularly how cash nets out, and so you'd see it in non-reciprocal relationships more frequently. The story about information and search costs I think is valid as well, and it fits in with my story just fine. You provide a list of things, and the gift-giver can either guarantee you a guaranteed utility gain by buying something on the list, OR they can optionally use your stated preferences and their personal information to present you with a gift that they predict you will like but that you simply did not know existed yet. As I said above, this pleasantly surprises us, since it is a successful forecast of our own preferences.
Hopefully this is not too weak of a social theory!
"folks pick their nose because it just feels good, not to clear out their noses."
"collect 'stylized facts' about holiday, wedding, birthday, and other gifts from across diverse cultures"
Isn't this difficult because in Asian societies, gifts of cash are the norm? In Korea, China, Japan, Taiwan, Singapore, Thailand and even parts of India, it's perfectly fine to give cash.
Well, yeah, that makes it difficult - but isn't that precisely the point that Robin's trying to make? A theory which explains why giving cash is considered tasteless in the US but fails to explain why it's acceptable in Korea is not a very useful theory at all. We would get much more insight from a general theory of what makes an acceptable gift which explains why a gift which is acceptable in Seattle isn't acceptable in Seoul, and vice versa.
"Jed, in the areas of social science I know best, most academic rewards are for collecting more data, and for building ad hoc models; there is little payoff for the sort of pattern-finding I'm describing. In those areas the strategy I describe gives great intellectual payoffs, if poor academic rewards."
That seems like a reasonable description. The main exception is those who stake their status on building a kind of 'guru' reputation, by building their specific pattern framework at the expense of others, and reinterpreting or selectively rewarding those finding that adhere to his model. Then there is a selection pressure in favour of more sane simplified patterns, albeit a weak one.
(Robin) "Eliezer, to understand an economy we seek patterns that compress, but I wouldn't claim one could build another economy that beats the first one only using such compressions of the first economy."
Not explicitly, perhaps. Yet it is inevitable that the professional opinions and expert testimony of economists do just this when they give predictions of the economic behaviors of super-intelligent, self improving ems. We've seen that over the last several weeks.
@frelkins: Indeed, I haven't run markets. Instead I try to anticipate what they'll do under different circumstances and take advantage of that. This has worked out well for me. Of course I look for the markets that are the easiest for me to outguess in a profitable way.
The context of my 98% (obviously a made up number, but I think in the right ballpark) wasn't markets per se, but I'll try to translate.
In wheat markets, for example, the farmers, the bakers, and the various intermediate transportation, processing and storage operations know a lot more than any wheat market theorist. Otherwise we wouldn't need a market, we could rely on central planning that uses our models. Surely this is an uncontroversial Austrian point.
If there are wheat market theorists I very much doubt they generate even 2% of the value of all the applied intelligence that goes into wheat related activities.
In financial markets the situation is much more peculiar, since the market participants to a large extent are trying to anticipate and respond effectively to each other, not exogenous phenomena. It isn't clear how to divide their subject knowledge from their theories about the market, and in fact widely held theories (such as Black-Scholes) tend to have a huge influence on the behavior of the financial markets, for good or ill.
So it isn't clear how to apply my 98% to the financial markets. However it is pretty clear that these markets failed to predict future facts that are extremely important to them. I would argue this is because the participants follow theories about the markets that are not grounded in enough concrete information.
Eliezer, to understand an economy we seek patterns that compress, but I wouldn't claim one could build another economy that beats the first one only using such compressions of the first economy.
Jed, in the areas of social science I know best, most academic rewards are for collecting more data, and for building ad hoc models; there is little payoff for the sort of pattern-finding I'm describing. In those areas the strategy I describe gives great intellectual payoffs, if poor academic rewards.
"Since I know you are an advocate of prediction markets, let me point out that this general point applies to them"
Forgive me, Jed, I don't know how many markets you run. My prediction market is 96% accurate so far. The 4% issue has never been a "small data" or "sampling" issue, if understand your point here.
Rather, that 4% has always been either an information issue or a trust issue. After the first information issue, I figured out a solution to that problem and it hasn't happened again. The trust issue's the be-otch.
"But why should we assume that most effects will have causes that explain more than one effect?"
Because otherwise, causal inference becomes very simple -- and it's not simple in practice.
When presented with a new problem, experts usually begin working over specific examples of solutions to similar past problems. Beginners tend to focus on the parameters of the problem itself. In other words, the experts know better how to compress the search space while minimizing important loss of detail. The beginner may complain that the expert is casually disregarding various elements of the problem that may end up being critical. The expert wonders if the beginner will ever get any solution at all. 
I of course agree that we should "seek theories that, with only a few free parameters, can explain a wide range of diverse puzzles." This is close to a restatement of Occam's razor as typically applied to scientific inquiry.
But in lots of scientific domains this only covers a small part of the actual practice. Most of biology requires explaining specific features of organisms, cells, or ecologies with specific appeals to history, environment, and the intersection of various larger principles.
Of course everyone wants to find ways to clear up the clutter by finding more general theories that can generate a lot of the specifics. But we often need to be more modest.
Conversely, until we have a lot of well described examples we typically do a terrible job formulating the general principles. There were general theories of biology before natural history -- they were comprehensive, reasonably simple, intuitively appealing, and wrong. The people who produced them were very smart -- I tend to think Aristotle was as smart as anybody in history -- but they didn't have enough, sufficiently detailed examples to work from.
"Trying to explain each puzzle with a new ad hoc story just doesn't get you very far" again is true. It may be a valuable caution, but to the extent that it implies that's the best available alternative, it is a straw man. The naturalists didn't mainly make up ad hoc stories (and the ones they did make up didn't last as contributions). However they also didn't have strong theories. They put a lot of work into accurately describing the details of the examples they found, and into finding somewhat patchy patterns of similarity and difference between them.
So I think we all agree that it's "[m]uch better to lay out lots of related puzzles on the table and then ask how to explain many of them all at once." But I think your way of presenting this implies that the process of laying out the puzzles is relatively cheap, and the process of explaining them is relatively expensive. I'd suggest that the opposite is true. 98% of the effort that bears long term fruit is accurately describing and deeply understanding the "puzzles" and the remaining 2% is finding the theories that explain their shared structure (which is often just a fraction of any given puzzle). Investing a lot in the theory building when you don't have enough examples is just building castles in the air.
Note that this pattern is common throughout science. Particle accelerators, field work or "wet" lab activities consume enormously more time, effort and resources than development of the theories they enable. They also produce enormously more specific knowledge than those theories can retrodict.
Reflecting a bit, if the pattern I think I'm seeing in your post(s) is indeed there, I think you and Eleizer have this in common: An implicit belief that a relatively small amount of information about examples is enough to let us find a really good theory, and this theory could then be used to act very effectively.
Since I know you are an advocate of prediction markets, let me point out that this general point applies to them. Unfortunately we are currently embroiled in an excellent counterexample to the power of prediction markets, and I think they failed due to exactly this sort of problem.
Obviously the financial markets, and the vast majority of individuals involved in the markets, failed to predict the consequences of the financial system's internal structural relationships (for example the knock on effects of credit default swaps), incentives, etc. Why? Because the individual relationships were deliberately opaque, and there were few relevant examples of what would happen in extreme circumstances. Fairly simple theories could predict to a large extent what actually happened, but other fairly simple theories (the more popular ones) predicted everything would be fine.
Furthermore, "stylized" examples were misleading, because they tended to describe events as filtered through peoples' preferred theories. Once we started to get really detailed examples of how things were actually working, some problems became obvious. However other important issues (such as the degrees and ways that market participants have been welded into a single mass by risk sharing) are still impossible to adequately theorize because proprietary opacity means we just don't know enough specifics to find the right theory.
So in this real example, markets with the most intelligence applied to them were unable to predict near future issues of the greatest importance to them, because they didn't have enough good data about their own workings!
Re US cashless gifting:
In the first 4 at least, with their elaborate rules for social gifting and where not giving appropriate gifts is a serious faux pas, I would offend no one with a cash gift of new, clean bills in a fancy red envelope with a gold ribbon, as is the custom. Whereas in the US and UK, I suspect I would offend most everyone with such.
Also in Asia, it's clearer what is appropriate because the rules are much more well-defined than in the US. Anyone who's taken an international business etiquette course can tell you about the time spent on gift-giving rules - This is just hard to compare across cultures in this case.
I read that and thought, "But why should we assume that most effects will have causes that explain more than one effect?" I'm struck by the reversed-mirror-image similarity to our discussion of whether intelligence compresses - perhaps because you are an economist and I am an intelligist.
all social science theories are abductive.
Another problem is that much social behavior is, indeed, fundamentally arbitrary; we do things because other people do them, regardless of the reason the behavior initiated.
Why do people drive on the right side of the road in some countries and on the left side in others? Why Microsoft? Why QWERTY? Why Beanie Babies?