- Dictator Game: I am strictly indifferent as to how much I send her or she sends me.
- Trust Game: I send her everything and am strictly indifferent as to how much she sends back; I expect she'd send me everything in the sender role and expect she'd prefer a split as a signal of caring but otherwise wouldn't much worry about it as it all winds up in the same place.
I do not understand the drawing of strong conclusions about couples from how much they send in the dictator game. Either I, or Sue, might choose amounts just based on amusement. And similarly in the recipient role in the trust game. Even if the stakes are very large relative to income, we jointly decide how things get spent afterwards. If the stakes are small, money in her pocket is money that isn't spent out of the joint account later, and vice-versa.
Carolina Castilla has a paper out in the May AER running trust and dictator games with spouses in India, with stakes of up to 80% of daily household income. Receivers send back a bit over half of what they receive; when playing as dictator, they send back half. Senders send over a bit over half. There's lots of proposed explanations for the less-than-optimal sending, none of which are the obvious "Maybe they don't trust the experimenter to actually triple the money or not to pocket some of what's in the envelope."
The main interesting finding (in the ungated and more extensive working paper) is that in households where the man spends a lot on tobacco, the wife sends over less money - potentially indicating problems in household bargaining. And, in households where the wife handles the kids' education expenses, the husband sends over more - presumably saving him the trouble of handing her a bigger share after the experiment to pay the school bills. None of that makes it into the the AER version, presumably because there are rather a few just-so stories floating around.
The World Bank's Development Blog discusses the paper, lauding the generally higher trust exhibited by married partners as compared to stranger partnerings. Where we don't know how well the subjects trust the experimenters in countries with no small issues with corruption, trust in the experimenter might limit the extent to which within-couple trust can be exhibited in the experiment.