A key assumption in many social sciences is that people have preferences, and that these are both knowable and stable. That's the point of surveys on every subject from whipped topping to the nature of happiness. We want to know what people want, or how they feel, or what they plan--so we ask, assuming the answers will be reliable.

But why should they be? People's moods rise and fall, their circumstances change all the time--and those shifts have a big impact on what they feel and what they want. Case in point: "Subjective well-being.'' Surveyers often ask people, for example, to pick a number between 0 and 10 to indicate how satisfied they are with their lives, all in all. The results have established the supposed facts that married people are happier than singles, that women in their 30's and 40's aren't too happy, and that beyond a basic floor of income, money doesn't buy happiness.

Trouble is, people's satisfaction with their lives seems to be sensitive to transient circumstances. Ask them three times, you'll get three different answers. For example, as Alpaslan Akay and Peter Martinsson found in this working paper last month, people's replies to the "how satisfied'' question changed depending on what day of the week they were asked. (Sunday, according to their analysis of data on German workers, is the glummest day.) Unless a poll asks the same people the same question over time, then, how do we know what it measures?