Just saw Kos was planning to read Freakonomics -- a fun, and at times quite insightful book by U. Chicago Professor Steven Levitt with Stephen Dubner of the NY Times Magazine. If you haven't read it, you probably won't find this diary entry very interesting. But if you have, I thought you might enjoy thinking or even arguing about a few errors and mis-drawn conclusions I identified... One of them is Levitt's conclusion that money doesn't affect the outcome of political races...
1) First, a few issues regarding the analysis of guns and swimming pools on pp. 149-150. Levitt compares the national ratios of number of pool deaths by children / the number of pools, against the number of children killed by guns / the number of guns, concluding that the likelihood of death by pool is 1 in 11,000 versus death by gun which is 1 in 1 million plus. However, in the hypothetical which precedes this comparison, he notes that "Molly" is forbidden from going to "Amy's" house. Therefore Molly would be visiting a gun-owning household, not a gun. I think there are approximately 110 million households in the US, and apparently 2 in 5 households own guns, according to a 2001 Harris Interactive study (the first Google-supplied link that I clicked). That means that there are roughly 44 million gun-owning households. The risk of death by gun for Molly is closer to 1 in 250,000. Still not close to the risk posed by a pool, but much riskier than Levitt concludes. (I do think that it's safe to assume that there is one pool per household!). While I don't have access to the data, it would be interesting to know if the swimming pool / gun comparison would also turn differently depending on the age of the children involved. I.e. do most children swimming pool deaths occur in children under the age of 5? What about gun deaths?
However, a friend of mine pointed out that Levitt's conclusion may be more seriously misleading. Levitt asserts that conventional instincts of parents are wrong because pools are more dangerous than guns; while it may well be the existence of this conventional wisdom which has led to guns being safer than pools. The perception that one activity is safe and the other unsafe is a self-fulfilling prophecy. Guns kill fewer children because parents are more cognizant of the dangers and act/parent accordingly. The statistics Levitt cites probably merely reflect successful parenting.
Levitt based his analysis on deaths-per-year, but the relevant statistic may be related to exposure rate or, even more importantly for children, to the unsupervised exposure rate. Children are encouraged to use pools for recreation, both supervised and unsupervised, while children are generally discouraged from using weaponry without adult supervision, particularly in the house. By the measures Levitt uses, almost any childhood unsupervised activity (swimming, riding bikes, horseback riding, etc.) will probably be more dangerous on a per unit basis than guns. The vastly different manner in which children are exposed to pools and firearms are such that simple ratios of deaths to guns and deaths to swimming pools are insufficient to speak to the danger or appropriate level of parental concern for either item. My friend and I would wager that average children are exposed to pools on an unsupervised basis for many of more hours than to guns on an annual basis. Presumably, if parents took Freakonomics to heart and children were encouraged to play with guns rather than pools, the incidence of gun deaths would be quite a bit higher.
2) In considering What Makes a Perfect Parent on p. 171, Levitt notes that the ECLS survey included direct interviews, making it likely that parents would lie about whether or not they spanked their kids. This makes complete sense to me, but spanking is not the only question likely to produce a false answer from a parent. Consider the questions about reading and books. An interviewer can see for themselves which children have lots of books in their home. But when the interviewer asks parents if they read to their children daily, it is highly likely that the parents will respond by saying "yes" whether true or not. One has to wonder how many parents truthfully read to their children daily, but don't have many children's books in their homes.
Levitt's analysis is that the factors which correlate to early educational success may be genetic rather than environmental. But a close look at his summary of the survey findings suggest that the factors which correlate to a child's success are really just the ones which the interviewer can readily observe or are least likely to be lied about - books in the house, speaking English, joining the PTA. Most of the factors which "surprisingly" don't correlate are the ones that are most likely to invite parental exaggeration or understatement: taking children regularly to museums, reading to them daily, and allowing them to frequently watch TV.
3) Levitt's analysis of names - particularly on pp. 201-204 and p. 197. He identifies the twenty white boy names that best signify high-education parents. Many of these names are Hebrew, but in his discussion of what drives names in and out of favor, he has missed one of the most obvious forces, at least for Jewish families. Traditional Jewish families typically do not name their children after living relatives. Accordingly, there is a natural cycle of grandchildren and great-grandchildren being named for relatives who have recently passed away. If one wants to find out some of the names which will be popular in 2015, at least among Jewish families, one of the most obvious things to do would be to look at the names that were most popular in 70-90 years ago.
While non-Jewish families do name children after living relatives, it is also probably likely that many of these families will be inspired to name children after recently deceased relatives. This is not meant to quarrel with the correlations Levitt identifies, but to suggest that other factors may also be at work and should not be left out of the discussion.
4) Analysis of money and politics beginning p. 10 - perhaps most important to Daily Kos. Levitt is persuasive in demonstrating that money and votes do not always correlate. But, by simply comparing money to votes he may be oversimplifying the relationship between two variables - or combining what really should be multiple variables into just two. From what I know, campaign money is overwhelmingly spent on advertising. But I suspect any campaign manager will tell you that "free media" is even more valuable than advertising. And incumbents always get more free media - in large part because they are already making news as officeholders long before the election is held. (By contrast, self-funding candidates without demonstrated real support may receive negative press coverage or be neglected.) Incumbents also typically enjoy the franking privilege while in office, again resulting in a disconnect between their spending and their actual marketing reach. So when Levitt writes that "A winning candidate can cut his spending in half and lose only 1 percent of the vote," he may be overlooking the possibility that his sample, winning candidate's combined paid and unpaid media exposure could have remained relatively constant.
Earlier he also speculates that "the appealing candidate raises much more money and wins easily. But was it the money that won him the votes, or was it his appeal that won the votes and the money?" An intriguing question - but I think it is only binary because it is based on a false premise. The incumbent raises money long before becoming a candidate for reelection. So, many donations to an incumbent have nothing to do with their upcoming campaign. PACs and people who donate to sitting Congressmen probably often do so because those politicians are in positions of direct power at that very moment. They are not contributing to a candidate who they want to win - but rather to an officeholder who has already won. They are buying influence or at least access after the initial race is over.
I haven't seen the data Levitt used, but it seems likely that the largest problem with his analysis is that he is inherently assuming that if spending and winning votes strongly correlated, then all votes would be up for grabs. By contrast, I suspect that in a typical general election very few votes are truly up for grabs. Levitt notes that a winning candidate cuts spending in half and only loses one percent of the vote, potentially showing that money is not a determinant. But consider a general election in which 5% of the voters are swing voters. A 1% change in winning margin may be equivalent to a 20% change in opinion among the voters who are sway-able. (I'm guessing Levitt did not use data from primary elections - but if he did, the results would be even more skewed since opposing an incumbent in a primary is usually considered heretical among the party faithful who are typically the only ones who turn out for mid-term congressional primaries.)
Similarly, consider an electorate that is sharply divided among party lines. If you are a Republican candidate, you will find that attracting votes from the ranks of Democrats is much harder than attracting left-leaning swing voters, which of course is harder than attracting right-leaning swing voters and much harder than attracting your core base of Republicans. A plurality or majority of voters in a given District usually belong to the incumbent's party. So they are unlikely to be drawn even to a free-spending challenger because most of them are not open to changing parties for a Congressional election. However, the free-spending challenger might be able to make significant in-roads among swing voters. In which case a 1% change in margin could be telling significant story. But by that standards Levitt applies, the only change which matters is a massive realignment of voters. Occasionally there are cross-over candidates who do cause realignments, whose ideology or appeal transcends traditional labels. However, his data only includes campaigns in which candidates are running against each other repeatedly, so it is almost inconceivable that a challenger who failed to cause such a realignment in his or her first campaign would somehow succeed in doing so in a subsequent campaign. (If the incumbent did something grossly corrupt or incompetent he or she likely wouldn't be running for reelection at all, and therefore would not appear in Levitt's data set.)