49 internautes sur 54 ont trouvé ce commentaire utile
not a natural
- Publié sur Amazon.com
Sasha Issenberg's book The Victory Lab is well written and, for the most part, quite interesting. It is, however, a good deal longer than need be to tell its story. This is due to the author's interest in not only explaining recent developments in mounting effective political campaigns, but also giving a good deal of attention to the history of such efforts, including background on the principal participants over the last forty years. I had not expected the historical material in a book subtitled "The Secret Science of Winning Campaigns," and it sometimes gets in the way of a streamlined presentation of technical issues. The book presupposes little or no knowledge of research design, statistics, or measurement theory. That's all to the good, but if the text included less chatty historical coverage and more discussion of the fundamentals of pertinent quantitative techniques, The Victory Lab would be a more satisfying read.
The two basic themes that undergird Issenberg's account are micro-targeting of prospective voters and random assignment of treatment and control status to permit interpretable comparisons. Micro-targeting means gaining access to individuals and small, homogeneous groups rather than using data aggregated to the precinct, county, or other geographical level. Micro-targeting enables political analysts and operatives to identify conveniently small groups that do not correspond to pre-drawn geographical or administrative boundaries and to use their peculiar characteristics to focus get-out-the-vote campaigns and messages tailored to enhance the appeal of a specific candidate. The shopworn,conventional, broad-brush alternative is to use existing data sets that aggregate measures to a higher level, such as the county.
As an example, after the 2004 presidential election that pitted George W. Bush against John Kerry, I had a data set consisting of all 3140 counties in the u.S. Using a variation on conventional multiple regression analysis, I was able to determine that the most powerful county-level predictor of the percentage voting for Bush was a fairly crude measure of "traditional family values." It was easy to identify the counties that were unusually high on the "traditional family values" measure, and in subsequent elections candidates might have reasoned that these counties would be favorably disposed to Republicans generally, not just to Bush, and they might have targeted their get-out-the-vote efforts accordingly. Sounds plausible, but the "traditional family values" variable had a very strong positive correlation with income, which could mean that many of the families were wealthy enough to employ live-in help. The impact of a get-out-the-vote drive, as a result, might be diminished for Republicans because it would also energize the low-income, perhaps Black or Latino, cooks, housekeepers, gardeners, and handymen, who were more likely to vote for a Democrat.
Similarly, the "traditional family values" variable had a strong positive correlation with voter turnout. This might lead one to the conclusion that resources devoted to increasing voter turnout would be wasted if they were expended on Republican voters who would turn out anyway. Instead, the prospective voters actually affected might be the hired help, perhaps low-income minorities who, again, would vote for Democrats.
However, micro-targeting of smaller, homogeneous aggregates or individuals themselves provides a way around this broad-brush difficulty, enabling us to concentrate on just those citizens we actually want to reach. As Issenberg makes clear, however, assembling data adequate to this task is time-consuming and very expensive, and it requires a long-term commitment to maintaining and updating data files, as well as employing analysts with the statistical and measurement skills needed to put it to good use. Nevertheless, micro-targeting -- reaching out and touching individuals and small groups -- is now a staple technique used by both parties.
Issenberg's second basic theme, use of randomized experiments, makes micro-targeting, as well as more conventional approaches, much more effective. For example, suppose we had used micro-targeting to identify clusters of affluent "traditional family values" voters who, contrary to convention, cannot be relied on to turn out on election day. Randomization assures that any treatments, such as mailings to non-voters that include a call to vote, will not be confounded with other variables, and we can vary the wording of the message, compare results of different messages from one cluster to another, and decide which is most effective. Subsequent work can then incorporate these insights.
It remains true, however, that micro-targeting and random assignment as used in The Victory Lab, are, for the most part, exercises in naked empiricism and guess work. Methodologically they may be exemplary, but substantively and theoretically they tend to be quite thin, relying heavily on expensive trial and error and indiscriminate data dredging. This may help to explain Democrats' disastrous performance in the 2010 midterm elections. Issenberg very briefly mentions 2010, but he does not begin to do it justice as a challenge to the efficacy of micro-targeting and random assignment. I think that a great deal of theoretical work, such as that reported by Robert Entman (2012) in Scandal and Silence, needs to be done to take more uncertainty out of even the most methodologically rigorous political campaigns.
The Victory Lab does a good job of documenting the introduction of methodological precision into data-driven election campaigns. However, Issenberg leaves the reader with the impression that political science is nothing but more and better analyses of voting behavior, and that a rich substantive and theoretical literature is not necessary to guiding the efforts of those who collect and analyze data. Both impressions are wrong.
Issenberg obliquely acknowledges that contemporary developments aimed at more effectively influencing voting behavior may be rightly interpreted as manipulation, unscrupulously useful to anyone with the resources to employ them, whatever their political objectives. I find this "value free" framework unsettling, hardly the stuff of democracy as most of us conceive it.
The Victory Lab is an interesting, even if unduly long and philosophically short-sighted book. I benefited from reading it and would recommend it to others.