The Internet offers a number of advantages as a survey mode: low marginal cost per completed response, capabilities for providing respondents with large quantities of information, speed, and elimination of interviewer bias. Those seeking these advantages confront the problem of representativeness both in terms of coverage of the population and capabilities for drawing random samples. Two major strategies have been pursued commercially to develop the Internet as a survey mode. One strategy, used by Harris Interactive, involves assembling a large panel of willing respondents who can be sampled. Another strategy, used by Knowledge Networks, involves using random digit dialing (RDD) telephone methods to recruit households to a panel of Web-TV enabled respondents. Do these panels adequately deal with the problem of representativeness to be useful in political science research? The authors address this question with results from parallel surveys on global climate change and the Kyoto Protocol administered by telephone to a national probability sample and by Internet to samples of the Harris Interactive and Knowledge Networks panels. Knowledge and opinion questions generally show statistically significant but substantively modest difference across the modes. With inclusion of standard demographic controls, typical relational models of interest to political scientists produce similar estimates of parameters across modes. It thus appears that, with appropriate weighting, samples from these panels are sufficiently representative of the U.S. population to be reasonable alternatives in many applications to samples gathered through RDD telephone surveys.

This article provides a basic report about subject recruitment processes for Web-based surveys. Using data from our ongoing Internet Survey of American Opinion project, two different recruitment techniques (banner advertisement and subscription campaigns) are compared. This comparison, together with a typology of Web-based surveys, provides insight into the validity and generalizability of Internet survey data. The results from this analysis show that, although Internet survey respondents differ demographically from the American population, the relationships among variables are similar across recruitment methods and match those implied by substantive theory. Thus, our research documents the basic methodology of subject acquisition for Web-based surveys, which, as we argue in our conclusion, may soon become the survey interview mode of choice for social scientists.

The practice of using point estimates produced by the King ecological inference technique as dependent variables in second-stage linear regressions leads to second-stage results that, in general, are inconsistent. This conclusion holds even when all assumptions behind King's ecological technique are satisfied. Second-stage inconsistency is a consequence of the fact that King-based point estimates of disaggregated quantities contain errors correlated with the true quantities the estimates measure. Our findings on second-stage inconsistency, as well as a fix that we propose, follow from econometric theory in conjunction with an analysis of simulated and real ecological data sets.

We take this opportunity to comment on Herron and Shotts (2003; hereinafter HS) because of its interesting and productive ideas and because of the potential to affect the way a considerable body of practical research is conducted. This article, and the literature referenced therein, is based on the suggestions in three paragraphs in King (1997, pp. 289–290). Because these paragraphs were not summarized in HS, we thought they might be a useful place to start.

We address in Herron and Shotts (2003; hereinafter HS) an increasingly common statistical practice called EI-R, in which point estimates generated by the King (1997) ecological inference technique are used as dependent variables in second-stage linear regressions. Although HS argue that EI-R slope estimates are inconsistent and suffer from attenuation bias, this characterization of EI-R is excessively conservative. Indeed, we show here that EI-R estimates can suffer from sign reversals, attenuation bias, and augmentation bias, that inference based on these estimates can be misleading, and that EI-R is unfixable and should not be used.

Since Herron and Shotts (2003a; hereinafter HS), Adolph and King (2003; hereinafter AK), and Herron and Shotts (2003b; hereinafter HS2), the four of us have iterated many more times, learned a great deal, and arrived at a consensus on this issue. This paper describes our joint recommendations for how to run second-stage ecological regressions, and provides detailed analyses to back up our claims.

In a 2000 article in American Journal of Political Science, James Snyder and Tim Groseclose develop and apply an innovative method for detecting and estimating the frequency and magnitude of party influence in congressional roll call voting. This paper presents a framework for assessing the coefficient that the authors interpret as “party influence.” The analysis reveals that, and shows why, the coefficient manifests two troublesome characteristics. The coefficient cannot discriminate between disparate types of party influence because the mapping between types of partison influence and signs of the coefficient is not one-to-one. Similarly, the coefficient has a responsiveness problem because a marginal increase in one party's influence can cause the estimate of the coefficient to increase, decrease, or remain constant. Because the literature on parties in Congress emphasizes majority-party strength, the inability of the coefficient to isolate party-specific effects is a serious drawback in the ongoing hunt for genuine party discipline.

Keith Krehbiel (2003) provides a thoughtful note that explains why the estimation strategy we proposed in Snyder and Groseclose (2000) cannot distinguish between majority-party influence and minority-party influence theories. We agree with Krehbiel's major points. In this note, we explain our disagreement with some of his minor points, and we suggest an additional assumption under which our method can distinguish between majority-party influence and minority-party influence.

Groseclose and Snyder's revised method (2003) for assessing relative party influence deserves high praise for its creativity and explicitness. Two limitations should also be noted, however. First, the method seems likely to overstate majority-party influence. Second, even if taken at face value, the finding of asymmetric party influence is weak when compared with a baseline model.