Theories demand much of data, often more than a single data collection can provide. For example, many important research questions are set in the past and must rely on data collected at that time and for other purposes. As a result, we often find that the data lack crucial variables. Another common problem arises when we wish to estimate the relationship between variables that are measured in different data sets. A variation of this occurs with a split half sample design in which one or more important variables appear on the “wrong” half. Finally, we may need panel data but have only cross sections available. In each of these cases our ability to estimate the theoretically determined equation is limited by the data that are available.

The problem of response instability in survey measures of policy positions has been studied for over 20 years without any apparent resolution. Two major interpretations remain: Philip Converse's nonattitudes model and a measurement error model. One reason why neither interpretation has as yet been rejected or well supported is that previous analyses have depended on three-wave panel data that do not contain sufficient information to assess the goodness-of-fit of the models and also provide unreliable estimates of the error variance for the issue questions. Using five-wave panel data, this article first re-estimates the measurement models for the issue positions to assess goodness-of-fit and then estimates models of response instability to help establish its determinants. Evidence consistent with both interpretations of response instability is found. It thus appears as if neither model can adequately deal with the empirical characteristics of opinion questions in panel data. In the conclusion, a third interpretation of the response instability problem is offered that better accounts for the empirical findings and is more consistent with our understanding of public opinion.

Systematic sampling and temporal aggregation are the practices of sampling a time series at regular intervals and of summing or averaging time series observations over a time interval, respectively. Both practices are a source of statistical error and faulty inference. The problems that systematic sampling and temporal aggregation create for the construction of strongly specified and weakly specified models are discussed. The seriousness of these problems then is illustrated with respect to the debate about superpower rivalry. The debate is shown to derive, in part, from the fact that some researchers employ highly temporally aggregated measures of U.S. and Soviet foreign policy behavior. The larger methodological lessons are that we need to devote more time to determining the natural time unit of our theories and to conducting robustness checks across levels of temporal aggregation.

The purpose of this article is to present a methodology for better gauging the nature and dynamics of social and political attitudes. Starting from a particular view of attitude assessment, we show how computer-assisted interviewing can help transform the survey interview from a passive to an interactive process. Since the ultimate test of a methodological approach is its substantive yield, we shall present both specific measurement procedures and illustrative substantive results. The methodology itself, however, is quite general and readily applied to other substantive areas, thanks both to the generality of the theory of measurement of which it is a part and to the flexibility of the technology in which it is embedded.

The Kalman filter is useful to estimate dynamic models via maximum likelihood. To do this the model must be set up in state space form. This article shows how various models of interest can be set up in that form. Models considered are Auto Regressive-Moving Average (ARMA) models with measurement error and dynamic factor models.

The filter is used to estimate models of presidential approval. A test of rational expectations in approval shows the hypothesis not to hold. The filter is also used to deal with missing approval data and to study whether interpolation of missing data is an adequate technique. Finally, a dynamic factor analysis of government entrepreneurial activity is performed.

Appendices go through the mathematical details of the filter and show how to implement it in the computer language GAUSS.

Many important social science models are estimated with subnational units, such as states, congressional districts, or counties as the units of observation. These analyses may require estimates of population characteristics for these units that are not readily available from published sources. Frequently the size and number of these units make direct data collection, such as household surveys, prohibitively expensive or completely infeasible. The need is to develop methods for using more readily available data to estimate these subgroup attributes. This article presents such a method based on national-level survey data.

In his 1962 paper, “Information Flow and the Stability of Partisan Attitudes,” Converse explained why moderately sophisticated voters are sometimes most susceptible to persuasion in election campaigns. Such people, Converse argued, pay enough attention to campaigns to be fairly heavily exposed to persuasive messages but lack the sophistication to be able to resist. The present article extends this model in several ways, showing that it can yield important insights into House elections, presidential elections, presidential primary elections, and the dynamics of presidential popularity.

On the theoretical side, this article introduces the notion of differential information flow, which is the idea that campaigns consist of multiple messages that may penetrate differentially far into the mass electorate. It shows that differential information flow can help explain nonobvious but theoretically interesting patterns of attitude change in several areas.

On the substantive side, this article pays particular attention to the dynamics of incumbent advantage in House elections, showing among other things, why senior incumbents are typically able to build up huge winning margins in weakly contested elections but are unable to maintain this advantage under challenge.

For more than three decades, social scientists have struggled with the statistical consequences of aggregation. Ever since Robinson (1950) first shocked a whole generation of social scientists with his demonstration of the “ecological fallacy,” much has been written about alleged fallacies, biases, pitfalls, and hazards of one kind or another lurking behind aggregate data and about strategies for circumventing them (Goodman, 1953, 1959; Blalock 1964; Scheuch 1966; Alker 1969; Shively 1969, 1974; Hannan 1971; Hammond 1973; Meckstroth 1974; Hanushek, Jackson, and Kain 1974; Hannan and Burstein 1974; Irwin and Lichtman 1976; Smith 1977; Langbein and Lichtman 1978). Intrigued—or alarmed—by the recurrent observation that correlations and regressions based on aggregate data (group means) often differ dramatically from those based on individual data, researchers have sought to answer the traditional question of ecological analysis: under what conditions can inferences to individual-level (micro) relationships be made from group-level (macro) data?