Edward R. Tufte

Visual explanations: images and quantities, evidence and narrative

Pick up any issue of tlie journal Science and take a moment to evaluate their typical composite graph. After struggling to correlate the A’s, B’s, C’s and D’s of the component pictures with those buried in the ponderous figure legend, try to match the shapes (circles, squares, triangles; filled or open) of the data points with the conditions they represent (also in the legend), and hope that the acronyms and abbreviations will make sense (if you manage to find them in the main text). Let us concede that we are conditioned to put up with a lot of poorly presented inforination. If you are in doubt, try submitting your own simple graphs to the journal with explicit labels on the curves, and be prepared for the editorial staff to amalgamate the items into confounding sets and to abstract the curve identifiers into the deadly legends. Another journal, Cell, elects to publish in fonts with no serifs ancl denies the wellfounded custom of using italics for Linnaean genus and species nomenclature. Perhaps the culprits believe that this “modern and crisp” style helps improve reader comprehension. In fact, controlled studies reveal the opposite. All of this is the more remarkable because editors are supposed to be engaged in an activity that improves clarity and logical explanation. I wish they would read Edward Tufte. Visiial Explanations is the third book in a series written and published by Tufte, a professor at Yale University. The first, The Visual Uisfilay of Quuiitital i 7 ~ Data (1983), was both instructive and entertaining. I missed his book EnvisioninSInformntioiz (1990), but want to read it as soon as I can. The present volume is beautifully produced. Several examples, both old and current, are displayed, first as originals and then after improvement by Tufte and his associates. The results are entirely convincing. Even an icon such as John Snow’s statistical graphic on the 1854 cholera epidemic in London is analyzed without trepidation. The author dives into the data around the Broad Street pump (the source of infected water in the epidemic) and shows how various displays will more or less support the proper interpretation of the mode of transmission of cholera. On the other hand, the change in incidence of disease after the climactic removal of tlie pump-handle has been overly interpreted by other commentators in terms of cause ancl effect. Tufte succinctly and intelligently takes the reader through the logical possibilities and coincidences, without denying that the correct and constructive conclusion was reached by Snow. “There are right ways and wrong ways to show data; there are displays that reveal the truth and displays that d o not.” Edward Tufte puts a fine point on this maxim by comparing the nineteenthcentury presentation of tlie cholera epidemic with the data displays behind the disastrous 1986 explosion of the Challenger space shuttle. The latter incident was attributed to the failure of two 0rings at low temperature, and the good professor shows us how the now-obvious indicatioiis of potential problems were hidden under unproductive data summations. The serious need for good visual and statistical thinking could hardly be exemplified more forcibly.

Determinants of the Outcomes of Midterm Congressional Elections

An explanatory model for the outcomes of midterm congressional elections is developed. Midterms are a referendum on the performance of the President and his administrations management of the economy. The explanatory model of midterm congressional elections is sufficiently powerful so as to yield honest and accurate pre-election predictions of the national two-party vote in midterm elections. These predictions have usually outperformed pre-election forecasts based on survey data. The model is extended by considering the translation of votes into seats, models of the electorate as a whole and of the individual voter, and the causes of the off-year loss by the Presidents party.

The Relationship between Seats and Votes in Two-Party Systems *

An enduring fact of life in democratic electoral systems is that the party winning the largest share of the votes almost always receives a still larger share of the seats. This paper tests three models describing the inflation of the legislative power of the victorious party and then develops explanations of the observed differences in the swing ratio and the partisan bias of an electoral system. The “cube law” is rejected as a description, since it assumes uniformity (which is not observed in the data) across electoral systems. Explanations for differences in swing ratio and bias are found in variations in turnout over districts, the extent of the “nationalization” of politics, and, most importantly, in who does the districting or reapportionment. The measures of swing ratio and partisan bias appear useful for the judicial evaluation of redistricting schemes and may contribute to the reduction of partisan and incumbent gerrymandering.

A Note of Caution in Causal Modelling

Many empirical investigations in the behavioral sciences today aim at tracing the causes of variations in some key dependent variable. The search for satisfying causal explanations is difficult because of the complexity of social phenomena, the crudeness of the measures of many important variables, and the prevalence of simultaneous cause and effect relations among variables. Although these difficulties remain, a number of important methodological contributions have clarified the conditions under which causal inferences can be made from non-experimental data. In particular the Simon-Blalock technique has recently gained considerable attention, and has been profitably used by a number of political scientists in their research. Examination of some of these applications does, however, reveal the need for a better understanding of the purposes and limitations of the technique. This paper reviews two studies: (1) the re-analysis of the Miller-Stokes data by Cnudde and McCrone, and (2) the analysis of the determinants of Negro political participation in the South by Matthews and Prothro. We shall argue that both these applications have two faults: (1) a failure to distinguish conclusions from assumptions, and (2) an inadequate correspondence between the assumptions made in constructing the mathematical models and our prior knowledge about the phenomena being studied. In addition, we shall use the first study to illustrate a principle of general importance in causal analysis: the investigator should check the possibility that different causal mechanisms occur in different subgroups of his data. And we shall use the second study to illustrate the difficulty of separating the effects of two highly correlated independent variables.

THE VISUAL DISPLAY OF QUANTITATIVE INFORMATION

Slik lyder tre av omtalene av Edward Tuftes The visual display of quantitative information. Siden første utgave kom ut i 1983, har den blitt sett på som en tidløs klassiker og bestselger innen informasjonsgrafikk. Siden har den hatt tre «oppfølgere»: Envisioning information (1990), Visual explanations (1997) og Beautiful Evidence (2006). Med sin høye kompetanse innen informasjonsgrafikk blir Edward Tufte i dag sett på som en av de fremste pioneerene innen faget, og han har blitt tildelt over 40 priser for sine verker. Han er professor ved Yale University, og underviser den dag i dag i statistikk, informasjonsdesign og politisk økonomi. I The visual display of quantitative information drøfter Tufte med et kritisk syn flere eksempler på informasjonsgrafikk skapt gjennom tidene. Budskapet summeres opp i klare retningslinjer på hva som er rett og galt innen faget – hvordan man skal visualisere kvantitativ informasjon korrekt. Boken har derfor blitt tatt godt i mot hos flere yrkesgrupper – den har nærmest blitt sett på som en bibel – dog har den og opplevd mye kritikk selv.

Political Statistics for the United States: Observations on Some Major Data Sources

Thirteen major data sources for the study of American politics are examined with regard to their conceptual orientation, error structure, and inferential utility. A great deal of ephemera and measurement without theory is discovered. Few of the documents contain any serious discussion of error structure, although some do report “standard errors” based on naive sampling models. In addition to suggestions for improving the compilation of political statistics, recommendations for a basic minimum library of data sources for American politics are made: The Almanac of American Politics and the Statistical Abstract of the United States , followed by the Guide to U.S. Elections .

Applied Multiple Regression/Correlation Analysis for the Behavioral Sciences.

Contents: Preface. Introduction. Bivariate Correlation and Regression. Multiple Regression/Correlation With Two or More Independent Variables. Data Visualization, Exploration, and Assumption Checking: Diagnosing and Solving Regression Problems I. Data-Analytic Strategies Using Multiple Regression/Correlation. Quantitative Scales, Curvilinear Relationships, and Transformations. Interactions Among Continuous Variables. Categorical or Nominal Independent Variables. Interactions With Categorical Variables. Outliers and Multicollinearity: Diagnosing and Solving Regression Problems II. Missing Data. Multiple Regression/Correlation and Causal Models. Alternative Regression Models: Logistic, Poisson Regression, and the Generalized Linear Model. Random Coefficient Regression and Multilevel Models. Longitudinal Regression Methods. Multiple Dependent Variables: Set Correlation. Appendices: The Mathematical Basis for Multiple Regression/Correlation and Identification of the Inverse Matrix Elements. Determination of the Inverse Matrix and Applications Thereof.