Validity of Four Types of States

Abstract

This chapter asks: Does the joint classification of the states by their level of human development and by their income equality produce a meaningful and valid typology? To answer this question, this chapter studies the joint effects of human development (0, 1) and income equality (0, 1) forming a typology defining the South, Heartland, postindustrial states, and balanced states. The latter are balanced in which they have both high human development (HD) and high equality, and they provide a baseline for comparisons with the South, its polar opposite, and with the other types of states. The four categories of this typology are meaningful and validly classify the states. This chapter studies the least-square means for the types of states. It studies how lower HD and lower equality interact to create polarity differences on macrolevel indicators of culture, postindustrial economies, and environmental pollution. Among states with lower HD, the South is worse off on these macroindicators than the Heartland, and both are worse off than the states with higher HD, the postindustrial and balanced states. The differences between the latter two are less salient but the income inequality of postindustrial states creates economic insecurity. Differences in HD have stronger average effects than differences in income equality. To ameliorate these tangible problems these differences can guide practical politics and voting.

Republicans are geniuses at getting people to vote against their own self-interests.

—Maureen Dowd (October 28, 2012, 13)

Notes

• 1. Woodard (2011).

• 2. Gelman et al. (2008).

• 3. HD taps a state’s wealth; income equality taps its distribution. HD also can be viewed as a comprehensive measure of a state’s socioeconomic status (SES), with health being the enhancing attribute. Effects are judged as favorable if they enhance HD broadly conceptualized.

• 4. Stinchcombe (2005, 39–41, 60–61, 250–251) provides examples of polarity differences.

• 5. This matrix typology (Gerring 2012, 146) groups the states into reasonably homogeneous groups thereby advancing the work of Stephan and Linz (2011), who do not study types of states.

• 6. Pinker reviews explanations of the red and blue political divide primarily in terms of selection effects and psychological predispositions and not by differences in HD and income inequality. See http://www.NYTimes.com. Accessed 24 October 2012.

• 7. Wikipedia discusses and lists the states forming the Confederacy. http://en.wikipedia.org/wiki/Confederate_States_of_America. Accessed 24 April 2012.

• 8. The political color of a state is based on a party’s average margin of victory in five presidential elections for the period 1992–2008. Table 3.2 reports which states are red, purple, or blue.

• 9. Fifield (2012, 2).

• 10. John McDermott of the Financial Times has provided “Notes from the Heartland” focusing on the following five states all of which are included in this chapter’s classification of Heartland states: South Dakota, North Dakota, Nebraska, Iowa, Kansas, and Missouri. Information from http://www.FT.com/Heartland. Accessed 17 October 2012.

• 11. Killian (2012, 159–180) characterizes Ohio as the ultimate swing state in part because of its deindustrialization and the recession: Many blue collar men in Ohio are cross-pressured “American First Democrats ”—nationalistic swing voters desiring job security who are hurting economically and receptive to the Republican candidates who promise economic recovery. But Obama won the electoral votes of Ohio in the 2008 and 2012 elections.

• 12. Fifield (2012, 2).

• 13. Killian (2012, 131–158) describes how Virginia is composed of liberal northern counties near Washington, D.C. and more conservative southern counties, resulting in many cross-pressured “Starbucks Moms and Dads ” who are swing voters. The northern counties include many high-income government workers who are economic and social liberals. The southern counties include many traditional southerners who are economic and social conservatives. These differences engender polarized politics. Obama won the electoral votes of Virginia in the 2008 and 2012 elections. In a closely contested gubernatorial election in 2013 liberal Democrat Terry McAuliffe won over conservative Republican Ken Cuccinelli by a margin of 47.9 to 45.5%. The latter used sentiments against the Affordable Care Act as a wedge issue and this tightened this election (Wall Street Journal, 2013, November 6, 1).

• 14. Credible samples should exhibit geographic spread (Sudman 1976, 26–27).

• 15. Killian (2012, 68–93) characterizes New Hampshire as a swing state in part because of its libertarian tradition. There are many moderate Republicans who are swing voters. These fiscal conservatives and social liberals may vote for moderate Democratic candidates because of the Republican’s rejection of women’s choice and that party’s shift to the extreme Right. Obama won the electoral votes of New Hampshire in the 2008 and 2012 elections.

• 16. Killian (2012, 94–130) characterizes Colorado as a swing state because it is composed of youthful, highly liberal urban areas (Boulder, Aspen, and Denver) along with extremely conservative areas (the frontier ranches and Colorado Springs with its evangelical organizations, military bases, and defense industries). Young people in their early twenties and thirties compose “The Facebook Generation ” composed of social liberals, environmentalists, economic centrists, voters dissatisfied with both political parties, and people registered to vote as unaffiliated. These independent voters preferred Obama and the Democrats in 2008 but became disenchanted because of the gridlock in Washington and now might prefer moderate Republicans. Even so, Obama won the electoral votes of Colorado in both 2008 and 2012.

• 17. Fifield (2012, 2).

• 18. Lazarsfeld ([1936] 1972, 99–106) primarily emphasizes typology construction and substruction (i.e., reducing the number of types systematically) and not the validity of the constructs that are most often based on the analyst’s intuition. Smith (2011, 452–453) defines five aspects of validity : fit, construct, internal, external, and statistical conclusion. The typology of states conforms to these criteria.

• 19. Texas is classified as a (0, 0) state but it includes high-tech clusters around Austin near the University of Texas and in Dallas (Moretti 2012, 86–88). Porter (2012, 1369) provides maps of areas in the South that have clusters of high development (e.g., Atlanta, Georgia) and of low development (e.g., the rural areas). North Carolina is classified as a (0, 0) southern state but it includes a high-tech cluster around Raleigh–Durham–Chapel Hill, near Duke University and the University of North Carolina. The Editorial Board of the New York Times (2013, A12) critiques North Carolina’s Republican governor and the Republican majority in the Legislature for: “tearing down years of progress in public education, tax policy, racial equality in the courtroom and access to the ballot.” See “The decline of North Carolina,” New York Times, July 10, A22.

• 20. The effects of economic inequality often trump the effects of average economic income, especially in comparisons among nation (Wilkinson and Pickett 2009, 499, 509). But in the U.S. average income often has strong effects on health and social problems. If income inequality is taken to be indicative of difference in socioeconomic status (SES), then their results and this chapter’s results are not contradictory, because HD can be viewed as a measure of SES and income inequality may operate through social stratification.

• 21. Richard Wilkinson emphasized the importance of uncovering the effects of different contexts of inequality in his lecture at the Harvard Inequality and Social Policy Seminar , 24 September 2012. However, this chapter reserves the term “contextual analysis” for the study of the relationship between individual-level variables in different macrolevel contexts (Ennis 1962, 180–211; Achen and Shively 1995, 219–233).

• 22. Etzioni (1968, 1983, 87–88) develops the concept of malleability for policy research: Effort is better spent on interventions that promise to more easily change the response variable.

• 23. Proc SurveyReg and Proc SurveyLogistic provide many of the estimates in this chapter rather than Proc Mixed or Proc Glimmix because with a level-2 response variable and a system of level-1 explanatory variables, the models do not converge if there is a random statement. If the random statement or its equivalent is deleted from the model, then this deletion creates a simple regression model with no capability for survey weights other than the Weight statement, which is not a good substitute. Consequently, when there is a level-2 response and level-1 covariates it is better to apply Proc SurveyReg or Proc SurveyLogistic using the sample weights rather than Proc Mixed or Proc Glimmix with no survey weights and no random statement. The level-2 variables are conceptualized as contextual properties of the individual respondents (Lazarsfeld and Menzel 1972, 230).

• 24. The linear probability model can be viewed as a special case of a generalized linear model , one with a binomial distribution and an identity link, or a normal model with an identity link. The research of Helevik (2009) justifies the use of linear probability models when the response is a dichotomy. Murname (2013, 395–398) models dichotomies using linear probability models.

• 25. SAS Usage Note 37228 describes the use of Proc NLMixed , Proc Logistic , Proc Genmod , and Proc Catmod for modeling differences in probabilities. Of these, only Proc SurveyLogistic has the capability of including survey weights.

• 26. Polarization between Democrats and Republicans regarding ideology implies that on the ideology continuum Democrats are strongly liberal and Republicans are strongly Conservative, with few taking a middle position; the distribution of ideology across degrees of party affiliation would be bimodal with little overlap. Consequently, the correlation between ideology and party affiliation would be very high. If the distribution flattens because some Democrats and Republicans move to the ideological center, then the correlation between ideology and party affiliation would become lower. The symmetric correlation between ideology and party affiliation quantifies the degree of sorting by party affiliation or by ideology. For the use of correlations as a measure of polarization , see Fiorina and Levendusky (2006, 58–63). For further discussions of polarization versus sorting see Abramowitz (2012, 45–61), Fiorina and Abrams (2009, 49–74), Levendusky (2009), McCarty et al. (2006), and Fiorina and Levendusky (2006, 49–71, 95–111) with comments by Abramowitz (2006, 72–85, 111–114) and by Jacobson (2006, 85–95). Attitudes about the Affordable Care Act (a.k.a. Obamacare ) are extremely sorted (or polarized) by party affiliation. Circa February 18–22, 2015 87% of Republicans disapprove and only 19% of Democrats disapprove: the polarity is 68 percentage points (Pew Research Center 2015). Since there are no generally accepted cut-points on the amount of polarity to distinguish polarization from sorting, this book most often uses the concept of polarity rather than sorting or polarization.

• 27. Smith (2011, 309–310) describes, and provides an example of, the step-down Bonferroni option of SAS’s Proc Multtest .

• 28. For microlevel response variables, Part III shall study the typology’s effects on moral conservatism and other political sentiments, and on voting. Because these response variables are at the microlevel, the effects are those of the macrolevel contexts on the individuals’ variables. Because these effects are much smaller than those for macrolevel responses, for microlevel responses the tables report probability values that may lack Bonferroni adjustments. This change in procedure reduces the likelihood of a false rejection of a hypothesis when there is in fact a statistically meaningful effect; statisticians refer to such mistakes as Type II errors .

• 29. These anthropologists are Woody and Rubie Watson .

• 30. This measure is derived from Skocpol and Williamson’s Fig. 3.1 (2012, 91).

• 31. These data on right-to-work laws are taken from The Economist (4 February 2012, 32).

• 32. These data are taken from Gillette’s map (2012, 68).

• 33. The estimated Tea Party proportions provided by Proc Survey Logistic are South = 1.000, Heartland = 0.899, postindustrial = 0.482, and balanced = 0.806. The proportions are very similar to the regression estimates in Table 7.2.

• 34. The estimated right-to-work proportions provided by Proc SurveyLogistic are South = 0.898, Heartland = 0.392, postindustrial = 0.240, and balanced = 0.099. These proportions are very similar to the regression estimates in Table 7.2.

• 35. The estimated anti-immigrant proportions provided by Proc SurveyLogistic are South = 0.979, Heartland = 0.439, postindustrial = 0.154, and balanced = 0.477. These proportions are very similar to the regression estimates in Table 7.2.

• 36. Lower HD is associated with fewer women legislators : the Heartland versus balanced difference δ_01–11 = −0.018 (B p = 0.0064).

• 37. Lower HD is associated with higher rates of military recruitment : the Heartland versus balanced difference δ_01–11 = +0.455 (B p < 0.0001).

• 38. Based on such indicators as these, Porter (2012, 1361–1363) characterizes many areas of the South as predisposed toward authoritarianism.

• 39. This pattern is broadly consistent with findings from research on educational interventions. Some examples: a preschool program has more salient effects on children from disadvantaged families than it does on children from advantaged families (Murname 2013, 409). School quality benefits children who are from disadvantaged minority families more than it does children from advantaged majority families (Coleman et al. 1966). Comprehensive school reforms benefit under-preforming schools more than over-preforming schools (Smith 2011, 259–380).

• 40. Lewis and Burd-Sharps (2010, 250–251) provide this measure of environmental cleanup sites and other measures of pollution.

• 41. If a scale is created by combining the indicators of pollution due to CO₂ emissions and pollution due to energy use, then it would have a reliability α = 0.89, very high. If the percent rural was added to this scale, then α = 0.79; a bit lower but still reliable, suggesting that these forms of pollution are associated with rural states.

• 42. The extremely high rate of mercury releases in pounds per 1000 of population (2008 data) characterizing the Heartland and the extremely low rate of the postindustrial states prevents the regression model that includes the covariates from estimating the least-squares means appropriately. Consequently, the effects of the macrolevel typology are estimated here without adjustments for the microlevel covariates.

• 43. During the 2012 campaign, some voters in the Ohio rustbelt approved of the Republican’s promise to loosen regulations on the coal and gas industries and open up more land for drilling. This would reduce energy costs and boost demand for steel pipes manufactured in Ohio.

• 44. Levi (2012, 11) opines “The world still generates 46% of its electricity from coal. ... Replacing coal-fired power stations with gas-fired equivalents slashes carbon dioxide emissions roughly in half.” On September 20, 2013 the U.S. Environmental Protection Agency (EPA) announced new regulations making it near impossible to build new coal-fired power plants in the U.S. Even so, climate activists desire the use of zero-emission renewable fuels and their opponents strive with some success to weaken these regulations.

• 45. When putative causal relationships are classified as level-1, stable association ; level-2, potential outcomes ; and level-3, dependency networks , most of the relationships in this book exemplify level-1. Cox and Wermuth (1996, 65–74) define these relationships; for examples and references see Smith (2011, 10–16).

• 46. Some social attributes exhibit rather consistent associational (i.e., not causal) relationships: (1) respondents with a critical class consciousness, more secular religious affiliation, or urban residence are associated with states scoring high on the postindustrial indicators, whereas Hispanics have negative associations with these postindustrial indicators; (2) respondents exhibiting a critical class consciousness, younger age, single marital status, or urban residence have negative associations with states scoring high on indicators of pollution due to CO₂ emissions and energy use in BTUs, whereas urban residents and Hispanics (more so than African Americans) are exposed to environments with more carcinogen emissions and lead releases.

• 47. Fiorina and Abrams (2009) study disconnections between political elites and the public.