Module 3 part 3 III. An Example: Measuring Potential Causes of Voter Turnout A Study at the Individual Unit of Analysis Let's return to our discussion of potential causes of voter turnout. At the end of Module Two, we discussed five potential causal influences on political participation: gender, race, age, education, and income. If we study the question of voter turnout at the individual unit of analysis, we are trying to determine what sort of factors make a person more (or less) likely to vote? When we go to measure these five independent variables, we find that three can be measured at the ratio level. Age can be measured in years lived; education can be measured in years of formal schooling; and income can be measured as the total number of dollars earned per year. All three such measures have ratio qualities: the measurement units are expressed in numbers with normal mathematical properties that capture discrete intervals of presence or absence. For instance, an 80-year-old is 4 times as old as a 20-year-old and 2 times as old as a 40-year-old. Furthermore, the complete absence of the variable at issue would be expressed by the use of a zero. However, just because we can measure these three concepts at the ratio level, it does not mean that we must do so. We could also measure all three independent variables at the ordinal level by grouping people into different ordinal categories. If you have ever responded to a written marketing questionnaire, you have probably seen all three of these concepts measured at the ordinal level. Do the sample questions below seem familiar? How many years old are you? 29 OR YOUNGER30-3940-4950-6465 OR OLDER What is the highest degree of education you have obtained? SOME OR NO HIGH SCHOOLHIGH SCHOOL DIPLOMA OR EQUIVALENCEATTENDED A TWO-YEAR OR A FOUR-YEAR COLLEGEGRADUATE OR PROFESSIONAL SCHOOL DEGREE Which of the following brackets best describes your personal annual income in 1998? UNDER $20,000$20,000 TO $34,999$35,000 TO $49,999$50,000 TO $74,999$75,000 AND UP Which approach to measuring these three independent variables is preferable? Should we take a ratio measure or an ordinal measure? The answer depends on the theoretical link between each independent variable and the dependent variable. For instance, if we think that each additional dollar of income has an equal effect on one's decision to vote, then we should stick to a ratio measure. However, if we think the important distinction is not the precise number of dollars but instead the ordinal degree of income, then we should measure income using some form of brackets that distinguishes low-income people from middle-income people and middle-income people from high-income people. So far, we have not examined how we might measure the other two independent variables presented in Module Two: gender and race. When we consider these concepts at the individual unit of analysis, we find that they are at the nominal level. Femaleness does not constitute "more gender" than maleness or vice versa; instead, people belong to different gender categories. Similarly, there is no meaningful measurement ordering of racial categories. We would not say that Asians have "less race" than Hispanics or that Caucasians have "more race" than Africans. In Module Five we will use tables to study the relationship between nominal and ordinal measures of each of our five independent variables and the dependent variable of voter turnout. Voter turnout will be measured by asking citizens whether or not they voted in the last election. The study in Module Five will be at the individual unit of analysis as we try to determine what factors influence a person's decision to vote. A Study at the Aggregate Unit of Analysis Suppose, however, that what sparked our interest in voter turnout was the difference in voter turnout across U.S. states. Instead of determining what factors influence a person's decision to vote, we instead ask: what conditions make the rate of voter turnout different from one state to another? If we change the unit of analysis from the individual person to the trends in different states in the U.S., we can move the entire research project from the individual to the aggregate level. When we change the unit of analysis, it implies that our measures need to be descriptive not of individual citizens but instead our measures need to capture the trend of activity across an entire state that is an aggregation of individuals. For instance, the measure of the dependent variable changes from an individual's yes/no answer to the question (did you vote in the last election?) to a statistic that captures the trend in voter turnout across an entire state. Normally state voter turnout in the U.S. is measured as the percentage of voting-age citizens who voted. Notice that this voter turnout rate is a ratio measure. An 45% turnout rate is 3 times as high as a 15% turnout and half as high as a 90% turnout rate. Also, 0% turnout would indeed mean no voter turnout. Similarly, our measures for gender and race (which were nominal measures at the individual unit of analysis) also become ratio measures when we try to describe the overall gender or racial makeup of an entire states. Gender gets measured as the percentage of the population that is male (or female). The precise measure for race depends on the distinction driving our hypothesis. In Module Two we noted that some researchers hypothesize that black citizens are less likely to vote than white citizens due to feelings of powerlessness and/or alienation produced by a climate of discrimination and indifference. If that is our theory, then at the aggregate level we might choose to test that hypothesis by measuring race as the percentage of the population that is African-American. Once again, when we shift the unit of analysis, concepts that were purely categorical distinctions for individual human beings can be measured at the ratio level once we try to measure statewide trends. When we turn to our other three independent variables (age, education, and income), we have some choices to make regarding how to measure these variables. We could measure all three concepts statewide by taking the average across the entire voting-age population: the average age of adults (in years), the average years of schooling of adults, and the average annual income. On the other hand, what if we feel strongly that what really matters is belonging to a certain ordinal category in each of these variables? If we believe that young people are markedly less likely to vote, we can test that hypothesis by measuring age statewide as the percentage of the adult population under 30. Similarly, if we think that limited education makes one less likely to vote, we can test that hypothesis statewide by measuring the percentage of the adult population that did not finish high school. Finally, if we think that high income makes one particularly likely to vote, we can test that hypothesis statewide by measuring the percentage of families with above average income. In Module Six we will test each of these hypotheses using data descriptive of the 50 states in the U.S. Before we turn to the analysis of real data on voter turnout in Modules Five and Six, we take time out to consider where and how one might look for data on political research questions. These nuts and bolts concerns are at the heart of Module Four.