Introduction to Research Methods in Political Science:
XIII. LONGITUDINAL ANALYSIS OF SURVEY DATA
Most sample surveys involve cross-sectional analysis. That is, they provide us with a snapshot of a sample of a population at a single point in time. Time, however, is itself one of the most important variables in politics. The study of change over time is called longitudinal analysis. Here we will focus on the study of change over time in public opinion and behavior, as measured through sample surveys. Longitudinal analysis of survey data can be subdivided into three types: trend analysis, cohort analysis, and panel studies.
The simplest type of longitudinal
analysis of survey data is called trend analysis, which examines overall change
over time. This figure, for example,
shows how newspaper readership declined between 1972 and 2010:
In the first year surveyed, about two-thirds of all respondents reported reading a newspaper on a daily basis. Thereafter, regular newspaper readership more or less continuously declined until, by the start of the new century, only about one-third were daily readers. (Note: look for overall trends, and do not get too caught up in short-term fluctuations. The slight rise in newspaper readership in 2002, for example, could indicate a reversal in the downward trend, but could also be due to short term forces or merely sampling error.)
Trend analysis has some significant limitations. While it can reveal change, it gives us little insight as to how or why the changes have taken place. First, individuals may change their attitudes or behaviors as they move through the life cycle. British Prime Minister Benjamin Disraeli (1804-1881) is supposed to have said that “a man who is not a liberal at 16 has no heart; a man who is not a conservative at 60 has no head.” It may be that people tend to change in predictable ways as they get older (and, depending on one’s perspective, sell out or mature). Second, people may change because of new circumstances. Technological developments, or a major crisis such as a war or depression, may result in similar changes that impact all age groups in a similar way. Finally, change may be the result of generational replacement. Different generations are shaped by different experiences. Events that occur as people are coming of age are likely to have a big impact on their political outlooks, and this impact may prove lasting, continuing to influence them for the rest of their lives. Even if no individual ever changed after reaching adulthood, overall change would occur as one generation dies out and another generation, having gone through different formative experiences, comes on the scene.
One way to sort some of this out is through cohort analysis. People born during the same time period are considered to form an age cohort. For example, respondents in their 20s in a 1980 survey belong to the same age cohort as respondents in their 50s in a 2010 survey. By comparing respondents from the same age cohort surveyed at different times, we can measure change over time in the attitudes and behavior within the cohort. In the examples that will be used here, respondents to the General Social Surveys from 1972 through 2010 have been divided into the following:
The GI Generation (born 1927 or earlier
— the earliest year of birth reported by respondents to any of the
General Social Surveys is 1883). Members
of this cohort were at least 18 years old by the end of 1945, the year World
War II ended. This generation lived
through the Great Depression of the 1930s, the New Deal, and the Second World
War. All Presidents of the
The Silent Generation (born 1928-1945). This generation came of age after World War II, but before the political turbulence of the late 1960s. The youngest members of the cohort turned 18 the year John Kennedy was assassinated. Because most members reached adulthood in relatively placid times and because it is a relatively small cohort, this group has sometimes been labeled the “silent” generation. At this writing (May 2013), it has yet to produce a president, though four members, Walter Mondale (born 1928), Michael Dukakis (1933), John McCain (1936), and John Kerry (1943), received major party presidential nominations.
The Baby Boomers (born
1946-1964). During the Depression and the Second World
Generation X (born 1965-1981). The “baby boom” was followed by the “baby bust” as birth rates plummeted in the mid-1960s. Called “Gen X” because, by some reckonings, this is the tenth generation of Americans since independence (hence the Roman numeral “X”), older members of this generation came of age during the administrations of Ronald Reagan and the first President Bush. Younger members attained adulthood during the Clinton Administration. Older members of this cohort are eligible to become president, but none have done so as yet.
The Millennium Generation (because its members came of age, and are still doing so, during or after the year 2000), sometimes called "Generation Y" (because Y comes after X). The oldest members of this cohort will not start to turn 35, and thus become eligible for the presidency, until the 2020 election.
For background purposes, the following figure graphs the changing composition of the General Social Survey from 1972 through 2010. The same information is presented in tabular form below the graph. The GI Generation provides half of the respondents to the 1972 survey, declining thereafter to less only three percent of all respondents in 2010. The Silent Generation is a relatively small cohort, never reaching more than about a third of all respondents, and reduced to about a sixth in 2010. The Baby Boomers reached their peak of representation in 1984 (46 percent), and still constitute a third of the total in 2010. The leading edge of Generation X entered the sample in 1983, and this generation is closing in on the Boomers as the largest cohort. Millennials are a small but rapidly growing part of the electorate.
As noted earlier, an overall trend, or lack thereof, may be explained in several ways. First may be a new development (a major crisis, or a new technology, for example) that effects all cohorts equally, either suddenly or gradually. If this is the case, trend lines should be similar for each cohort. Second, there may be life cycle factors at work. If Disraeli is correct, an analysis of ideology would show each cohort starting out liberal and gradually becoming more conservative. The trend line for each cohort will be in a conservative direction, but the change for each generation will lag that of the one preceding it. Third, a trend may occur as a result of generational replacement; there may be little within-cohort change over time, but an older generation may pass from the scene and be replaced by a new generation with different attitudes or behaviors. Of course, two or more of these factors may combine to produce the overall pattern.
To see how cohort analysis can be applied to study change over time, consider the next figure, which breaks trends in daily newspaper readership down by age cohort. Clearly, the overall trend away from daily newspaper reading is due almost entirely to generational replacement. Habits established early in life seem to persist, but older readers are dying out. Each generation is less devoted to the daily newspaper than its predecessor.
While cohort analysis allows us to extract more information from our data than an overall trend analysis, it still suffers from some serious limitations. While people in their twenties who were surveyed in the 1970s are drawn from the same age cohort as people in their 40s interviewed in the 1990s, they are not the same people. Both surveys are subject to random sampling error, and this may produce some of the changes we observe.
For this reason, the “gold standard” for longitudinal analysis of survey data is the panel study. In a panel, the same people are interviewed at two or more points in time. Since the sample is the same, any changes we observe are not a result of random sampling error.
Panel studies, however, have problems of their own. For one thing, they are generally very expensive, since great effort has to be expended to keep track of respondents. For another, despite our best efforts, we will not be successful in all of our attempts to recontact respondents, especially if the study is conducted over a long period of time. Those who drop out of the panel (by moving, dying, refusing to continue, etc.) might have differed in their attitudes and behaviors from those who remain. Finally, there is the problem of reactivity. When respondents are interviewed, their interest in politics may be piqued. If they know that they will be interviewed again, they may even tend to study up on politics so as not to appear ignorant. By the end of the study, what started off as a representative sample may have become something of an elite group.
cross sectional analysis
1. Start SPSS, and open gsscums.sav. Open the General Social Survey 1972-2010 Subset codebook. To facilitate analysis for the exercises in this Topic, all attitudinal and behavioral variables in this subset have been recoded to form dichotomies, with valid values of 0 and 100. Variables were recoded to divide respondents into groups that are as nearly equal in size as possible. The vertical axes of the line charts can be interpreted as the percent of respondents choosing one side of the dichotomy. In the case of newspaper readership, this represents the percent reporting daily readership of a newspaper.
Weight cases by weight. Create line charts, repeating the analysis of figures 1 and 3 in this topic, but replacing news with tvhours (the percent reporting that they watch more than two hours of television per day) another measure of attitude. Is there an overall trend over time? Are there differences within and/or between cohorts? How would you explain the patterns you observe - as generational replacement, stage in the life cycle, or a change that impacts all generations in a similar fashion?
Do the same for conpress (the percent expressing at least some confidence in the press), and for other variables of your choosing.
The remainder of the exercises for this topic use data from the American National Election Study 2000-2004 Panel Study Subset. Start SPSS, and open “anespanl.sav.” Open the American National Election Study 2000-2004 Panel Study Subset codebook.
2. Did some people recall their 2000 president votes differently in 2002 and 2004 than they had right after the election? Crosstabulate p200000 and p200004.
3. Use Select Cases to analyze only those who in 2000 said that they had voted for Bush. Crosstabulate how, in 2004, they recalled voting in 2000 by whether their feelings about Bush (as measured by feeling thermometers) got warmer, cooler, or stayed the same between the 2000 post-election survey and 2004.
4. The panel study includes feeling thermometers for George W. Bush and Ralph Nader in each election year. Compute new variables measuring changes in how respondents felt about each of these people over time. Select independent variables that you think might explain these changes. For example, did Democrats (more than independents and Republicans) become less warm in their feelings toward Ralph Nader as a result of the election? Compare means to test these hypotheses.
5. We might hypothesize that the traumatic events of September 11, 2001 changed the way we regard other individuals and the government. Examine changes over time in the several "trust" measures described in the codebook. Were changes uniform across the sample, or did some groups change more or in different ways than others?
For Further Study
Nelson, Elizabeth N., and Edward E. Nelson, “Research Design and Methods of Analysis for Change Over Time,” California Opinions on Women's Issues — 1985-1995 http://www.ssric.org/trd/modules/cowi/chapter6. August 15, 1998. Accessed April 15, 2013.
Palmquist, Ruth A., “Survey Methods,” https://www.ischool.utexas.edu/~palmquis/courses/survey.html. Accessed
“Longitudinal Research in the Social Sciences,” Social Research Update. http://sru.soc.surrey.ac.uk/SRU28.html. Spring, 2000. Accessed
sayings have been attributed to various others. See Mark T. Shirey, “Unquote,” http://www.geocities.ws/unmark/unquote.html.
 Note: in the analyses that follow, and in the dataset (gsscums.sav) provided for the exercises at the end of the Topic, a cohort has been excluded in any year for which it was represented by fewer than 100 respondents.
April 30, 2013 .
© 2003---2013 John L. Korey. Licensed under a Creative Commons Attribution-Noncommercial-Share Alike 3.0 Unported License.