Longitudinal Research

Longitudinal research may best be defined by contrasting it with cross-sectional research. CROSS-SECTIONAL research refers to research in which data are collected for a set of cases (individuals or aggregates such as cities or countries) on a set of variables (e.g., frequency of illegal behavior, attitudes toward globalization) and in which data collection occurs specifically (a) at a single time and (b) for a single time point or a single interval of time (hereafter, both will be referred to as periods). Analysis of purely cross-sectional data can examine differences between cases but not changes within cases. Different disciplines define longitudinal research differently, but a broad definition would include research in which (a) data are collected for more than one time period, (b) possibly but not necessarily involving collection of data at different time periods, and (c) permitting analysis that, at a minimum, involves the measurement and analysis of change over time within the same cases (individual or aggregate). Longitudinal data have been collected at the national level for more than 300 years, since 1665, and at the individual level since 1759, but longitudinal research in the social sciences has really flourished since the 1970s and 1980s (Menard, 2002).

Although some researchers would consider the collection of data at different time periods to be a defining characteristic of longitudinal research, it is also possible in principle to collect retrospective data, in which the data are collected at a single period but for more than one period. For example, a respondent may be asked to report on all of the times he or she has been a victim of crime, or has committed a crime, over a span of years up to and including her or his lifetime. In contrast to retrospective studies, prospective longitudinal data collection involves the repeated collection of data with usually short recall periods: crimes committed or victimizations experienced in the past week, month, or year or data on attitudes, beliefs, and sociodemographic characteristics at the instant the data are collected.

Longitudinal research serves two primary purposes: to describe patterns of change and to establish the direction (positive or negative, from Y to X or from X to Y) and magnitude (a relationship of magnitude zero indicating the absence of a relationship) of causal relationships. Change is typically measured with respect to one of two continua: chronological time (for historical change) or age (for developmental change). Sometimes, it is difficult to disentangle the two. If older individuals are less criminal than younger individuals, is this because crime declines with age, or is it possible that older individuals were always less criminal (even when they were younger) and younger individuals will remain more criminal (even as they get older), or some combination of the two? With cross-sectional data, it is not possible to disentangle the effects of history and development. Even with longitudinal data, it may be difficult, but with data on the same individuals both at different ages and different time periods, it at least becomes possible.

There are several different types of longitudinal designs (Menard, 2002), as illustrated in Figure 1. In each part of Figure 1, the columns represent years, and the rows represent subjects, grouped by time of entry into the study. Thus, subjects enter the population or sample (“rows”) at different times (“columns”), and subjects who have entered the study at different times may be in the study at the same time (more than one row outlined in the same column), except in the repeated cross-sectional design, in which different subjects are studied at each different time (no two rows outlined in the same column).

...