Lecture
It is usually contrasted with the cross-sectional method, which involves comparing psychological measures at one and the same time in people of different age groups and links changes in the dependent variable to the age index.
The classical longitudinal study means a «continued study» with repeated recording of measures on one and the same person or one group of people.
In general terms, the longitudinal method should be taken to include a group of methods characterized by the presence of several repeated measurements of one or more variables of interest to the researcher, carried out on the same or similar groups of subjects.
Example: L. Terman's longitudinal study of intellectual development on a sample of intellectually gifted children.
In a longitudinal study, hypotheses about development are tested.
In modern designs another goal is also achieved – tracing the time-delayed effects of experimental interventions – which makes it possible to test causal hypotheses.
The key concept is the cohort. A group of people who experienced similar events during a specified period of time – a commonality within the sample defined by the criterion of year of birth.
Cohort = period of measurement (calendar year) – age (number of years of the subjects since birth).
This formula demonstrates the linear dependence of these indices.
Attempts to overcome this linear dependence:
1) K. Mason. He attempts to solve the problem by statistical means – by creating models that eliminate the statistical collinearity (complete mathematical dependence) among age, cohort, and time period.
2) Approaches that involve a theoretical justification for excluding from consideration the influence of one of the three variables on the developmental trajectories being detected, or for reinterpreting them.
In the ideal case, the effects of cohort and period are replaced by operationalized characteristics that allow precise control of the effects of these variables, which makes it possible to disentangle the effects of age (development), period (historical effects), and cohort.
The goals of conducting longitudinal studies:
- establishing the functional form of intra-individual developmental trajectories/curves;
- assessing inter-individual differences in intra-individual developmental trajectories by constructing causal models;
- increasing the precision of measuring the experimental effect by controlling intra-individual variability;
- testing hypotheses about the direction of causal relationships and assessing their strength.
Two types of designs:
1) Prospective panel designs
- data collection for the relevant time intervals is carried out during those very intervals, i.e. «here and now».
2) Retrospective panel designs
Data are collected with respect to various time periods in the past.
The formulation of hypotheses about intra-individual variability includes statements about the functional form of developmental curves: they may be nonlinear, linear, or discontinuous.
The developmental curves that are detected make it possible to reject preliminary hypotheses and to formulate new ones.
Prospective designs (Menard's classification):
1. The total population design;
Examination of the entire population of interest to the researcher throughout the whole target time period. Problem: changes in the composition of the population, attrition of subjects – this leads to non-identity of the samples across different time intervals.
2. The repeated cross-sectional measurement design;
Measurement of indices on random samples that differ for each time period. This design does not allow a fine-grained analysis of developmental patterns for individual cohorts or the refinement of causal hypotheses. A pity.
3. The revolving panel longitudinal design;
It allows the researcher to conduct multiple measurements on a particular sample of subjects over a certain span of time and then to replace part (or all) of the subjects; this makes it possible to overcome the limitations associated with attrition and to increase reliability. It is fruitful when studying specific groups with strictly defined age constraints.
4. The (true) panel longitudinal design
It involves collecting data for all time periods on the material of multiple cohorts. Its advantage is the possibility of carrying out any type of longitudinal data analysis and the precise quantification and disentangling of cohort and time effects.
Threats to validity. In addition to the usual ones, there are special threats:
1. An insufficient number of time points and excessively long intervals between them.
2. A threat to longitudinal validity, understood as the invariance of the measured constructs throughout the conduct of the longitudinal study.
Example: aggression in an 8-year-old child and in an 18-year-old.
If construct invariance is present, one must also verify the identity of the metric of the scales used.
To what extent did the structure of the property being studied change, and to what extent is this reflected in changes in the structure of the instruments used to operationalize the variables.
3. Attrition of subjects.
The problem of missing data. Do these subjects drop out at random or not at random?
Strategies for maintaining contact with subjects should be used.
Statistical analysis of the longitudinal data obtained.
The choice of analytic strategies depends on the longitudinal design, the hypotheses, and the researcher's level of statistical training. Areas of application of longitudinal studies:
1) Experimental and quasi-experimental research.
Here researchers are interested in short-term changes and differences between groups after an intervention has been carried out. Here the samples are large. For such research, multivariate models of developmental curves are the most informative.
2) Studies of growth and development.
Long-term patterns of change and their systematic connections with various characteristics of the subjects. Here the samples are smaller. In this case, autoregressive time-series analysis is best suited.
1. The classical approach – ANOVA – the general linear model. Goal: identifying between-group differences in the means of indices that change over time. Drawbacks: the impossibility of quantifying intra-individual changes in indices, difficulty in dealing with missing data, and so on.
2. Hierarchical linear modeling (HLM) and structural equation modeling (SEM). They solve the problems of the classical method.
Modern methods of statistically processing longitudinal data make it possible to reveal hidden trends and non-obvious developmental curves that cannot always be represented by ordinary graphs and diagrams.
Comments