Methodological issues have been the source of much debate and discussion within the field of developmental psychology. Research strategies which have been commonly used have been demonstrated to be flawed with respect to internal validity. This is in part due to the study of age as a factor necessitating the use of quasi-experimental designs. The two most commonly used designs include the longitudinal and cross-sectional designs. Results of these studies are confounded, however, by influences occurring within the individuals studied and the environment. The major outcome of these studies include normative age-graded influences, normative history-graded influences, nonnormative influences, and the so-called terminal decline. These influences will be discussed first followed by a critique of the longitudinal and cross-sectional methodologies. This critique will emphasize how they relate to the influences stated above. In addition, proposed methodological solutions will be discussed.
Sources of Influence on Human Development
Baltes, Reese, and Lipsitt (1980) have identified three major influences that impact on the dynamic interaction of the individual and the context. They include normative age-graded influences, normative history-graded influences, and nonnormative life event influences. The relationship between these influences is proposed to be dynamic and reciprocal. In other words, these influences are in a continuous state of change and they influence one another respectively. It should be noted that this interrelationship between the three is proposed to be different during different phases of the life cycle. For example, nonnormative life events are proposed to be particularly significant during middle and late adulthood (Baltes et al., 1980). Each of these influences will be discussed in greater detail below. See Figure One for a representation of these influences.
Normative age-graded influences are those influences within the life course that are correlated with chronological age. For example, marriage and retirement are two normative age-graded influences. These influences are the result of either biological or environmental determinants or an interaction of the two. Puberty and menopause would be examples of biological determinants; graduation and retirement would be examples of environmental determinants. Most environmental determinants fall into one of three categories: family life cycle, education, and occupational. Age-related events are considered normative if they occur with great frequency and are similar with respect to duration and timing for the majority of the population within a culture. It should be noted that each culture or subculture has its own set of age-graded normative influences. Thus, for a young girl to be pregnant at 15 years of age would be nonnormative in much of the American culture but normative for other cultures.
Normative history-graded influences are those influences within the life course that are correlated with historical time and are experienced by the majority of a culture. For example,, wars and epidemics are considered history-graded events. They are normative in that they are experienced by the majority of the population during a given time. In addition, the majority of a cohort (individuals all born the same year) experience the event in similar ways. They may however affect cohorts differentially. For example, the 1950 cohort experienced and still experiences the Vietnam War differently than the 1970 cohort. History-graded normative events are both the result of biological and environmental determinants. For example, an epidemic would be classified as a biological determinant and and economic depression an environmental determinants. Famine resulting from economic depression would represent an example of an interaction between the two determinants. The current AIDs crisis and the computer revolution could be viewed as normative history-graded events.
The impact of the interaction of these three influences on the developing individual is what defines life-span development. It is the impact of this interaction that accounts for the consistency with respect to individual life cycles as well as the increasing heterogeneity of a cohort as people age. In other words, all cohorts will share similar experiences as other members of their cohort (normative age-graded and history-graded influences), thus producing regularity with respect to their development. Conversely, as each member of a cohort continuously experiences different nonnormative life events throughout their life course, interindividual differences increase across the life-span.
The term terminal decline or terminal drop has been defined by Riegel and Riegel (1972) as, "a sudden drop in performance occurring within 5 years prior to death" (p. 306). As a phenomena, terminal decline has been observed by many researchers including Jarvik and Falek (1963), and Lieberman (1965, 1966). Much of the research relating to terminal drop has been in the area of intellectual functioning (Jarvik & Falek, 1963; Kleemeier, 1961, cited in Riegel & Riegel, 1972). Five years prior to death a noticeable decline in intellectual performance has been observed. Cross-sectional studies have attributed this decline in performance to decline with age. However, as there is an increased incidence of mortality with increased age, the overall decrease in intellectual performance as a group may be simply the result of sampling bias. When only survivors are examined, intellectual performance remains unchanged (Riegel & Riegel, 1972). A hypothetical example is provided in Figure Two. Thus, the issue of terminal decline is relevant to methodology as well as an influence on the individual life course.
Traditional Designs
The four influences described above have an impact on the results of research studies examining human development. A research study's results may not accurately portray the way that the individual develops but rather may simply reflect methodological artifact. Three commonly used designs include the cross-sectional design, the longitudinal design, and the time lag design. These designs are described as unifactorial designs, with age as the single factor (Campbell & Stanley, 1963). However, each of these designs are noted for low internal validity. For example, the cross-sectional design is confounded by cohort effects. Each of the designs will be discussed below. Included in this discussion will be an analysis of the internal validity problems as they relate to each design. Proposed solutions will be presented in the next section of this paper. To aid in the understanding of the various designs, Figure Three has been included.
The cross-sectional method has been defined by Baltes (1968) as follows: "Samples (S1 - Sn) of different ages (A1 - An) are observed on the same dependent variable once (O1) at the same time of measurement (T1)" (p. 146). In other words, two or more cohorts are tested at one time to see if differences exist across ages. This design is represented by the first column in Figure Three. Cook and Campbell (1979) argue that this is not a true design but rather separate samples. As such, there are many threats to internal validity. The major threat to internal validity in a separate sample quasi-experimental study is selection. The samples may be different on any number of variables other than the one under investigation. In the cross-sectional study, age differences may be confounded with differences in generations or cohorts. All members of a cohort share similar experiences in relation to normative history-graded influences. Thus, the researcher is not able to differentiate between maturational differences and cohort differences. An example may be useful in clarifying this point.
A researcher might choose to conduct a study examining differences in spending habits across the life-span. The hypothesis might be as follows - as individuals age they become more conservative in their spending habits. The researcher would then randomly select samples from various age cohorts; for example: individuals born in 1910, 1920, 1930, 1940, 1950, 1960, 1970. These groups would then be tested for differences in spending habits. Subsequently, the researcher finds differences in spending habits across age with increasing conservatism correlated with increasing age. The researcher concludes that an age difference has been demonstrated. However, age is confounded with a cohort effect. In particular, the older groups experienced the depression (in different ways) whereas the younger groups did not. This, not age, may account for the differences in spending habits.
As demonstrated above, the cross-sectional design confounds maturation with cohort. Therefore, it can only be used descriptively. Differences in age groups or cohort can be described but the differences can not be definitively explained.
It should be noted that the selective sampling with the cross-sectional method can also be problematic. For example, selective sampling is a problem when examining intellectual performance with age, specifically as it relates to terminal drop. The studies conducted reporting a drop in intelligence with increasing age may be simply the reporting of a selection bias. This bias has been described above and is represented in Figure Two. When evaluating the results of cross-sectional studies, care should be taken to examine the size and representativeness of the selected samples.
The longitudinal method is defined by Baltes (1968) as follows: "One sample (S1 is observed several times (O1 - On) on the same dependent variable at different age levels (A1 - An), and therefore by definition at different times of measurement (T1 - Tn)" (p. 146). In other words, one group of individuals within one cohort is tested at least twice over time. The design is represented by the first row in Figure Three. Cook and Campbell (1979) would define this method as a time-series design. As such, it suffers from many threats to internal validity with history being the most serious threat. History is defined as those events that occur between time of testing. In the longitudinal method, age differences or differences in maturation are confounded with history effects. What occurs in the environment represents an experimental treatment. In other words, normative history-graded influences are confounded with age differences. An example is provided below.
Let us presume that a researcher had decided to study spending habits across the life-span and this research was begun shortly after the turn of the century. A group of individuals was initially studied at 20 years of age in 1910. A follow-up test was then conducted every ten years for the next 50 years. Once again, increased conservatism concerning spending was found to be correlated with increased age. However, age is confounded with a normative history-graded event. In this example, the event was the depression of the early 1930s. Therefore, the depression acted as a treatment effect.
As demonstrated above, the longitudinal method confounds history and maturation. Therefore, as a methodology it can also only be used descriptively.
There are also several threats to selection with the longitudinal method. First, the longitudinal method rarely meets the criteria of selective sampling (Baltes, 1968). For example, individuals who volunteer to participate in a longitudinal study are usually of higher intelligence and socioeconomic status (Baltes, 1968). Second, longitudinal studies suffer from selective survival. Individuals who survive (or at least don't drop out of the study) may be qualitatively different than those who do not (Jarvik & Falek, 1963). This selective survival, however, is a characteristic of the population under study. Third, longitudinal studies also suffer from selective drop-out/experimental mortality (Campbell & Stanley, 1963). It is theorized, in the longitudinal method, that the same group of individuals will be repeatedly tested. Thus, leading to a homogeneity of groups across testing time. However, as subjects drop out or die, the groups, in fact, become heterogeneous. Subject attrition due to drop-out is, however, not a characteristic of the population under study. Thus, the longitudinal method suffers from many selection biases.
Testing effects are also a problem with the longitudinal method. This is particularly evident in studies where subjects have been retested many times. For example, the Berkeley Growth Study tested the majority of subjects approximately 38 times over a period of 18 years (Bayley, 1948, cited in Baltes, 1968).
It should be clear from the description above that the longitudinal method suffers from many threats to internal validity. It should also be noted that the longitudinal method is very time-consuming and expensive to conduct.
The time lag design is used less often in developmental research so it will only be briefly discussed in this paper. It is of primary interest to the social psychologist. The time lag design has been defined by Schaie (1965) as examining "whether there are differences in a give characteristic for samples of equal age but drawn from different cohorts measured at different times" (p. 95). In other words, only one age is studied but across different cohorts at different times. The time lag design is represented by a diagonal in Figure Three. The time lag design could also be defined by Cook and Campbell (1979), as a separate sample design. As such, it also confounded by differences in generations or cohorts. According to Schaie (1970), the time lag method is designed to measure cultural change but confounds environmental treatments or normative history-graded influences with differences between cohorts.
The three designs described above represent the three conventional strategies used to study age differences. As all suffer from major threats to internal validity, alternative strategies have been proposed.
Alternative Design Strategies
The alternative design strategies can be divided into three categories: the longitudinal/cross-sectional bifactorial strategy proposed by Baltes (1968), the sequential strategies proposed by Schaie (1965), and the multivariate procedures (Bock, 1979; Nesselroade, 1970). The appropriateness and usability of each method has been widely debated within the field of developmental psychology.
The cross-sectional method and the longitudinal method are unifactorial methods with age the only factor. Baltes (1968) proposes a bifactorial method with age and cohort as the two factors. Specifically, this method calls for the joint use of cross-sectional and longitudinal strategies for the study of age differences. It is proposed that through this method, a quantification and direct assessment of interindividual differences (between cohorts) and intraindividual differences (across age) in age-related change can be examined. In other words, this model represents a complete matrix for the study of age-change (See Figure Four). Therefore, when the goal of the investigator is the description identification of age-changes, the Baltes bifactorial model is most appropriate (Schaie & Baltes, 1975).
Schaie (1965) proposes a General Developmental Model of sequential designs for the analysis of age-related changes. In addition to being a descriptive model, it is also, according to Schaie (1965) a model of theory building. As such, it is functionally different the the Baltes (1968) model. The model is not only used to describe age-changes but to develop explanations of developmental change (Schaie, 1965). This is in part due to the model being trifactorial, with age, cohort, and time of measurement as the factors. Baltes (1968), however, disagrees with the usability of a trifactorial model. He states that it is redundant, immeasurable, and does not function as an explanatory model.
The three designs proposed by Schaie (1965) include the cohort-sequential, time-sequential, and cross-sequential designs. See Figure Three for a representation of these three designs. As stated, these designs are best used when an explanation of age-related changes is being investigated; they are not primarily descriptive. The purpose of these designs is to separate out the variance that is accounted for by normative history-graded influences as well as cohort effects. Baltes (1968), however, as stated previously, disagrees with the usage of a trifactorial model.
The cohort-sequential method (also called the longitudinal-sequential design (Baltes, Cornelius, & Nesselroade, 1979)) is designed to measure all cohorts at all ages. In other words, this method consists of longitudinal sequences for two or more cohorts. It should be used when a researcher's primary interest is to make generalizations concerning cohort differences. For example, the 1920, 1930, 1940, 1950 cohorts could be examined longitudinally and sequentially to see if a spending habits differences is consistent across cohorts over time. In other words, the cohort-sequential design could be used to examine whether a spending conservatism difference is consistent between the various cohorts. At the same time, inferences can be made about age changes across the life-span. This method would thus aid in differentiating age differences from cohort differences. However, normative history-graded influences still serve as a confound with this design.
The time-sequential method (also called the cross-sectional-sequential design (Baltes et al., 1979)) is designed to measure samples of all ages at all times of measurement. In other words, it consists of cross-sectional sequences at two or more times of testing. For example, the 1920, 1930, 1940, and 1950 cohorts could be examined cross-sectionally in 1970 and once again in 1980. The purpose of this design is to examine the effect of normative history-graded effects on various age groups. Any normative history-graded cultural shifts between 1970 and 1980 would be consistent for all groups. Thus, the variance due to environmental or cultural shifts could be separated from age changes across the life-span. However, cohort effects are still a confound with this design.
When both cohort effects and normative history-graded effects are thought to play a role, it is suggested that a cross-sequential design be implemented (Schaie, 1965). This can be represented in Figure 3 by any rectangular area. The cross-sequential method is designed to measure all cohorts at all times of measurement. thus, one can examine not only age changes but also cohort effects and normative history-graded influences. It is assumed that these effects are additive and thus the amount of variance due to each can be divided out. However, if interactions occur these effects can not be partitioned out. Thus, the assumptions underlying the cross-sequential method may be false.
In addition to the above design, it has been suggested that multivariate procedures be applied to the study of human development (Bock, 1979; Nesselroade, 1970). The traditional designs, such as those described above, examine only one dependent variable. As such, their external validity is low. In other words, a set of variables serves better to define a psychological construct than a single variable. In addition, a multivariate procedure enables interrelationships between variable and constructs to be examined. However, as a wide range of multivariate techniques can be used in the study of development, they will not all be discussed here. Briefly stated, those with the greatest applicability include factor analysis, principle components analysis, multifactor analysis of variance (MANOVA), and multivariate analysis of covariance (MANCOVA). The later two methodologies can be used on all the designs described above as long as more than one dependent variable is being studied and the basic assumptions underlying each test is met. Therefore, this methodology is extremely functional in relation to the study of human development.
In summary, three influences have been identified that effect human development. They are normative age-graded influences, normative history-graded influences, and nonnormative life event influences. These three influences impact on methodological strategies within developmental psychology. Conventional designs however are inadequate in dealing with these influences. The are unable to separate out the variability due to history-graded influences, cohort differences, or age-change. Three strategies have been suggested: a bifactorial model (Baltes, 1968), a trifactorial model (Schaie, 1965), and multivariate procedures (Nesselroade, 1970). It should be noted that the later can be used in conjunction with the bifactorial or trifactorial models. Decisions concerning which procedure or design to use should be based upon what psychological construct or variable is under study (Schaie & Baltes, 1975). Additionally, the use of multiple dependent variables would further increase the internal and external validity of a study. The use of these multiple variables would thus necessitate the use of a multivariate data analysis.
1998 copyright Linda M. Woolf
To the next section of the paper.A discussion of cognitive changes associated with old age within the framework of the first three sections.
Back to the Introduction.