Designs for Studying Development

Once the general design is chosen, most gerontolo­gists must decide how to measure possible changes or age differences that emerge as people develop. For
example, if we want to know how people continue (or fail) to use imagery or lists in remembering gro­cery items as they get older, we will want to use a design that is particularly sensitive to developmen­tal differences. Such designs are based on three key variables: age, cohort, and time of measurement. Once we have considered these, we will examine the specific designs for studying development.

Age, Cohort, and Time of Measurement. Every study of adult development and aging is built on the combi­nation of three building blocks: age, cohort, and time of measurement (Cavanaugh & Whitbourne, 2003).

Age effects reflect differences caused by underly­ing processes, such as biological, psychological, or sociocultural changes. Although usually represented in research by chronological age, age effects are inherent changes within the person and are not caused by the passage of time per se.

Cohort effects are differences caused by experi­ences and circumstances unique to the generation to which one belongs. In general, cohort effects cor­respond to the normative history-graded influences discussed earlier. However, defining a cohort may not be easy. Cohorts can be specific, as in all people born in one particular year, or general, such as the baby-boom cohort. As described earlier, each gen­eration is exposed to different sets of historical and personal events (such as World War II, home com­puters, or opportunities to attend college). Later in this section we consider evidence of how profound cohort effects can be.

Time-of-measurement effects reflect differences stemming from sociocultural, environmental, his­torical, or other events at the time the data are obtained from the participants. For example, data about wage increases given in a particular year may be influenced by the economic conditions of that year. If the economy is in a serious recession, pay increases probably would be small. In contrast, if the economy is booming, pay increases could be large. Clearly, whether a study is conducted during a recession or a boom affects what is learned about pay changes. In short, the point in time in which a researcher decides to do research could lead him or her to different conclusions about the phenomenon being studied.

23

Table 1.1

Three Basic Building Blocks of
Developmental Research

Time of Measurement Cohort 2000 2010 2020 2030 1950 50 60 70 80 1960 40 50 60 70 1970 30 40 50 60 1980 20 30 40 50 Cohort is represented by the years in the first column, time of measurement by the years across the top, and age by the values in the cells.

The three building-block variables (age, cohort, and time of measurement) can be represented in a single chart, such as the one shown in Table 1.1. Cohort is represented by the years in the first col­umn, time of measurement is represented by the years across the top, and age is represented by the numbers in the individual cells. Note that age is computed by subtracting the cohort year from the time of measurement.

In conducting adult development and aging research, investigators have attempted to identify and separate the three effects. This has not been easy, because all three influences are interrelated. If one is interested in studying 40-year-olds, one must necessarily select the cohort that was born 40 years ago. In this case age and cohort are confounded, because one cannot know whether the behaviors observed occur because the participants are 40 years old or because of the specific life experiences they have had as a result of being born in a par­ticular historical period. In general, confounding is any situation in which one cannot determine which of two or more effects is responsible for the behaviors being observed. Confounding of the three effects we are considering here is the most serious problem in adult development and aging research.

What distinguishes developmental researchers from their colleagues in other areas of psychology is a fundamental interest in understanding how people change. Developmental researchers must look at the ways in which people differ across time. Doing so

23 CHAPTER 1
necessarily requires that they understand the distinc­tion between age change and age difference. An age change occurs in an individual’s behavior over time. Leah’s or Sarah’s memory at age 75 may not be as good as it was at age 40. To discover an age change one must examine the same person (in this case, Leah or Sarah) at more than one point in time. An age differ­ence is obtained when at least two different people of different ages are compared. Leah and Sarah may not remember as many grocery items as another person of age 40. Even though we may be able to document substantial age differences, we cannot assume they imply an age change. We do not know whether Leah or Sarah has changed since she was 40, and of course we do not know whether the 40-year-old will be any different at age 75. In some cases age differences reflect age changes, and in some cases they do not.

If what we really want to understand in devel­opmental research is age change (what happens as people grow older), we should design our research with this goal in mind. Moreover, different research questions necessitate different research designs. We next consider the most common ways in which researchers gather data about age differences and age changes: cross-sectional, longitudinal, time lag, and sequential designs.

Cross-Sectional Designs. In a cross-sectional study, developmental differences are identified by testing people of different ages at the same time. Any single column in Table 1.2 represents a cross-sectional design. Cross-sectional designs allow researchers to examine age differences but not age change.

Cross-sectional research has several weaknesses. Because people are tested at only one point in their development, we learn nothing about the continu­ity of development. Consequently, we cannot tell whether someone who remembers grocery items well at age 50 (in 2000) is still able to do so at age 80 (in 2030), because the person would be tested at age 50 or 80, but not both. Cross-sectional stud­ies also are affected by cohort effects, meaning that differences between age groups (cohorts) may result as easily from environmental events as from developmental processes. Why? Cross-sectional stud­ies assume that when the older participants were younger, they resembled the people in the younger

age groups in the study. This isn’t always true, of course, which makes it difficult to know why age differences are found in a cross-sectional study. In short, age and cohort effects are confounded in cross-sectional research.

Despite the confounding of age and cohort and the limitation of being able to identify only age differences, cross-sectional designs dominate the research literature in gerontology. Why? The reason is a pragmatic one: Because all the measurements are obtained at one time, cross-sectional research can be conducted more quickly and inexpensively than research using other designs. In addition, one particular variation of cross-sectional designs is used the most: the extreme age groups design.

Suppose you want to investigate whether people’s ability to remember items at the grocery store dif­fers with age. Your first impulse may be to gather a group of younger adults and compare their per­formance with that of a group of older adults. Typically, such studies compare samples obtained in convenient ways; younger adults usually are college students, and older adults often are volunteers from senior centers or church groups.

Although the extreme age groups design is very common (most of the studies cited in this book used this design), it has several problems (Hertzog & Dixon, 1996). Three concerns are key. First, the samples are not representative, so we must be very careful not to read too much into the results; find­ings from studies on extreme age groups may not generalize to people other than ones like those who
participated. Second, age should be treated as a continuous variable, not as a category (“young” and “old”). Viewing age as a continuous variable allows researchers to gain a better understanding of how age relates to any observed age differences. Finally, extreme age group designs assume the measures used mean the same thing across both age groups. Measures may tap somewhat different constructs, so the reliability and validity of each measure should be checked in each age group.

Despite the problems with cross-sectional designs in general and with extreme age groups designs in particular, they can provide useful information if used carefully. Most importantly, they can point out issues that may provide fruitful avenues for subse­quent longitudinal or sequential studies, in which case we can uncover information about age changes.

Longitudinal Designs. In a longitudinal study, the

same individuals are observed or tested repeatedly at different points in their lives. As the name implies, a longitudinal study involves a lengthwise account of development and is the most direct way to watch growth occur. A longitudinal design is represented by any horizontal row in Table 1.3. A major advan­tage of longitudinal designs is that age changes are identified because we are studying the same people over time.

Usually the repeated testing of longitudinal studies extends over years, but not always. In a microgenetic study, a special type of longitudinal design, participants are tested repeatedly over a

Table 1.3 Longitudinal Design Time of Measurement Cohort 2000 2010 2020 2030 1950 50 60 70 80 1960 40 50 60 70 1970 30 40 50 60 1980 20 30 40 50 Cohort is represented by the years in the first column, time of measurement by the years across the top, and age by the values in the cells.

span of days or weeks, typically with the aim of observing change directly as it occurs. For example, researchers might test children every week, start­ing when they are 12 months old and continuing until 18 months. Microgenetic studies are par­ticularly useful when investigators have hypoth­eses about a specific period when developmental change should occur (Flynn, Pine, & Lewis, 2006). In this case, researchers arrange to test individuals frequently before, during, and after this period, hoping to see change as it happens (e. g., Opfer & Siegler, 2004).

Microgenetic studies are particularly useful in tracking change as a result of intervention. For example, older adults could be given a series of measures of memory ability and interviewed about their use of memory strategies. A series of training sessions about how to improve memory could be introduced including additional memory tests and interviews, followed by a posttest to find out how well the participants learned the skills they were trained. The microgenetic method would look in detail at the performance of those who learned and improved after training compared to those who did not and search for differences in either the pattern of performance in the memory tests or in the details in the interviews for reasons why some people improved whereas others did not. This would pro­vide a vivid portrait of change over the period of the intervention.

If age changes are found in longitudinal studies, can we say why they occurred? Because only one cohort is studied, cohort effects are eliminated as an explanation of change. However, the other two potential explanations, age and time of measure­ment, are confounded. For example, suppose we wanted to follow the 1980 cohort over time. If we wanted to test these individuals when they were 20 years old, we would have to do so in 2000. Consequently, any changes we identify could result from changes in underlying processes or factors related to the time we choose to conduct our mea­surement. For instance, if we conducted a longitu­dinal study of salary growth, the amount of salary change in any comparison could stem from real
change in the skills and worth of the person to the company or from the economic conditions of the times. In a longitudinal study we cannot tell which of these factors is more important.

Longitudinal studies have three additional poten­tial problems. First, if the research measure requires some type of performance by the participants, we may have the problem of practice effects. Practice effects result from the fact that performance may improve over time simply because people are tested over and over again with the same measures. Second, we may have a problem with participant dropout because it is difficult to keep a group of research participants intact over the course of a longitudinal study. Participants may move, lose interest, or die. Participant dropout can result in two different out­comes. We can end up with positive selective sur­vival if the participants at the end of the study tend to be the ones who were initially higher on some variable (e. g. the surviving participants are the ones who were the most healthy at the beginning of the study). In contrast, we could have negative selec­tive survival if the participants at the conclusion of the study were initially lower on an important vari­able (e. g. the surviving participants may have been those who initially less healthy).

The third problem with longitudinal designs is that our ability to apply the results to other groups is limited. The difficulty is that only one cohort is fol­lowed. Whether the pattern of results that is observed in one cohort can be generalized to another cohort is questionable. Thus researchers using longitudinal designs run the risk of uncovering a developmental process that is unique to that cohort.

Because longitudinal designs necessarily take more time and usually are expensive, they have not been used very often. However, researchers now recognize that we badly need to follow individuals over time to further our understanding of the aging process. Thus, longitudinal studies are becoming more common.

Sequential Designs. Thus far, we have considered two developmental designs, each of which has problems involving the confounding of two effects. These

Table 1.4 Sequential Design Time of Measurement Cohort 2000 2010 2020 2030 1950 50 60 70 80 1960 40 50 60 70 1970 30 40 50 60 1980 20 30 40 50 Cohort is represented by the years in the first column, time of measurement by the years across the top, and age by the values in the cells.

effects are age and cohort in cross-sectional designs, and age and time of measurement in longitudinal designs. These confounds create difficulties in inter­preting behavioral differences between and within individuals, as illustrated in the How Do We Know? feature. Some of these interpretive dilemmas can be alleviated by using more complex designs called sequential designs, which are shown in Table 1.4. Keep in mind, though, that sequential designs do not cure the confounding problems in the three basic designs.

Sequential designs represent different combina­tions of cross-sectional or longitudinal studies. In the table, a cross-sequential design consists of two or more cross-sectional studies conducted at two or more times of measurement. These multiple cross-sectional designs include the same age ranges; however, the participants are different in each wave of testing. For example, we might compare perfor­mances on intelligence tests for people between ages 20 and 50 in 1980 and then repeat the study in 1990 with a different group of people aged 30 to 60.

Table 1.4 also depicts the longitudinal sequential design. A longitudinal sequential design consists of two or more longitudinal designs that represent two or more cohorts. Each longitudinal design in the sequence begins with the same age range and follows people for the same length of time. For example, we may want to begin a longitudinal study of intellec­tual development with a group of 50-year-olds in 1980, using the 1930 cohort. We would then follow this cohort for a period of years. In 1990 we would begin a second longitudinal study on 50-year-olds, using the 1940 cohort, and follow them for the same length of time as we follow the first cohort. This design helps clarify whether the longitudinal effects found in a single longitudinal study are cohort — specific or are more general findings.

Although sequential designs are powerful and pro­vide by far the richest source of information about developmental issues, few researchers use them, because they are costly. Trying to follow many people over long periods of time, generating new samples, and conducting complex data analyses are expensive and time consuming. Clearly, this type of commitment to one project is not possible for most researchers.