Badri and Burchinal (1985) were fortunate in finding data over a period of 27 years for enrollment of boys and girls at the elementary, intermediate, and secondary levels for most of northern Sudan. Enrollment data were available for 1956, 1967, 1977, and 1982/83 from the Ministry of Education. They constructed ratios showing the proportion of girls enrolled relative to boys at each level of schooling for each time period. A low ratio indicated that few females were enrolled relative to males; a higher number indicated female enrollment was moving closer to that for males. A ratio of 1.00 would indicate that the same percentages of girls as boys were enrolled at each level. For simplicity, ratios are presented only for the secondary level. In 1956, the first year for which data were available, the ratio was 0.18; it rose to 0.23 in 1967; further to 0.47 in 1977; and still further to 0.61 in 1982/83. With four points spread out over 27 years, Badri and Burchinal could confidently conclude that enrollment for females relative to males had increased substantially over the 27 years for which data were available. A clear linear or straight-line trend was established.

But what could have been concluded if data were available for only 1956 and 1982/83? With just these data, the authors could have concluded that a significant rise in female/male enrollment ratios had occurred over the 27 year period. But with only two points, the question of year-to-year changes could not have been addressed.   Enrollment ratios could have declined over some years within the 27-year span   To avoid these kinds of problems, researchers try to get data for as many points in time as possible. This strengthens the basis of analysis and gives greater support to any conclusions that are drawn.

Longitudinal designs

Social scientists also study changes in variables as they occur. These investigations, called longitudinal studies, are based on data collection at a given point in time, followed by successive additional waves of data collection at future times. The right side of Figure 5.1 illustrates this design. Time intervals between data collection may vary from weeks, to months or even years, depending on the purpose of the study and the money or staff available for data collection.

In Figure 5.1, illustration D represents the general pattern for a longitudinal investigation. The investigator starts with data collected at some current or present time (T₁) and then collects additional data at successive times in the future, represented by T₂) through T_n).   Three widely used longitudinal designs are described next.

Panel designs. A powerful way of conducting a longitudinal study is to collect data at successive times from the same sample. The time periods involved can range from weeks to many years. Kadri and others (2000) studied the irritability of Muslim men during Ramadan. Their study extended over six weeks, starting with data collection a week before Ramadan, followed with repeated data collection each week during Ramadan, and ending with final data collection one week after Ramadan. King and Byerlee (1977) wanted to know how the spending patterns of rural families in Sierra Leone varied throughout the year. To find out, they arranged for a sample of farmers to be interviewed twice a week for a year. Some panel studies extend over years. A study of the effects of malnutrition on children in a Mexican village, for example, extended over 24 years (Chavez, Martinez and Soberanes, 2000).

Panel designs are particularly useful for detecting changes in variables, but they have the disadvantage of the loss of respondents over time. As the initial sample gets smaller, it becomes less like the original sample, and hence comparisons among the waves of data collection can become less meaningful. When the time period is relatively short, over several months for example, losses generally are not too great. Over longer periods of time losses in respondents inevitably occur. Causes include death, illness, moving away, lose interest, or refuse to cooperate after several periods of data collection. Careful researchers report losses in samples used in panel designs and discuss the possible effects of the losses on the findings.

Designs based on independent samples. A weaker form of the longitudinal design is based on comparisons among two or more successive, but new, independent samples from the population being studied. Researchers are forced to use this design when the population being investigated is undergoing rapid change. Investigators collecting data on trends in household composition in camps for displaced persons, for example, face this problem. They know the population changes daily: new arrivals come, some previous residents leave, and some die. A panel approach simply would not work: the loss of respondents, even in a month, would be very large. The alternative is to select a random sample of households at one time and record the composition of the households in the sample and repeat the same survey with a new random sample some time later. The comparison of results from the two samples would show what, if any changes had occurred in the composition of the households.