If you are following the same people (let's say you start with 1000 people at baseline) over time and you make measurements cross-sectionally (every two years - measure all 1000 people that you start with MINUS those loss to follow up),  it is considered a longitudinal study.  That is, we can use the repeated measurements of the same people to analyze.   If you are measuring different people at multiple timed points, you do not have the longitudinal data.

e.g. In the US, we have National health and nutrition examination survey (NHANES) which collect data from a representative sample of US population periodically.  Each timed period, they are selecting different sample, thus, not longitudinal data.  We only have cross-sectional data of different samples at each timed period.

Dear Edbert Chia,

I agree with you on the idea that longitudinal studies and cross-sectional  studies are two completely different things in terms of goals and procedure. The former is interested in getting access to intrapersonal change over time whereas the latter is interested in getting access to interindividual differences at a given point of their development. 

Even though both longitudinal and cross-sectional  studies are often used in developmental psychololy, only the former is consistent with the very nature of developmental psychology, to get access to intraindividual change over time.

 Let us suppose that you have a sample of, say, 1000 people who were born in the same year, 2000, for example, and that you are interested in seeing if their cognitive development changes over time. In order to attain this goal, you have to evaluate your sample' cognitive abilities, for example, at T1 (2018), T2 (2020), T3 (2022), and so forth. So, in a longitudinal design, individuals of a certain age, 10-year-olds, for instance, at T1, are evaluted when their are 12-year-olds (T2), 14-year-olds (T3), and so forth. Thus, you have repeated evalutions ( at T1, T2, T3, and so forth) of the same individuals with the same age. Needless to say, the more a longitudinal study is long, the more likely you are to lose some individuals of the initial sample.

In contradistinction, in a cross-sectional design, you have individuals of differrent age  (e.g.,10, 12, 14- year olds) who are evaluated at the same time (at T1, for example). In a nutshell, in longitudinal studies, individuals who belong to a certain cohort or have the same age are assessed at several points in their development (at T1, T2, T3, and so forth). In cross-sectional studies, individuals of different age (e.g., 10, 12, 14- years olds) are assessed only once (at T1, for example) .Of course, we can also have repeated cross-sectional sudies. When this is the cause, cross-sectional sudies are converted into cross-sectional sequential designs.

I hope I has got your question and that this helps,

Best regards,

Orlando

The simplest longitudinal descriptive study consists of two repeated cross-sectional studies on the same population or samples, looking for the same measurements.

Yes - repeated cross-sectional analysis can be longitudinal as you are repeatedly measuring something. Two examples eg using the European Social Survey to measure country change  with countries at level 3, wave or year at level 2,  and respondents at level 1 (and they are not the same people being repeatedly measured). EG 2: examining changing school performance with schools at level 3, graduating cohort at level 2 and pupils at level 1. The cohorts and there many be many of them are repeated measures of schools. 

There is an extended example of the latter in the last chapter of this

https://www.researchgate.net/publication/260771330_Developing_multilevel_models_for_analysing_contextuality_heterogeneity_and_change_using_MLwiN_3_Volume_1_updated_March_2017_Volume_2_is_also_on_RGate

Different sorts of structures are discussed here

http://www.bristol.ac.uk/cmm/software/mlwin/mlwin-resources.html#mlmbkgrd

and here is a paper on their analyses

https://www.researchgate.net/publication/259638850_Two_Multilevel_Modeling_Techniques_for_Analyzing_Comparative_Longitudinal_Survey_Datasets

Article Two Multilevel Modeling Techniques for Analyzing Comparative...

Book Developing multilevel models for analysing contextuality, he...

@Edbert Chia and Kelvin Jones and others

I entirely agree with Kelvin Jones and other answerers to this question: yes, you can use repeated cross sectional data to undertake a longitudinal study. For me this problem is more general and leads to interval-censored event history analysis (Courgeau and Najim, 1996). For example we used the French INSEE Demographic Panel Survey (Courgeau, Lelièvre and Wolber, 1998) which contains all the information in the individual census schedules since 1968 coupled with the corresponding registration data to estimate residential mobility. We also used the survey on social, geographical and wealth in 19th and 20th centuries in France (Courgeau, 1993) to estimate individual migration history, as the place of residence is only known at the time of vital events.

However in order to use such data some conditions are important to verify. We undertook a test of the hypotheses on the complete retrospective data from the 3B survey conducted in 1981. It is then possible to transform a complete data set into an interval-censored data set, and to test a number of methods which allow for the fragmented nature of the data (Courgeau and Najim, 1996). We give here the main results of such a confrontation.

The first condition concerns the density in time of the events permitting the location of the individual in geographical or occupational space. The higher is this density, the more chance that events that are close to others in time will not be omitted.

The second condition concerns the dependence that may exist between the events defining the individual’s spatial location and his or her geographical or occupational mobility. This dependence modifies the timing of the phenomenon studied, but has little effect on its intensity.

The interested reader may find here enclosed the more detailed results of this test, and will find in my ResearchGate website more papers on this topic. There is still a lot of work in hand to correct these errors.

References

Courgeau D. (1993). An attempt to analyse individual migration histories from data on place of usual residence at the time of certain vital events. France during the nineteenth century. In Methods in Historical Demography, Reher, Schofield eds, Oxford: Clarendon Press, 206-222.

Courgeau D., Lelièvre E., Wolber O. (1998). Reconstruire des trajectoires de mobilité résidentielle. Eléments d’une analyse biographique de l’EDP. Economie et Statistique, 316-317, 163-173.

Courgeau D., Najim J. (1996). Interval-censored event history analysis. Population-E, 8, 191-208.

Yes, it can be, in certains circunstances. Not allways, but it is frequently possible.

@Manuel Lourenço Nunes

Yes it is possible that a repeated cross sectional survey give similar results than an event history survey. However the delay of time between these repetitions may be short to give such results. Unfortunately many of these surveys have a longer delay than one or two years. For example the INSEE French panel survey uses data from different censuses: 1954, 1962, 1968, 1975, 1982, 1990, and this delay vary from 6 to 8 years: the results we obtained in 1998 (see previous post) were not really satisfactory.

I strongly agree with Wayway. Repeated measure on regular interval on the same population.