For example, one should be cautious while trying to indicate causal mechanisms based on the data collected at one point in time... but what else?
Thank you very much
beside it doesn't help to determine cause and effect relationship, it is also susceptible to bias due to low response and misclassification due to recall bias.
In addition to what had been mentioned, cross-sectional research should be treated carefully when it is referenced to the time been taken, age or other characteristics of participants, and criteria reference. For example, in Large Scale Students Assessments such as PISA study, as the research doesn't consider repeating the measure (i.e., students' performance), therefore when comparing the outcome with other contextual factors we should use sophisticated statistical models to avoid bias caused by the nature of the study (and its sample). We need to consider to control for contextual factors while comparing for others (assuming that we can find out value-added factors). Most of what we can say is association between contextual factors and the outcome in a context and conditions as mentioned above (e.g., age of student, criteria of the test or the curriculum reference, etc.). Moreover, if the cross-sectional study is pure quantitative, from my point of view, follow-up qualitative research should follow to avoid the limitations and to assure the quality of the data that been taken at a point and analyzed (for example, mixed-methods can be considered as the best here).
You would need to be more interested in a total or mean for a finite population, than in following individual members of that population. (It is a strength for cross-sectional surveys that the members and size of the population are allowed to change.)
Witold -
I worked at a statistical agency which is responsible for providing official energy statistics on a periodic basis - for a number of energy establishment surveys - typically collecting sample data monthly, and census data annually. These cross-sectional surveys tracked markets, but did not necessarily follow individual respondents, though that could be done in some cases. Panel data may he used for both purposes, but we were only giving periodic "snapshots" of energy market related data.
In one set of surveys, mixing the two purposes at once (a cross-sectional view, with some nonresponse imputation looking at individual respondent time series data) was done, which unnecessarily complicated processing and interpretation of results, especially understanding of variance. But it is difficult to stop a bad habit that had been embedded in that system, so I eventually wrote the paper at the following URL:
https://www.researchgate.net/publication/275365705_When_Prediction_is_Not_Time_Series_Forecasting
That was for a set of establishment surveys using model-based sampling and estimation. Many cross-sectional surveys use model-assisted design-based methods, or just design-based methods, and that means that for imputation to be consistent with estimation methodology, there are even further considerations, as you may know.
Perhaps you are more interested in panel surveys, and have expertise with that, but I have limited experience there.
Cheers - Jim
Research When Prediction is Not Time Series Forecasting
In a cross-sectional study, data are collected on the whole study population at a single point in time to examine the relationship between disease (or other health related state) and other variables of interest.
Cross-sectional studies therefore provide a snapshot of the frequency of a disease or other health related characteristics in a population at a given point in time. This methodology can be used to assess the burden of disease or health needs of a population, for example, and is therefore particularly useful in informing the planning and allocation of health resources.
Generally its weaknesses are
- Difficult to determine whether the outcome followed exposure in time or exposure resulted from the outcome.
- Not suitable for studying rare diseases or diseases with a short duration.
- As cross-sectional studies measure prevalent rather than incident cases, the data will always reflect determinants of survival as well as aetiology.1
- Unable to measure incidence.
- Associations identified may be difficult to interpret.
- Susceptible to bias due to low response and misclassification due to recall bias.
- Badges
- Science topic
- Similar topics
- Psychology