In general, epidemiological studies can have experimental or observational designs. They are distinguished with respect to whether the major factor of interest in the study is under the control of the investigator or not.

In experimental studies, the investigator actually introduces a factor or intervenes in the environment of the study subjects to see what impact the intervention has on the study subjects compared to a group of subjects that didn't have the intervention. In observational studies the investigators do not directly intervene, but instead, develop methods for describing events that occur naturally without direct intervention and effect on study subjects.

There are three major types of observational study designs - cross-sectional, group-comparison, and longitudinal. These designs differ principally with respect to (1) the number of groups included in the study and the criteria for choosing them and (2) the number of points in time and reference period(s) for gathering the data.

The distinguishing features of the three types of observational designs are presented in the table below.

Table 1. Types of study designs (L. A. Aday "Designing and conducting health surveys")

Characteristics
	Groups		Time periods
Types of study design	# of groups	Criteria for the selection of groups	# of periods of data collection	Reference periods for data collection
Observational
Cross-sectional	1	Population of interest	1	Present (and recall of past)
Group- comparison (case-control)	2+	Population subgroups with and without characteristics of interest	1	Present and recall of past
Longitudinal (Prospective)	1 or 2+	Population or subgroups that are and are not likely to develop characteristic of interest	2+	Present and Future

Cross-sectional studies are conducted at a given point in time, and are used to characterize generally a single group representative of some population of interest. The reference period for characteristics that study subjects are asked to report may, however, be either for that point in time or for some reasonable period of time that they can recall in the past. They, for example, are appropriate to determine the prevalence of specified morbid conditions, or to reveal current patterns of health services utilization.

Cross-comparison studies explicitly focus on two or more groups. The groups are chosen based on the criterion that one has a characteristic of interest and the other does not. As is the case with the cross-sectional design, data are collected at one point in time, and the reference period for asking study subjects questions could be either the present or some period of time in the past. With analytical group-comparison designs (called "case-control or retrospective studies" in epidemiological studies) there is an effort to explicitly look back in time at the factors that could have given rise to one group having the characteristic (for example, a particular disease) and the other not having the characteristic.

Longitudinal studies, in contrast, assess changes over time. They employ procedures for repeated sampling of a population. Longitudinal studies focus on a population or subgroups, some members of which will be exposed to or experience certain events over time, while others will not; and data are collected at more than one point in time, the reference period is prospective rather than retrospective. Longitudinal designs can be cohort studies or panel studies. In the cohort approach a new sample of individuals is selected randomly for each set of measurements (cohort studies of given age groups are commonly used by demographers). In the case of the panel approach the sample selected initially (same individuals) is followed over time. A panel study allows maximum explanation of change over time, but they expensive and highly time-consuming as well as possibly produce serious problems of non-response trough attrition. Panel studies require complex statistical analysis. To avoid these problems and still capture many of mentioned benefits, surveyors frequently attempt in a single respondent contact to reconstruct retrospectively the history of interest, for example through a pregnancy history. In this instance the validity of the data will be subject to recall problems.

There are three principal methods of gathering data in surveys.