In causal inference, the concept of temporality (or directionality) has not been fully clarified. Starting from causal thinking, this paper divides the time axis in nature into three time domains and two time points by the occurrence timings of both a real cause and a real effect. This has anchored that causal inference can only be realized in the third domain. The measured temporalities can be divided into five types: cross-first-to-third-domain longitudinal (or experimental temporalities), cross-second-to-third-domain longitudinal, within-domain longitudinal, within-domain reversely longitudinal, and within-domain transversal (or observational temporalities). This new classification encompasses all measurement strategies, either for first or multiple measurements, or timely and delayed measurements. Except that the actual measurement for the cause occurs either before its occurrence (only in experiment) or within the second domain, all other measurements are similar to the act of historical reconstruction or "archaeology" , where the importance of measured temporalities may be inferior to the accuracy of the measurements. From the point of view that research design should integrate bias design, this new classification for measured temporalities based on the time axis in Nature, which has a clear meaning and helps to judge the possible biases in the observation methods, provides a basis for correct causal inferences.

Epidemiological analysis describes and compares the characteristics of a certain number of people to make causal inferences. The formation of the study population is always the first step. In this paper, we first define the concepts of cross-sections at both individual level and population level and introduce the three assumptions needed in the measurements in observational studies, i. e. the true values of the attributes are stable with time, the attribute variables are independent and the individuals are independent during the measuring process. We also determine that the causal inference research should be unified based on the time of the occurrence or beginning of a postulated cause, or exposure, should be in. Then, based on the dual roles of the population cross-section with causal thinking, we propose that research designs can be classified into two types with different characteristics: history reconstruction research and future exploration research. Finally, we briefly analyze the research design framework and the relationship between estimated effects and different designs. The discussion of the formation of a study population from the perspective of causal thinking can make a foundation for the classification of causal inference research design with appropriate effect parameters, which needs to be further studied.