Arrows → can be interpreted as "tells us something about". ## Data → Sample Measurements should be relevant, valid, unbiased, and with low variance upon repetition: - Is what is being measured accurately reflecting what we are interested in? - Asking a person the same questions again should lead to the same answers; Measurements should ideally be idempotent. - The data should be [valid](Exploratory%20Data%20Analysis%20(EDA)%20checklist.md) and lack systematic biases or clustered data (location, time, etc.). - Surveys and polls should avoid [well-known pitfalls](Survey%20and%20Polling%20Pitfalls.md). ## Sample → Study Population The study should have **internal validity** to draw conclusions about the study population: - Did the sampling use a randomized process? - Were subjects properly shuffled to avoid bias? ## Study Population → Target Population The study should have **external validity** to draw conclusions about the target population: - Is the study population the same as the target population? - Is there any evidence that the conclusions on the study population cannot be generalized to the target population? Sometimes there is no distinction between sample and study population, between study and target population, or even all three (if your data covers the whole target population). In those cases, some of the questions above can be skipped, and possibly only the Data → Sample (= Target Population) requirements need to be checked. Source: The Art of Statistics by David Spiegelhalter