8. Data sources and collection

Summary

Where data come from and how they are collected determines the conclusions that can be drawn from them. Having a census (a complete survey of the units of analysis) is not the same as having a sample (a partial survey); the latter carries the uncertainty inherent in not having all the data. Yet that is precisely the setting where data analysis plays its largest part.

A sample is usually the most feasible way to obtain data. Even in the era of big data, however much data there is, it tends to be a partial representation of reality, of the processes that generate it. The sample is the rule, not the exception.

Sampling is a procedure by which a set of units of analysis is drawn, at random, from a larger (finite or infinite) set of units. In its simplest version it consists of drawing them directly, although this may not be the best way to do it. Later chapters present other methods that, in essence, seek to improve the representativeness of the data obtained with respect to the whole set, or equivalently, to reduce errors due to sampling itself. Randomness plays a fundamental role here: it is the channel through which every other kind of error due to subjectivity is meant to be eliminated, while errors from operational issues must also be avoided. It is also worth noting that sampling is not only a technique applied to real data; samples can also be obtained from a simulation or from synthetic data, which is useful for experimentation.

A sampling procedure (from here on always understood as random) necessarily yields random variables; the goal is to characterize them while taking their random nature into account. Because each time a sample is taken, the results of a mean or a standard deviation (or any other measure) for a given variable will differ. Every single time, it will be different. These results can themselves be treated as new variables and characterized as well. This is precisely one of the great pillars of data analysis.

To recap, what matters from here on is the ability to deal with a sample and draw conclusions about the variables from it. The following chapters introduce the theory and concepts of probability, which will be of great help.

Further reading

  • Origen de los datos (in Spanish). Kelmansky, Estadística para todos, pp. 51–52.
  • Muestreo (in Spanish). Kelmansky, Estadística para todos, pp. 29–35.
  • Fuentes de datos. Aguirre, Niño and Simonetti, Estadística aplicada en las ciencias sociales y humanas (in Spanish), pp. 18–20.
  • Métodos de muestreo. Lind, Marchal and Wathen, Estadística aplicada a los negocios y la economía (in Spanish), 15th ed., pp. 266–268.
  • Muestras aleatorias. Canavos, Probabilidad y estadística. Aplicaciones y métodos (in Spanish), pp. 214–217.
  • Métodos de recolección de datos para una investigación (in Spanish). Mariela Torres, Karim Paz and Federico G. Salazar, Boletín Electrónico No. 3, Faculty of Engineering, Universidad Rafael Landívar.