Econometrics borrowed a tool astronomers built to track the heavens
The least squares method was devised in the early nineteenth century to pin down the motions of planets and the shape of the Earth. Economists later adopted it to measure things like how much another year of school raises wages, and it became the backbone of econometrics, the statistics of real economies.
Econometrics applies statistical methods to economic data to put numbers on relationships that theory only describes. Its roots include the political arithmetic of William Petty and Gregory King and later work by Francis Edgeworth and Vilfredo Pareto. Its workhorse remains the linear regression, in its simplest form fitting the best straight line through a scatter of data points, which is still the usual starting point even though many other tools now exist.
One early use came in 1889, when the British statistician Udny Yule compared English county census data from 1871 and 1881 to estimate whether public assistance affected poverty rates. By modern standards his study had problems, since poverty might drive assistance as much as the reverse. Another textbook case is Okun's law, linking economic growth to unemployment: in one estimate, a percentage point of extra growth corresponds to unemployment falling by about 1.77 points, other things equal.
Theorists look for estimators with good properties: no systematic bias, efficiency and consistency as data grows. The hard part is that economists rarely run controlled experiments and mostly study observational data, so they hunt for natural experiments and use techniques such as instrumental variables, regression discontinuity and difference-in-differences to tease out cause and effect. A classic labour economics model relates the logarithm of a person's wage to years of education, lumping every other influence into an error term.
The toolkit has widened. Linear discriminant analysis arrived in 1936 to predict categories, logistic regression for yes-or-no outcomes followed in the 1940s, and generalised linear models came in the early 1970s. For most of the twentieth century methods stayed linear because anything else demanded too much computing, until faster machines in the 1980s opened the way to non-linear approaches.
Source: Econometrics
