The History of Statistics

The word statistics is derived from the Latin word “status” or the Italian word
“statista,” and meaning of these words is “political state” or “government.”
Shakespeare used the word statist is his drama Hamlet (1602). In the past, statistics
was used by rulers. The application of statistics was very limited, but rulers and kings
needed information about land, agriculture, commerce, populations of their states to
assess their military potential, their wealth, taxation and other aspects of
government.

Gottfried Achenwall used the word statistik at a German university in 1749 to mean
the political science of different countries. In 1771 W. Hooper (an Englishman) used
the word statistics in his translation of Elements of Universal Erudition written by
Baron B.F Bieford. In his book, statistics was defined as the science that teaches us
the political arrangement of all the modern states of the known world. There is a big
gap between the old statistics and modern statistics, but old statistics is also used as
a part of present-day statistics.

During the 18th century, English writers used the word statistics in their works, so
statistics has developed gradually during the last few centuries. A lot of work was
done at the end of the nineteenth century.

At the beginning of the 20th century, William S Gosset developed the methods for
decision making based on small sets of data. During the 20th century, several
statisticians were active in developing new methods, theories and applications of
statistics. These days, the availability of electronics is certainly a major factor in the
modern development of statistics.

In the mid-nineteenth century, humanity was in a state of apotheosis with scientific

discoveries, a great wave of optimism seized Europe with the new possibilities. It
seemed like a matter of time before we learned all the laws that govern nature. We
have made great progress in physics, biology, astronomy, etc. that justified this
excess of optimism. It seemed that if we had good measurements we could describe
and predict anything.

A little earlier, in the eighteenth century, when making some measurements of the
position of the planets, there were some slight deviations. We expected the planet to
be in a position and it was slightly in the “wrong place.” There were two possible
explanations for this phenomenon. Either the model would be wrong or our
equipment that collects this data is not good / sensitive enough. The model looked
pretty good so it should be the fault of the instruments, and so they started producing
ever more accurate equipment. In analyzing these small deviations, they noticed that
he always followed a certain distribution. Laplace devoted an entire volume of his
astronomical predictions just to dealing with errors. Today we know it as normal
distribution.

Galton box is a machine that tries to demonstrate that different distributions

converge to the normal distribution. We call this curious fact the central limit
theorem.

So, we started to improve the quality of the equipment and as expected, the errors
decreased, and this encouraged to create better and better equipment, but as the
equipment became more and more accurate, something bizarre started to happen,
the errors instead. To decrease, started to increase! And the scientists of the time
began to wonder why and the answer is that our world is not deterministic but
stochastic, that is, events have an intrinsically random feature! Even if we had the
best data collections and knew the perfect model of nature, yet it would not guarantee
a good prediction, as there are random factors that we do not control.

Early examples of statistical inference

5th century B.C. — Athenians estimated the height of ladders necessary to scale the
walls of Platea by having multiple soldiers count the bricks, then multiplying the most
frequent count (the mode) by the height of a brick. Al-Kindi (801-873 A.D.) wrote
“Manuscript on Deciphering Cryptographic Messages” which showed how to use
frequency analysis to decipher encrypted messages. John Graunt in Natural and
Political Observations Made Upon the Bills of Mortality estimated the population of
London in 1662 as 384,000 using records on the number of funerals per year
(13,000), the death rate (3 people per 11 families per year), and average family size
(8).

Probability

By 1700 the mathematics of many games of chance was well understood; major
contributors were Fermat, Pascal, Huygens, Leibniz. But this early work did not
consider the inferential problem: How, from one or several outcomes of the game,
could one learn about the properties of the game and how could one quantify the
uncertainty of the inferred knowledge of these properties? It was Jacob Bernoulli
(1654-1705) who began down this road. His weak law of large numbers was
published in 1713 (posthumously), and he put a lower bound on the probability that
X/N (the proportion of successes in N repeated trials) was within a specified distance
from the true probability of success. De Moivre (1667-1754) refined Bernoulli’s bound
and stumbled upon a discrete version of the normal curve as an approximation to
binomial probabilities.

Error distributions

In 1755 Thomas Simpson introduced the notion of error curves, including what are
now called probability density functions. His pdf of choice was triangular. Others
proposed error curves of different shapes: semicircular (Lambert 1765), exponential
(Laplace 1774), parabolic (Lagrange 1776), normal (Laplace 1778, although it was
not called the normal distribution until 1873).

The Mean

We needed new techniques to deal with this new world, and so came statistics as we
know it, a tool for understanding a stochastic world. In the case of measurements
with random characteristics, for example, we should work with some measure of
centrality, such as the average:

The mean of two numbers was a well-known concept to the ancient Greeks. Extended
to more than two numbers in the late 1500s in order to estimate the locations of
celestial bodies. Shown to minimize the sum of squared deviations by Thomas
Simpson in 1755.

The Median

Originated in Edward Wright’s 1599 book, Certaine Errors in Navigation, concerning

the determination of a location with a compass. Further advocated by Ruder Boskovic
in his 1755 book on the shape of the earth, in which he showed that the median
minimizes the sum of absolute deviations. The term “median” was coined by Galton
in 1881.
References:

http://homepage.divms.uiowa.edu/~dzimmer/alphaseminar/Statistics-history.pdf

https://www.emathzone.com/tutorials/basic-statistics/history-of-statistics.html

https://www.slideshare.net/AledelChristianAleja/a-brief-history-of-statistics

https://www.hup.harvard.edu/catalog.php?isbn=9780674403413

https://towardsdatascience.com/a-brief-history-of-statistics-36cfdac9439f