Spaceflight went quiet for thirty years. Then 2021 broke the space race record.

An analysis of every orbital launch attempt from Sputnik 1 in October 1957 to 2022: 4,630 flights, 62 operators, 370 rockets and 158 launch sites.

4,630

launch attempts analysed, spanning 65 years of spaceflight

111

launches a year in the current era, against 81 at the height of the space race

89.9%

of all missions succeeded, 4,162 of 4,630 attempts

80%

of every launch ever flown was carried out by just ten organisations

The launch rate collapsed in 1979 and stayed flat for thirty six years. The era that began in 2016 averages 111 launches a year, well above the 81 a year flown at the height of the space race.

Orbital launch attempts per year, 1957 to 2022Line chart. Launches climb through the 1960s, peak at 119 in 1971, fall to around 40 a year in the 2000s, then rise sharply after 2015 to a record 157 in 2021. 0 40 80 120 160 1960 1970 1980 1990 2000 2010 2020 Space race 81 a year Retrenchment 56 a year New space 111 a year 1971 peak, 119 2021 record, 157
Orbital launch attempts per year. Era boundaries are taken from the series itself: the fall is 1978 to 1979 and the rise is 2015 to 2016. 2022 is a partial year in the source data and is not a decline.

Method

How the analysis was built

The project follows the KDD process: selection, preprocessing, transformation, data mining, then interpretation. The source is a single record of global launch attempts, cleaned in pandas, with dates parsed to a yearly index and mission outcomes reduced to a binary success flag for rate comparisons.

The mining step is unsupervised. Latent Dirichlet Allocation was run over mission names and rocket names to test whether launch programmes separate into recoverable themes without being told what to look for. They do, and the topics split cleanly along Cold War and modern constellation lines.

A second stage attaches a 95 per cent Wilson interval to every operator's success rate, because a rate computed from two launches and one computed from 1,777 are not the same claim, and tests by permutation whether launch experience actually predicts reliability. It mostly does not, which is covered in the analysis.

Where to go next