All models are wrong

Box (1979) includes the following section heading and first two sentences

All models are wrong but some are useful Now it would be very remarkable if any system existing in the real world could be exactly represented by any simple model. However, cunningly chosen parsimonious models often do provide remarkably useful approximations (pp 202-203)

Box was writing in the context of the models developed by statisticians. Others have expanded this to include more general models and theories developed by scientists. For example, Harai writes

The real test of ‘knowledge’ is not whether it is true, but whether it empowers us. Scientists usually assume that no theory is 100 per cent correct. Consequently, truth is a poor test for knowledge. The real test is utility. A theory that enables us to do new things constitutes knowledge.

Embrace the epistemic humility (or perhaps epistemic-fluency)#

So (2017) explains that Box takes this humble approach to suggest and approach to statistics that sees models as "recursive mechanisms for generative exploration" (p. 669). Box assumes that are models must be wrong, yet they allow the isolate certain findings and in doing so tweak the model. A process So (2017) describes as an iterative process where "truth" is "at some asymptomatic point that can never be reached. But along the way, the modeling process yields productive insights" (p. 669).

It's about the journey and not the destination.

Which links nicely to ideas of practice like bad_set-casa-gather-weave, not-how-bad-you-start and epistemic-fluency.

References#

Box, G. E. P. (1979). Robustness in the Strategy of Scientific Model Building. In R. Launer & G. Wilkinson (Eds.), Robustness in Statistics (pp. 201--236). Academic Press.

So, R. J. (2017). "All Models Are Wrong." PMLA, 132(3), 668--673.