Post #20
268

И это хороший паттерн мышления для статистика
- ❤ 6
DE Showing posts older than #21 · Back to latest



“Religion is a culture of faith; science is a culture of doubt” – Richard Phillips Feynman
The consequences of neglecting ergodic theory in social, behavioral, and medical fields may have substantial epistemic and practical consequences. In the absence of quantitative examination at the individual level, the consequences could range from zero if we are lucky to find one of the few ergodic processes in nature (19), to catastrophic if a process is quite nonergodic. In clinical research, diagnostic tests may be systematically biased and our classification systems may be at least partially invalid.


we cannot know a priori whether or not a process is ergodic; we cannot test the data for ergodicity. However, we can analyze repeated measures data on multiple individuals and use the ergodic theorem to determine that a process of interest is not ergodic.
Quite simply, comparisons of the first and second moments (mean and variance) of intraindividual and interindividual distributions can inform us about the accuracy of generalizations between groups and individuals.
An intuitive example is provided by Hamaker (14), who describes the correlation between typing speed and typos. At the group level, the correlation is negative, as experienced typists are both faster and more proficient. However, within individuals, the correlation is positive — the faster a given individual types, the greater the number of mistakes she or he will make relative to their own performance at slower speeds. Thus, the aggregation of the data produces an example of Simpson’s paradox, and we would commit an ecological fallacy by concluding that the relationship observed at the group level represents any of the individuals in the group.

Although sample size does not force within-person and between–person processes to resemble one another, it does influence our ability to statistically infer that these two differ or that one person differs from another.
Ergodicity only applies to identical processes: all people
must have all of the same parameters. Moreover, even if all people do
share all the same parameters, the process may still be non-ergodic.
Homogeneity is too weak a condition. Everyone could be the same, and
yet the phenomena of interest may still not be ergodic.
