Historically, logically and methodologically it seems that NHSTs and p values have been shown to have many deficiencies, to say the least.
https://www.researchgate.net/publication/260062295_Null_hypothesis_significance_tests_A_mix-up_of_two_different_theories_The_basis_for_widespread_confusion_and_numerous_misinterpretations
The ASA in 2017 held a conference informally called"the woodstock of significance".
http://ww2.amstat.org/meetings/ssi/2017/recordedsessions.cfm
What are the alternatives to NHSTs?
¿Que opciones existen a las NHST ahora que posiblemente la ASA va a abandonar el uso de estas pruebas?
Historicamente, logicamente y metodologicamente parece que se ha demostrado que las NHST tienen muchas deficiencias, por decir lo menos.
https://www.researchgate.net/publication/260062295_Null_hypothesis_significance_tests_A_mix-up_of_two_different_theories_The_basis_for_widespread_confusion_and_numerous_misinterpretations
El ASA el año 2017 realizo una conferencia a la que se le ha llamado informalmente "el woodstock de la significancia"
http://ww2.amstat.org/meetings/ssi/2017/recordedsessions.cfm
Que alternativas hay a las NHST?
Within the current scientific system, there is no viable alternative.
It's like asking for alternatives to airbags in order to avoid traffic jams, where individual car traffic is generally wanted and supported. It's like changing the color of a court plaster, disregarding the reason for the wound.
We need to change our education. A PhD must actually be able to give a correct definition of "probability", and must be able to explain correctly what "P(D|H)" means. We need (much) more information-theoretic and philosophical background (what is information? what is knowledge?).
That change won't be easy, because many weird and illogical concepts are already cemented in many influential brains (current bosses, editors and reviewers), old habits of successful professors may turn out to be wrong or misleading (at best). Thus, I thing a change will come only from a next generation that had a better education - and we still don't have (enough) people to initiate a change in the curricula and to teach. In fact, instead I see a trend towards a more superficial education, bringing more students faster into jobs.
Within the current scientific system, there is no viable alternative.
It's like asking for alternatives to airbags in order to avoid traffic jams, where individual car traffic is generally wanted and supported. It's like changing the color of a court plaster, disregarding the reason for the wound.
We need to change our education. A PhD must actually be able to give a correct definition of "probability", and must be able to explain correctly what "P(D|H)" means. We need (much) more information-theoretic and philosophical background (what is information? what is knowledge?).
That change won't be easy, because many weird and illogical concepts are already cemented in many influential brains (current bosses, editors and reviewers), old habits of successful professors may turn out to be wrong or misleading (at best). Thus, I thing a change will come only from a next generation that had a better education - and we still don't have (enough) people to initiate a change in the curricula and to teach. In fact, instead I see a trend towards a more superficial education, bringing more students faster into jobs.
Thanks for your answer.
You said:
"We need to change our education. A PhD must actually be able to give a correct definition of "probability", and must be able to explain correctly what "P(D|H)" means."
I have some epistemological background because I was fortunate that my master's coordinator had the good idea to include that course. But when I have to read some articles about the statistical questioning of NHSTs I don't fully understand them. How could I learn the theory of probability? What previous knowledge would I need? The statistics I learned were like most people learn, like a cookbook, with little reasoning.
On the other hand, I believe that it is possible to achieve change in the next 10 years. If statistical reasoning is simplified and popularized. I think 100 years of statistical errors can't be changed in 1 year. People like you, with all your knowledge, can help this. I believe that in every change there are three parts: discovery, technical diffusion, media diffusion and propaganda. The last two steps are the weak point of science. But that can be changed.
----------------------------------------------------------------------------------------------------------
Tengo algunas bases epistemologicas debido a que tuve la suerte que mi coordinador de maestria tuvo la buena idea de incluir ese curso. Pero cuando tengo que leer algunos articulos sobre los cuestionamientos estadisticos a las NHST no los entiendo por completo. Como podria aprender la teoria de la probabilidad? Que conocimientos previos yo necesitaria?. La estadistica que aprendi fue como la que aprenden la mayoria, tipo libro de cocina, con poco razonamiento.
Por otra parte, yo creo que si es posible lograr cambios en los proximos 10 años. Si se simplifica y populariza el razonamiento estadistico. Creo que 100 años de errores estadisticos no se pueden cambiar en 1 año. Personas como usted, con todo su conocimiento, pueden ayudar a esto. Yo creo que en todo cambio hay tres partes: el descubrimiento, la difusion tecnica, la difusion en medios y la propaganda. Los ultimos dos pasos son el punto de flaco de la ciencia. Pero eso se puede cambiar.
This could be a start: https://plato.stanford.edu/entries/probability-interpret/
It is important to really read the works given in the bibliography. Especially the Port Royal Logic, Laplace, Carnap, Gavalotti, Hajèk, Jaynes, Jeffreys, Kolmogorov, Mellor, Ramsey. - And, unfortunately: reading is not enough. One really needs to think about all that. Question the concepts. Read again. Find the weak spots and the contradictions. Read again. It's work.
« If statistical reasoning is simplified and popularized »
I think the current problem with p-values and so on is because statistical reasoning was simplified (too much)...
I'm not sure that statistical reasoning can be simplified. It requires mainly logic and reflexion, not much more (statistical tools, that's another problem), but it is inherently subtle — and even more, it's application to reality and concrete cases is subtle, much more than the theory itself. I think user's understanding of it should be increased instead.
- Badges
- Science method
- Similar topics
- Methodology