Professional Documents
Culture Documents
This is the most long-standing worry lodged at Bayesians, and it was first asked by Clark
Gleymour in 1980.
Suppose a hypothesis H is proposed which turns out to explain some already well-known data
E. Can H ever be confirmed by E? What do you guys think? Obviously, yes it can diba?
We can observe many, many examples demonstrating this in the field of science. It’s not really
controversial to say that scientific theories can be confirmed by facts that are already known.
Just to name two very famous examples:
1. Newton's theory (H) could explain Kepler's planetary motion (E) and how tides behave
(E).
2. Einstein’s General Relativity (H) is supported by the perihelion precession of Mercury (E).
So, ano ‘yung issue rito? Let’s go back to the Bayes’ Theorem.
Don’t worry, kahit ako hindi ko pa masyadong maintindihan ‘yung mga math parts ng topic na ‘to
haha pero ang mahalaga lang na i-observe ay ‘yung denominator kung saan nakalagay si
probability of evidence.
So, basically kung alam na natin dati pa ang isang established piece of evidence, we’ll have to
assign it a probability value of 1 kaya kung old evidence, magiging 1 ‘yung denominator. If the
denominator is 1, tapos ‘yung taas din ay 1 edi parang walang nangyayari.
Old Evidence has 0 power to confirm the hypothesis because the prior probability is getting
multiplied by 1 over 1. The prior probability (which is ‘yung before evidence) is guaranteed to
stay the same as the posterior probability (after evidence).
According sa lecturer dun sa vids, “subjectivity is the s-word in science” daw haha diba kasi in
science, usually deemphasized ang subjective views in favor or objectivity in order to reduce
quote-unquote biases. Kailangan dapat maging objective. So hindi ba objective ang
Bayesianism? Hindi naman sa ganun.
One unique thing talaga about the Bayesian Theorem is the freedom it gives to us to evaluate
our own personal and subjective degrees of belief. It’s one of its core principles. So as long as
coherent ang lahat ng beliefs mo, almost anything goes talaga.
Kung para sa’yo hindi totoo na may pandemya, pwede raw under Bayesianism ahaha basta nag-
aalign lang din siya sa iba mong beliefs.
Pero ayun nga, because of this leeway, Bayesians are questioned as to how it can ever be
scientifically rational kung pwedeng magkaka-iba-iba kayo ng paniniwala.
Well, despite all that subjectivity, Bayesians think they can reach scientific agreement. The
initial subjectivity, they argue, starts to disappear when enough good evidence comes in (aka
washing out or swamping of prior probabilities).
Pero hindi ba’t parang medyo wishful thinking ang ganito? It’s kind of too idealistic to believe
that "if perfect evidence comes in and if two people agree on how the evidence bears on the
theory, then they'll come to agreement." So, ‘yung isang criticism about Bayesianism.
If Bayesianism will be used as a theory of scientific reasoning, many people think it needs to
make room for scientific constraints on the values that get plugged into the equation.
The idea is to use evidence and scientific value to impose substantive constraints. Basically,
we have to tweak Bayesianism until it becomes a scientifically informed version of itself. This is
called tempered personalism.
So in defense of subjectivity, Kuhn has said that a certain amount of subjectivity is healthy for
science (like it has brought about some paradigm shifts and scientific revolutions in the past).
Bayesians obviously subscribe to this notion, and they argue that other approaches involve
subjective factors but tend not to admit it. Kaya nga S-word of science daw ang subjectivity.
So isang example daw ng ayaw umamin ay ang classical statistics, one of Bayesian statistics’
major competitors. Basically, it’s the stat that we usually learn in our typical statistics classes:
have a null hypothesis that you will reject or fail to reject based on some threshold such as the
5% threshold (the alpha level). CS and this 5% alpha level thing became really popular even if it’s
kind of just arbitrary because it’s objective pero hindi naman sa lahat ng pagkakataon ay
nagamit ito nang tama → which sabi rin ni Kasser leading to bad science. Bayesians argue that
classical statistics has to admit that they cannot attach probabilities to hypotheses, only rules
to reject.
Conclusion
Lastly, we prepared a self-paced Kahoot quiz just to serve as a review for what we discussed
today. We decided to make it self-paced instead of real-time to be inclusive to people who
maybe don’t have the stable internet connection to join a game in real-time. Tsaka hassle pa
‘yon hahaha. You can answer it by yourself until April 20, which is ‘yung expiration date ng quiz.
The game pin is on the screen, but we’ll put it on our Course Notes docs as well. ‘Yun lang.
Thank you so much for listening!