
so I’m gonna try to see how it goes so
the study behind it was published August
2015 in a psychologist conducted a
hundred exact replications published
studies only 36 of the replicates itself
that is the same statistically
significant results the first even worse
only 47 of them thoughts the phone in
the night confidence intervals for the
original studies now if the original
specs were done correctly because 95 of
the reputations to follow the
comprehensible because that’s what a
confident civil is the only 47 did other
studies to show fairly certain age in
other scientific fields the larger
problems have come in psychology and
medicine which is especially worried so
this seems like a pretty big problem for
science a lot of science this is how
many things in a fix-it lady larger
sample sizes permutation of trials make
stricter criteria for p-values all of
these ideas and more are worth exploring
the theme of this accident is
questioning the north I will argue that
the probably goes deeper than just
specific research as alternative
besides busy these continues with Bayes
theorem here it is so the first term is
T of the key given me this is the
probability assuming that our hypothesis
is true that you can get we’re doing
science this is how confident we are in
our hypothesis before we go to our
second Nexus pap this is the probability
that we get the other to be done so we
put all these steps together it’s our
lot P of each giving me this is the
posterior probability that hypothesis is
true so we’re doing science this is how
population in our hypothesis after we’ve
done our study so it turns out that we
can use any hypothesis here even though
methods so that’s a frequentist
statistics that’s except they only use
one term on the left
P D given H naught Q H given you know
these are not the same thing the
probability that I’m wearing a shirt
human wearing a red shirt is 100% the
probability that wearing a red shirt
given that I’m Granger is not understand
in fact seem to be quite so this is the
main problem in frequency it’s measuring
the wrong thing instead of everything
of our hypothesis given our evidence
unless we probability of getting our
evidence giving some hypothesis this is
just totally backwards so what’s my
third example plus my mate at 4.30 much
thinner and one of them is weighted so
lent heads 80% of the time so we’ll take
the point of random and then try to
determine using evidence of pointless
whether the point be pissed
well what are the fingertips is those
waiting for the point and it’s heads so
well let’s try our frequences method
first our hypothesis we put the twenties
week which means of the null hypothesis
is fair so what’s the probability given
our know that hast given the coin is
fair then we would get the day we got
that we get as well insensitive
percent of the time so now use the
standard scientific significance
P is less than point zero five then this
data is not statistically significant
which means you can’t learn anything
from the frigid detector with a Bayesian
method can make a conclusion let’s use
Bayes theorem the traffic database so
again we will do each tournament of time
so first we community this is the
probability doing our hypothesis which
is that the point is waiting that we
would get the heads so in the way home
Len has a you for some time so this is
just 80 percent of point
next is P of each this is the prior
probability that would have to be the
clock since there’s four points and one
of them is wait at this is just one last
piece this is the probability of getting
heads before we do so there’s a poor
chance we take away that let has 80
percent there’s a three quarter chance
when we get one with Fairfax in each of
those one that’s 50 percent of time so
we take all these numbers put them back
to the Bayes theorem to do the matter
turns out there’s a 35% chance
so the point we just split is the way to
talk okay
you’re learning with amazing that the
country we were implement interest we
changed our probability weighted in 25%
35% this is the main advantage of
asia-pacific
he frequented statistics all infants are
black and white either a result a
significant report isn’t all things to
do after they put appointments he said
that they need to collect more did on
the other hand of Bayesian to make
inferences in shades of gray
Bayesian can also go collect our data to
be more sure but right now they have to
the question how likely is it that the
point you just love is a later point so
it turns out the figuring out the prior
probability of our hypothesis to be is a
lot harder than it was usually than he
was not available so you know doing
science the prior probability is usually
subjected to tension adjustment for
researchers rather distributed imagine
see this means that different scientists
for their packaging for the prior
probabilities about the same hypothesis
so this might be bad system if what’s
science fair publish were probabilities
and hypotheses be triggered because
those who depend on subjective
probabilities individual researchers so
depending on who did the study together
another experiment themselves our
objective increased objective results
that shouldn’t depend on researcher so
let’s go back acquaintance except this
time we don’t know
and don’t know prior probability we’re
just sticking point
totally Atlanta only knows that it’s
either way tonight has made seven times
understand so once we do then we
calculate is called with base batteries
the base factor depends only on the
experimental result degree two
hypotheses say each one h9 the base
factor is P P given h1 so with our phone
if each one is not possible to the point
its way to speech not fair then the
probability people get ahead my weight
is 8% eight probably isn’t a spare fifty
percent of five so base factors just one
point six so what can you do is well we
can multiply our base factor by our
primary nerd calculate where posterior
won’t we shouldn’t need after we see
evidence
so that’s so Bayes factor that’s what we
actually need a science that’s actually
record so instead trying to make
inferences or get results that the
participates
we want to resolve to change our beliefs
in the direction of truth this was the
base factor of the students so science
we need to know what about your data
before campaigns at all there are times
harder where we ended up on diesel take
the following problem
10 out of every thousand residents of
these tournament with breast cancer 90
excess power of the remaining 990 women
89 will also test object a woman test
possible he wants to know whether she’s
breast cancer for sure or these from the
tentacles
so this problem presented over $4,000
well this is actually not too far solved
using Bayes theorem and really Seoul say
the hypothesis it’s a woman’s breast
cancer any evidence that she tested
positive so we can just use Bayes
thinner as we do I want the gross stuff
this time
turns out there’s about a 90% chance
here’s the problem a 1/4 of doctors
answer this question correct this is a
multiple choice question by problem
another full quarter doctors so that the
woman has
that’s a huge thing it makes total sense
these doctors who made this mistake
because they all their lives so is the
probability both the show
but also we can’t always go for dinner
each women only got past what so it
seems similar bandages to use amazing
statistics rather than the standard of
treatment in statistics use bioscience
if sons you may sit to visit people will
be better
Beasley’s and they do better the proper
kind of problem that we just saw science
based on all the information disposal
not just
mythix really make conclusions that work
just black and white with the country’s
brightest and most normal be able to
have scientifically informed beliefs
about how these beliefs about the way
the one who actually did
[Applause]
the study behind it was published August
2015 in a psychologist conducted a
hundred exact replications published
studies only 36 of the replicates itself
that is the same statistically
significant results the first even worse
only 47 of them thoughts the phone in
the night confidence intervals for the
original studies now if the original
specs were done correctly because 95 of
the reputations to follow the
comprehensible because that’s what a
confident civil is the only 47 did other
studies to show fairly certain age in
other scientific fields the larger
problems have come in psychology and
medicine which is especially worried so
this seems like a pretty big problem for
science a lot of science this is how
many things in a fix-it lady larger
sample sizes permutation of trials make
stricter criteria for p-values all of
these ideas and more are worth exploring
the theme of this accident is
questioning the north I will argue that
the probably goes deeper than just
specific research as alternative
besides busy these continues with Bayes
theorem here it is so the first term is
T of the key given me this is the
probability assuming that our hypothesis
is true that you can get we’re doing
science this is how confident we are in
our hypothesis before we go to our
second Nexus pap this is the probability
that we get the other to be done so we
put all these steps together it’s our
lot P of each giving me this is the
posterior probability that hypothesis is
true so we’re doing science this is how
population in our hypothesis after we’ve
done our study so it turns out that we
can use any hypothesis here even though
methods so that’s a frequentist
statistics that’s except they only use
one term on the left
P D given H naught Q H given you know
these are not the same thing the
probability that I’m wearing a shirt
human wearing a red shirt is 100% the
probability that wearing a red shirt
given that I’m Granger is not understand
in fact seem to be quite so this is the
main problem in frequency it’s measuring
the wrong thing instead of everything
of our hypothesis given our evidence
unless we probability of getting our
evidence giving some hypothesis this is
just totally backwards so what’s my
third example plus my mate at 4.30 much
thinner and one of them is weighted so
lent heads 80% of the time so we’ll take
the point of random and then try to
determine using evidence of pointless
whether the point be pissed
well what are the fingertips is those
waiting for the point and it’s heads so
well let’s try our frequences method
first our hypothesis we put the twenties
week which means of the null hypothesis
is fair so what’s the probability given
our know that hast given the coin is
fair then we would get the day we got
that we get as well insensitive
percent of the time so now use the
standard scientific significance
P is less than point zero five then this
data is not statistically significant
which means you can’t learn anything
from the frigid detector with a Bayesian
method can make a conclusion let’s use
Bayes theorem the traffic database so
again we will do each tournament of time
so first we community this is the
probability doing our hypothesis which
is that the point is waiting that we
would get the heads so in the way home
Len has a you for some time so this is
just 80 percent of point
next is P of each this is the prior
probability that would have to be the
clock since there’s four points and one
of them is wait at this is just one last
piece this is the probability of getting
heads before we do so there’s a poor
chance we take away that let has 80
percent there’s a three quarter chance
when we get one with Fairfax in each of
those one that’s 50 percent of time so
we take all these numbers put them back
to the Bayes theorem to do the matter
turns out there’s a 35% chance
so the point we just split is the way to
talk okay
you’re learning with amazing that the
country we were implement interest we
changed our probability weighted in 25%
35% this is the main advantage of
asia-pacific
he frequented statistics all infants are
black and white either a result a
significant report isn’t all things to
do after they put appointments he said
that they need to collect more did on
the other hand of Bayesian to make
inferences in shades of gray
Bayesian can also go collect our data to
be more sure but right now they have to
the question how likely is it that the
point you just love is a later point so
it turns out the figuring out the prior
probability of our hypothesis to be is a
lot harder than it was usually than he
was not available so you know doing
science the prior probability is usually
subjected to tension adjustment for
researchers rather distributed imagine
see this means that different scientists
for their packaging for the prior
probabilities about the same hypothesis
so this might be bad system if what’s
science fair publish were probabilities
and hypotheses be triggered because
those who depend on subjective
probabilities individual researchers so
depending on who did the study together
another experiment themselves our
objective increased objective results
that shouldn’t depend on researcher so
let’s go back acquaintance except this
time we don’t know
and don’t know prior probability we’re
just sticking point
totally Atlanta only knows that it’s
either way tonight has made seven times
understand so once we do then we
calculate is called with base batteries
the base factor depends only on the
experimental result degree two
hypotheses say each one h9 the base
factor is P P given h1 so with our phone
if each one is not possible to the point
its way to speech not fair then the
probability people get ahead my weight
is 8% eight probably isn’t a spare fifty
percent of five so base factors just one
point six so what can you do is well we
can multiply our base factor by our
primary nerd calculate where posterior
won’t we shouldn’t need after we see
evidence
so that’s so Bayes factor that’s what we
actually need a science that’s actually
record so instead trying to make
inferences or get results that the
participates
we want to resolve to change our beliefs
in the direction of truth this was the
base factor of the students so science
we need to know what about your data
before campaigns at all there are times
harder where we ended up on diesel take
the following problem
10 out of every thousand residents of
these tournament with breast cancer 90
excess power of the remaining 990 women
89 will also test object a woman test
possible he wants to know whether she’s
breast cancer for sure or these from the
tentacles
so this problem presented over $4,000
well this is actually not too far solved
using Bayes theorem and really Seoul say
the hypothesis it’s a woman’s breast
cancer any evidence that she tested
positive so we can just use Bayes
thinner as we do I want the gross stuff
this time
turns out there’s about a 90% chance
here’s the problem a 1/4 of doctors
answer this question correct this is a
multiple choice question by problem
another full quarter doctors so that the
woman has
that’s a huge thing it makes total sense
these doctors who made this mistake
because they all their lives so is the
probability both the show
but also we can’t always go for dinner
each women only got past what so it
seems similar bandages to use amazing
statistics rather than the standard of
treatment in statistics use bioscience
if sons you may sit to visit people will
be better
Beasley’s and they do better the proper
kind of problem that we just saw science
based on all the information disposal
not just
mythix really make conclusions that work
just black and white with the country’s
brightest and most normal be able to
have scientifically informed beliefs
about how these beliefs about the way
the one who actually did
[Applause]
Please follow and like us: