The Probability of God: A Bayesian Analysis

by Glen Davis, Chi Alpha @ Stanford, http://www.xastanford.org/

inspired by Stephen Unwin's book The Probability of God
This analysis is based on Bayes' Theorem, which can be used to determine what you ought to believe about
something based on what else you already know or believe. See the other sheets for more info.
Prior Probability of God's existence

Evidentiary Area
Morality
Evil
Suffering
Answered/Unanswered Prayer
Reports of Miracles
Religious Experiences
Origin/Nature of the Universe
Logical Necessity/Impossibility
Other Category
Other Category

D Factor
1/
1/
1/
1/
1/
1/
1/
1/
1/
1/

Overall Probability of God's Existence:

50%

1
1
1
1
1
1
1
1
1
1

Likelihood of God
50%
50%
50%
50%
50%
50%
50%
50%
50%
50%
50%

The last number is how likely you ought to think it is that God exists
given what else you believe about the universe.
If this approach intrigues you, I recommend Unwin's book as a
great place to begin digging deeper.

INSTRUCTIONS
Each evidentiary area subsumes many arguments
God's existence. Hover your mouse over the evide
see a description of it. Think about it, and then dec
the evidence in that category is more likely a resul
or of a universe without God.

These are recommended value increments, but yo

whatever you like.
10: Much more likely if there is a God.
2: More likely if there is a God.
1: Equally likely regardless of God (put this if u
1/2: More likely if there is no God.
1/10: Much more likely if there is no God.

Place your conclusion in the cell under D Facto

Cells with red corners have comments--hover y

learn more!

nalysis

ght to believe about

re info.

NS
ary area subsumes many arguments for and against
ce. Hover your mouse over the evidence category to
ion of it. Think about it, and then decide whether or not
n that category is more likely a result of God existing
se without God.

ommended value increments, but you can put down

like.
re likely if there is a God.
if there is a God.
ely regardless of God (put this if unsure).
ly if there is no God.
more likely if there is no God.

onclusion in the cell under D Factor.

d corners have comments--hover your mouse to

Bayes Theorem is an identity stating that the probability of A given that B has occurred is equal to the probability of B
given that A has occurred times the probability of A occurring divided by the probability of B occurring.
Confused?
Think of it this way: suppose you have a bowl of ten marbles. Nine are red and one is blue.
Now pull a marble at random from the bowl--how likely is it to be blue?
Clearly 10%.

Now imagine two bowls of ten marbles. One has nine red and one blue, and the other has eight blue and two
red.
Shuffle the bowls around as if playing some magic trick, and then pick a marble at random.
You pick a blue marble.
Which bowl did you likely pick from--the one that's almost all red or the one that's almost all blue?
You probably got your marble from the bowl that's mostly blue. It doesn't take a genius to figure that out.

Using Bayes' theorem, we can show exactly how likely it is. Here is the problem expressed in terms of elements of Bayes' theo
P(A given B)
P(B given A)
P(A)
P(B)

?
80%
50%
45%

What is the probability that I took a marble from the mostly blue bowl if the marble I picked is blue?
What is the probability that I picked a blue marble if I picked from the mostly blue bowl?
What is the probability that I picked from the mostly blue bowl?
What is the probability that the marble I picked is blue?

So here's how we would figure things out:

P(A given B)

89%

So there's roughly a nine in ten chance that you pulled from the mostly blue bowl.
What's the significance of this? We inferred a reliable probability of something we did not
directly measure by looking only at related data!
Surprisingly, this approach has proven quite useful in many non-mathematical applications
(although not without controversy). It's possible to apply Bayes' Theorem to all sorts of
relatively fuzzy subjects if you can quantify some of the related data, even if only in a vague
and somewhat subjective way. See the next sheet for an example of that.
You can learn a lot more about Bayes (and some controversy about the proper application
of his theorem) on the Wikipedia (www.wikipedia.org).

Incidentally, the astute will have noticed that the formula on the first sheet is different from the
one that I used above. They are equivalent to one another (I leave it to you to demonstrate
this if you're really nervous about it). The different formula on the first page was occassioned
by the need to integrate user data into the equation. Stephen Unwin chose to do that using
the variable D, and made the appropriate transformations to Bayes' to make it work. I think it
makes it much more accessible for non-math types.

Incidentally, the astute will have noticed that the formula on the first sheet is different from the
one that I used above. They are equivalent to one another (I leave it to you to demonstrate
this if you're really nervous about it). The different formula on the first page was occassioned
by the need to integrate user data into the equation. Stephen Unwin chose to do that using
the variable D, and made the appropriate transformations to Bayes' to make it work. I think it
makes it much more accessible for non-math types.

equal to the probability of B

B occurring.

has eight blue and two

m.

all blue?

figure that out.

erms of elements of Bayes' theorem.

wl if the marble I picked is blue?

he mostly blue bowl?

A Non-Mathematical Example of Bayes' Theorem

Here's a non-mathematical example that gets across Bayes' theorem. While the other sheet is more helpful for the
math, this may be more helpful for understanding the basic concept.
Question One: You see me walking down the street with tears in my eyes. How likely am I to be sad?
Well, pretty likely. Based on the information you have, thats a reasonable assumption. People cry when they are
really sad or really happy (usually in places like wedding chapels and hospital delivery rooms) or when they have
something in their eye, but sadness causes more tears than anything else.
But say you get to know me well, and realize that I have pesky contacts that keep causing me to tear up for no good
reason at all. In fact, I cry far more often due to my contacts than due to sorrow or joy.
Question Two: You see me walking down the street with tears in my eyes. How likely am I to be sad?
This is the same as question one, but now we have more information that can help us give a better answer.
One way we can try to answer this question is using Bayes theorem. To do that, we have to try to quantify our
knowledgeput numbers on it. This is going to be somewhat subjective (you might choose to assign different
numbers than me), and yet oddly objective (Im making my assumptions plain and putting them out in the open for
discussion and possible refutation).
Lets say a = me crying.
Lets say b = me being sad.
What is P(b|a)? That's statistical notation for "the probability (P) of b given that a is true." In other words, how likely is
it that Im sad (b) given that Im crying (a)?
This is where Bayes theorem becomes useful. It tells us that P(b|a)=P(a|b)P(b)/P(a)
In English: the probability that I am sad given that you see me crying is going to be equal to the probability that I
would cry if I was sad times the probability that Im sad divided by the probability that Im crying.
Now we obviously don't have hard numbers that will help us make an inference. Nobody (to my knowledge) has ever
followed me around and recorded how frequently I cry or regularly assessed my emotional state to develop a
normative model of my personal psychology.
Does that mean that we're completely stuck?
No, not at all.
If we can get some round numbers we'll still be able to get some information. That information will be useful if our
numbers are reasonably accurate.
So let's think for a minute. Knowing me, you'd say the following numbers are close:
* If I was really sad, theres probably a 75% chance that I would cry. [thats P(a|b)]
* Im only really sad about 1% of the time, though. [thats P(b)]
* And I have tears in my eyes about 5% of the time. [thats P(a)]
Now lets plug the data into Bayes theorem: .75*.01/.05=.15
If our numbers were perfectly accurate, that would mean that theres a 15% chance if you see me crying that Im
actually sad. The rest of the time it's either bad contacts or rapturous joy (or some other unidentified cause).
The numbers are probably not perfectly accurate, so relying on the 15% figure as opposed to a 25% figure is a little
silly. The conclusion, though, is solid: "If I see Glen crying, it's probably not because he's sad."
That's a somewhat silly example, but it gets the core concepts across.
You can use Bayesian analysis for all sorts of things IF you can come up with reasonable numbers for the related

* Im only really sad about 1% of the time, though. [thats P(b)]

* And I have tears in my eyes about 5% of the time. [thats P(a)]
Now lets plug the data into Bayes theorem: .75*.01/.05=.15
If our numbers were perfectly accurate, that would mean that theres a 15% chance if you see me crying that Im
actually sad. The rest of the time it's either bad contacts or rapturous joy (or some other unidentified cause).
The numbers are probably not perfectly accurate, so relying on the 15% figure as opposed to a 25% figure is a little
silly. The conclusion, though, is solid: "If I see Glen crying, it's probably not because he's sad."
That's a somewhat silly example, but it gets the core concepts across.
You can use Bayesian analysis for all sorts of things IF you can come up with reasonable numbers for the related
evidence. You can come up with reasonable numbers in all sorts of ways--experimentation, research, or interviewing
experts.
To the extent that your numbers are externally justifiable your analysis is persuasive to others. To the extent that your
numbers are subjective your analysis is persuasive only to you (and to those who mostly agree with you).
But here's the thing--you ought to believe the results of your analysis until you are persuaded to change some of
your numbers.
What Bayes' Theorem is doing in that case is helping you be consistent: "Given that you believe X, Y, and Z, W is the
most reasonable expectation for you to hold of the universe."
And that's what's happening on the first sheet. You're not necessarily creating an argument that will crush all in your
path--you're discovering how much confidence you should place in the God hypothesis BASED ON WHAT YOU
ALREADY BELIEVE ABOUT ALL THE RELEVANT EVIDENCE.
And so I encourage you to think deeply about the arguments for and against the existence of God. If you'd like to see
a helpful discussion in the context of Bayes' theorem, I encourage you to go read Stephen Unwin's book The
Probability of God.

s more helpful for the

be sad?

e cry when they are

) or when they have

e to tear up for no good

o be sad?

better answer.

ry to quantify our
o assign different
m out in the open for

her words, how likely is

he probability that I
ng.

my knowledge) has ever

ate to develop a

n will be useful if our

e me crying that Im
entified cause).

a 25% figure is a little

"

mbers for the related

e me crying that Im
entified cause).

a 25% figure is a little

"

mbers for the related

esearch, or interviewing

. To the extent that your

ee with you).
to change some of

eve X, Y, and Z, W is the

at will crush all in your

ED ON WHAT YOU

God. If you'd like to see

nwin's book The