Probability Kinematics and Commutativity

Carl G. Wagner∗

Department of Mathematics
The University of Tennessee
Knoxville, TN 37996–1300

Abstract

The so-called “non-commutativity” of probability kinematics has
caused much unjustified concern. When identical learning is properly
represented, namely, by identical Bayes factors rather than identical
posterior probabilities, then sequential probability-kinematical revi-
sions behave just as they should. Our analysis is based on a vari-
ant of Field’s reformulation of probability kinematics, divested of its
(inessential) physicalist gloss.

1 Introduction

The much remarked “non-commutativity” of probability kinematics ( Do-
motor 1982, Skyrms 1986, van Fraassen 1989, Döring 1999, Lange 2000) has
evoked varying degrees of concern. In this paper it is shown that when identi-
cal learning is properly represented, namely, by identical Bayes factors rather
than identical posterior probabilities, then sequential probability-kinematical
revisions behave just as they should.

Our analysis, which unifies and extends results in Field 1978, Diaconis
and Zabell 1982, and Jeffrey 1988, is based on a variant of Field’s reformula-
tion of probability kinematics, divested of its (inessential) physicalist gloss.
In § 2 a brief review of probability kinematics is presented. In § 3 we ex-
tend Field’s Theorem to countable partitions, showing that the uniformity

∗Research supported by the National Science Foundation (SES-9984005)

1



of Bayes factors in the representation of identical learning is sufficient for the
commutativity of probability kinematics. In § 4 it is shown that under mild
restrictions such uniformity is also necessary for commutativity. In § 5 we
discuss the methodological and philosophical implications of these theorems.

The notational conventions of this paper are as follows: If q is a revision
of the probability measure p and A and B are events, the Bayes factor (or
odds factor) βq,p(A : B) is the ratio

βq,p(A : B) :=
q(A)

q(B)

/
p(A)

p(B)
(1.1)

of new-to-old odds, and the probability factor (or relevance quotient) πq,p(A)
is the ratio

πq,p(A) := q(A)/p(A)(1.2)

of new-to-old probabilities. When q comes from p by conditionalization on
the event E, then (1.1) is simply the likelihood ratio p(E|A)/p(E|B). More
generally,

βq,p(A : B) = πq,p(A)/ πq,p(B),(1.3)

a simple, but useful, identity.

2 Probability Kinematics

Let (Ω, A, p) be a probability space, and suppose that E = {Ei} is a count-
able family of pairwise disjoint events, with p(Ei) > 0 for all i. A probability
measure q is said to come from p by probability kinematics on E (Jeffrey
1965, 1983) if there exists a sequence (ei) of positive real numbers summing
to one, such that

q(A) =
∑

i

eip(A|Ei), for all A ∈ A.1(2.1)

If E = {E}, then q(A) = p(A|E), and so probability kinematics is a general-
ization of ordinary conditionalization.

Formula (2.1) is clearly equivalent to the conjunction of the conditions

q(Ei) = ei, for all i, and(2.2)

q(A|Ei) = p(A|Ei), for all A ∈ A and for all i.(2.3)

2



Then (2.1) defines the appropriate revision of p in the light of new evidence
when the total evidence, old as well as new,2 prompts us to revise the proba-
bilities of the events Ei as specified by (2.2), but we learn nothing new about
the relevance of any Ei to other events. Condition (2.3), which captures the
latter state of affairs, is called the rigidity (or sufficiency) condition.

Having revised p to q by the probability-kinematical formula (2.1) above,
let us consider a subsequent revision of q to r by the formula

r(A) =
∑

j

fjq(A|Fj), for all A ∈ A,(2.4)

where F = {Fj} is a countable family of pairwise disjoint events such that
q(Fj) > 0 for all j, and (fj) is a sequence of positive real numbers summing
to one. Now imagine reversing the order of the foregoing, first revising p to,
say, q′ by the formula

q′(A) =
∑

j

fjp(A|Fj), for all A ∈ A,(2.5)

and then revising q′ to, say, r′ by the formula

r′(A) =
∑

i

eiq
′(A|Ei), for all A ∈ A.(2.6)

Symbolically,

p
E,ei−−−→ q

F,fj

∣∣∣ �F,fj� r
q′ −−−→

E,ei
r′

.(2.7)

Unless E = {E} and F = {F }, in which case r′(A) = r(A) = p(A|EF ),
it may well be the case that r′ 	= r. The possibility of such “non-commutati-
vity” has been the source of much confusion and unjustified concern. In the
next two sections we lay the foundations for clarifying this issue with two
theorems, delineating their methodological and philosophical implications in
§5.

3



3 Field’s Theorem

A proper analysis of the commutativity issue requires consideration of the
more general probability-kinematical revision schema

p
E−−−→ q

F

∣∣∣ �F� r
q′ −−−→

E
r′

,(3.1)

where the sequence
(
r′(Ei)

)
may differ from

(
q(Ei)

)
, and the sequence

(
q′(Fj)

)
from

(
r(Fj)

)
. Field (1978) was the first to identify conditions sufficient to

ensure that r′ = r in this setting, in the special case where E and F are
finite. In fact, Field’s result holds for all countable families E and F, but
requires a different proof.

Theorem 3.1. Given the probability revision schema (3.1), if the Bayes fac-
tor identities

βr′,q′(Ei1 : Ei2 ) = βq,p(Ei1 : Ei2 ), for all i1, i2,(3.2)

and

βq′,p(Fj1 : Fj2 ) = βr,q(Fj1 : Fj2 ), for all j1, j2,(3.3)

hold, then r′ = r.

Proof. It follows from (2.1) and (2.2) that for all A, A′ ∈ A

q(A) /q(A′) =
∑

i

q(Ei)

p(Ei)
p(AEi)

/ ∑
i

q(Ei)

p(Ei)
p(A′Ei).(3.4)

Dividing the numerator and denominator of the right-hand side of (3.4) by
q(E1)/p(E1), and setting A

′ = Ω, then yields the formula

q(A) =
∑

i

Bip(AEi)

/ ∑
i

Bip(Ei),(3.5)

4



where

Bi := βq,p(Ei : E1).(3.6)

With B′i := βr′,q′(Ei : E1), bj := βr,q(Fj : F1), and b
′
j := βq′,p(Fj : F1), it

follows from (3.5), along with analogous formulas for r, q′, and r′ that

r(A) =
∑
i,j

Bibjp(AEiFj)

/ ∑
i,j

Bibjp(EiFj)(3.7)

and

r′(A) =
∑
i,j

B′ib
′
jp(AEiFj)

/ ∑
i,j

B′ib
′
jp(EiFj).(3.8)

Since (3.2) implies (indeed, is equivalent to) B′i = Bi and (3.3) implies (in-
deed, is equivalent to) b′j = bj, it follows that r

′ = r.

Remark 3.1. Field’s proof of Theorem 3.1, with E and F finite and E =
{E1, . . . , Em}, involved reformulating (2.1) as

q(A) =
m∑

i=1

Gip(AEi)

/
m∑

i=1

Gip(Ei),
3(3.9)

where Gi is the geometric mean

Gi :=

(
m∏

k=1

βq,p(Ei : Ek)

)1/m
.(3.10)

Remark 3.2. By (1.3), the Bayes factor identities (3.2) and (3.3) are equiva-
lent, respectively, to the probability factor proportionalities

πr′,q′(Ei) ∝ πq,p(Ei), for all i, and(3.11)
πq′,p(Fj) ∝ πr,q(Fj), for all j.4(3.12)

Remark 3.3. Jeffrey (1988) showed that if E and F are finite and r and r′

are defined by (3.1), then the probability factor identities

πr′,q′(Ei) = πq,p(Ei), for all i, and(3.13)

πq′,p(Fj) = πr,q(Fj), for all j,(3.14)

imply that r′ = r. In light of the above remark, Jeffrey’s result is a corollary
of Theorem 3.1.

5



In the special case of schema (3.1) represented by (2.7), Bayes factor
and probability factor identities are equivalent to each other, as well as to an
especially salient pair of conditions known as Jeffrey-independence conditions,
namely,

q′(Ei) = p(Ei), for all i, and(3.15)

q(Fj) = p(Fj), for all j.(3.16)

Theorem 3.2. If in (3.1) it is the case that
(
r′(Ei)

)
=

(
q(Ei)

)
, then the

conditions (3.2), (3.13), and (3.15) are equivalent. If it is the case that(
q′(Fj)

)
=

(
r(Fj)

)
, then the conditions (3.3), (3.14), and (3.16) are equiva-

lent.

Proof. Straightforward algebraic verification.

Remark 3.4. Diaconis and Zabell (1982), who coined the term “Jeffrey inde-
pendence,” proved in the context of (2.7), with E and F finite, that conditions
(3.15) and (3.16) imply that r = r′. In view of Theorem 3.2, this result is
also a corollary of Theorem 3.1. Note that the p-independence of E and F
(i. e., p(EiFj) = p(Ei)p(Fj), for all i, j) entails Jeffrey independence.

Remark 3.5. Formula (3.7), and consequently Theorem 3.1, can be general-
ized to arbitrary finite sequences of probability-kinematical revisions. We
leave the details as an exercise for interested readers.

4 A Partial Converse of Field’s Theorem

The Bayes factor identities (3.2) and (3.3) are not in general necessary for r′

and r in the revision schema (3.1) to coincide. For example, if F = E = {Ei}
and r(Ei) = r

′(Ei) for all i, then r
′ = r, no matter what values are assigned to

q(Ei) and q
′(Ei). However, Field’s Theorem does admit of a partial converse.

To motivate the statement of conditions under which (3.2) and (3.3) are
necessary for commutativity, it is useful to reiterate the conditions under
which the probabilities in schema (3.1) are well-defined. In order to imple-
ment the formulas for q, q′, r, and r′, we must have, respectively, p(Ei) > 0,
p(Fj) > 0, q(Fj) > 0, and q

′(Ei) > 0, for all i and j, or, equivalently, that

∀i ∃j : p(EiFj) > 0, and(4.1)
∀j ∃i : p(EiFj) > 0. 5(4.2)

6



The identities (3.2) and (3.3) turn out to be necessary for commutativity
under a mild strengthening of the well-definedness conditions (4.1) and (4.2).

Theorem 4.1. Let r and r′ be defined by the probability revision schema
(3.1), and suppose that

∀i1 ∀i2 ∃j : p(Ei1 Fj)p(Ei2 Fj) > 0, and(4.3)
∀j1 ∀j2 ∃i : p(EiFj1 )p(EiFj2 ) > 0.(4.4)

If r′ = r, then the Bayes factor identities (3.2) and (3.3) hold.

Proof. Since q comes from p by probability kinematics on E, the rigidity
condition (2.3) implies that

q(EiFj) = q(Ei)p(Fj|Ei), for all i, j.(4.5)

Similarly,

q(EiFj) = q(Fj)r(Ei|Fj), for all i, j,(4.6)
q′(EiFj) = q

′(Fj)p(Ei|Fj), for all i, j, and(4.7)
q′(EiFj) = q

′(Ei)r
′(Fj|Ei), for all i, j,(4.8)

and so

q(Ei)p(Fj|Ei) = q(Fj)r(Ei|Fj), for all i, j, and(4.9)
q′(Ei)r

′(Fj|Ei) = q′(Fj)p(Ei|Fj), for all i, j.(4.10)

Given arbitrary i1 and i2, let j be such that (4.3) holds. It then fol-
lows inter alia from the relevant rigidity conditions that r(Ei2 Fj) > 0 and
r′(Ei2 Fj) > 0. Setting i = i1, i2 in (4.9) and solving for q(Ei1 ) and q(Ei2 )
then yields the formula

βq,p(Ei1 : Ei2 ) =
p(Ei2 Fj)

p(Ei1 Fj)

r(Ei1 Fj)

r(Ei2 Fj)
.(4.11)

Similarly, (4.10) yields the formula

βr′,q′(Ei1 : Ei2 ) =
p(Ei2 Fj)

p(Ei1 Fj)

r′(Ei1 Fj)

r′(Ei2 Fj)
.(4.12)

7



Hence if r′ = r, then the Bayes factor identity (3.2) holds.
Given arbitrary j1 and j2, let i be such that (4.4) holds. It then follows

from (4.9) and (4.10) by an argument similar to the above that

βq′,p(Fj1 : Fj2 ) =
p(EiFj2 )

p(EiFj1 )

r′(EiFj1 )

r′(EiFj2 )
, and(4.13)

βr,q(Fj1 : Fj2 ) =
p(EiFj2 )

p(EiFj1 )

r(EiFj1 )

r(EiFj2 )
.(4.14)

Hence if r′ = r, then the Bayes factor identity (3.3) holds.

Remark 4.1. Note that when F = E, conditions (4.3) and (4.4) fail to hold,
thus allowing the possibility (illustrated in the example cited at the beginning
of this section) that r′ = r even though (3.2) and (3.3) fail to hold. On the
other hand, (4.3) and (4.4) always hold when E and F are qualitatively
independent (for all i, j, EiEj 	= ∅) and p is strictly coherent (p(A) > 0 for
all nonempty events A).

Remark 4.2. As noted in Remark 3.4, if in the probability revision schema
(3.1),

(
r′(Ei)

)
=

(
q(Ei)

)
and

(
q′(Fj)

)
=

(
r(Fj)

)
, then the Jeffrey indepen-

dence conditions (3.15) and (3.16) imply that r′ = r. Interestingly, given (4.3)
and (4.4), if for some i, r′(Ei) 	= q(Ei), or for some j, q′(Fj) 	= r(Fj), then
Jeffrey independence not only fails to ensure that r′ = r, but actually ensures
that r′ 	= r. This follows from Theorem 4.1 and the easily verified fact that
(3.15) and (3.16), along with (3.2) and (3.3), imply that

(
r′(Ei)

)
=

(
q(Ei)

)
and

(
q′(Fj)

)
=

(
r(Fj)

)
. In particular, the p-independence of E and F, since

it implies both (4.3) and (4.4) and (as noted in Remark 3.4 above) Jef-
frey independence, also ensures that r′ 	= r unless

(
r′(Ei)

)
=

(
q(Ei)

)
and(

q′(Fj)
)

=
(
r(Fj)

)
.

Remark 4.3. Diaconis and Zabell (1982) proved for E and F finite that in the
special case of (3.1) represented by (2.7) Jeffrey independence is necessary
for r′ = r. In view of Theorem 3.2, this result is a corollary of Theorem 4.1.

5 Sequential Probability Kinematics: All is

Cool

That r′ may fail to coincide with r in the probability-kinematical revision
schema (2.7) has been cause for concern among several commentators (see

8



Lange 2000 for some sample quotations). Their concern appears to be
based on implicit acceptance of two principles relating to the general revision
schema (3.1), reproduced below:

p
E−−−→ q

F

∣∣∣ �F� r
q′ −−−→

E
r′

.

I. If what is learned from the experience prompting the revisions of p to
q, and of q′ to r′, is the same, and if what is learned from the experience
prompting the revisions of q to r, and of p to q′, is the same, then it ought
to be the case that r′ = r.

II. Identical learning underlying the revisions of p to q and of q′ to r′

ought to be reflected in the posterior probability identities

r′(Ei) = q(Ei), for all i,(5.1)

and identical learning underlying the revisions of q to r and of p to q′ in the
identities

q′(Fj) = r(Fj), for all j.(5.2)

The first of these principles is unexceptionable. To paraphrase van Fraass-
en (1989), two persons who undergo identical learning experiences on the
same day, but in a different order, ought to agree in the evening if they
had exactly the same opinions in the morning. But the second principle
is mistaken, losing sight of the fact that posterior probabilities assigned to
events in the families E and F are based on the total evidence, old as well
as new, and thus incorporate elements of the relevant priors.6

What we need is a numerical representation of what is learned from
new evidence alone, with prior probabilities factored out. It is a staple of
Bayesianism that ratios of new-to-old odds furnish the correct representation
of the desired type (Good, 1950, 1983; Jeffrey 2000).7 Accordingly, Princi-
ple II needs to be modified by replacing (5.1) and (5.2), respectively, by
the Bayes factor identities (3.2) and (3.3). So modified, Principle II is both
sufficient (Theorem 3.1), and in a substantial number of cases, necessary
(Theorem 4.1) for the satisfaction of Principle I.

9



Remark 5.1. Field (1978) came close to our modification of Principle II, but
made sensory stimulus, eo ipso, the source of Bayes factors, with identical
stimuli prompting probability revisions by identical Bayes factors. A coun-
terexample of Garber (1980) shows this physicalist view to be untenable:
Imagine that you glance briefly in dim light at an object known to be blue
or green, resulting in your becoming slightly more confident than you were
before that the object is blue. Repeated glances producing the identical sen-
sory stimulus will then result in your approaching certainty that the object
is blue. Our formulation, in which identical learning gives rise to identical
Bayes factors, is immune to Garber’s counterexample. We learn nothing new
from repeated glances and so all Bayes factors beyond the first are equal to
one.8

Remark 5.2. If F = E in the probability revision schema (2.7), then r′ = q
and r = q′, and so it is always the case that r′ 	= r, except in the uninteresting
case in which q′ = q. Lange (2000) has furnished a lucid example suggesting
that this never involves a violation of Principle I

¯
since the relevent revisions

are not based on identical learning.9 Note that this claim follows from the
principle that identical learning ought to be reflected in identical Bayes fac-
tors: Since q′ 	= q, either q′ 	= p or q 	= p. In the former case, the identity
βr′,q′(Ei1 : Ei2 ) = βq,p(Ei1 : Ei2 ) can not hold for all i1, i2, since r

′ = q. In the
latter, the identity βr,q(Ei1 : Ei2 ) = βq′,p(Ei1 : Ei2 ) cannot hold for all i1, i2,
since r = q′.

Remark 5.3. Note that in Theorem 3.1 the probabilities p, q, q′, r, and r′

are assumed to be well-defined and in place at the outset. Then r′ = r
if the Bayes factor identities (3.2) and (3.3) hold. Suppose that only p, q,
and q′ were in place at the outset. Does (3.2) then furnish a recipe for
constructing a probability r′ that would be the appropriate revision of q′ if
in the probabilistic state q′ one were to learn precisely what prompted the
revision of p to q? And does (3.3) function analogously in the construction
of a probability r? Only if, in the first instance,∑

i

q(Ei) q
′(Ei)/ p(Ei) < ∞(5.3)

and in the second,

∑
j

q(Fj) q
′(Fj)/ p(Fj) < ∞, 10(5.4)

10



since a probability r′ satisfies (3.2) if and only if

r′(Ei) =
q(Ei)q

′(Ei)

p(Ei)

/∑
i

q(Ei)q
′(Ei)

p(Ei)
(5.5)

and a probability r satisfies (3.3) if and only if

r(Fj) =
q(Fj)q

′(Fj)

p(Fj)

/∑
j

q(Fj)q
′(Fj)

p(Fj)
.(5.6)

But this raises an intriguing question, with which we conclude this paper. If
(5.3), and hence (5.4), fails to hold, as may be the case,11 does this mean that
the learning prompting the revision of p to q (respectively, p to q′) cannot
identically occur in the probabilistic state q′ (respectively, q)?

Notes

1. In the standard exposition of probability kinematics the family E = {Ei}
is taken to be a partition of Ω, so that, in addition to pairwise disjointness of
the events Ei, one has E1∪E2∪· · · = Ω. Standardly, (ei) is a sequence of non-
negative real numbers summing to one, and it is assumed that “zeros are not
raised”, i. e., that p(Ei) = 0 implies that ei

(
= q(Ei)

)
= 0. Finally, it is stip-

ulated that 0·p(A|Ei) = 0 if p(Ei) = 0, so that eip(A|Ei) is well-defined even
when p(A|Ei) isn’t. Given the standard format, our family E simply com-
prises those Ei in the partition for which ei > 0. Conversely, our format yields
the standard one by associating to our family E = {E1, E2, . . . }, if it fails to
be a partition, the partition {E0, E1, E2, . . . }, where E0 := Ω−(E1∪E2∪. . . ),
and setting e0 = 0. When one deals with sequential probability-kinematical
revisions in the standard format, conventions involving the values of expres-
sions involving benign sorts of undefinedness multiply, and can easily obscure
non-benign cases of undefinedness (see, especially, §4 below, where certain
positivity conditions are crucial). Our format minimizes the possibility of
such confusion.

Probability kinematics may arise through ordinary conditionalization.
Suppose, for example, that A is the σ-algebra generated by E = {Ei} along
with arbitrary hypotheses H1, H2, . . . , and an event E ⊂

⋃
Ei such that

p(EEi) > 0 for all i. Let A
′ be the σ-subalgebra of A generated by E and

11



the Hi. If q comes from p by conditioning on E and for all A ∈ A′ and for all
i, A and E are conditionally p-independent, given Ei (p(A|EEi) = p(A|Ei)),
then on A′, q comes from p by probability kinematics on E. The whole point
of probability kinematics is of course that the experience prompting revision
of the probabilities of events in the family E often fails to be representable
as the occurrence of an event E. But it is sometimes useful, when attempt-
ing to explore certain aspects of probability kinematics, to entertain such
a fictional “phenomenological event” E. At least when A is finite, this is
always a formal possibility (see Diaconis and Zabell 1982, Theorem 2.1, for
the whole story).

2. This is an exceedingly important point. As will be seen in §5 below,
the mistaken apprehension that the probabilities q(Ei) are based solely on
new evidence is responsible for much of the confusion surrounding the “non-
commutativity” of probability kinematics.

3. Actually, Field expresses Gi in the form e
αi , where αi := log Gi, interpret-

ing the αi as “input parameters” associated with a given sensory stimulus.
Each instance of this stimulus prompts a probability revision of the type (3.9)
involving these input parameters. See Remark 5.1 in §5 for further discussion
of this idea.

4. As an illustration, let us show the equivalence of (3.11) and (3.2). The
proportionality (3.1) asserts the existence of a positive constant c such that,
for all i, πr′,q′(Ei) = cπq,p(Ei). Then (3.2) follows immediately from (1.3).
Conversely, it follows from (3.2) with i1 = i and i2 = 1, along with (1.3),
that, for all i, πr′,q′(Ei) = cπq,p(Ei), where c = πr′,q′(E1)/πq,p(E1).

5. From (4.1) it follows that p(Ei) > 0 for all i, and from (4.2) that p(Fj) > 0
for all j. Since q(Fj) =

∑
i eip(Fj|Ei), with all ei > 0, it follows from (4.2)

that q(Fj) > 0 for all j. Since q
′(Ei) =

∑
j fjp(Ei|Fj), with all fj > 0,

it follows from (4.1) that q′(Ei) > 0 for all i. If, on the other hand, (4.2)
fails to hold, then by the above formula for q(Fj), there exists a j such that
q(Fj) = 0. And if (4.1) fails to hold, then by the above formula for q

′(Ei),
there exists an i such that q′(Ei) = 0.

6. Suppose, for example, that a ball is chosen at random from an urn con-
taining 9999 green balls and one blue ball, and you get to examine it fleetingly
in a dim light. It would be folly to assess the probability that the ball is blue
based only on your sensory impression, ignoring the composition of the urn.
Indeed, unless you have solid grounds for regarding the sensory impression

12



that your glance produces as being much more likely if the ball is blue than
if it is green, your prior probability should undergo little, if any, revision.

7. Notice that several familiar measures of probability change lack the req-
uisite feature of effacing all traces of the prior. For example, knowing noth-
ing about p and q except, say, that q(E) − p(E) = 1/4, one can conclude
that p(E) � 3/4, and knowing nothing except, say, that q(E)/p(E) = 2,
one can conclude that p(E) � 1/2. Suppose, on the other hand, that
E = {E1, . . . , Em} and nothing is known about p and q except that
βp,q(Ei : E1) = βi, where (βi) is a given sequence of positive real num-
bers with β1 = 1. From this information nothing whatsoever can be in-
ferred about p. For given any such (βi) and any prior p with p(Ei) > 0
for all i, there exists a probability q such that βq,p(Ei : E1) = βi, namely,
q(Ei) = βip(Ei) /

∑
i βip(Ei) .

8. A way to see this is to entertain an imaginary phenomenological event E
capturing the visual content of the glance (see note 1, supra). Then if q(·) =
p(·|E), and r(·) = q(·|E), βq,p

(
B(lue) : G(reen)

)
= p(E|B)/p(E|G) > 1,

by assumption. But βr,q(B : G) = q(E|B)/q(E|G) = p(E|EB)/p(E|EG) =
1/1 = 1.

9. Lange speaks of experience rather than learning, but we mean the same
thing by these terms, namely, considered experience (in light of ambient mem-
ory and prior probabilistic commitment), rather than the isolated sensory
experience that Field saw as the source of Bayes factors. We have adopted
our terminology from Jeffrey 2000 to minimize confusion with the latter sort
of experience.

10. Actually, (5.3) and (5.4) are equivalent, since the sums in question are
each equal to ∑

i,j

q(Ei)

p(Ei)

q′(Fj)

p(Fj)
p(EiFj).

11. Let Ω = {1, 2, . . . }, Ei = {2i−1, 2i}, and Fj = {4j −3, 4j −2, 4j −1, 4j}
for i, j = 1, 2, . . . . Let p(2i − 1) = p(2i) = 7/2 · 8i, let q come from p by
probability kinematics on {Ei} with q(Ei) = 1/2i, and let q′ come from p
by probability kinematics on {Fj} with q′(Fj) = 2/3j. Then, since Fj =
E2j−1 ∪̇ E2j,

∑
j

q(Fj)q
′(Fj)

p(Fj)
=

∑
j

2

21

(
16

3

)j
= ∞.

13



References

[1] Diaconis, Persi and Sandy L. Zabell (1982), “Updating Subjective Prob-
ability”, Journal of the American Statistical Association 77(380): 822–
830.

[2] Domotor, Zoltan (1980), “Probability Kinematics and Representation
of Belief Change”, Philosophy of Science 47: 384–403.

[3] Döring, Frank (1999), “Why Bayesian Psychology is Incomplete”, Phi-
losophy of Science 66: S379–S389.

[4] Field, Hartry, (1978), “A Note on Jeffrey Conditionalization”, Philoso-
phy of Science 45: 361–367.

[5] Garber, Daniel, (1980), “Field and Jeffrey Conditionalization”, Philos-
ophy of Science 47: 142–145.

[6] Good, Irving J. (1950), Probability and the Weighing of Evidence. Lon-
don: Charles Griffin.

[7] Good, Irving J. (1983), Good Thinking. Minneapolis: University of Min-
nesota Press.

[8] Jeffrey, Richard C. (1965), The Logic of Decision. New York: McGraw-
Hill; (1983), 2nd ed. Chicago: University of Chicago Press.

[9] Jeffrey, Richard C. (1988), “Conditioning, Kinematics, and Exchange-
ability”, in Brian Skyrms and William L. Harper (eds.), Causation,
Chance, and Credence, vol. 1. Dordrecht: Kluwer, 221–255.

[10] Jeffrey, Richard C. (2000), “Epistemology Probabilized”, manuscript,
available at http://www.princeton.edu/~bayesway.

[11] Lange, Marc (2000), “Is Jeffrey Conditionalization Defective in Virtue
of Being Non-Commutative? Remarks on the Sameness of Sensory Ex-
perience”, Synthese 123: 393–403.

[12] Skyrms, Brian (1986), Choice and Chance, 3rd ed. Belmont: Wadsworth
Publishing Co.

[13] van Fraassen, Bas (1989), Laws and Symmetry. Oxford: Clarendon
Press.

14