a clear step-by-step procedure?

CSGnet Archive CSG2018
13

[Bruce Nevin (20191101.16:30 ET)]

Probably Alex is no longer looking for a clear step-by-step specification of the Test, but on the assumption that the question will arise again I have put Billʽs specification (as rephased by Phil Runkel) on the Discourse forum under Methodlogy at http://discourse.iapct.org/t/the-test-for-the-controlled-variable/134.

···

/Bruce

On Wed, May 2, 2018 at 12:13 AM Martin Taylor mmt-csg@mmtaylor.net wrote:

[Martin Taylor 2018.04.30.15.05]

  On 2018/04/30 2:29 PM, Alex Gomez-Marin > wrote:

i’d still appreciate a clear TCV step-by-step.

…

                    Hey,

Adam and I are trying to take the TCV seriously,
but we would like to have it in a clear logical
step-by-step way so that it is concrete. I know
these has been around there in many papers, but
I wonder if someone could just share the
“cooking recipe” so as to be clear when a
variable can be ruled out as a
controlled variable, and when it can still be
one (although one is never sure it must
be). Thanks, Alex

As with must such questions, I doubt that there is a black-and-white

all-purpose answer to this. So let’s take a PCT-based approach: (1)
ask what you are trying to achieve, (2) Ask how your perception of
the situation differs from how you would like it to be (what you are
trying to achieve), (3) Ask if you have a means to make your
perception closer to its reference value (what you are trying to
achieve). You are starting with question (3), which leads to a
different way of thinking. It is like saying “I have this thing
called a screwdriver . Now what can I do with it?”. The answer would
be “if you have a screw of the right type, you can use it and the
screw to attach one thing to another.” “But what is ‘the right
type’?” “If the things are wood, you want a wood screw, but if they
are metal, you want a metal screw.”

Enough analogy. The TCV is a tool that can be used for related but

different purposes, and the step-by-step instructions depend on what
you want to achieve. All of them depend on finding a way to disturb
some possibly controlled perception by way of disturbing a variable
in the environment of the organism that might be responsible for the
perception you hypothesise. But after than, there are various
possibilities. The Runkel sequence that was linked a while back is
one of them. (Bruce Nevin [Bruce Nevin 2018-05-01_08:20:48 ET]
reposted it today.

You say you want to know whether a particular variable V in the

environment does or does not correspond to a controlled perception.
You do not say whether this variable lies on a continuum of similar
hypotheses such as V = Xp+Yq , where p and q
are exponents that may take on a range of values your theory
permits, or whether it is of a category that differs cleanly from
other categories that might sensibly be hypothesized to be “the”
controlled variable (the wood screw versus the metal screw).

Powers required two preliminary tests as prerequisites to performing

the TCV (or as part of it), so I’ll include them as steps 1 and 2,
in either order. I will call the hypothesized corresponding
controlled quantity in the environment “V”.

1. Could the subject plausibly perceive the value V?

2. Could the subject act deliberately to influence the value V?



The answers to these questions may be evident, but if not, then

finding the answers provides two steps of the TCV. For example, in
the case of the speed-curvature relation, it is not plausible that
the subject could perceive the power relation while acting in a way
that produces it, so it cannot be a controlled variable, and must be
a side effect of controlling something else.

From here on, the steps of the TCV depend on whether the question is

a Yes-No (YN) question, a Forced-Choice question (FC) or an unforced
choice question (UC). These three choices apply mainly to cases in
which the hypothesised variable is categorical. If V is continuous,
the TCV becomes an optimization problem. In each situation, you need
to find a way of disturbing the hypothesized V, but your choice
depends on the question.

It doesn't matter what your question, the TCV will always have one

or all of three problems. The first problem is that the subject’s
reference value may change for the controlled perception that is the
object of your Test, which has the effect of adding noise into your
measurements; the second is that the subject may stop actively
controlling that perception during your test; and the third is the
contextual dependence problem that is simplified in the X+Y+Z
example below, so you can never be sure that you have captured all
the inputs to the perceptual function that produces your target
perception.

All three problems are worse in the wild than in the lab, but

there’s really nothing you can do about that. In the lab you can do
what psychologists have been doing for a century or more – either
deliberately increase contextual variability in a random way or try
to keep the context as stable and as bland as you can. The first
increases the noise, but gives you much more confidence you are
right if you get a clear answer, while the second increases the
likelihood you will get a clear answer, at the cost of lost
generalizability.

Despite these issues, the TCV can still be useful as a guide, and as

an analogy of what we do when interacting with other people. So
let’s look at the steps that follow the preliminaries, first for
Yes-No ("is this particular environmental variable perceived by a
perception that is being controlled?).

3(YN). Find some action that you are sure will influence V if it is

not being controlled. If you can estimate the magnitude of the
effect your influence should have on V in the absence of control, so
much the better. If not, and you can measure both the magnitude of
your influence as you randomly change it and the magnitudes of the
variable and if possible the supposed influence of the controller on
V, you are still in good shape, because:

4(YN)a If you apply your influence abruptly, and the value of V

changes rapidly but then tends to return toward its previous value,
then V is related to a controlled perception. It is controlled. If
it returns toward it earlier value only after you remove your
influence, or does not return at all, it is not controlled.

4(YN)b If the effect of your influence on V is less than expected,

or if the effect of your influence is larger when you prevent the
supposed controller either from perceiving the magnitude of V or
from influencing it, then V is an environmental variable related to
a controlled perception. It is controlled.

4(YN)c If your randomly varying influence magnitude changes have a

high correlation with the changes in the magnitude of V, and a low
correlation with the influence of the supposed controller, V is not
controlled. If your influence is highly negatively correlated with
the influence of the supposed controller and only slightly
correlated with V, then V is related to a controlled perception, and
is controlled.

In the YN situation, you don't care whether V actually is the

variable perceived by the perception that is controlled. To say that
V is controlled is simply to say that its variation under
disturbances is countered to some extent by the actions of the
controller. That will be the case for say, V = X + Y + Z if the
controlled perception is of X + Y and the other variable, Z, changes
only slightly compared to the variation in X and Y. But if that’s
all you want to know, YN is your way to go.

The categorical (FC) situation is easier. Your question is which of

a defined list of environmental variables corresponds to a
controlled perception. Implicitly you are asserting that exactly one
of the list is correct.

3(FC) Arrange to disturb all of the hypothesized environmental

correlates of controlled perceptions with a randomly varying
influence that may be the same for all or may be individualized. If
you are correct that one and only one is the environmental variable
corresponding to a controlled perception, then all but the correct
one will vary in a way highly correlated with the changes in your
influence while the correct one will have a low correlation with
your influence.

In the unforced choice condition (UC), you add the possibility "none

of the above". If that is the correct answer, then all of the
hypothesized environmental correlates will have a high correlation
with your influence.

Finally you have the continuous variants of FC and UFC. Here, the

form of the hypothesised controlled perception is assumed, but you
don’t know what environmental variable parameter values correspond
to it. Is it X+Y, 1.2X+0.8Y or something with a ratio between those
possibilities? In this case, you only have to disturb X and Y with
independent random variations of your influence on them, and find
the X/Y ratio of disturbance scale that gives the lowest correlation
between the disturbance and the environmental variable.

I don't know if these steps are sufficient for your purposes, but

they might serve as a guide. Ask what it is you want to achieve, ask
how your current state differs from that, and ask whether the TCV is
the right tool to allow you to achieve what you want – or at least
get closer than you are – and if so, which form of the TCV is
appropriate.

Martin