Positive Feedback & Economic Regulation (was Re: Popular Cover/Feature Article)

[amatic]

[From Adam Matic 2011.06.29.2000 gmt+1]

Rick Marken (2011.06.29.0830)

Yes, that's what everyone has been saying; that's what "correlation does not
imply causation" mean. But I have made two two points relative to this:

1) Taxes rates are a manipulated variable and thus qualify as an independent
variable from a research perspective. These changes are made at essentially
random points in time. So while there is no experimental control,
variations in tax rate qualify as a quasi-experimental manipulation which
means that you can _cautiously_ conclude that changes in a dependent
variable, such as unemployment, that are correlated with changes in tax rate
are caused by the changes in tax rate.

AM:
Oh, well, you can also cautiously paint your car red and press the gas
pedal while praying it will go faster. It's possible that those two
are causally connected since the sample was random.
I thought you said this reasoning didn't apply to control systems.

2) Since every correlation between socioeconomic variables that has ever
been observed is inconsistent with the predictions of what will happen
according to conservative/ libertarian ideology, it would seem to me that
this ideology is a poor perspective from whence to start developing theories
of how economies work. Of course, it's always a bad idea to get attached to
ideologies; but it seems to me that the conservative/ libertarian ideology
is, based on the existing evidence, one of the worst of a bad lot.

AM: Again, I do not defend conservative (or libertarian for that
matter, even though I do endorse it. And they are just so too
different to equate them like that with a backslash) ideology. What
they predict is not my concern.
On the other hand, Austrian economics does not predict correlations
such as those mentioned so far, neither positive or negative. There is
lots of data consistent with Austrian theory.

Best, Adam

[MartinT]

[Martin Taylor 2011.06.29.15.01]

[From Rick Marken (2011.06.28.2150)]

        Martin Taylor

(2011.06.28.10.52)–

Rick Marken (2011.06.27.2230)–

              RM: The term "correlation does

not imply causation" means that non-experimentally
obtained relationships between variables do not imply
causation.

MT: That is not what it means…

        The statement "correlation does not imply causation" simply

means that a correlation between A and B does not mean there
is a causal link between A and B.

      But it does mean that (according to the causal model that is

the basis of experimental research in the physical and life
sciences) if the correlation between A and B was observed
under experimental conditions.

You can make this condition more general. If you can be sure that

there exists no influence in common between A and B and yet A and B
are correlated, however slightly, then you can be sure that either A
influences B or B influences A. The experimental condition is one
example of this, since the experimenter ensures that the independent
variable is influenced only by the experimenter’s decisions.

        A and B may have no

causal link whatever, but if they don’t, then both must have
a causal link to some other variable.

      This is true if the relationship between A and B was observed

non-experimentally.

It's true always. The restriction to non-experimental observations

merely says that you won’t observe a correlation between A and B if
the experimenter ensures that A is determined by the experimenter
alone and there is no causal link between A and B.

Of course, even if there is a direct causal link between A and B,

the experimenter may measure a correlation of zero.

        Consider the

correlations between a simple index of income inequality
(the ratio of the to 20% to the bottom 20%) and 29 different
social indices across 23 developed nations or 50 US states,
from the book “The Spirit level: Why equality is better for
everyone” by Wilkinson and Pickett, Penguin 2010:`

        `

      This is great. Thanks for posting them.



      Of course, these correlations are based on purely

non-experimental data. It is not even quasi-experimental data
because the main “predictor variable”, income inequality, was
not manipulated purposefully (unlike, for example, top
marginal tax rates) and there was, of course, no control of
variables other than those being measured.

True. All it says is that at least one of the following is true:

Some unspecified influence acts on all the correlated variables.

Some set of correlated influences act on subsets of the correlated

variables.

Some of the correlated variables directly influence others of the

correlated variables.

A combination of the above statements exists.

      But, still, I think it's interesting that _all_ the

correlations show that the relationship between income
inequality and aggregate measures of quality of life (life
expectancy, infant mortality, homicide rate, teen birth rate,
education level, etc.) always go in the “wrong” direction:
increases in income inequality are always associated with
decreases in quality of life measures.

More to the point, variations in income equality account for over

75% of the variance in the combined index of the 10 variables that
were available from all 23 nations. That doesn’t mean income
equality causes those effects on quality of life, but it does mean
that if we instituted policies that tended to reduce income
inequality, there’s a pretty good chance those policy changes would
improve many of the measures of life quality.

As the book authors take pains to point out, this is independent of

tax policy, Japan having low inequality because of low variation in
basic income, whereas Scandinavian countries have low inequality
because of high taxes on high incomes. They are at similar levels on
the index of life quality.

Martin

[rsmarken]

[From Rick Marken (2011.06.29.1900)]

Martin Taylor (2011.06.29.15.01) –

Rick Marken (2011.06.28.2150)]

        MT: A and B may have no

causal link whatever, but if they don’t, then both must have
a causal link to some other variable.

      RM: This is true if the relationship between A and B was observed

non-experimentally.

It’s true always.

In a perfect experiment, where A is the IV and B the DV and ALL variables except A and B are held constant (controlled) then any observed correlation between A and B means that A was in some way responsible for the concomitant variations in B. I don’t want to use the word “cause” because we know that in a control loop where A is a disturbance to a CV and B the output of the control system, the observed correlation between A and B does not reflect a lineal causal connection between A and B. But A is, via its effect on the CV, the only variable responsible for the correlated variations in B.

Whether or not the system under study is a causal or a control system, an experiment where you manipulate an IV (which may be a disturbance to a CV if the system under study happens to be a control system) and observe variations in a DV (which may be the output that affects the same CV affected by the IV, if the system is a control system) under controlled conditions, any observed correlation between IV and DV means that the IV (and no other variable) is somehow responsible for the concomitant variations in the DV. The “somehow” is determined by testing to see if a CV is involved. If a CV is not involved – as in the case of an observed relationship between variations in voltage (IV) and current (DV) – then the relationship between the IV and DV can be considered causal. If there is a CV involved then the relationship reflects characteristics of the feedback function connecting DV to CV rather than characteristics of the functional path from IV to DV.

Experiments on control systems are done in the same way as experiments on causal systems: you manipulate an IV under controlled conditions and measure variations in a DV. The extra added attraction in experiments on systems that might possibly be control systems is that you also measure variations in a suspected CV. If variations in the IV result in far less variation in the suspected CV than is expected on physical grounds and if this lack of expected variation can be accounted for by variation in the DV that opposes the effect of the IV, then the suspected CV can be considered a CV and the relationship between IV and DV can be explained as the actions of a control system (the DV) resisting disturbances (IV) to the CV. Discovery of the CV is central to experiments on suspected control systems and, of course, irrelevant to experiments on causal systems.

Best

Rick

···

The restriction to non-experimental observations

merely says that you won’t observe a correlation between A and B if
the experimenter ensures that A is determined by the experimenter
alone and there is no causal link between A and B.

Of course, even if there is a direct causal link between A and B,

the experimenter may measure a correlation of zero.

–
Richard S. Marken PhD
rsmarken@gmail.com
www.mindreadings.com

[MartinT]

[Martin Taylor 2011.06.29.23.09]

[From Rick Marken (2011.06.29.1900)]

        Martin Taylor

(2011.06.29.15.01) –

Rick Marken (2011.06.28.2150)]

                  MT: A and B

may have no causal link whatever, but if they
don’t, then both must have a causal link to some
other variable.

                RM: This is true if the relationship between A and B

was observed non-experimentally.

It’s true always.

      In a _perfect_ experiment, where A is the IV and B the DV and

ALL variables except A and B are held constant (controlled)
then any observed correlation between A and B means that A was
in some way responsible for the concomitant variations in B. I
don’t want to use the word “cause” because we know that in a
control loop where A is a disturbance to a CV and B the output
of the control system, the observed correlation between A and
B does not reflect a lineal causal connection between A and B.
But A is, via its effect on the CV, the only variable
responsible for the correlated variations in B.

Two questions:

1. Is this supposed to contradict my statement that it is always

true that if A and B are correlated, then either one influences the
other or both have a causal link to some other variable?

2. Do you know of any control system in which there is not a direct

link (which is stronger than simply a causal connection) between the
disturbance and the output? Here’s a diagram of the canonical
control loop. I think it applies to all control loops, with
appropriate elaboration of the different segments of the pathways:

![loopFunctions7.jpg|473x518](upload://picaYS9iHfF5qyO3XTygEefdch6.jpeg)

If you accept this as a diagram of a control loop, not only is the

output influenced by the disturbance, but it is completely
determined by the disturbance and reference inputs together. Do you
know of any control loop in which this is not the case?

      Whether or not the system under study is a causal or a control

system, an experiment where you manipulate an IV (which may be
a disturbance to a CV if the system under study happens to be
a control system) and observe variations in a DV (which may be
the output that affects the same CV affected by the IV, if the
system is a control system) under controlled conditions, any
observed correlation between IV and DV means that the IV (and
no other variable) is somehow responsible for the
concomitant variations in the DV.

Why do you insert "(and no other variable)"?

 If the correlation between IV and DV is not 1.00, then there may

well be some other variable involved. In a control loop, the output
is determined jointly by the reference and the disturbance, so if
you know that the system is a control loop, you know that at least
two variables are involved. And don’t forget about time. If the
value of the DV is determined by more than one value of the IV (as
is the case in a control loop with an integrating output function)
then there are many values involved. The correlation between DV(now)
and IV(now) is affected by the contributions of DV(then) where
“then” means at many previous moments. Those past values of DV are
“other variables” that share with DV(now) the responsibility for the
current value of IV.

      The "somehow" is determined by testing to see if a CV is

involved. If a CV is not involved – as in the case of an
observed relationship between variations in voltage (IV) and
current (DV) – then the relationship between the IV and DV
can be considered causal. If there is a CV involved then the
relationship reflects characteristics of the feedback function
connecting DV to CV rather than characteristics of the
functional path from IV to DV.

True, but it hardly has anything to do with the statement on which

you appear to be commenting, does it?

      Experiments on control systems are done in the same way as

experiments on causal systems: you manipulate an IV under
controlled conditions and measure variations in a DV. The
extra added attraction in experiments on systems that might
possibly be control systems is that you also measure
variations in a suspected CV. If variations in the IV result
in far less variation in the suspected CV than is expected on
physical grounds and if this lack of expected variation can be
accounted for by variation in the DV that opposes the effect
of the IV, then the suspected CV can be considered a CV and
the relationship between IV and DV can be explained as the
actions of a control system (the DV) resisting disturbances
(IV) to the CV. Discovery of the CV is central to experiments
on suspected control systems and, of course, irrelevant to
experiments on causal systems.

Do you really want to say that control systems are acausal? I don't

believe it. I have always understood you to treat them as physical
systems.

Martin

[rsmarken]

[From Rick Marken (2011.06.29.2210)]

Martin Taylor (2011.06.29.23.09)–

Rick Marken (2011.06.29.1900)]

                  MT: A and B

may have no causal link whatever, but if they
don’t, then both must have a causal link to some
other variable.

                RM: This is true if the relationship between A and B

was observed non-experimentally.

MT: It’s true always.

      RM: In a _perfect_ experiment, where A is the IV and B the DV and

ALL variables except A and B are held constant (controlled)
then any observed correlation between A and B means that A was
in some way responsible for the concomitant variations in B. I
don’t want to use the word “cause” because we know that in a
control loop where A is a disturbance to a CV and B the output
of the control system, the observed correlation between A and
B does not reflect a lineal causal connection between A and B.
But A is, via its effect on the CV, the only variable
responsible for the correlated variations in B.

MT: Two questions:

1. Is this supposed to contradict my statement that it is always

true that if A and B are correlated, then either one influences the
other or both have a causal link to some other variable?

Yes. But I now see that I kind of jumped the gun. I thought you were talking about the “third variable” problem, which is the main basis for saying that correlation does not imply causality for correlations based on non-experimental data. But I see (from your second question below) that you were referring to the case that exists in closed loop control when the correlation between A (disturbance, IV) and B (output, DV) exists because A and B both have a causal link to another variable, the CV.

                  So take my little discussion as an expansion rather than a contradiction of your point that it's always true that "A and B

may have no causal link whatever, but if they
don’t, then both must have a causal link to some
other variable".

Best

Rick

···

2. Do you know of any control system in which there is not a direct

link (which is stronger than simply a causal connection) between the
disturbance and the output? Here’s a diagram of the canonical
control loop. I think it applies to all control loops, with
appropriate elaboration of the different segments of the pathways:

<img alt="" src="cid:part1.01060300.02080200@mmtaylor.net" width="237" height="259">



If you accept this as a diagram of a control loop, not only is the

output influenced by the disturbance, but it is completely
determined by the disturbance and reference inputs together. Do you
know of any control loop in which this is not the case?

      Whether or not the system under study is a causal or a control

system, an experiment where you manipulate an IV (which may be
a disturbance to a CV if the system under study happens to be
a control system) and observe variations in a DV (which may be
the output that affects the same CV affected by the IV, if the
system is a control system) under controlled conditions, any
observed correlation between IV and DV means that the IV (and
no other variable) is somehow responsible for the
concomitant variations in the DV.

Why do you insert “(and no other variable)”?

 If the correlation between IV and DV is not 1.00, then there may

well be some other variable involved. In a control loop, the output
is determined jointly by the reference and the disturbance, so if
you know that the system is a control loop, you know that at least
two variables are involved. And don’t forget about time. If the
value of the DV is determined by more than one value of the IV (as
is the case in a control loop with an integrating output function)
then there are many values involved. The correlation between DV(now)
and IV(now) is affected by the contributions of DV(then) where
“then” means at many previous moments. Those past values of DV are
“other variables” that share with DV(now) the responsibility for the
current value of IV.

      The "somehow" is determined by testing to see if a CV is

involved. If a CV is not involved – as in the case of an
observed relationship between variations in voltage (IV) and
current (DV) – then the relationship between the IV and DV
can be considered causal. If there is a CV involved then the
relationship reflects characteristics of the feedback function
connecting DV to CV rather than characteristics of the
functional path from IV to DV.

True, but it hardly has anything to do with the statement on which

you appear to be commenting, does it?

      Experiments on control systems are done in the same way as

experiments on causal systems: you manipulate an IV under
controlled conditions and measure variations in a DV. The
extra added attraction in experiments on systems that might
possibly be control systems is that you also measure
variations in a suspected CV. If variations in the IV result
in far less variation in the suspected CV than is expected on
physical grounds and if this lack of expected variation can be
accounted for by variation in the DV that opposes the effect
of the IV, then the suspected CV can be considered a CV and
the relationship between IV and DV can be explained as the
actions of a control system (the DV) resisting disturbances
(IV) to the CV. Discovery of the CV is central to experiments
on suspected control systems and, of course, irrelevant to
experiments on causal systems.

Do you really want to say that control systems are acausal? I don't

believe it. I have always understood you to treat them as physical
systems.

Martin

–
Richard S. Marken PhD
rsmarken@gmail.com
www.mindreadings.com

[Martin_Lewitt]

[Martin Lewitt 2011 June 30 0521 PDT]

[From Rick Marken (2011.06.29.0815)]

        Martin Lewitt (2011

June 29 0620 PDT)–

        I really think you need to look at the data again, taxes

were much higher in the 1950s, yet life expectancy, infant
mortality, and education level are much higher today.

      And inflation adjusted GDP, population and internet access is

much greater today than in the 1950s. Some things just get
bigger and better over time due to technological (and the
usual reproductive) advances. These are among the confounding
variables that are cancelled out when you look at, say,
unemployment rate as a function of actively produced changes
in tax rate that have made at essentially random points in
time.

So, now things that "always go in the wrong direction", always go up

“over time”? Should that give the N. Koreans hope? Are changes in
tax rate made at “random points in time”? There are major federal
elections every two years, the swings in control of congress and
thus tax policy are often reactive to events and conditions, which
suggest timings in changes in tax policy are not “random”, but
responsive to other forces. Even if the changes in tax policy
since the 1950s were random, they hardly prove that the recognition
by Congress that tax policy was causing too much investment was
driven by unproductive tax avoidance, such a real estate investment
trusts was wrong, or that economic growth would have been as robust
as it has been without those tax decreases.

Even the economic theory supporting tax reduction, doesn't presume a

simple mechanistic causal effect. Under current theory, economic
and business decisions are based upon decision maker expectations
about the future. A tax cut of uncertain duration due to constant
political disputation and class warfare demagoguery, might have
quite a different effect than perceived to be long term consensus
commitment to an environment allowing more returns to research and
investment. “Random” changes in tax policy would contribute to
decision maker uncertainty. If the changes occur too frequently,
they might abort the response we are looking for. No one is arguing
that investment produces instantaneous results, research and
development cycles vary from a few years to one or more decades.
These are the time frames and delays over which you would have to
look for correlations even in a linear system.

Perhaps this raises a question appropriate for PCT, is it the

present or the future state of variables that are being controlled?
Is the relationship complicated by present variables that control
future variables?

      The nice thing about the continuous improvement in quality of

life that results from improvements in technology (for you,
anyway) is that when your reactionary policies turn the US
into a third world country it will be country with a microwave
in every kitchen and an Xbox in every bedroom (and an A bomb
in every silo).

That hardly sounds like a third would country, but also doesn't

sound like continuous improvement, since it doesn’t allow for the
nonlinear and unpredictable advances in technology that will have
occurred. Stalling at microwaves and xbox’s would represent
stagnation. You can’t take the continuous improvement for granted.

Martin L

···

On 6/29/2011 9:14 AM, Richard Marken wrote:

      Best



      Rick

–

  Richard S. Marken PhD

  rsmarken@gmail.com

  [www.mindreadings.com](http://www.mindreadings.com)

[Bill_Powers1]

[From Bill Powers (2011.06.30.0630 mdt)]

Rick Marken (2011.06.29.1900) --

RM: In a _perfect_ experiment, where A is the IV and B the DV and ALL variables except A and B are held constant (controlled) then any observed correlation between A and B means that A was in some way responsible for the concomitant variations in B. I don't want to use the word "cause" because we know that in a control loop where A is a disturbance to a CV and B the output of the control system, the observed correlation between A and B does not reflect a lineal causal connection between A and B. But A is, via its effect on the CV, the only variable responsible for the correlated variations in B.

BP: It seems to me that one of the basic difficulties in this discussion arises from the use of the term "cause." As soon as you say "A causes B," you have to add "... all else being held constant," which of course is impossible. In the real world, B = f(A1,A2,A3,...An) with various magnitudes of coefficients in front of each argument. If you hold A2 through An constant, then A1 causes B. If you hold A1 constant instead of A2, then A2 causes B. The cause of B is whichever argument or combination of arguments you choose not to hold constant. It doesn't help to substitute "is resposible for" for "causes."

Behind the idea of causation is a conviction that you can isolate just one significant cause, while the other variables are unimportant and have only minor effects. It's a naive wish for simplicity that creates the concept of "the cause of B." In general there is no one cause of B. The only way to make it seem that A1 causes B is to keep all the other A's from showing their natural variations. This is a very handy way to stack the deck to help a weak theory. You simply hold all variables constant but the one your theory says is the important one. It then is guaranteed to be the only important one.

RM: Experiments on control systems are done in the same way as experiments on causal systems: you manipulate an IV under controlled conditions and measure variations in a DV. The extra added attraction in experiments on systems that might possibly be control systems is that you also measure variations in a suspected CV. If variations in the IV result in far less variation in the suspected CV than is expected on physical grounds and if this lack of expected variation can be accounted for by variation in the DV that opposes the effect of the IV, then the suspected CV can be considered a CV and the relationship between IV and DV can be explained as the actions of a control system (the DV) resisting disturbances (IV) to the CV. Discovery of the CV is central to experiments on suspected control systems and, of course, irrelevant to experiments on causal systems.

BP: In PCT, or system analysis in general, we don't have to keep all the system variables but one constant. We have to let qi, p, e, and qo vary in order to have a working control system. If we can measure d and r, we don't have to keep them constant, either. We simply record the values of all the variables we can find, and show that all the dependent variables can be calculated from the values of d and r, the independent variables. If there are unpredicted variations, they must arise because of other independent variables we have failed to notice or that are too numerous to keep track of.

The less unpredicted variation there is in the dependent variables, the more sure we can be that we have accounted for all the main variables of importance. We can use the measured values of the variables to deduce the forms, or at least plausible forms, of the various functions connecting the variables. As long as the equations containing those functions and variables continue to predict the future values of the dependent variables, we can be satisfied that we understand the system well enough for now, without having to designate any one cause or effect.

So I don't see any use for the term "cause" in a scientific discussion of behavior, or anything else for that matter. It's an ancient concept which is no longer needed. Informal usages, of course, will remain, and we will have to deal with others who still think the term is meaningful, but when we want precision we can talk about functional relationships among multiple variables. We no longer have to hold all else equal from the Big Bang to the present.

Best,

Bill P.

[Bill_Powers1]

[From Bill Powers (2011.06.30.0710 MDT)]

Martin Taylor 2011.06.29.23.09 –

MMT: Two questions:

1. Is this supposed to contradict my statement that it is always true
   that if A and B are correlated, then either one influences the other or
   both have a causal link to some other variable?

2. Do you know of any control system in which there is not a direct link
   (which is stronger than simply a causal connection) between the
   disturbance and the output? Here’s a diagram of the canonical control
   loop. I think it applies to all control loops, with appropriate
   elaboration of the different segments of the pathways:

If you accept this as a diagram of a control loop, not only is the output
influenced by the disturbance, but it is completely determined by the
disturbance and reference inputs together. Do you know of any control
loop in which this is not the case?

What we can be sure of here is that the “direct connection”
from d to qo does not determine the relationship between qo and d. In
fact,

qo = G(r - P(d + E(qo))).

If the functions are separable we have qo being some composite function
of r and d, the function including all three functions in the
control loop, not just the two functions in the direct connection. If
this is a good high-gain control system, in fact, we will find that the
observed form of the overall connection between d and qo is dominated by
the inverse of the environmental feedback function. While the qualitative
causal path appears to go from d to qo, the form of the functional path
is not determined by the two functions in that path, but primarily by the
feedback function.

This is all the more reason to minimize the use of the term
“cause.”

Best,

Bill P.

(Attachment 5926ff.jpg is missing)