Gus,

I am trying to replicate your data.  I do not trust you or your pals. I am
not a James Bond fan and do not know what myths you guys are acting out
today. So please be explicit and let me replicate what you do. That way I
will know you are not cheating or doing something incompetent.

"Gus Gassmann" <[EMAIL PROTECTED]> wrote in message
[EMAIL PROTECTED]">news:[EMAIL PROTECTED]...
>
>
> [EMAIL PROTECTED] wrote:
>
> > Gus,
> >
> > You refused to show your cards (explain your data) and now you are
blaming
> > the victim of your fraud.  You appear to be sufficiently threatened to
act
> > like a conman in front of all the people on this newslist. If you are
not
> > threatened by the truth, explain the data. I wasted a lot of time with
Steve
> > because he did not bother to tell us the data he interpreted were
> > inappropriate. That was deception. It was important enough to him to
claim
> > victory that he would cheat. You are doing the same thing. Put up or
shut
> > up.
>
> That's bullshit, and you know it.

Actually I still am not sure what you have done.

>
> I described the method to you in gory detail, but let's state it one more
time,
> for the record: I generate x1 and x2 uniformly on [-1, 1] and compute y =
x1 +
> x2.
> Thus y is caused by x1 and x2, and y is the effect. You seem not to agree
> with this statement, argiung that it somehow depends on the actual data
> I give you. I can't agree with your position on this.

No, I just do not trust that your data is what you describe above. Its a
matter of integrity and competence. You and you pals appear to lack both.
And at some point along the way I and at least one other person got the
impression you were uniformly sampling from a normal distribution. Never was
clear what you were doing, because your adolescent Oedipal strivings kept
confusing the issues. The reason I want explicit instructions from you is so
that I can replicate the data generation.

>
> Next I construct a grid over [-0.5,+0.5] x [-0.5, +0.5]. The grid is a
square
> grid or, yes, equal-sized smaller squares, of any size. I originally
placed a
> 100x100 grid, but it works just as well with a 10 x 10 grid. The way I
look
> at it, if I can get one point in each of the cells of the grid, then the
random
> variables are reasonably close to a two-dimensional uniform distribution.

So you are creating a factorial subsample. But why have you truncated the
data down to +/- .5? Your x values are from -1 to +1 and uniform.  Why not
go all the way up to +/- 1 and do a cross tabulation of y means from there?
Excessive truncation attenuates the correlations.


> (Since the gridsize can be varied, I can make the agreement with a true
> uniform distribution as close as possible.)

>
> While generating x1 and x2 I keep track of the pairs (x2,y) that fall into
the
> areas of my grid and I keep the first such pair in each small square. I
also
> store the corresponding x1. I continue until the entire grid has been
filled.

Here is where I get lost. Why not just create the data and then cross
tabulate x2 and y by trimming y?  There will be no need to trim the tails
off x1 and x2 since they are already uniform.


>
> I am not going to give you the data a third time, but let's look at what
> I end up with: A distribution of triples (x1, x2, y) that
> - are uniformly distributed in (x2, y)
> - therefore satisfy the CR relations
> - were generated in such a way that y is an effect, not a cause.

Here you lose me. Y is always the effect since we start with the model
y=x1+x2. Trimming the tails off y will not change it from being the effect.
There is no reason to trim the tails off x1 and x2 since they are already
uniform.


>
> The conclusion is simple: Without knowing what the data represent,
> you cannot _correctly_ infer causation.
>


WHOA.... show me the rde values. Show me C. Show me the stats and how you
got them. Your conclusion does not have any logical justification. Simply
creating uniform distributions does not make something a cause or effect.
The distribution is not the issue. Getting those cases where extremes of the
causes are combined is the issue. When you have those observations, what
does CR show. Please show your work. I do not trust you.

Bill

Bill>
>
>
>
>



.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
.                  http://jse.stat.ncsu.edu/                    .
=================================================================

Reply via email to