Comments for Statistics and Other Things

Comment on Temporary Comment by RomanM

RomanM — Fri, 05 Apr 2013 18:15:19 +0000

Yes. I sometimes use this site to fine tune a post and determine what it looks like. If I mark it as “Private”, no one other than myself can see it.

In this case, I had asked someone else to look at it so I had to post it publicly with password protection.

Comment on Temporary Comment by None

None — Fri, 05 Apr 2013 17:39:36 +0000

I assume that was the post at climateaudit then ?

Comment on Temporary Comment by RomanM

RomanM — Thu, 04 Apr 2013 20:47:39 +0000

Not yet.

Comment on Temporary Comment by None

None — Thu, 04 Apr 2013 20:44:54 +0000

Can I get the password ?

Comment on Wegman and the Ankle-Biters by stacyglen

stacyglen — Sun, 18 Nov 2012 17:41:35 +0000

The point was not to blame the grad student. It was to explain that what happened was an honest error of judgment. While on principle copying was always an issue, there was no enforcement in the absense of a complaint. There was no complaint AT THE TIME as there was no thief of ideas, just recycling standard phases. The attacks were selective enforcement. If it were standard to enforce at the time of publication, you would have expected the journal would have picked it up at the spell checking stage, before publication. This is standard to running tools to check for this in student works now. Withdrawing the paper politically motivated. The appropriate response was an correction. .

Comment on EIV/TLS Regression – Why Use It? by RomanM

RomanM — Mon, 13 Aug 2012 21:53:32 +0000

With regard to the y-intercept, there is no y-intercept. In order to do the regression properly, you need to center all of the variables at zero. Then after the regression has been calculated, the subtracted means and regression coefficients will provide the intercept. The calcx.tls function was written by myself to carry out the calculations. Depending on how much math you know, you might look here to see what needs to be done for the procedure. For the two variable case, you can try Deming regression function. I am not sure why you think that "the associated p-values and coefficient signs more important than the coefficient itself". Generally, the coefficients (and their signs) are quite important because they carry information about the type and magnitude of the relationship. The p-values are important in determining the uncertainty level of the estimated coefficient values. They are concerned with somewhat different aspects of the problem. A real difficulty with TLS/EIV is that it is necessary to know a priori the relative variability of the variables themselves when doing an analysis. Even then, the "residual" estimates of each are perfectly correlated. IMHO, the results in cases where the variables are also scaled to the same variance are simply unreliable particularly in the multivariate (three or more variable) case. That is part of what I meant by the term “out to lunch”.

Comment on EIV/TLS Regression – Why Use It? by Edward

Edward — Mon, 13 Aug 2012 19:43:22 +0000

Greetings,
I have run the attached script in R in order to walk through the comparison between OLS and EIV/TLS. I’ve learned a lot about R in a short period of time. The claim is made that the EIV/TLS data is “out to lunch”, but what about the y-intercept? How does the calcx.tls output the intercept. Also, aren’t the associated p-values and coefficient signs more important than the coefficient itself, where it simply reflects the units of each parameter? Lastly, I can’t seem to find much info on calcx.tls function. Any web/thread leads would be helpful. Thanks.
-EC

Comment on Wegman and the Ankle-Biters by kim2ooo

kim2ooo — Fri, 15 Jun 2012 12:09:13 +0000

Reblogged this on Climate Ponderings.

Comment on EIV/TLS Regression – Why Use It? by almostcertainly

almostcertainly — Thu, 07 Jun 2012 19:56:12 +0000

Dear Dr.;
I have a lay question, triggered by the Gergis Significance thread at CA.
If they are (at the end of the last blackboard) going to merely weight the various series in the big weighted summation, why do they bother to pre-select the ones that they analyze? The proxies with no contribution would just get a low weight, right?
Maybe there is some problem of the ‘unhelpful’ proxy series diluting the ‘significance’ of the outcome? Or what?
I am having trouble guessing why they bother to toss (or ‘screen out’ ) unpopular series ‘a priori’.
Please forgive my ‘lay’ usage of words which have very specific meanings in your field.
Maybe you will leave some breadcrumbs in your embedded commentary over at CA.
TIA
RR aka AC

Comment on Anomaly Regression – Do It Right! by tchannon

tchannon — Tue, 05 Jun 2012 13:12:45 +0000

Thanks Roman, yes I wondered whether that was just the plot. R is inaccessible, so I ignore it.
I’ll wait for a more apt moment to raise what could be a long subject.