Prediction

[TheodoreM.Bones]

Corrections and enhancements added. The example worked here is for pulse rate.

[TheodoreM.Bones]

In another application of this predict worksheet, it processes real data unadorned with any noise added in an example for the pulse rate from just 8 observations of the St-Segment, QRS Interval & Pulse Rate made with a hand held portable electrocardiograph (EKG). The options of studying the other traits must include re-setting display parmeters on graphs. Predictions outwards from 8 observations randomly made in time as far ahead as a year (365 days) may be made with confidence.

[TheodoreM.Bones]

Previous autocorrelations for extrapolation (predict) outside the range of the sample data tended to overreach and gave the expected mean of the data, as would be the case with the Central Limit Theorem. Some confidence was there, but there was no way of telling.

Here a smaller range is made up with the autocorrelation function and the entire new sample is subject to a t-test to determine the confidence of data obtained thereby with the raw sample based on the means and the variances of both.

[TheodoreM.Bones]

Previous autocorrelations for extrapolation (predict) outside the range of the sample data tended to overreach and gave the expected mean of the data, as would be the case with the Central Limit Theorem. Some confidence was there, but there was no way of telling.

Here a smaller range is made up with the autocorrelation function and the entire new sample is subject to a t-test to determine the confidence of data obtained thereby with the raw sample based on the means and the variances of both.

Corrected some typos on v1 & v2.

[TheodoreM.Bones]

The preceeding worksheet was not the right one.

Clinicians often wish to have data on, for example, cardiac stroke volume or blood pressure predicted ahead a short period of time and an autocorrelation method is developed here allowing extrapolation/prediction on the data assuming it to be a time series. The data thusly obtained is compared with the original samples with/without data smoothing. A check for normal probability distribution is provided. Predictions as far ahead as one year (365 days) may be made - with caution.

[TheodoreM.Bones]

A valued paper that appeared in the British medical "Lancet" is studied and programmed. The paper has been cited over 8900 times. There has bedn no mention of it on this Forum. It dismisses the ordinary statistical methods.

[IRstuff]

On 3/16/2009 1:42:28 PM, bones7xx wrote:
>A valued paper that appeared
>in the British medical
>"Lancet" is studied and
>programmed. The paper has
>been cited over 8900 times.
>There has bedn no mention of
>it on this Forum. It
>dismisses the ordinary
>statistical methods.

I don't believe that is a correct interpretation of the author's statements:
http://www-users.york.ac.uk/~mb55/meas/ba.htm

What he complains about is incorrectly using correlation coefficients to demonstrate agreement between two different instruments, instruments designed to measure the same quantities should have correlated answers. But, the correlation coefficient does not prove that they're actually agreeing.

TTFN,
Eden

[TheodoreM.Bones]

Yes, it took 10 pages to get that point across in the Lancet article. I did not attempt to use corr for that reason. I attempted to find data that were out of line with the precision model, whichever that would be, using their method, not commnn statistics, but compzred it with them.

Many thanks.

[TheodoreM.Bones]

Can't locate or log on go the Web page you gave. York University, I presume.

Cheers.

[IRstuff]

On 3/16/2009 7:12:42 PM, bones7xx wrote:
>Can't locate or log on go the
>Web page you gave. York
>University, I presume.
>
>Cheers.

I don't seem to have a problem with that. Here's the link to the PDF version: http://www-users.york.ac.uk/~mb55/meas/ba.pdf

It's supposedly a slightly later and updated version of the same paper you reference in your sheet.

TTFN,
Eden

[TheodoreM.Bones]

Thanks. That was successful. I tried the corr function on the lnnormal plot and the LR vslue from the t-test, and got quite good agreement. The authors showed in their paper that the corr function was blind to obvious errors or outliers in 2 sets of data. Thus the corr values were not going to support the repeatibility or to compare 2 observations.

[TheodoreM.Bones]

After thinking it over, I added a check on the difference data like the authors themselves presented, 2 standard deviations above and below the mean for a presumed 95% reliability.

It seems to me that this sort of analysis is the basis of the USA Framingham Heart Study that predicts adverse heart events on the basis of the Pulse Pressure, which is the difference between the systolic & diastolic blood pressures.

Risks are then computed from normal probability between these limits. Coherence has nothing to
do with it.

A requirement for Bland/Altman is that the two data sets be equal in length. The standard Student's t-test does mot require that, nor does normal probability.

Bland/Altman is a yardstick process. It can check on instruments, etc., where one data set is a presumed master standard like a certified water flow meter.

Evaluations and risks are the provenance of probability. Data is takn as it comes although
outliers may be trimmed arbitrarily.

[TheodoreM.Bones]

Printed objectios to the Bland/Altman method were found. Regression followeed by autocorrrelation is give after the Bland/Altman
method is worked on one example, showing no actual conclusions may come out of it except to
provide a normal probability ysrdstick for contemplation. It is necessary to assume some physical understanding of the data given by Bland/Altmsn, like thinking two water meters are
connected in series and a time series set of data
follows. The two meters should read the same save for wear and tear of a second meter compared to a archived master meter reading made at the exact same time.