Resampling an analasis
Locked
 
|
This post was updated on .
CONTENTS DELETED
The author has deleted this message.
|
|
|
I think re-sampling won't gain anything that you can't have directly.
"Noise" is almost always some version of the variance; given variance, you have the standard deviation; dividing Mean by SD gives you a z-score.
Potential conceptual problems arise if variances vary -- It could matter if decide
to pool the four variances. For some data, it would be right to compute each z
based on its own sample; other data, not.
If variances differ by a lot, you may have to decide the meaning of a higher mean in
a stage which yields a smaller z-score because the variance is huge. (If means increase directly with increased variances, is the need for (log?) transformation implied?)
-- Rich Ulrich From: SPSSX(r) Discussion <[hidden email]> on behalf of Juleke2 <[hidden email]>
Sent: Wednesday, December 13, 2017 6:17:01 PM To: [hidden email] Subject: Resampling an analasis Dear all,
I need som help with bootstrapping or resampling in SPSS. For an analysis I'm calculating the signal to noise ratios (SNRs) of a measurement in 4 different disease stages. The signal and noise are calculated from 100 tests with this measurement. The mean signal and mean noise for each stage are then divided, resulting in a signal-to-noise ratio for each stage. We want to compare these ratio's between the stages, only, because there is no distribution of the ratio (it's only one ratio for each stage calculated from the signal and noise data (100 tests) which do have a distribution, it is not possible to define the (significant) differences between the ratios of the different stages. However, if I can somehow resample the signal and noise and the calculation of the ratio for each stage, a distribution of the ratio can be formed. Does anyone perhaps know how to this kind of bootstrapping or resampling? Many thanks in advance! -- Sent from: http://spssx-discussion.1045642.n5.nabble.com/ ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD |
|
|
In reply to this post by Juleke2
I saw Rich's reply but I want to ask about the S/N ratio computation itself. My visualization of this is that each of the 100 trials yields a signal number and a noise number. Thus, you have 100 S/N ratios at each assessment point for each patient. What I don't understand (and maybe it's because I don't know anything about signal/noise ratios) is why you compute the mean S/N ratio as (mean S)/(mean N) rather than as mean(S/N).
Gene Maguin -----Original Message----- From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Juleke2 Sent: Wednesday, December 13, 2017 6:17 PM To: [hidden email] Subject: Resampling an analasis Dear all, I need som help with bootstrapping or resampling in SPSS. For an analysis I'm calculating the signal to noise ratios (SNRs) of a measurement in 4 different disease stages. The signal and noise are calculated from 100 tests with this measurement. The mean signal and mean noise for each stage are then divided, resulting in a signal-to-noise ratio for each stage. We want to compare these ratio's between the stages, only, because there is no distribution of the ratio (it's only one ratio for each stage calculated from the signal and noise data (100 tests) which do have a distribution, it is not possible to define the (significant) differences between the ratios of the different stages. However, if I can somehow resample the signal and noise and the calculation of the ratio for each stage, a distribution of the ratio can be formed. Does anyone perhaps know how to this kind of bootstrapping or resampling? Many thanks in advance! -- Sent from: http://spssx-discussion.1045642.n5.nabble.com/ ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD |
|
|
I'm with Gene in that more information is needed. I have a good
background in signal detection theory (SDT) and know how its methods can be applied to diagnostic tests and other situations
but it unclear to me what a signal to noise ratio means in the
measurement of a disease state. Some of the sources I have
looked at for SNR such as the Wikipedia entry on it
(see: https://en.wikipedia.org/wiki/Signal-to-noise_ratio ) seem to assume that one knows the level of signal (e.g., in
sound, the energy of a tone) and the level of noise (e.g., the
energy level of white noise background that the tone is presented
in). These are direct measurements and not tests in the
traditional sense, especially for the presence or absence
of disease; see the Wikipedia entry on Detection theory for an indication of the difference (https://en.wikipedia.org/wiki/Detection_theory ). The decision matrix used in SDT will be familiar to those who do diagnostic testing and deal with concepts such as sensitivity (in SDT, the "Hit rate") and specificity (in SDT, the "correct
rejection rate"), for more on this
see: https://en.wikipedia.org/wiki/Sensitivity_and_specificity. In SDT and medical testing the statistics that are computed assume that samples come from population probability distributions
and measures of discriminability (i.e., ability to distinguish between
the presence and absence of a "signal") such as d' in SDT and
Area under the Curve (AuC or A') in diagnostic testing can
vary as a function of sampling error but resampling procedures
can be used to determine what the population parameters are
(in SDT, the distance in standard deviation units between the signal and noise distributions).
So, I have to ask: in the OP's situation what does a signal and noise mean since the situation does not involve engineering
data (where SNR appear to be commonly used). If possible,
can the OP identify which part(s) of the table at the following
link is/are relevant:
https://en.wikipedia.org/wiki/Sensitivity_and_specificity#Confusion_matrix If the above table is not relevant, can the OP provide more info On Thu, Dec 14, 2017 at 9:26 AM, Maguin, Eugene <[hidden email]> wrote: I saw Rich's reply but I want to ask about the S/N ratio computation itself. My visualization of this is that each of the 100 trials yields a signal number and a noise number. Thus, you have 100 S/N ratios at each assessment point for each patient. What I don't understand (and maybe it's because I don't know anything about signal/noise ratios) is why you compute the mean S/N ratio as (mean S)/(mean N) rather than as mean(S/N). |
«
Return to SPSSX Discussion
|
1 view|%1 views
| Free forum by Nabble | Edit this page |