Ratio in Data analysis

[Manoj Diwakar]

Dear All,

Have a good day !! This email finds you well 

When to take the Ratio of data ? and how to decide ..

Take an example if I am have two continuous bio markers then should we go with these two variables or should we take the ratio.. ? 

Just wants to more inputs and discussion 

Out of this

Is anybody in Pittsburgh USA ?

With regards 

Manoj 

University of Pittsburgh 

Thanks and Regards
Dr.Manoj Kumar Diwakar
Assistant Professor
Jawaharlal Nehru University
JNU, New Delhi, India
+91-9990346151
i touch

[Rich Ulrich]

If you have two scores that are very highly correlated, you

obtain the 'full information' in uncorrelated packages – namely,

the principal components as in a PCA – by taking the sum 

and difference of the standardized variables.  Or use the

simple sum and difference if they have similar variation on

the same scale: that gives NEARLY the same thing, and is easier

for the reader to describe and grasp.  This works for variables

that are distributed as Normal, or close to it.

Bio-markers are apt to be distributed much more like log-

normal than like normal.  The log of the ratio is the log(numerator)

minus log(denominator), and the log of the product is the sum 

of the logs – again, relatively uncorrelated to the extent that the

terms added and subtracted have similar variances.

If you aren't dealing with HIGH correlations, then you are in

the realm of whatever someone figures should make sense.

Rich Ulrich

---

From: meds...@googlegroups.com <meds...@googlegroups.com> on behalf of Manoj Diwakar <manojdiw...@gmail.com>
Sent: Thursday, April 18, 2024 4:28 PM
To: meds...@googlegroups.com <meds...@googlegroups.com>
Subject: {MEDSTATS} Ratio in Data analysis

 

--
--
To post a new thread to MedStats, send email to MedS...@googlegroups.com .
MedStats' home page is http://groups.google.com/group/MedStats .
Rules: http://groups.google.com/group/MedStats/web/medstats-rules

---
You received this message because you are subscribed to the Google Groups "MedStats" group.
To unsubscribe from this group and stop receiving emails from it, send an email to medstats+u...@googlegroups.com.
To view this discussion on the web, visit https://groups.google.com/d/msgid/medstats/CADTHf4ug8QKJ6mgFk5-FL%3Dx58sCimqbBq_S41ciyPvr9K1FWpw%40mail.gmail.com.

[MANOJ KUMAR]

Thanks Dr. Rich. for explaining this to me. 

To view this discussion on the web, visit https://groups.google.com/d/msgid/medstats/DS0PR10MB7017903D5373E539BCECDE3F97132%40DS0PR10MB7017.namprd10.prod.outlook.com.

--

Best Regards,

Dr. Manoj Kumar Diwakar, M.Sc., M.Phil.,Ph.D. (Statistics)
Assistant Professor
Centre for Economic Studies & Planning (CESP), School of Social Sciences (SSS-II),
Jawaharlal Nehru University, New Delhi-110067, India. 

Email id: manojkumar@jnu.ac.in 

 Mobile-09990346151

Area of Specialisation: Statistics, Econometric and Applied Mathematics 

Research Methodology -Quantitative Methods, Health Economics, Clinical Trial-Biostatistics

Data analysis and Software: SAS, SPSS, R, STATA, SPSS AMOS

[Venkata Putcha]

I would  prefer average if the two biomarkes measured on same observation.

Best wishes 

V

--

[Sreenivas.V]

I would prefer average difference for practical/clinical decisions while ratio for etiological purposes

Best wishes 

Sreenivas 

Dr. V. Sreenivas, M.Sc, PhD, FRSS (UK), FSMS

Formerly Professor of Biostatistics
All India Institute of Medical Sciences
New Delhi 110029

https://scholar.google.co.in/citations?hl=en&user=6unT1oYAAAAJ

https://www.ncbi.nlm.nih.gov/pubmed/?term=”V. Sreenivas” or “Sreenivas. V” OR “V Sreenivas” OR “Sreenivas V” OR “S Vishnubhatla” or “S. Vishnubhatla” OR “Vishnubhatla. S” OR “Vishnubhatla S” OR "Vishnubhatla Sreenivas" OR "Sreenivas Vishnubhatla" 

To view this discussion on the web, visit https://groups.google.com/d/msgid/medstats/CAHF%3D%3DL89oe9DW-iL0o0GEvpy4u-L7Uc9iG4oJDaACKSSyUx6%2BA%40mail.gmail.com.

[Diana Kornbrot]

1. I would do both ratio and mean difference

But absolutely essential whichever is chosen need EFFECT SIZE and distribution of  chosen parameter spread sand asymmetry

To view this discussion on the web, visit https://groups.google.com/d/msgid/medstats/CACzM0-MGcOFTbMo_vALPfCQF__bfktEa%2BVJz7O3MB6TAxTD1iQ%40mail.gmail.com.

[MANOJ KUMAR]

Thanking  Diana/ Sreenivas/Venkata/Rich 

. I still want to know which one is more efficient or robust than the others.

I have read " The data used for assessing the biomarker candidate is represented as a ratio of BiomarkerUP_regu over Biomarker_down_regula. This method intrinsically normalizes the data by adjusting for the differences in the denominator to look at how the numerator affects the outcome. Therefore, the ratio would be more specific to individual patient's data in our study population of research." 

To view this discussion on the web, visit https://groups.google.com/d/msgid/medstats/A7153933-9E12-4939-B292-8CBBD8458276%40gmail.com.

[Rich Ulrich]

Your quotation seems to focus on UP.

This seems to look like the same problem posed by measuring

outcome when you have a PRE score.  That is, your UP seems to

delegated as the more important, the relevant outcome.

Talking about 'robust' or 'efficient' is not important if you are

measuring the wrong thing.  UP versus DOWN is like POST versus PRE.

There is a lot of discussion of 'change' scores for outcome – When 

are they appropriate? The three main choices are (a) raw outcome,

(b) simple change score, and (c) regressed change score.  Your source

is concerned that raw outcome (UP) contains too much irrelevant

contribution per individual, and 'normalizes' by taking the ratio.

Ratios are involved when the natural relation is multiplicative.  That

is, would you normally want to describe UP as "twice DOWN" or do

points of score seem appropriate?  Least-square models want you

to use additive, not multiplicative terms, so taking the log is ordinary

and converts the model to additive. 

The regressed change can be computed by simple regression on PRE or 

by models that include other variables.  The goal is to find the outcome

that is sensitive/robust/efficient IN RELATION TO the intervention or

condition that is being studied. 

When the 'coefficient of variation' is small, then you get very similar

results whether you take the ratio or the subtract, to see how much

more UP is than DOWN.  Since what you quote specifically says 'ratio',

they seem to assume that the simple differences would be misleading
because of the scaling and unequal variation for the extremes.

The use of regressed-change is potentially more 'efficient' – in terms of

smaller variance for the outcome – than either the outcome alone or the

change score:  that is, than either UP or the ratio.  That might be

used in an analysis by taking log(UP) as the criterion, with log(DOWN)

as the covariate.

Rich Ulrich

---

From: meds...@googlegroups.com <meds...@googlegroups.com> on behalf of MANOJ KUMAR <manoj...@jnu.ac.in>
Sent: Monday, April 22, 2024 9:35 AM
To: meds...@googlegroups.com <meds...@googlegroups.com>
Subject: Re: {MEDSTATS} Ratio in Data analysis

 

To view this discussion on the web, visit https://groups.google.com/d/msgid/medstats/CA%2ByqjKsH2%3DuJGt%2BeUDAJtp7DWX4GBvOFbmt5Ti0_TM51%3DLd0cA%40mail.gmail.com.