Your problem is essentially the same as
constructing age-related reference centiles (or
reference intervals) - you want to model the
distribution of one continuous variable (Y)
conditional on the value of another (X).
This can be done by finding a suitable function
for regressing Y on X, and (if necessary) a
second model for the variability (SD) of the
residuals. Those functions may well not be straight lines. See for example
Calculating reference intervals for laboratory measurements.
Wright EM, Royston P.
Stat Methods Med Res. 1999 Jun;8(2):93-112.
Doug
At 09:29 23/01/2011, Óscar Martín Camacho wrote:
>Dear members,
>
>Thank you for your replies.
>I have been advised to group my data in
>intervals around the weights of my interest.
>Then I could assume these data to be
>representative of the distribution for the exact
>value (65kg, 70kg,...) from where calculating the percentiles.
>
>This would be the classical approach but I have
>found very few values around some of the points
>of interest. I wonder if I could use probability
>with these data, may be conditional or joint
>distributions to answer questions such as:
>Having a 80kg man, what would need to be his
>blood pressure for this man to be in the lowest
>25% blood pressure group among the 80kg men?
>
>P((BP<=X inters W=80)/W=80)=0.25 ???
>
>I would greatly appreciate any more suggestions.
>Thank you all for your replies.
>
>Best regards,
>Oscar.
>
>
>--- El sáb, 22/1/11, Óscar Martín Camacho <[log in to unmask]> escribió:
>
>
>De: Óscar Martín Camacho <[log in to unmask]>
>Asunto: Rv: Percentiles
>Para: [log in to unmask]
>Fecha: sábado, 22 de enero, 2011 13:51
>
>
>
>
>
>
>
>Dear list members,
>
>It seems I was not clear. Thank you for letting me know.
>Simplifying, I have 2 continuous variables
>Weight and Blood Pressure taken to 1100 men.
>They are presented in 1 decimal place units: one
>observation was 1 man with 79.7 kg and 115 mmHg.
>With these data. How could I calcilate the
>percentiles 25, 50 and 75 with respect to blood
>pressure of a 80 kg man. My table would looks like:
>
> Weight percentile 25 (blood pressure)
> 80 kg
>
>
>
>
>
>
>
>--- El sáb, 22/1/11, Óscar Martín Camacho <[log in to unmask]> escribió:
>
>
>De: Óscar Martín Camacho <[log in to unmask]>
>Asunto: Percentiles
>Para: [log in to unmask]
>Fecha: sábado, 22 de enero, 2011 12:34
>
>
>Hello,
>I am trying to characterise descriptively the
>weight of a group of men with respect to their
>systolic blood pressure. An extract of the two variables can be seen below:
>Weight (kg) Blood P(mm)Hg
>81.4 123
>64.7 119
>72.1 115
>There are 1100 in total.
>I would like to make a table with percentiles
>for these groups of men from 65kg to 80kg by
>intervals of 5kg with respect to blood pressure. Example:
>Weight Percentile25
>percentile50 percentile75
>65kg 109mmHg 114mmHg 118mmHg
>70kg 108mmHg 116mmHg 120mmHg
>75kg 109mmHg 117mmHg 120mmHg
>80kg 112mmHg 119mmHg 122mmHg
>
>My problem is that the weights are not in these
>same measures so I am not sure how this interpolation could be made.
>I would appreciate any suggestions about how to proceed.
>Best regards,
>Oscar
>
>
>
>
>You may leave the list at any time by sending the command
>
>SIGNOFF allstat
>
>to [log in to unmask], leaving the subject line blank.
>
>
>
>You may leave the list at any time by sending the command
>
>SIGNOFF allstat
>
>to [log in to unmask], leaving the subject line blank.
_____________________________________________________
Doug Altman
Professor of Statistics in Medicine
Centre for Statistics in Medicine
University of Oxford
Wolfson College Annexe
Linton Road
Oxford OX2 6UD
email: [log in to unmask]
Tel: 01865 284400 (direct line 01865 284401)
Fax: 01865 284424
www: http://www.csm-oxford.org.uk/
CONSORT 2010 Statement published in March 2010
www.consort-statement.org
EQUATOR Network - resources for reporting research
www.equator-network.org/
You may leave the list at any time by sending the command
SIGNOFF allstat
to [log in to unmask], leaving the subject line blank.
|